American Journal of Epidemiology

Neta, Gila; Brownson, Ross C; Chambers, David A

doi: 10.1093/aje/kwx323pmid: 29036569

Abstract The field of epidemiology has been defined as the study of the spread and control of disease. However, epidemiology frequently focuses on studies of etiology and distribution of disease at the cost of understanding the best ways to control disease. Moreover, only a small fraction of scientific discoveries are translated into public health practice, and the process from discovery to translation is exceedingly slow. Given the importance of translational science, the future of epidemiologic training should include competency in implementation science, whose goal is to rapidly move evidence into practice. Our purpose in this paper is to provide epidemiologists with a primer in implementation science, which includes dissemination research and implementation research as defined by the National Institutes of Health. We describe the basic principles of implementation science, highlight key components for conducting research, provide examples of implementation studies that encompass epidemiology, and offer resources and opportunities for continued learning. There is a clear need for greater speed, relevance, and application of evidence into practice, programs, and policies and an opportunity to enable epidemiologists to conduct research that not only will inform practitioners and policy-makers of risk but also will enhance the likelihood that evidence will be implemented. dissemination research, epidemiologic methods, implementation research, implementation science, translational science The field of epidemiology has been broadly defined as “the study of how disease spreads and can be controlled” (1) or “the branch of medicine which deals with the incidence, distribution, and possible control of diseases and other factors relating to health” (2). Using the latter part of this definition, Galea (3) made the case for what he called a consequentialist epidemiology, suggesting that our inordinate focus on the etiology and distribution of disease comes at a major cost to the field and to public health in general. He called for a “more rigorous engagement with the second part of our vision for ourselves—the intent for us to intervene” or control disease (3, p. 1185). Heeding this call, Brownson and numerous other leaders in the field published a paper entitled “Charting a Future for Epidemiologic Training” (4). They identified macro-level trends influencing the field of epidemiology and developed a set of recommended competencies for future epidemiologic training. Among these noted trends was the emergence of translational sciences, specifically dissemination and implementation research, also known collectively as implementation science (the term we use in this article), whose goal is to move evidence into practice as effectively and efficiently as possible. One may ask why a distinct field of research is needed in order to more effectively translate research into practice. Numerous studies have estimated that the amount of time it takes for even a fraction of original research to be implemented into practice is approximately 17 years (5–7). Moreover, life-saving measures are not always implemented despite years of evidence, as has been shown for hospital-acquired infections (8), heart disease (9, 10), diabetes (11), asthma (12), and cancer (13). This consistent failure to translate research findings into practice has resulted in an estimated 30%–40% of patients not receiving treatments of proven effectiveness and 20%–25% of patients receiving care that is not needed or potentially harmful (14–16). It is increasingly clear that the processes of dissemination and implementation are not passive but require active strategies to ensure that the evidence is effectively understood, adopted, implemented, and maintained in practice settings (5, 17). Furthermore, epidemiologists are critical in making this happen; we are among the primary generators of evidence, and that evidence is only as useful as it is effectively generated and communicated to stakeholders, including public health practitioners, policy-makers, providers, and patients. Even when evidence is relevant and effectiveness is demonstrated, implementation of an effective intervention may not occur. The questions are: 1) Will it be adopted?, 2) Will practitioners be equipped to deliver it?, and 3) Of those who are equipped, will they choose to deliver it or receive institutional support to do so—and if so, what portion of the population will actually benefit? Brownson et al. (4) recommended increasing the competency of epidemiologists in implementation science in order to contribute effectively to the integration of evidence into practice. Our purpose in this paper is to provide epidemiologists with a primer in implementation science. We describe the basic principles of implementation science, highlight key components for conducting research, articulate roles for epidemiologists in this field, provide brief examples of implementation studies that encompass epidemiologic principles, and offer additional resources and opportunities for continued learning. WHAT IS IMPLEMENTATION SCIENCE? “Implementation science” is one of several terms that have been used to describe the science of putting knowledge or evidence into action and of understanding what, why, and how evidence or evidence-based practices work in the real world (18, 19). “Implementation science” is the predominant term used in the United States and Europe (18), and it has been defined in various ways by different agencies and organizations. For the purpose of this article, we are using the definition described in the journal Implementation Science—the study of methods to promote the integration of research findings and evidence into health-care policy and practice (20). This includes the study of how best to spread or sustain evidence, as well as the testing of strategies that can best facilitate the adoption and integration of evidence-based interventions into practice. In response to the need for such research, the National Institutes of Health has issued funding announcements to build the knowledge base on how to improve upon these processes (21–23). In these funding announcements, implementation science is broken down into 2 components, dissemination research and implementation research, which are defined as follows: Dissemination research is the scientific study of targeted distribution of information and intervention materials to a specific public health or clinical practice audience. The intent is to understand how best to spread and sustain knowledge and the associated evidence-based interventions. Implementation research is the scientific study of the use of strategies to adopt and integrate evidence-based health interventions into clinical and community settings in order to improve patient outcomes and benefit population health. While we recognize that there may be some overlap, the general distinction between these two subfields is that dissemination research is focused on spread across multiple settings, while implementation research is focused on specific local efforts to implement in targeted settings (e.g., schools, worksites, clinics). Implementation science broadly focuses on the how questions: How do we get evidence to drive practice and get evidence-based interventions to become standard care so that everyone who can benefit from them has access? A critical part of understanding that process is to examine the intermediary dissemination and implementation outcomes that are requisite to achieve population health impact. These include the awareness, acceptability, reach, adoption, appropriateness, feasibility, fidelity, cost, and sustainability of efforts to disseminate and/or implement evidence in practice settings (24, 25). Otherwise, it can be difficult to discern whether, for example, an intervention does not work because it is not effective in a particular population or setting or because it was not disseminated and/or implemented properly in a given context. And if implementation works in a given case, what can we learn from those successful strategies to enhance implementation of other public health measures? For the purpose of this paper, we will use the term “implementation science” to refer to the overall field and the term “implementation studies” for investigations within this broad area of research. KEY COMPONENTS OF IMPLEMENTATION SCIENCE Implementation science seeks to address gaps in the provision of evidence-based practices and is rooted in theories and methods from a variety of fields; it hinges on transdisciplinary collaboration, with an emphasis on engaging stakeholders; it focuses on the strategies used to disseminate and implement evidence and evidence-based interventions, and understanding how and why they work; it uses rigorous and relevant methodologies; and it emphasizes the importance of external validity and scalability. Table 1 summarizes key components of an implementation science study. These components were adapted from previously published work that assessed the most critical aspects of successful grant proposals in implementation science at the National Institutes of Health (26) and the Canadian Institutes of Health Research (27), taking into account key elements defined by the Department of Veterans Affairs (28). To provide context for each of these components, we walk through the specific example of implementing the hepatitis B vaccine in order to prevent liver cancer. Table 1. Key Components of an Implementation Studya Study Component . Why It Matters . Selected Resources . Research objective Research question addresses a gap in the provision of an evidence-based intervention, practice, or policy EPOC Cochrane Review Group: http://epoc.cochrane.org/ The Community Guide: https://www.thecommunityguide.org/ Evidence-based practice Sufficient evidence of effectiveness and an appropriate fit for a given context Research-Tested Intervention Programs (NCI): https://rtips.cancer.gov/rtips/index.do Effective Interventions (HIV prevention): https://effectiveinterventions.cdc.gov/ National Registry of Evidence-based Programs and Practices (behavioral health; SAMHSA): https://www.samhsa.gov/nrepp Guidelines International Network: http://www.g-i-n.net/home Theoretical justification Conceptual model and theoretical justification supports the choice of intervention and informs the design, the variables to be measured, and the analytical plan Tool to assist researchers in selecting models that best fit the research question: http://dissemination-implementation.org/index.aspx Consolidated Framework for Implementation Research: http://www.cfirguide.org/ RE-AIM Framework: http://re-aim.org/ Stakeholder engagement Clear engagement process and demonstrated support from relevant stakeholders to ensure the feasibility that implementation can be studied Review of participatory research (72): Community-Campus Partnerships for Health: https://ccph. memberclicks.net/ Community Tool Box: http://ctb.ku.edu Research Toolkit: http://researchtoolkit.org/ Community Research Partners: http:// communityresearchpartners.net/ Implementation strategy Implementation strategy or strategies for implementing the evidence-based practice are justified and well-described Recommendations for specifying and reporting implementation strategies (73) Compilation of strategies (74) Team expertise Appropriate multidisciplinary expertise on the study team is demonstrated, including qualitative and/or quantitative expertise, to ensure rigorous data collection and analysis Networking communities Research to Reality (NCI): https://researchtoreality.cancer.gov Implementation Science Exchange: https://impsci.tracs.unc.edu Society for Implementation Research Collaboration: https:// societyforimplementationresearchcollaboration.org/ Study design Study design is justified and feasible given the study context (e.g., feasibility of randomization) Reviews of study designs (40, 69) Mixed-methods designs (41) Effectiveness-implementation hybrid designs (42) Webinars on study designs (NCI): https://cyberseminar.cancercontrolplanet.org/implementationscience/ Measurement Implementation outcome measures should be included, conceptually justified, well-defined, and informed by existing measurement instruments, and should cover concepts of both internal and external validity Society for Implementation Research Collaboration: https://societyforimplementationresearchcollaboration.org/ Grid-Enabled Measures Database (NCI): https://www.gem-beta.org/Public/Home.aspx https://whatiskt.wikispaces.com/home Grid-Enabled Measures Database methods paper (75) Study Component . Why It Matters . Selected Resources . Research objective Research question addresses a gap in the provision of an evidence-based intervention, practice, or policy EPOC Cochrane Review Group: http://epoc.cochrane.org/ The Community Guide: https://www.thecommunityguide.org/ Evidence-based practice Sufficient evidence of effectiveness and an appropriate fit for a given context Research-Tested Intervention Programs (NCI): https://rtips.cancer.gov/rtips/index.do Effective Interventions (HIV prevention): https://effectiveinterventions.cdc.gov/ National Registry of Evidence-based Programs and Practices (behavioral health; SAMHSA): https://www.samhsa.gov/nrepp Guidelines International Network: http://www.g-i-n.net/home Theoretical justification Conceptual model and theoretical justification supports the choice of intervention and informs the design, the variables to be measured, and the analytical plan Tool to assist researchers in selecting models that best fit the research question: http://dissemination-implementation.org/index.aspx Consolidated Framework for Implementation Research: http://www.cfirguide.org/ RE-AIM Framework: http://re-aim.org/ Stakeholder engagement Clear engagement process and demonstrated support from relevant stakeholders to ensure the feasibility that implementation can be studied Review of participatory research (72): Community-Campus Partnerships for Health: https://ccph. memberclicks.net/ Community Tool Box: http://ctb.ku.edu Research Toolkit: http://researchtoolkit.org/ Community Research Partners: http:// communityresearchpartners.net/ Implementation strategy Implementation strategy or strategies for implementing the evidence-based practice are justified and well-described Recommendations for specifying and reporting implementation strategies (73) Compilation of strategies (74) Team expertise Appropriate multidisciplinary expertise on the study team is demonstrated, including qualitative and/or quantitative expertise, to ensure rigorous data collection and analysis Networking communities Research to Reality (NCI): https://researchtoreality.cancer.gov Implementation Science Exchange: https://impsci.tracs.unc.edu Society for Implementation Research Collaboration: https:// societyforimplementationresearchcollaboration.org/ Study design Study design is justified and feasible given the study context (e.g., feasibility of randomization) Reviews of study designs (40, 69) Mixed-methods designs (41) Effectiveness-implementation hybrid designs (42) Webinars on study designs (NCI): https://cyberseminar.cancercontrolplanet.org/implementationscience/ Measurement Implementation outcome measures should be included, conceptually justified, well-defined, and informed by existing measurement instruments, and should cover concepts of both internal and external validity Society for Implementation Research Collaboration: https://societyforimplementationresearchcollaboration.org/ Grid-Enabled Measures Database (NCI): https://www.gem-beta.org/Public/Home.aspx https://whatiskt.wikispaces.com/home Grid-Enabled Measures Database methods paper (75) Abbreviations: EPOC, Effective Practice and Organisation of Care; HIV, human immunodeficiency virus; NCI, National Cancer Institute; RE-AIM, Reach, Effectiveness, Adoption, Implementation, and Maintenance; SAMHSA, Substance Abuse and Mental Health Services Administration. a The study components selected for this table were based on those described by Proctor et al. (26) as key components for an implementation science grant. Open in new tab Table 1. Key Components of an Implementation Studya Study Component . Why It Matters . Selected Resources . Research objective Research question addresses a gap in the provision of an evidence-based intervention, practice, or policy EPOC Cochrane Review Group: http://epoc.cochrane.org/ The Community Guide: https://www.thecommunityguide.org/ Evidence-based practice Sufficient evidence of effectiveness and an appropriate fit for a given context Research-Tested Intervention Programs (NCI): https://rtips.cancer.gov/rtips/index.do Effective Interventions (HIV prevention): https://effectiveinterventions.cdc.gov/ National Registry of Evidence-based Programs and Practices (behavioral health; SAMHSA): https://www.samhsa.gov/nrepp Guidelines International Network: http://www.g-i-n.net/home Theoretical justification Conceptual model and theoretical justification supports the choice of intervention and informs the design, the variables to be measured, and the analytical plan Tool to assist researchers in selecting models that best fit the research question: http://dissemination-implementation.org/index.aspx Consolidated Framework for Implementation Research: http://www.cfirguide.org/ RE-AIM Framework: http://re-aim.org/ Stakeholder engagement Clear engagement process and demonstrated support from relevant stakeholders to ensure the feasibility that implementation can be studied Review of participatory research (72): Community-Campus Partnerships for Health: https://ccph. memberclicks.net/ Community Tool Box: http://ctb.ku.edu Research Toolkit: http://researchtoolkit.org/ Community Research Partners: http:// communityresearchpartners.net/ Implementation strategy Implementation strategy or strategies for implementing the evidence-based practice are justified and well-described Recommendations for specifying and reporting implementation strategies (73) Compilation of strategies (74) Team expertise Appropriate multidisciplinary expertise on the study team is demonstrated, including qualitative and/or quantitative expertise, to ensure rigorous data collection and analysis Networking communities Research to Reality (NCI): https://researchtoreality.cancer.gov Implementation Science Exchange: https://impsci.tracs.unc.edu Society for Implementation Research Collaboration: https:// societyforimplementationresearchcollaboration.org/ Study design Study design is justified and feasible given the study context (e.g., feasibility of randomization) Reviews of study designs (40, 69) Mixed-methods designs (41) Effectiveness-implementation hybrid designs (42) Webinars on study designs (NCI): https://cyberseminar.cancercontrolplanet.org/implementationscience/ Measurement Implementation outcome measures should be included, conceptually justified, well-defined, and informed by existing measurement instruments, and should cover concepts of both internal and external validity Society for Implementation Research Collaboration: https://societyforimplementationresearchcollaboration.org/ Grid-Enabled Measures Database (NCI): https://www.gem-beta.org/Public/Home.aspx https://whatiskt.wikispaces.com/home Grid-Enabled Measures Database methods paper (75) Study Component . Why It Matters . Selected Resources . Research objective Research question addresses a gap in the provision of an evidence-based intervention, practice, or policy EPOC Cochrane Review Group: http://epoc.cochrane.org/ The Community Guide: https://www.thecommunityguide.org/ Evidence-based practice Sufficient evidence of effectiveness and an appropriate fit for a given context Research-Tested Intervention Programs (NCI): https://rtips.cancer.gov/rtips/index.do Effective Interventions (HIV prevention): https://effectiveinterventions.cdc.gov/ National Registry of Evidence-based Programs and Practices (behavioral health; SAMHSA): https://www.samhsa.gov/nrepp Guidelines International Network: http://www.g-i-n.net/home Theoretical justification Conceptual model and theoretical justification supports the choice of intervention and informs the design, the variables to be measured, and the analytical plan Tool to assist researchers in selecting models that best fit the research question: http://dissemination-implementation.org/index.aspx Consolidated Framework for Implementation Research: http://www.cfirguide.org/ RE-AIM Framework: http://re-aim.org/ Stakeholder engagement Clear engagement process and demonstrated support from relevant stakeholders to ensure the feasibility that implementation can be studied Review of participatory research (72): Community-Campus Partnerships for Health: https://ccph. memberclicks.net/ Community Tool Box: http://ctb.ku.edu Research Toolkit: http://researchtoolkit.org/ Community Research Partners: http:// communityresearchpartners.net/ Implementation strategy Implementation strategy or strategies for implementing the evidence-based practice are justified and well-described Recommendations for specifying and reporting implementation strategies (73) Compilation of strategies (74) Team expertise Appropriate multidisciplinary expertise on the study team is demonstrated, including qualitative and/or quantitative expertise, to ensure rigorous data collection and analysis Networking communities Research to Reality (NCI): https://researchtoreality.cancer.gov Implementation Science Exchange: https://impsci.tracs.unc.edu Society for Implementation Research Collaboration: https:// societyforimplementationresearchcollaboration.org/ Study design Study design is justified and feasible given the study context (e.g., feasibility of randomization) Reviews of study designs (40, 69) Mixed-methods designs (41) Effectiveness-implementation hybrid designs (42) Webinars on study designs (NCI): https://cyberseminar.cancercontrolplanet.org/implementationscience/ Measurement Implementation outcome measures should be included, conceptually justified, well-defined, and informed by existing measurement instruments, and should cover concepts of both internal and external validity Society for Implementation Research Collaboration: https://societyforimplementationresearchcollaboration.org/ Grid-Enabled Measures Database (NCI): https://www.gem-beta.org/Public/Home.aspx https://whatiskt.wikispaces.com/home Grid-Enabled Measures Database methods paper (75) Abbreviations: EPOC, Effective Practice and Organisation of Care; HIV, human immunodeficiency virus; NCI, National Cancer Institute; RE-AIM, Reach, Effectiveness, Adoption, Implementation, and Maintenance; SAMHSA, Substance Abuse and Mental Health Services Administration. a The study components selected for this table were based on those described by Proctor et al. (26) as key components for an implementation science grant. Open in new tab Research objective The first component of any strong research study is a clear research objective or set of specific aims; what distinguishes an implementation study is that the objective and aims address a gap in the provision of care or quality. Rather than focus on an increase in disease-specific incidence or mortality, an implementation study might focus on a clear gap in the provision of evidence-based practices to address disease. For example, where an epidemiologic study may focus on understanding the etiology of the increase in incidence of liver cancer in the United States, an implementation study may focus on understanding the barriers to and facilitators of implementing widespread hepatitis B vaccination or hepatitis C treatment and seek to develop and test strategies to address those barriers to implementation. Epidemiologic investigation identifying not only the incidence and mortality trends but also the most vulnerable populations and settings where gaps in hepatitis vaccination or treatment may exist would be critical to shaping the research objective for an implementation study. Evidence-based practice The second component has to do with the practice or intervention to be implemented and its evidence base regarding efficacy or effectiveness. Beyond the strength of evidence, additional critical questions to consider when selecting an evidence-based intervention to address a care or quality gap include: 1) How well does the intervention fit with the study population?, 2) What are the available resources to implement the intervention in the study settings?, and 3) How feasible is the intervention’s use in the given context?. While the nature of the evidence is important, implementation science further prioritizes the importance of the context for that evidence, including the population, setting, and political and policy environment. In our driving example of hepatitis B vaccination and liver cancer prevention, we might ask which population should be targeted (e.g., patient populations, clinics, public health departments, ministries of health), what resources are available to implement a vaccination program, and how feasible a vaccination program is in a given context (e.g., developing country, schools, clinics, pharmacies). A useful tool to help guide researchers in thinking through relevant issues to consider when developing or selecting an intervention is the Pragmatic Explanatory Continuum Indicator Summary, called PRECIS-2 (https://www.precis-2.org/). PRECIS-2 was initially created to help trialists design trials that are fit for their purpose, whether that is to understand efficacy or the effectiveness of an intervention (29). The tool identifies 9 issues or domains for which design decisions can influence the degree to which a trial is explanatory (i.e., addresses efficacy) versus pragmatic (i.e., addresses effectiveness). The domains address features of the target of the intervention (e.g., participant eligibility, patient-centeredness of the primary outcome) as well as of the delivery of the intervention (e.g., setting, organization, delivery and adherence methods, follow-up protocol). Understanding the generalizability of the evidence for a given intervention, as well as the context in which it was proven effective, can help to inform decisions about selecting the appropriate practice or intervention to implement. Theoretical justification Just as epidemiologists may rely on causal directed acyclic graphs to determine which variables should be included in a regression model as possible confounders and which should be considered as potential mediators or effect modifiers (30, 31), researchers conducting implementation studies should be informed by theories, conceptual frameworks, and models, which serve to explain phenomena, organize conceptually distinct ideas, and help visualize relationships that cannot be observed directly. Although theories, conceptual frameworks, and models are distinct concepts with unique definitions and characteristics, they all serve a similar purpose, and thus we will henceforth refer to them singularly as “models” (32). In implementation science, models not only serve to inform which variables are relevant to measure and analyze but also can serve to inform the development or selection of an evidence-based practice or intervention, as well as the development or selection of a strategy for implementing that intervention. Theoretical models in implementation science have been categorized into 2 categories: Explanatory models describe whether and how implementation activities will affect a desired outcome, and process models inform which implementation strategies should be tested (33). We recognize that the selection of strategies will depend on the current stage of implementation for a given context. For example, if an intervention or practice has yet to be adopted, strategies may focus on influencing decision-makers by educating them about the value of the intervention. Alternatively, if an intervention has been adopted and implemented but the challenge is how best to sustain it, the strategies would focus on sustainment. A review of models in implementation science identified 61 different models that were characterized by construct flexibility (how loosely or rigidly defined are the concepts in the model), socioecological level (e.g., individual, organization, community, system), and the degree to which they addressed dissemination versus implementation processes (32). While each model had distinguishing features, common elements among them included an emphasis on the importance of change and characterization of the nature of change, the significance of context at both the local and external levels, and the recognition that most change requires active and deliberate facilitation (e.g., local champions, tools, training). Furthermore, barriers to dissemination and implementation exist across settings and sites and could include factors relating to leadership, resources, technology, and inertia. Some of the most widely used models in implementation science include Roger’s Theory of Diffusion of Innovations (34), the RE-AIM (Reach, Effectiveness, Adoption, Implementation, and Maintenance) Framework (35), and the Consolidated Framework for Implementation Research (36). Table 1 lists useful resources that can help investigators select an appropriate model for their research. Some important factors to consider when selecting a model include the research question (whether it is addressing dissemination and/or implementation), the socioecological level of change (e.g., provider, clinic, organization, system), relevant characteristics of context, time frame, and availability of measures (37). Looking at the example of hepatitis B vaccination, important influences might exist at the national or state level (e.g., ministries of health, state health departments), the organizational level (e.g., integrated delivery systems), the clinic level, the provider level, and the patient level. For example, the culture of a health-care organization may not be open to adding an additional program, or a particular ministry of health may not prioritize prevention practices. Frameworks can help to identify these various key influences. Stakeholder engagement Many areas of science do not require engagement from stakeholders (e.g., patients, physicians, clinics, community members), particularly in basic and exploratory research, but when studying implementation, engagement is a necessity. In order to maximize the likelihood that stakeholders will implement an intervention, it is necessary to understand their needs and challenges to ensure that a given intervention and the approach to implementing it meet their contextual conditions, that they have the necessary resources to implement the intervention, and that they are able to sustain it (38). Review committees for implementation science proposals strongly weigh demonstrated support from relevant stakeholders for given projects. Reviewers assess the level of stakeholder engagement both by letters of support and by a clear record of collaboration between the researchers and stakeholders involved in implementing the evidence-based practice. As we think about key stakeholders in the example of hepatitis B vaccination and liver cancer prevention, we might focus on ministries of health, state health departments, schools, clinics, or pharmacies. For example, if the ministry of health in a given country of interest does not prioritize prevention, the ministry might not be the right locus for implementation; other stakeholders to consider in that situation might be local schools or pharmacies. The key is connecting to the appropriate stakeholders within a given setting who may be the lynchpins to widespread implementation. The more we work with these key stakeholders, the better we can identify the needs of and barriers to implementation, and thus develop feasible and sustainable solutions. Implementation strategy Implementation studies often focus on identifying, developing, testing, and/or evaluating strategies that enhance uptake of an intervention, program, or practice. Implementation strategies are the techniques used to ensure or enhance the adoption, implementation, and sustainability of an evidence-based practice. In a review of implementation strategies, Powell et al. (39) found 73 distinct strategies that could be grouped into 6 categories defined by key processes: planning, educating, financing, restructuring, managing quality, and attending to the policy context. Examples of implementation strategies for each of these categories include strategies with which to build stakeholder buy-in (planning), train practitioners to deliver an intervention (educating), modify incentives (financing), revise professional roles (restructuring), conduct audit and feedback (managing quality), and create credentialing or licensure standards (attending to the policy context). Implementation science focuses on understanding whether and/or how these strategies work to foster implementation and on understanding the mechanisms behind these strategies. In the example of hepatitis B vaccination, we might consider strategies related to educating patients about the value of the vaccine in cancer prevention, or strategies to help finance resource-strapped community health centers to facilitate both the implementation and sustainment of vaccination programs. Team expertise As with any research endeavor, having appropriate expertise on an implementation science team is critical and depends on the research questions at hand. Unlike epidemiology, in implementation science it is common to see mixed-methods studies that require the expertise of both quantitative and qualitative researchers. The qualitative component of a study can help to inform the findings from quantitative analysis, providing valuable data that can explain how or why X strategy to implement Y intervention did or did not work. Conversely, qualitative data can help to shape the quantitative component of a study (e.g., qualitative analysis to understand barriers to adoption can inform which strategies are developed or selected to be quantitatively tested). In addition, key implementation outcomes may require specific expertise to be represented on the team. For example, costs and the cost-effectiveness of implementation strategies are commonly measured outcomes that would require the inclusion of an economist or cost-effectiveness analyst for effective evaluation. And unlike many other scientific endeavors, stakeholders (e.g., patients, providers, community representatives) also can be a critical part of the study team, as they can inform both the strategies used and the most feasible research designs for a given study population. Finally, having someone with previous experience in conducting implementation science on the study team or as a consultant or mentor is often seen as a critical asset by reviewers. In the example of hepatitis B, epidemiologists seeking to close the gap in vaccination rates globally may wish to team with implementation scientists or qualitative researchers to investigate existing barriers to implementation using focus groups, or to identify strategies to facilitate the adoption of vaccination programs, or to conduct a mixed-methods analysis to understand why a tested strategy did or did not work. Study design There are a variety of rigorous study designs that have been developed and used in implementation science. These include both experimental (e.g., randomized controlled trial, cluster-randomized controlled trial, pragmatic trial, stepped wedge trial, dynamic wait-listed control trial) and quasi-experimental (e.g., nonequivalent groups, pre-/post-, regression discontinuity, interrupted time series), nonexperimental or observational (including designs from epidemiology), mixed-methods (i.e., the collection and integration of qualitative and quantitative data), qualitative methods (e.g., focus groups, semistructured interviews), and system science (e.g., system dynamics, agent-based modeling) approaches (40, 41). Additionally, study designs that simultaneously test intervention effectiveness as well as implementation are called hybrid designs (42). While epidemiologists may be familiar with many of these designs, mixed-methods approaches and qualitative analyses likely will be unfamiliar, and thus it would be critical for epidemiologists to team with qualitative researchers in conducting implementation studies, which often rely on these methods. Notably, there is a wide range of acceptable study designs, and we have seen successful grant applications for all of these, not only randomized controlled trials. Selecting the appropriate study design for an implementation science study depends on the study question and the available evidence, as well as the study circumstances, such as whether randomization is possible. For example, if the study question addresses why or how dissemination or implementation occurs, a design that includes qualitative assessment might be required. If study participants will not accept randomization, then a quasi-experimental design might be indicated. A variety of resources exist to help researchers learn more about study designs for implementation science, including the Implementation Science Exchange hosted by the North Carolina Translational and Clinical Sciences Institute at the University of North Carolina at Chapel Hill (https://impsci.tracs.unc.edu) and webinars hosted by the Implementation Science Program in the Division of Cancer Control and Population Sciences at the National Cancer Institute, as referenced in Table 1. In the example of hepatitis B vaccination, the unit of analysis for implementation may be at the clinic level, and not all clinics may want to be randomized. Thus, a quasi-experimental design may be most appropriate. Measurement Implementation science requires specific measurement of constructs related to the key implementation outcomes described above (e.g., awareness, acceptability, reach, adoption, appropriateness, feasibility, fidelity, cost, and sustainability), which are informed by implementation science theories, models, or frameworks. There are online databases of existing measures and measurement tools for implementation science, including the National Cancer Institute’s Grid-Enabled Measures Database Dissemination and Implementation Initiative and the Society for Implementation Research Collaboration's Instrument Review Project and Instrument Repository, both listed in Table 1. Additionally, several systematic reviews of implementation science measurement instruments have been published (43–48). In an implementation study of hepatitis B vaccination in a low-resource setting, for example, the question is not only whether vaccination is effective at reducing the risk of liver cancer but also whether the vaccination program was adopted (e.g., measuring the rate of uptake), implemented with fidelity (e.g., measuring the dose and complete rate), or sustained (e.g., measuring whether clinics continued to vaccinate 12 months after the program was initiated). WHAT ROLE CAN EPIDEMIOLOGISTS PLAY? The interface between epidemiology and implementation science should be bidirectional, as illustrated in Figure 1. As a collaborative and interdisciplinary enterprise, implementation science relies on the work of epidemiologists to inform the evidence base, identify the gaps in health status, contribute to methods for design and measurement, and inform program and policy evaluation. On the other hand, the work of epidemiologists can be magnified through direct engagement in implementation science, enhancing the real-world relevance of epidemiologic research and increasing the likelihood that scientific findings will be useful or consequential (3). There are numerous ways in which epidemiologists contribute to and lead implementation studies, often as part of an interdisciplinary team (49). We provide several examples to illustrate the nexus between epidemiologic methods and implementation science. Defining evidence Evidence relevant to implementation science can come in multiple forms, including data on the size of a public health problem, the causes of the problem, and effective interventions. In particular, the notion of an evidence-based intervention is central to implementation science because the focus is often on the scale-up and spread of evidence-based programs and policies (50). As in many types of epidemiologic studies, both internal validity and external validity play key roles in an implementation study. Internal validity is threatened by multiple types of systematic error, and error rates are influenced by both study design and study execution. Too little attention has been paid to external validity (i.e., the degree to which findings from a study or set of studies can be generalizable to and relevant for populations, settings, and times other than those in which the original studies were conducted) (51). The epidemiologic skills (addressing systematic error) to describe the quality of intervention evidence related to internal validity and the potential for generalizability (external validity) are crucial for implementation research (52–55). Contributing to systems approaches Complex systems thinking is needed to address our most vexing public health issues (56) and is a central theme in implementation science (57). Systems thinking in implementation science may be operationalized in several ways. At a conceptual level, systems approaches take into account context involving multiple interacting agents and study processes that are nonlinear and iterative. Numerous systems methods (e.g., agent-based modeling, systems dynamics modeling), often developed in other disciplines, are becoming more commonly used in implementation science to predict the future impact, or impact at full scale, of new interventions. Epidemiologists are often well-equipped for dealing with multiple interacting factors and are therefore valuable team members in systems studies. Viewing causality in a new light It has been argued that the classical framework for causal thinking articulated by Hill has been critical to understanding causation of chronic diseases but does not fully represent causation for complex public health issues (58). The classic Hill criteria (59) and those outlined in the 1964 US Surgeon General’s Report (60) have proven highly useful in etiological research but may be less useful for studies of the effectiveness of interventions or the scale-up of effective programs and policies in implementation studies—namely that factors occur across multiple levels (from biological to macrosocial) and these factors often influence one another (61). Numerous epidemiologic skills contribute to an expanded view of causation (62–67), including those involving multilevel modeling, developing unbiased estimators of both cluster and individual-level effects over multiple time points, and developing causal diagrams to help understand complex relationships (30, 68). Determining appropriate study designs Designs for implementation science have greater variation than those used for more traditional epidemiologic research (often studying etiology, efficacy, or effectiveness) (69). The policy and political context for a given study may influence a researcher’s ability to randomize an intervention in practice settings, given potential concerns about cost, feasibility, or convenience. Therefore, while some study questions may allow for use of a randomized design, often implementation studies rely on a suite of alternative quasi-experimental designs, including interrupted time series, multiple-baseline (where the start of intervention conditions is staggered across settings/time), or regression discontinuity (where the intervention status is predetermined by an established cutoff or threshold) designs (40), which can allow for the estimation of causal effects. When randomization is possible in an implementation study, it is often at the group level rather than at the individual level (as in a clinical trial). Epidemiologists bring competencies that benefit implementation science in formulating research questions, determining the range of designs available, and assessing the trade-offs in various designs. Measuring intermediate outcomes As we described previously, in many cases the outcomes in an implementation science study are different than those in a more traditional epidemiologic study where one commonly measures clinical outcomes or changes in health status. Proximal measures of implementation processes and outcomes are often assessed (e.g., organizational climate or culture, the uptake of an evidence-based practice). Epidemiologists can play important roles in developing and testing new measures and in leading efforts to determine: which outcomes should be tracked and how long it will take to show progress; how to best determine criterion validity (how a measure compares with some “gold standard”); how to best measure effect modifiers across a range of settings (e.g., schools, worksites); and how common, practical measures can be developed and shared so researchers are not constantly reinventing measures. We have enumerated 5 ways in which epidemiologists can and do contribute to implementation science. Ideally, all or most of these can contribute to a given study. In Table 2, based on a review of grant applications, we provide a selection of National Institutes of Health–funded implementation science studies to illustrate how epidemiology played a role for each. Table 2. Selected National Institutes of Health–Funded Grant Proposals in Implementation Science With Epidemiologic Contributions Project Title, Year (Funding Mechanism) . Principal Investigator (Department(s)/Center; Institution) . Study Setting/Design . Study Aims/Objectives . Example(s) of Epidemiologic Contributions . “Test and Treat TB: a Proof-of-Concept Trial in South Africa,” 2014 (R21a grant) Ingrid Valerie Bassett, MD, MPH (Infectious Diseases, Medicine; Massachusetts General Hospital, Boston, Massachusetts) Mobile, community-based HIV screening program in Durban, South Africa 3-arm randomized trial Aim 1: Assess the feasibility, yield, and clinical impact of a test-and-treat TB strategy integrated with ongoing rapid HIV screening Aim 2: Assess the cost and cost-effectiveness of the strategy Defining the evidence: identifying the scope of the problem to inform the study and intervention design “Enhancing Evidence-Based Diabetes Control Among Local Health Departments,” 2017 (R01b grant) Ross Brownson, PhDc (Epidemiology, Prevention Research Center; Washington University in St. Louis, St. Louis, Missouri) 30 local health departments in Missouri Qualitative needs assessment Group-randomized Aim 1: Refine and test measures to assess the adoption of evidence-based programs and policies in local health departments, building on pilot work Aim 2: Conduct a qualitative needs assessment of 15 local health departments to understand factors influencing the adoption of evidence-based programs and policies for diabetes control Aim 3: Conduct a group-randomized experimental study of 30 local health departments in Missouri to evaluate the effectiveness of active dissemination and implementation approaches Measuring intermediate outcomes: adapt and test valid and reliable dissemination measures Determining appropriate study designs: determine trade-offs in various randomized designs (cluster level) “SIngle-saMPLE Tuberculosis Evaluation to Facilitate Linkage to Care: The SIMPLE TB Trial,” 2016 (R01 grant) Adithya Cattamanchi, MD, MAS (Medicine; University of California, San Francisco, San Francisco, California) Cluster-randomized trial 20 community health centers in Uganda Aim 1: Determine comparative effectiveness of a TB diagnostic strategy designed for low-resource settings versus standard care to improve TB diagnosis and treatment initiation rates Aim 2: Evaluation of factors influencing adoption and maintenance of intervention components using mixed methods Aim 3: Perform economic and epidemic modeling to estimate the cost-effectiveness and epidemiologic impact of strategy Viewing causality in a new light: intervening at a health systems level to impact the epidemiology of TB in low-resource settings “Online Social Networks for Dissemination of Smoking Cessation Interventions,” 2011 (R01 grant) Nathan Cobb, MD (Institute for Tobacco Research; Truth Initiative Foundation, Washington, DC) Facebook (Facebook, Inc., Menlo Park, California) Randomized trial (factorial design) Aim 1: Identify and characterize structural characteristics of an effective dissemination strategy for an evidence-based tobacco cessation intervention Aim 2: Identify and characterize network characteristics of participants (i.e., modifiable variables) that influence dissemination and behavior change Defining evidence: epidemiology informs intervention development Viewing causality in a new light and determining study design: considering multilevel factors through social network analysis “A User-Friendly Epidemic-Economic Model of Diagnostic Tests for Tuberculosis,” 2012 (R21 grant) David Dowdy, MD, PhDc (Epidemiology; Johns Hopkins University, Baltimore, Maryland) Dynamic modeling Aim 1: Develop a combined epidemic-economic model of TB diagnosis Aim 2: Project the impact and cost-effectiveness of strategies for scaling up TB diagnostics Aim 3: Disseminate the structure and findings of a TB diagnostic model for use, adaptation, and improvement by the global TB control community Contributing to a systems approach: epidemiologic methods contribute to developing a systems model Viewing causality in a new light: considering multilevel factors “Addressing Hepatitis C and Hepatocellular Carcinoma: Current and Future Epidemics,” 2013 (R01 grant) Holly Hagan, PhDc, MPH (College of Nursing; New York University, New York, New York) Agent-based modeling Aim 1: Synthesize evidence on HCV epidemiology, prevention, progression, and treatment to inform an agent-based model Aim 2: Develop an agent-based model to estimate the effects of scaling up various evidence-based HCV interventions Aim 3: Determine the combination of interventions for particular budget and epidemiologic scenarios Defining evidence and contributing to a systems approach: epidemiologic evidence and methods contribute to developing an agent-based model Viewing causality in a new light: considering multilevel factors “Translating Molecular Diagnostics for Cervical Cancer Prevention into Practice,” 2016 (R01 grant) Patti Gravitt, PhDc (Global Health, Epidemiology; George Washington University, Washington, DC) 3 communities in the Loreto region of Peru Focus groups, key informant interviews, surveys Interrupted time-series design Aim 1: Conduct community census to establish cervical cancer screening registries and identify key members for working groups to facilitate Participatory Action Research approach Aim 2: Develop screening intervention monitoring and evaluation criteria, objectives, methods, and instruments for ongoing monitoring and evaluation of implementation, guided by the Consolidated Framework for Implementation Research Aim 3: Evaluate the comparative effectiveness of each strategy using qualitative and quantitative methods iteratively executed in both pre- and postimplementation phases embedded in the overall design, allowing real-time adaptation of the programs for broader sustainability Aim 4: Synthesize a formal cost-effectiveness evaluation to inform policy-makers Defining evidence to inform intervention Determining appropriate study design “A Retail Policy Laboratory: Modeling Impact of Retailer Reduction on Tobacco Use,” 2013 (R21 grant) Douglas Luke, PhD (Center for Public Health Systems Science; Washington University in St. Louis, St. Louis, Missouri) Agent-based modeling Aim 1: Build Tobacco Town, a simulation of a realistic community, to model tobacco retailer density and individual tobacco purchasing Aim 2: Use the model built in aim 1 as a retail policy laboratory to explore and compare the potential effects on behavior of a suite of real-world retailer reduction policy approaches. The effects of the retailer density policies on vulnerable populations will also be examined, particularly for low-income residents and minorities. Contributing to a systems approach: parameters of Tobacco Town informed by epidemiologic evidence of tobacco use and purchasing “Dissemination and Implementation of a Corrective Intervention to Improve Mediastinal Lymph Node Examination in Resected Lung Cancer,” 2013 (R01 grant) Raymond Osarogiagbon, MD (School of Public Health; University of Memphis, Memphis, Tennessee) 12 hospitals in western Tennessee, northern Mississippi, and eastern Arkansas Multiple-baseline study design Aim 1: Recruit institutions and surgeons to participate in an implementation study Aim 2: Evaluate effectiveness of intervention in a diverse mix of institutions to maximize external validity of the intervention Aim 3: Process evaluation of dissemination and implementation using RE-AIM Framework Defining evidence: data from the SEER Program on pathological lymph node staging informed the scope and magnitude of the problem “Sustainable Financial Incentives to Improve Prescription Practices for Malaria,” 2012 (R21 grant) Wendy Prudhomme-O’Meara, PhD (Medicine, Infectious Diseases; Duke University, Durham, North Carolina) 18 rural health facilities in western Kenya Cluster-randomized trial Objective: test whether financial incentives offered at the facility level improve targeting of antimalarial medications to patients with parasitologically diagnosed malaria Viewing causality in a new light “Adapting Patient Navigation to Promote Cancer Screening in Chicago’s Chinatown,” 2012 (R01 grant) Melissa Simon, MD (Obstetrics and Gynecology; Northwestern University at Chicago, Chicago, Illinois) Chinatown section of Chicago, Illinois Quasi-experimental (pre-/posttest, time-series) design Aim 1: Adapt and expand to broader cancer control education a patient navigator intervention designed for low-income Latinas and African Americans to Chicago’s Chinatown population Aim 2: Evaluate implementation of the patient navigator intervention using qualitative methods Aim 3: Evaluate adapted intervention’s impact on screening rates and support longer-term epidemiologic surveillance systems Aim 4: Compare adapted intervention’s rate of timely follow-up after abnormal test results with that of comparable control group receiving standard care Viewing causality in a new light: identifying multilevel barriers and facilitators to screening, including social, economic, cultural, and psychosocial barriers and community facilitators Project Title, Year (Funding Mechanism) . Principal Investigator (Department(s)/Center; Institution) . Study Setting/Design . Study Aims/Objectives . Example(s) of Epidemiologic Contributions . “Test and Treat TB: a Proof-of-Concept Trial in South Africa,” 2014 (R21a grant) Ingrid Valerie Bassett, MD, MPH (Infectious Diseases, Medicine; Massachusetts General Hospital, Boston, Massachusetts) Mobile, community-based HIV screening program in Durban, South Africa 3-arm randomized trial Aim 1: Assess the feasibility, yield, and clinical impact of a test-and-treat TB strategy integrated with ongoing rapid HIV screening Aim 2: Assess the cost and cost-effectiveness of the strategy Defining the evidence: identifying the scope of the problem to inform the study and intervention design “Enhancing Evidence-Based Diabetes Control Among Local Health Departments,” 2017 (R01b grant) Ross Brownson, PhDc (Epidemiology, Prevention Research Center; Washington University in St. Louis, St. Louis, Missouri) 30 local health departments in Missouri Qualitative needs assessment Group-randomized Aim 1: Refine and test measures to assess the adoption of evidence-based programs and policies in local health departments, building on pilot work Aim 2: Conduct a qualitative needs assessment of 15 local health departments to understand factors influencing the adoption of evidence-based programs and policies for diabetes control Aim 3: Conduct a group-randomized experimental study of 30 local health departments in Missouri to evaluate the effectiveness of active dissemination and implementation approaches Measuring intermediate outcomes: adapt and test valid and reliable dissemination measures Determining appropriate study designs: determine trade-offs in various randomized designs (cluster level) “SIngle-saMPLE Tuberculosis Evaluation to Facilitate Linkage to Care: The SIMPLE TB Trial,” 2016 (R01 grant) Adithya Cattamanchi, MD, MAS (Medicine; University of California, San Francisco, San Francisco, California) Cluster-randomized trial 20 community health centers in Uganda Aim 1: Determine comparative effectiveness of a TB diagnostic strategy designed for low-resource settings versus standard care to improve TB diagnosis and treatment initiation rates Aim 2: Evaluation of factors influencing adoption and maintenance of intervention components using mixed methods Aim 3: Perform economic and epidemic modeling to estimate the cost-effectiveness and epidemiologic impact of strategy Viewing causality in a new light: intervening at a health systems level to impact the epidemiology of TB in low-resource settings “Online Social Networks for Dissemination of Smoking Cessation Interventions,” 2011 (R01 grant) Nathan Cobb, MD (Institute for Tobacco Research; Truth Initiative Foundation, Washington, DC) Facebook (Facebook, Inc., Menlo Park, California) Randomized trial (factorial design) Aim 1: Identify and characterize structural characteristics of an effective dissemination strategy for an evidence-based tobacco cessation intervention Aim 2: Identify and characterize network characteristics of participants (i.e., modifiable variables) that influence dissemination and behavior change Defining evidence: epidemiology informs intervention development Viewing causality in a new light and determining study design: considering multilevel factors through social network analysis “A User-Friendly Epidemic-Economic Model of Diagnostic Tests for Tuberculosis,” 2012 (R21 grant) David Dowdy, MD, PhDc (Epidemiology; Johns Hopkins University, Baltimore, Maryland) Dynamic modeling Aim 1: Develop a combined epidemic-economic model of TB diagnosis Aim 2: Project the impact and cost-effectiveness of strategies for scaling up TB diagnostics Aim 3: Disseminate the structure and findings of a TB diagnostic model for use, adaptation, and improvement by the global TB control community Contributing to a systems approach: epidemiologic methods contribute to developing a systems model Viewing causality in a new light: considering multilevel factors “Addressing Hepatitis C and Hepatocellular Carcinoma: Current and Future Epidemics,” 2013 (R01 grant) Holly Hagan, PhDc, MPH (College of Nursing; New York University, New York, New York) Agent-based modeling Aim 1: Synthesize evidence on HCV epidemiology, prevention, progression, and treatment to inform an agent-based model Aim 2: Develop an agent-based model to estimate the effects of scaling up various evidence-based HCV interventions Aim 3: Determine the combination of interventions for particular budget and epidemiologic scenarios Defining evidence and contributing to a systems approach: epidemiologic evidence and methods contribute to developing an agent-based model Viewing causality in a new light: considering multilevel factors “Translating Molecular Diagnostics for Cervical Cancer Prevention into Practice,” 2016 (R01 grant) Patti Gravitt, PhDc (Global Health, Epidemiology; George Washington University, Washington, DC) 3 communities in the Loreto region of Peru Focus groups, key informant interviews, surveys Interrupted time-series design Aim 1: Conduct community census to establish cervical cancer screening registries and identify key members for working groups to facilitate Participatory Action Research approach Aim 2: Develop screening intervention monitoring and evaluation criteria, objectives, methods, and instruments for ongoing monitoring and evaluation of implementation, guided by the Consolidated Framework for Implementation Research Aim 3: Evaluate the comparative effectiveness of each strategy using qualitative and quantitative methods iteratively executed in both pre- and postimplementation phases embedded in the overall design, allowing real-time adaptation of the programs for broader sustainability Aim 4: Synthesize a formal cost-effectiveness evaluation to inform policy-makers Defining evidence to inform intervention Determining appropriate study design “A Retail Policy Laboratory: Modeling Impact of Retailer Reduction on Tobacco Use,” 2013 (R21 grant) Douglas Luke, PhD (Center for Public Health Systems Science; Washington University in St. Louis, St. Louis, Missouri) Agent-based modeling Aim 1: Build Tobacco Town, a simulation of a realistic community, to model tobacco retailer density and individual tobacco purchasing Aim 2: Use the model built in aim 1 as a retail policy laboratory to explore and compare the potential effects on behavior of a suite of real-world retailer reduction policy approaches. The effects of the retailer density policies on vulnerable populations will also be examined, particularly for low-income residents and minorities. Contributing to a systems approach: parameters of Tobacco Town informed by epidemiologic evidence of tobacco use and purchasing “Dissemination and Implementation of a Corrective Intervention to Improve Mediastinal Lymph Node Examination in Resected Lung Cancer,” 2013 (R01 grant) Raymond Osarogiagbon, MD (School of Public Health; University of Memphis, Memphis, Tennessee) 12 hospitals in western Tennessee, northern Mississippi, and eastern Arkansas Multiple-baseline study design Aim 1: Recruit institutions and surgeons to participate in an implementation study Aim 2: Evaluate effectiveness of intervention in a diverse mix of institutions to maximize external validity of the intervention Aim 3: Process evaluation of dissemination and implementation using RE-AIM Framework Defining evidence: data from the SEER Program on pathological lymph node staging informed the scope and magnitude of the problem “Sustainable Financial Incentives to Improve Prescription Practices for Malaria,” 2012 (R21 grant) Wendy Prudhomme-O’Meara, PhD (Medicine, Infectious Diseases; Duke University, Durham, North Carolina) 18 rural health facilities in western Kenya Cluster-randomized trial Objective: test whether financial incentives offered at the facility level improve targeting of antimalarial medications to patients with parasitologically diagnosed malaria Viewing causality in a new light “Adapting Patient Navigation to Promote Cancer Screening in Chicago’s Chinatown,” 2012 (R01 grant) Melissa Simon, MD (Obstetrics and Gynecology; Northwestern University at Chicago, Chicago, Illinois) Chinatown section of Chicago, Illinois Quasi-experimental (pre-/posttest, time-series) design Aim 1: Adapt and expand to broader cancer control education a patient navigator intervention designed for low-income Latinas and African Americans to Chicago’s Chinatown population Aim 2: Evaluate implementation of the patient navigator intervention using qualitative methods Aim 3: Evaluate adapted intervention’s impact on screening rates and support longer-term epidemiologic surveillance systems Aim 4: Compare adapted intervention’s rate of timely follow-up after abnormal test results with that of comparable control group receiving standard care Viewing causality in a new light: identifying multilevel barriers and facilitators to screening, including social, economic, cultural, and psychosocial barriers and community facilitators Abbreviations: HCV, hepatitis C virus; HIV, human immunodeficiency virus; MAS, master of advanced studies; MD, medical doctor; MPH, master of public health; PhD, doctor of philosophy; RE-AIM, Reach, Effectiveness, Adoption, Implementation, and Maintenance; SEER, Surveillance, Epidemiology, and End Results; TB, tuberculosis. a The R21 grant mechanism at the National Institutes of Health is for funding smaller exploratory or developmental research. b The R01 grant mechanism at the National Institutes of Health is for funding larger research projects. c Degree in epidemiology. Open in new tab Table 2. Selected National Institutes of Health–Funded Grant Proposals in Implementation Science With Epidemiologic Contributions Project Title, Year (Funding Mechanism) . Principal Investigator (Department(s)/Center; Institution) . Study Setting/Design . Study Aims/Objectives . Example(s) of Epidemiologic Contributions . “Test and Treat TB: a Proof-of-Concept Trial in South Africa,” 2014 (R21a grant) Ingrid Valerie Bassett, MD, MPH (Infectious Diseases, Medicine; Massachusetts General Hospital, Boston, Massachusetts) Mobile, community-based HIV screening program in Durban, South Africa 3-arm randomized trial Aim 1: Assess the feasibility, yield, and clinical impact of a test-and-treat TB strategy integrated with ongoing rapid HIV screening Aim 2: Assess the cost and cost-effectiveness of the strategy Defining the evidence: identifying the scope of the problem to inform the study and intervention design “Enhancing Evidence-Based Diabetes Control Among Local Health Departments,” 2017 (R01b grant) Ross Brownson, PhDc (Epidemiology, Prevention Research Center; Washington University in St. Louis, St. Louis, Missouri) 30 local health departments in Missouri Qualitative needs assessment Group-randomized Aim 1: Refine and test measures to assess the adoption of evidence-based programs and policies in local health departments, building on pilot work Aim 2: Conduct a qualitative needs assessment of 15 local health departments to understand factors influencing the adoption of evidence-based programs and policies for diabetes control Aim 3: Conduct a group-randomized experimental study of 30 local health departments in Missouri to evaluate the effectiveness of active dissemination and implementation approaches Measuring intermediate outcomes: adapt and test valid and reliable dissemination measures Determining appropriate study designs: determine trade-offs in various randomized designs (cluster level) “SIngle-saMPLE Tuberculosis Evaluation to Facilitate Linkage to Care: The SIMPLE TB Trial,” 2016 (R01 grant) Adithya Cattamanchi, MD, MAS (Medicine; University of California, San Francisco, San Francisco, California) Cluster-randomized trial 20 community health centers in Uganda Aim 1: Determine comparative effectiveness of a TB diagnostic strategy designed for low-resource settings versus standard care to improve TB diagnosis and treatment initiation rates Aim 2: Evaluation of factors influencing adoption and maintenance of intervention components using mixed methods Aim 3: Perform economic and epidemic modeling to estimate the cost-effectiveness and epidemiologic impact of strategy Viewing causality in a new light: intervening at a health systems level to impact the epidemiology of TB in low-resource settings “Online Social Networks for Dissemination of Smoking Cessation Interventions,” 2011 (R01 grant) Nathan Cobb, MD (Institute for Tobacco Research; Truth Initiative Foundation, Washington, DC) Facebook (Facebook, Inc., Menlo Park, California) Randomized trial (factorial design) Aim 1: Identify and characterize structural characteristics of an effective dissemination strategy for an evidence-based tobacco cessation intervention Aim 2: Identify and characterize network characteristics of participants (i.e., modifiable variables) that influence dissemination and behavior change Defining evidence: epidemiology informs intervention development Viewing causality in a new light and determining study design: considering multilevel factors through social network analysis “A User-Friendly Epidemic-Economic Model of Diagnostic Tests for Tuberculosis,” 2012 (R21 grant) David Dowdy, MD, PhDc (Epidemiology; Johns Hopkins University, Baltimore, Maryland) Dynamic modeling Aim 1: Develop a combined epidemic-economic model of TB diagnosis Aim 2: Project the impact and cost-effectiveness of strategies for scaling up TB diagnostics Aim 3: Disseminate the structure and findings of a TB diagnostic model for use, adaptation, and improvement by the global TB control community Contributing to a systems approach: epidemiologic methods contribute to developing a systems model Viewing causality in a new light: considering multilevel factors “Addressing Hepatitis C and Hepatocellular Carcinoma: Current and Future Epidemics,” 2013 (R01 grant) Holly Hagan, PhDc, MPH (College of Nursing; New York University, New York, New York) Agent-based modeling Aim 1: Synthesize evidence on HCV epidemiology, prevention, progression, and treatment to inform an agent-based model Aim 2: Develop an agent-based model to estimate the effects of scaling up various evidence-based HCV interventions Aim 3: Determine the combination of interventions for particular budget and epidemiologic scenarios Defining evidence and contributing to a systems approach: epidemiologic evidence and methods contribute to developing an agent-based model Viewing causality in a new light: considering multilevel factors “Translating Molecular Diagnostics for Cervical Cancer Prevention into Practice,” 2016 (R01 grant) Patti Gravitt, PhDc (Global Health, Epidemiology; George Washington University, Washington, DC) 3 communities in the Loreto region of Peru Focus groups, key informant interviews, surveys Interrupted time-series design Aim 1: Conduct community census to establish cervical cancer screening registries and identify key members for working groups to facilitate Participatory Action Research approach Aim 2: Develop screening intervention monitoring and evaluation criteria, objectives, methods, and instruments for ongoing monitoring and evaluation of implementation, guided by the Consolidated Framework for Implementation Research Aim 3: Evaluate the comparative effectiveness of each strategy using qualitative and quantitative methods iteratively executed in both pre- and postimplementation phases embedded in the overall design, allowing real-time adaptation of the programs for broader sustainability Aim 4: Synthesize a formal cost-effectiveness evaluation to inform policy-makers Defining evidence to inform intervention Determining appropriate study design “A Retail Policy Laboratory: Modeling Impact of Retailer Reduction on Tobacco Use,” 2013 (R21 grant) Douglas Luke, PhD (Center for Public Health Systems Science; Washington University in St. Louis, St. Louis, Missouri) Agent-based modeling Aim 1: Build Tobacco Town, a simulation of a realistic community, to model tobacco retailer density and individual tobacco purchasing Aim 2: Use the model built in aim 1 as a retail policy laboratory to explore and compare the potential effects on behavior of a suite of real-world retailer reduction policy approaches. The effects of the retailer density policies on vulnerable populations will also be examined, particularly for low-income residents and minorities. Contributing to a systems approach: parameters of Tobacco Town informed by epidemiologic evidence of tobacco use and purchasing “Dissemination and Implementation of a Corrective Intervention to Improve Mediastinal Lymph Node Examination in Resected Lung Cancer,” 2013 (R01 grant) Raymond Osarogiagbon, MD (School of Public Health; University of Memphis, Memphis, Tennessee) 12 hospitals in western Tennessee, northern Mississippi, and eastern Arkansas Multiple-baseline study design Aim 1: Recruit institutions and surgeons to participate in an implementation study Aim 2: Evaluate effectiveness of intervention in a diverse mix of institutions to maximize external validity of the intervention Aim 3: Process evaluation of dissemination and implementation using RE-AIM Framework Defining evidence: data from the SEER Program on pathological lymph node staging informed the scope and magnitude of the problem “Sustainable Financial Incentives to Improve Prescription Practices for Malaria,” 2012 (R21 grant) Wendy Prudhomme-O’Meara, PhD (Medicine, Infectious Diseases; Duke University, Durham, North Carolina) 18 rural health facilities in western Kenya Cluster-randomized trial Objective: test whether financial incentives offered at the facility level improve targeting of antimalarial medications to patients with parasitologically diagnosed malaria Viewing causality in a new light “Adapting Patient Navigation to Promote Cancer Screening in Chicago’s Chinatown,” 2012 (R01 grant) Melissa Simon, MD (Obstetrics and Gynecology; Northwestern University at Chicago, Chicago, Illinois) Chinatown section of Chicago, Illinois Quasi-experimental (pre-/posttest, time-series) design Aim 1: Adapt and expand to broader cancer control education a patient navigator intervention designed for low-income Latinas and African Americans to Chicago’s Chinatown population Aim 2: Evaluate implementation of the patient navigator intervention using qualitative methods Aim 3: Evaluate adapted intervention’s impact on screening rates and support longer-term epidemiologic surveillance systems Aim 4: Compare adapted intervention’s rate of timely follow-up after abnormal test results with that of comparable control group receiving standard care Viewing causality in a new light: identifying multilevel barriers and facilitators to screening, including social, economic, cultural, and psychosocial barriers and community facilitators Project Title, Year (Funding Mechanism) . Principal Investigator (Department(s)/Center; Institution) . Study Setting/Design . Study Aims/Objectives . Example(s) of Epidemiologic Contributions . “Test and Treat TB: a Proof-of-Concept Trial in South Africa,” 2014 (R21a grant) Ingrid Valerie Bassett, MD, MPH (Infectious Diseases, Medicine; Massachusetts General Hospital, Boston, Massachusetts) Mobile, community-based HIV screening program in Durban, South Africa 3-arm randomized trial Aim 1: Assess the feasibility, yield, and clinical impact of a test-and-treat TB strategy integrated with ongoing rapid HIV screening Aim 2: Assess the cost and cost-effectiveness of the strategy Defining the evidence: identifying the scope of the problem to inform the study and intervention design “Enhancing Evidence-Based Diabetes Control Among Local Health Departments,” 2017 (R01b grant) Ross Brownson, PhDc (Epidemiology, Prevention Research Center; Washington University in St. Louis, St. Louis, Missouri) 30 local health departments in Missouri Qualitative needs assessment Group-randomized Aim 1: Refine and test measures to assess the adoption of evidence-based programs and policies in local health departments, building on pilot work Aim 2: Conduct a qualitative needs assessment of 15 local health departments to understand factors influencing the adoption of evidence-based programs and policies for diabetes control Aim 3: Conduct a group-randomized experimental study of 30 local health departments in Missouri to evaluate the effectiveness of active dissemination and implementation approaches Measuring intermediate outcomes: adapt and test valid and reliable dissemination measures Determining appropriate study designs: determine trade-offs in various randomized designs (cluster level) “SIngle-saMPLE Tuberculosis Evaluation to Facilitate Linkage to Care: The SIMPLE TB Trial,” 2016 (R01 grant) Adithya Cattamanchi, MD, MAS (Medicine; University of California, San Francisco, San Francisco, California) Cluster-randomized trial 20 community health centers in Uganda Aim 1: Determine comparative effectiveness of a TB diagnostic strategy designed for low-resource settings versus standard care to improve TB diagnosis and treatment initiation rates Aim 2: Evaluation of factors influencing adoption and maintenance of intervention components using mixed methods Aim 3: Perform economic and epidemic modeling to estimate the cost-effectiveness and epidemiologic impact of strategy Viewing causality in a new light: intervening at a health systems level to impact the epidemiology of TB in low-resource settings “Online Social Networks for Dissemination of Smoking Cessation Interventions,” 2011 (R01 grant) Nathan Cobb, MD (Institute for Tobacco Research; Truth Initiative Foundation, Washington, DC) Facebook (Facebook, Inc., Menlo Park, California) Randomized trial (factorial design) Aim 1: Identify and characterize structural characteristics of an effective dissemination strategy for an evidence-based tobacco cessation intervention Aim 2: Identify and characterize network characteristics of participants (i.e., modifiable variables) that influence dissemination and behavior change Defining evidence: epidemiology informs intervention development Viewing causality in a new light and determining study design: considering multilevel factors through social network analysis “A User-Friendly Epidemic-Economic Model of Diagnostic Tests for Tuberculosis,” 2012 (R21 grant) David Dowdy, MD, PhDc (Epidemiology; Johns Hopkins University, Baltimore, Maryland) Dynamic modeling Aim 1: Develop a combined epidemic-economic model of TB diagnosis Aim 2: Project the impact and cost-effectiveness of strategies for scaling up TB diagnostics Aim 3: Disseminate the structure and findings of a TB diagnostic model for use, adaptation, and improvement by the global TB control community Contributing to a systems approach: epidemiologic methods contribute to developing a systems model Viewing causality in a new light: considering multilevel factors “Addressing Hepatitis C and Hepatocellular Carcinoma: Current and Future Epidemics,” 2013 (R01 grant) Holly Hagan, PhDc, MPH (College of Nursing; New York University, New York, New York) Agent-based modeling Aim 1: Synthesize evidence on HCV epidemiology, prevention, progression, and treatment to inform an agent-based model Aim 2: Develop an agent-based model to estimate the effects of scaling up various evidence-based HCV interventions Aim 3: Determine the combination of interventions for particular budget and epidemiologic scenarios Defining evidence and contributing to a systems approach: epidemiologic evidence and methods contribute to developing an agent-based model Viewing causality in a new light: considering multilevel factors “Translating Molecular Diagnostics for Cervical Cancer Prevention into Practice,” 2016 (R01 grant) Patti Gravitt, PhDc (Global Health, Epidemiology; George Washington University, Washington, DC) 3 communities in the Loreto region of Peru Focus groups, key informant interviews, surveys Interrupted time-series design Aim 1: Conduct community census to establish cervical cancer screening registries and identify key members for working groups to facilitate Participatory Action Research approach Aim 2: Develop screening intervention monitoring and evaluation criteria, objectives, methods, and instruments for ongoing monitoring and evaluation of implementation, guided by the Consolidated Framework for Implementation Research Aim 3: Evaluate the comparative effectiveness of each strategy using qualitative and quantitative methods iteratively executed in both pre- and postimplementation phases embedded in the overall design, allowing real-time adaptation of the programs for broader sustainability Aim 4: Synthesize a formal cost-effectiveness evaluation to inform policy-makers Defining evidence to inform intervention Determining appropriate study design “A Retail Policy Laboratory: Modeling Impact of Retailer Reduction on Tobacco Use,” 2013 (R21 grant) Douglas Luke, PhD (Center for Public Health Systems Science; Washington University in St. Louis, St. Louis, Missouri) Agent-based modeling Aim 1: Build Tobacco Town, a simulation of a realistic community, to model tobacco retailer density and individual tobacco purchasing Aim 2: Use the model built in aim 1 as a retail policy laboratory to explore and compare the potential effects on behavior of a suite of real-world retailer reduction policy approaches. The effects of the retailer density policies on vulnerable populations will also be examined, particularly for low-income residents and minorities. Contributing to a systems approach: parameters of Tobacco Town informed by epidemiologic evidence of tobacco use and purchasing “Dissemination and Implementation of a Corrective Intervention to Improve Mediastinal Lymph Node Examination in Resected Lung Cancer,” 2013 (R01 grant) Raymond Osarogiagbon, MD (School of Public Health; University of Memphis, Memphis, Tennessee) 12 hospitals in western Tennessee, northern Mississippi, and eastern Arkansas Multiple-baseline study design Aim 1: Recruit institutions and surgeons to participate in an implementation study Aim 2: Evaluate effectiveness of intervention in a diverse mix of institutions to maximize external validity of the intervention Aim 3: Process evaluation of dissemination and implementation using RE-AIM Framework Defining evidence: data from the SEER Program on pathological lymph node staging informed the scope and magnitude of the problem “Sustainable Financial Incentives to Improve Prescription Practices for Malaria,” 2012 (R21 grant) Wendy Prudhomme-O’Meara, PhD (Medicine, Infectious Diseases; Duke University, Durham, North Carolina) 18 rural health facilities in western Kenya Cluster-randomized trial Objective: test whether financial incentives offered at the facility level improve targeting of antimalarial medications to patients with parasitologically diagnosed malaria Viewing causality in a new light “Adapting Patient Navigation to Promote Cancer Screening in Chicago’s Chinatown,” 2012 (R01 grant) Melissa Simon, MD (Obstetrics and Gynecology; Northwestern University at Chicago, Chicago, Illinois) Chinatown section of Chicago, Illinois Quasi-experimental (pre-/posttest, time-series) design Aim 1: Adapt and expand to broader cancer control education a patient navigator intervention designed for low-income Latinas and African Americans to Chicago’s Chinatown population Aim 2: Evaluate implementation of the patient navigator intervention using qualitative methods Aim 3: Evaluate adapted intervention’s impact on screening rates and support longer-term epidemiologic surveillance systems Aim 4: Compare adapted intervention’s rate of timely follow-up after abnormal test results with that of comparable control group receiving standard care Viewing causality in a new light: identifying multilevel barriers and facilitators to screening, including social, economic, cultural, and psychosocial barriers and community facilitators Abbreviations: HCV, hepatitis C virus; HIV, human immunodeficiency virus; MAS, master of advanced studies; MD, medical doctor; MPH, master of public health; PhD, doctor of philosophy; RE-AIM, Reach, Effectiveness, Adoption, Implementation, and Maintenance; SEER, Surveillance, Epidemiology, and End Results; TB, tuberculosis. a The R21 grant mechanism at the National Institutes of Health is for funding smaller exploratory or developmental research. b The R01 grant mechanism at the National Institutes of Health is for funding larger research projects. c Degree in epidemiology. Open in new tab CONCLUSION While it has been widely asserted that epidemiologists’ core function is to observe and analyze the distribution and control of diseases, we may disproportionately focus on the etiological questions at the cost of addressing the solutions. Given the emerging trend in translational sciences and the call by leaders in the field to take a more consequentialist approach (3, 4), we have provided a primer in implementation science as a means of catalyzing the attention of epidemiologists towards increasing the likelihood that research findings will be useful or consequential. We have enumerated several ways in which epidemiologists can engage in and drive implementation science, highlighting the critical role that epidemiology plays in informing the evidence base as well as evaluating the ultimate impact of health interventions. In Figure 1, we summarize the mutually beneficial relationship between the fields of epidemiology and implementation science and show how together these fields can ultimately affect population health. Epidemiology is critical to identifying patterns of disease distribution which can pinpoint public health problems, as well as understanding and measuring associated causes of and potential solutions or interventions with which to address those problems. These findings can lead to actionable information that may inform policy and practice decisions. However, effective interventions are only as useful as they are adopted, implemented, and sustained in practice. Implementation science is critical to identifying strategies that can drive adoption, implementation, and sustainability, ultimately leading to sustained practice change. In this way, the two fields contribute in parallel to improving population health. But these fields also can contribute to and collaborate with one another to the same end. Epidemiology can inform or drive implementation science by supplying evidence on causes of disease and effective interventions as well as informing study methods, measurement, and designs. Implementation science can enhance epidemiology by informing the research questions epidemiologists seek to answer as well as the measures and methods we use (70, 71). Together we can have a greater impact on improving population health. Figure 1. Open in new tabDownload slide Interrelationship between epidemiology and implementation science for maximization of population health impact. Figure 1. Open in new tabDownload slide Interrelationship between epidemiology and implementation science for maximization of population health impact. ACKNOWLEDGMENTS Author affiliations: Implementation Science, Division of Cancer Control and Population Sciences, National Cancer Institute, Bethesda, Maryland (Gila Neta, David A. Chambers); Prevention Research Center in St. Louis, Brown School, Washington University in St. Louis, St. Louis, Missouri (Ross C. Brownson); Division of Public Health Sciences, Department of Surgery, School of Medicine, Washington University in St. Louis, St. Louis, Missouri (Ross C. Brownson); and Alvin J. Siteman Cancer Center, School of Medicine, Washington University in St. Louis, St. Louis, Missouri (Ross C. Brownson). Conflict of interest: none declared. Abbreviations PRECIS-2 Pragmatic Explanatory Continuum Indicator Summary REFERENCES 1 Merriam-Webster, Inc . Definition of epidemiology. Springfield, MA : Merriam-Webster, Inc. ; 2017 . https://www.merriam-webster.com/dictionary/epidemiology. Accessed September 19, 2017. 2 Oxford Living Dictionaries . Definition of epidemiology in English. New York, NY : Oxford University Press ; 2017 . https://en.oxforddictionaries.com/definition/epidemiology. Accessed September 19, 2017. 3 Galea S . An argument for a consequentialist epidemiology . Am J Epidemiol . 2013 ; 178 ( 8 ): 1185 – 1191 . Google Scholar Crossref Search ADS PubMed WorldCat 4 Brownson RC , Samet JM, Chavez GF, et al. . Charting a future for epidemiologic training . Ann Epidemiol . 2015 ; 25 ( 6 ): 458 – 465 . Google Scholar Crossref Search ADS PubMed WorldCat 5 Balas EA , Boren SA. Managing clinical knowledge for health care improvement . Yearb Med Inform . 2000 ;( 1 ): 65 – 70 . Google Scholar OpenURL Placeholder Text WorldCat 6 Morris ZS , Wooding S, Grant J. The answer is 17 years, what is the question: understanding time lags in translational research . J R Soc Med . 2011 ; 104 ( 12 ): 510 – 520 . Google Scholar Crossref Search ADS PubMed WorldCat 7 Hanney SR , Castle-Clarke S, Grant J, et al. . How long does biomedical research take? Studying the time taken between biomedical and health research and its translation into products, policy, and practice . Health Res Policy Syst . 2015 ; 13 : 1 . Google Scholar Crossref Search ADS PubMed WorldCat 8 Shekelle PG , Pronovost PJ, Wachter RM, et al. . Advancing the science of patient safety . Ann Intern Med . 2011 ; 154 ( 10 ): 693 – 696 . Google Scholar Crossref Search ADS PubMed WorldCat 9 Freeman AC , Sweeney K. Why general practitioners do not implement evidence: qualitative study . BMJ . 2001 ; 323 ( 7321 ): 1100 – 1102 . Google Scholar Crossref Search ADS PubMed WorldCat 10 Howitt A , Armstrong D. Implementing evidence based medicine in general practice: audit and qualitative study of antithrombotic treatment for atrial fibrillation . BMJ . 1999 ; 318 ( 7194 ): 1324 – 1327 . Google Scholar Crossref Search ADS PubMed WorldCat 11 Titler MG . The evidence for evidence-based practice implementation. In: Hughes RG, ed. Patient Safety and Quality: An Evidence-Based Handbook for Nurses . Rockville, MD : Agency for Healthcare Research and Quality, US Department of Health and Human Services ; 2008 : 113 – 161 . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC 12 Massie J , Efron D, Cerritelli B, et al. . Implementation of evidence based guidelines for paediatric asthma management in a teaching hospital . Arch Dis Child . 2004 ; 89 ( 7 ): 660 – 664 . Google Scholar Crossref Search ADS PubMed WorldCat 13 Bryant J , Boyes A, Jones K, et al. . Examining and addressing evidence-practice gaps in cancer care: a systematic review . Implement Sci . 2014 ; 9 ( 1 ): 37 . Google Scholar Crossref Search ADS PubMed WorldCat 14 McGlynn EA , Asch SM, Adams J, et al. . The quality of health care delivered to adults in the United States . N Engl J Med . 2003 ; 348 : 2635 – 2645 . Google Scholar Crossref Search ADS PubMed WorldCat 15 Institute of Medicine (US) Committee on Quality of Health Care in America . Crossing the Quality Chasm: A New Health System for the 21st Century . Washington, DC : National Academies Press ; 2001 . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC 16 Levine DM , Linder JA, Landon BE. The quality of outpatient care delivered to adults in the United States, 2002 to 2013 . JAMA Intern Med . 2016 ; 176 ( 12 ): 1778 – 1790 . Google Scholar Crossref Search ADS PubMed WorldCat 17 Proctor EK . Leverage points for the implementation of evidence-based practice . Brief Treat Crisis Interv . 2004 ; 4 ( 3 ): 227 – 242 . Google Scholar Crossref Search ADS WorldCat 18 Straus SE , Tetroe J, Graham I. Defining knowledge translation . CMAJ . 2009 ; 181 ( 3-4 ): 165 – 168 . Google Scholar Crossref Search ADS PubMed WorldCat 19 Peters DH , Adam T, Alonge O, et al. . Implementation research: what it is and how to do it . BMJ . 2013 ; 347 : f6753 . Google Scholar Crossref Search ADS PubMed WorldCat 20 Eccles MP , Mittman BS. Welcome to Implementation Science [editorial]. Implement Sci . 2006 ; 1 : 1 . Google Scholar Crossref Search ADS WorldCat 21 National Institutes of Health . Department of Health and Human Services. Part 1. Overview information. Dissemination and Implementation Research in Health (R01). https://grants.nih.gov/grants/guide/pa-files/PAR-16-238.html. Published May 10, 2016. Accessed April 11, 2017. 22 National institutes of Health . Department of Health and Human Services. Part 1. Overview information. Dissemination and Implementation Research in Health (R03). https://grants.nih.gov/grants/guide/pa-files/PAR-16-237.html. Published May 10, 2016. Accessed April 11, 2017. 23 National Institutes of Health . Department of Health and Human Services. Part 1. Overview information. Dissemination and Implementation Research in Health (R21). https://grants.nih.gov/grants/guide/pa-files/PAR-16-236.html. Published May 10, 2016. Accessed April 11, 2017. 24 Proctor E , Silmere H, Raghavan R, et al. . Outcomes for implementation research: conceptual distinctions, measurement challenges, and research agenda . Adm Policy Ment Hlth . 2011 ; 38 ( 2 ): 65 – 76 . Google Scholar Crossref Search ADS WorldCat 25 Proctor EK , Brownson RC. Measurement issues in dissemination and implementation research. In: Brownson R, Colditz G, Proctor E, eds. Dissemination and Implementation Research in Health: Translating Science to Practice . New York, NY : Oxford University Press ; 2012 : 261 – 280 . Google Scholar Crossref Search ADS Google Preview WorldCat COPAC 26 Proctor EK , Powell BJ, Baumann AA, et al. . Writing implementation research grant proposals: ten key ingredients . Implement Sci . 2012 ; 7 : 96 . Google Scholar Crossref Search ADS PubMed WorldCat 27 Graham ID . Guide to Knowledge Translation Planning at CIHR: Integrated and End-of-Grant Approaches. Ottawa, Ontario, Canada : Canadian Institutes of Health Research ; 2015 . http://www.cihr-irsc.gc.ca/e/45321.html#a4. Updated March 19, 2015. Accessed March 3, 2017. 28 Stetler CB , Mittman BS, Francis J. Overview of the VA Quality Enhancement Research Initiative (QUERI) and QUERI theme articles: QUERI Series . Implement Sci 2008 ; 3 : 8 . Google Scholar Crossref Search ADS PubMed WorldCat 29 Loudon K , Treweek S, Sullivan F, et al. . The PRECIS-2 tool: designing trials that are fit for purpose . BMJ . 2015 ; 350 : h2147 . Google Scholar Crossref Search ADS PubMed WorldCat 30 Greenland S , Pearl J, Robins JM. Causal diagrams for epidemiologic research . Epidemiology . 1999 ; 10 ( 1 ): 37 – 48 . Google Scholar Crossref Search ADS PubMed WorldCat 31 Ahrens KA , Cole SR, Westreich D, et al. . A cautionary note about estimating effects of secondary exposures in cohort studies . Am J Epidemiol . 2015 ; 181 ( 3 ): 198 – 203 . Google Scholar Crossref Search ADS PubMed WorldCat 32 Tabak RG , Khoong EC, Chambers DA, et al. . Bridging research and practice: models for dissemination and implementation research . Am J Prev Med . 2012 ; 43 ( 3 ): 337 – 350 . Google Scholar Crossref Search ADS PubMed WorldCat 33 Sales A , Smith J, Curran G, et al. . Models, strategies, and tools. Theory in implementing evidence-based findings into health care practice . J Gen Intern Med . 2006 ; 21 ( suppl 2 ): S43 – S49 . Google Scholar PubMed OpenURL Placeholder Text WorldCat 34 Rogers EM . Diffusion of Innovations . 4th ed. New York, NY : The Free Press ; 1995 . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC 35 Glasgow RE , Vogt TM, Boles SM. Evaluating the public health impact of health promotion interventions: the RE-AIM framework . Am J Public Health . 1999 ; 89 ( 9 ): 1322 – 1327 . Google Scholar Crossref Search ADS PubMed WorldCat 36 Damschroder LJ , Aron DC, Keith RE, et al. . Fostering implementation of health services research findings into practice: a consolidated framework for advancing implementation science . Implement Sci . 2009 ; 4 : 50 . Google Scholar Crossref Search ADS PubMed WorldCat 37 Chambers DA . Guiding theory for dissemination and implementation research: a reflection on models used in research and practice. In: Beidas RS, Kendall PC, eds. Dissemination and Implementation of Evidence-Based Practices in Child and Adolescent Mental Health . New York, NY : Oxford University Press ; 2014 : 9 – 21 . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC 38 Chambers DA , Azrin ST. Partnership: a fundamental component of dissemination and implementation research . Psychiatr Serv . 2013 ; 64 ( 6 ): 509 – 511 . Google Scholar Crossref Search ADS PubMed WorldCat 39 Powell BJ , McMillen JC, Proctor EK, et al. . A compilation of strategies for implementing clinical innovations in health and mental health . Med Care Res Rev . 2012 ; 69 ( 2 ): 123 – 157 . Google Scholar Crossref Search ADS PubMed WorldCat 40 Landsverk J , Brown CH, Chamberlain P, et al. . Design and analysis in dissemination and implementation research. In: Brownson R, Colditz G, Proctor E, eds. Dissemination and Implementation Research in Health: Translating Science to Practice . New York, NY : Oxford University Press ; 2012 : 225 – 260 . Google Scholar Crossref Search ADS Google Preview WorldCat COPAC 41 Palinkas LA , Aarons GA, Horwitz S, et al. . Mixed method designs in implementation research . Adm Policy Ment Health . 2011 ; 38 ( 1 ): 44 – 53 . Google Scholar Crossref Search ADS PubMed WorldCat 42 Curran GM , Bauer M, Mittman B, et al. . Effectiveness-implementation hybrid designs: combining elements of clinical effectiveness and implementation research to enhance public health impact . Med Care . 2012 ; 50 ( 3 ): 217 – 226 . Google Scholar Crossref Search ADS PubMed WorldCat 43 Brennan SE , Bosch M, Buchan H, et al. . Measuring team factors thought to influence the success of quality improvement in primary care: a systematic review of instruments . Implement Sci . 2013 ; 8 : 20 . Google Scholar Crossref Search ADS PubMed WorldCat 44 Chaudoir SR , Dugan AG, Barr CH. Measuring factors affecting implementation of health innovations: a systematic review of structural, organizational, provider, patient, and innovation level measures . Implement Sci . 2013 ; 8 : 22 . Google Scholar Crossref Search ADS PubMed WorldCat 45 Chor KH , Wisdom JP, Olin SC, et al. . Measures for predictors of innovation adoption . Adm Policy Ment Health . 2015 ; 42 ( 5 ): 545 – 573 . Google Scholar Crossref Search ADS PubMed WorldCat 46 Emmons KM , Weiner B, Fernandez ME, et al. . Systems antecedents for dissemination and implementation: a review and analysis of measures . Health Educ Behav . 2012 ; 39 ( 1 ): 87 – 105 . Google Scholar Crossref Search ADS PubMed WorldCat 47 Gagnon MP , Attieh R, Ghandour EK, et al. . A systematic review of instruments to assess organizational readiness for knowledge translation in health care . PLoS One . 2014 ; 9 ( 12 ): e114338 . Google Scholar Crossref Search ADS PubMed WorldCat 48 Lewis CC , Fischer S, Weiner BJ, et al. . Outcomes for implementation science: an enhanced systematic review of instruments using evidence-based rating criteria . Implement Sci . 2015 ; 10 : 155 . Google Scholar Crossref Search ADS PubMed WorldCat 49 Hall KL , Vogel AL, Stipelman B, et al. . A four-phase model of transdisciplinary team-based research: goals, team processes, and strategies . Transl Behav Med . 2012 ; 2 ( 4 ): 415 – 430 . Google Scholar Crossref Search ADS PubMed WorldCat 50 Rabin BA , Brownson R. Developing the terminology for dissemination and implementation research. In: Brownson R, Colditz G, Proctor E, eds. Dissemination and Implementation Research in Health: Translating Science to Practice . New York, NY : Oxford University Press ; 2012 : 23 – 51 . Google Scholar Crossref Search ADS Google Preview WorldCat COPAC 51 Green LW , Ottoson JM, García C, et al. . Diffusion theory and knowledge dissemination, utilization, and integration in public health . Annu Rev Public Health . 2009 ; 30 : 151 – 174 . Google Scholar Crossref Search ADS PubMed WorldCat 52 Slack MK , Draugalis JR. Establishing the internal and external validity of experimental studies . Am J Health Syst Pharm . 2001 ; 58 ( 22 ): 2173 – 2181 . Google Scholar PubMed OpenURL Placeholder Text WorldCat 53 Cole SR , Stuart EA. Generalizing evidence from randomized clinical trials to target populations: the ACTG 320 trial . Am J Epidemiol . 2010 ; 172 ( 1 ): 107 – 115 . Google Scholar Crossref Search ADS PubMed WorldCat 54 Persaud N , Mamdani MM. External validity: the neglected dimension in evidence ranking . J Eval Clin Pract . 2006 ; 12 ( 4 ): 450 – 453 . Google Scholar Crossref Search ADS PubMed WorldCat 55 Gordis L . More on causal inferences: bias, confounding, and interaction. In: Epidemiology . 5th ed. Philadelphia, PA : Saunders Elsevier ; 2013 : 262 – 278 . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC 56 Diez Roux AV . Complex systems thinking and current impasses in health disparities research . Am J Public Health . 2011 ; 101 ( 9 ): 1627 – 1634 . Google Scholar Crossref Search ADS PubMed WorldCat 57 Glasgow RE , Chambers D. Developing robust, sustainable, implementation systems using rigorous, rapid and relevant science . Clin Transl Sci . 2012 ; 5 ( 1 ): 48 – 55 . Google Scholar Crossref Search ADS PubMed WorldCat 58 Glass TA , Goodman SN, Hernán MA, et al. . Causal inference in public health . Annu Rev Public Health . 2013 ; 34 : 61 – 75 . Google Scholar Crossref Search ADS PubMed WorldCat 59 Hill AB . The environment and disease: association or causation? Proc R Soc Med . 1965 ; 58 : 295 – 300 . Google Scholar PubMed OpenURL Placeholder Text WorldCat 60 Surgeon General’s Advisory Committee on Smoking and Health . Smoking and Health: Report of the Advisory Committee to the Surgeon General of the Public Health Service . (PHS publication no. 1103). Washington, DC : Public Health Service, US Department of Health, Education, and Welfare ; 1964 . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC 61 Galea S , Riddle M, Kaplan GA. Causal thinking and complex system approaches in epidemiology . Int J Epidemiol . 2010 ; 39 ( 1 ): 97 – 106 . Google Scholar Crossref Search ADS PubMed WorldCat 62 Hernán MA . A definition of causal effect for epidemiological research . J Epidemiol Community Health . 2004 ; 58 ( 4 ): 265 – 271 . Google Scholar Crossref Search ADS PubMed WorldCat 63 Hernán MA , Robins JM. Instruments for causal inference: an epidemiologist’s dream? Epidemiology . 2006 ; 17 ( 4 ): 360 – 372 . Google Scholar Crossref Search ADS PubMed WorldCat 64 Pearl J . An introduction to causal inference . Int J Biostat . 2010 ; 6 ( 2 ): Article 7 . Google Scholar Crossref Search ADS PubMed WorldCat 65 Parascandola M , Weed DL. Causation in epidemiology . J Epidemiol Community Health . 2001 ; 55 ( 12 ): 905 – 912 . Google Scholar Crossref Search ADS PubMed WorldCat 66 Rothman KJ , Greenland S. Causation and causal inference in epidemiology . Am J Public Health . 2005 ; 95 ( suppl 1 ): S144 – S150 . Google Scholar Crossref Search ADS PubMed WorldCat 67 Greenland S . For and against methodologies: some perspectives on recent causal and statistical inference debates . Eur J Epidemiol . 2017 ; 32 ( 1 ): 3 – 20 . Google Scholar Crossref Search ADS PubMed WorldCat 68 Joffe M , Gambhir M, Chadeau-Hyam M, et al. . Causal diagrams in systems epidemiology . Emerg Themes Epidemiol . 2012 ; 9 ( 1 ): 1 . Google Scholar Crossref Search ADS PubMed WorldCat 69 Brown CH , Curran G, Palinkas LA, et al. . An overview of research and evaluation designs for dissemination and implementation . Annu Rev Public Health . 2017 ; 38 : 1 – 22 . Google Scholar Crossref Search ADS PubMed WorldCat 70 Westreich D . From exposures to population interventions: pregnancy and response to HIV therapy . Am J Epidemiol . 2014 ; 179 ( 7 ): 797 – 806 . Google Scholar Crossref Search ADS PubMed WorldCat 71 Westreich D , Edwards JK, Rogawski ET, et al. . Causal impact: epidemiological approaches for a public health of consequence . Am J Public Health . 2016 ; 106 ( 6 ): 1011 – 1012 . Google Scholar Crossref Search ADS PubMed WorldCat 72 Cargo M , Mercer SL. The value and challenges of participatory research: strengthening its practice . Annu Rev Public Health . 2008 ; 29 : 325 – 350 . Google Scholar Crossref Search ADS PubMed WorldCat 73 Proctor EK , Powell BJ, McMillen JC. Implementation strategies: recommendations for specifying and reporting . Implement Sci . 2013 ; 8 : 139 . Google Scholar Crossref Search ADS PubMed WorldCat 74 Powell BJ , Waltz TJ, Chinman MJ, et al. . A refined compilation of implementation strategies: results from the Expert Recommendations for Implementing Change (ERIC) project . Implement Sci . 2015 ; 10 : 21 . Google Scholar Crossref Search ADS PubMed WorldCat 75 Rabin BA , Purcell P, Naveed S, et al. . Advancing the application, quality and harmonization of implementation science measures . Implement Sci . 2012 ; 7 : 119 . Google Scholar Crossref Search ADS PubMed WorldCat Published by Oxford University Press on behalf of the Johns Hopkins Bloomberg School of Public Health 2018. This work is written by (a) US Government employee(s) and is in the public domain in the US. Published by Oxford University Press on behalf of the Johns Hopkins Bloomberg School of Public Health 2018.

Pan, Xi;Luo, Ye;Roberts, Amy Restorick

doi: 10.1093/aje/kwx377pmid: 29370335

Abstract Exposure to secondhand smoke (SHS) is known to be harmful to health. However, the association between household SHS and cognitive function among middle-aged and older women in China is understudied. Lagged dependent variable regression was used to examine the association between household SHS exposure and the cognitive function of married women who had been exposed to SHS, using data from 2 waves of the China Health and Retirement Longitudinal Study (CHARLS, 2011–2013). Controlling for age, educational attainment, geographic residence, household expenditures, and chronic conditions (i.e., hypertension, diabetes, and depressive symptoms), the results indicated that longer SHS exposure was associated with a greater decline in memory over 2 years. After comparing differences across age groups, this pattern was significant for women aged 55–64 years. Furthermore, those who were illiterate, lived in rural areas, and reported depressive symptoms had a greater decline in memory. With evidence linking household SHS exposure with a higher risk of cognitive decline, effective education and public health intervention programs are urgently needed. Stronger tobacco control regulations and education about the dangers of household SHS are viable strategies to reduce the impending dementia epidemic in China. aging, cognitive function, household secondhand smoke, longitudinal study, women Editor’s note: An invited commentary on this article appears on page 919, and the authors’ response appears on page 922. The prevalence of cognitive impairment is rising in China. In 2011, national figures estimated that around 9% of older persons in China had cognitive impairment (1). The World Health Organization projects that, by 2050, 10 million older Chinese will have dementia (2). Cognitive impairment refers to reduction of ability in various tasks, such as memory, learning, concentration, language, and reasoning, that might affect a person’s everyday life (3, 4). People with mild cognitive impairment have a 10-fold increased risk of later developing greater levels of cognitive impairment common in Alzheimer’s disease and other dementias (5). Furthermore, the lifetime risk of cognitive impairment is disproportionately greater in women because they live longer and are more vulnerable to cognitive decline than men from a neurobiological standpoint (6, 7). Recent evidence has demonstrated that exposure to secondhand smoke (SHS), also called passive smoking, is associated with greater risk of cognitive impairment in older adults (8–10). SHS is the major source of indoor air pollution worldwide (11) and contains hundreds of carcinogenic chemicals. The ingestion of carcinogens may lead to cognitive impairment by increasing oxidative stress (5, 12, 13). Cross-sectional evidence has demonstrated that SHS exposure within the home is associated with a higher risk of Alzheimer’s disease and other dementias (8–10, 14, 15). Specifically, longer exposure to SHS has been reported to increase the risk of Alzheimer’s disease and vascular dementia (9, 10, 16). China is the world’s largest producer and consumer of tobacco. The prevalence of smoking in China is high, with 350 million active smokers and 740 million passive smokers in 2010 (17). Chinese women are more likely to be passive smokers and are heavily exposed to SHS at home because many live in households with husbands or other family members who smoke cigarettes. In 2010, nearly 50% of nonsmoking Chinese women were exposed to household SHS, and SHS exposure was even more common in rural China (18). However, there is a paucity of research addressing the relationship between SHS and cognitive function among older women, especially those residing in China. There is even less research using a longitudinal design in this area of inquiry. The primary aim of this study was to address this gap by examining the association between household SHS exposure and changes in the cognitive function of older, married women over the course of 2 years, using longitudinal, national survey data from China. The longitudinal data and our analytic approach provided an opportunity to contribute new knowledge about correlates of cognitive impairment among older Chinese women, particularly concerning environmental factors. Examining the relationship between SHS and cognitive function is also important for understanding the environmental etiology of adverse cognitive outcomes among older women in China. The secondary aim of the study was to examine whether the association between SHS exposure and cognitive function varies across different age cohorts. Prior longitudinal research has suggested that cognitive decline is already evident in middle age (ages 45–49 years) (19), and as such, the inclusion of a middle-aged group in this study is a crucial addition to the literature. Further, a descriptive study on the prevalence of age-associated memory impairment has found that the age-specific prevalence rate of age-associated memory impairment is highest in adults aged 60–64 years (45.7%) and lowest in those aged 75–78 years (20). Also, studies of active smoking and cognitive impairment among the Chinese population suggest that older current smokers (with an average age of 63 years) were more likely to develop cognitive impairment compared with the oldest smokers or never-smokers in China (1, 21). Thus, it is critical to investigate whether the probability of cognitive decline is also high among the older women who were exposed to SHS within similar age ranges, and to examine whether the risk of cognitive decline varied across different age cohorts ranging from mid-life to the oldest. Comparing the influence of SHS exposure across these age groups could help to identify whether certain age groups are at higher risk of cognitive decline than others. This study is vitally important to inform policy and programs aimed at reducing household SHS exposure and ultimately cognitive impairment in China. METHODS Data Data for this study were from 2 waves of the China Health and Retirement Longitudinal Study (CHARLS, 2011–2013), which included interviews with a nationally representative sample of adults aged 45 years or older, as well as their spouses when possible. The sample was obtained through multistage cluster sampling, with an overall response rate of 80.5% (nearly 64.15% in rural areas) (22, 23). The national baseline survey of CHARLS was conducted between June 2011 and March 2012 and included 17,705 respondents from 10,257 households. At baseline, there were 6,248 women who never smoked cigarettes and were aged 45 years or older, married, and living with spouses who had either smoked cigarettes in the past or smoked at the time of interview. Of these, 5,119 respondents were interviewed again during the second wave of data collection in 2013. Our final sample was composed of 2,037 respondents who had completed both surveys and provided complete data for all study variables. Measures Cognitive function CHARLS included items for cognitive function similar to those used in the Health and Retirement Study, which were components of the Telephone Interview of Cognitive Status battery (24). The Telephone Interview of Cognitive Status is a telephone version of the Mini-Mental State Examination and assesses cognitive function (25). An analysis by McArdle et al. (26) of the Health and Retirement Study data suggested 2 factors to adequately capture cognitive function: one factor related to episodic memory and a second factor related to other tasks of the Telephone Interview of Cognitive Status battery. Based upon this suggestion and recommendations from previous studies using the CHARLS data (27, 28), we constructed 2 measures of cognitive function. In CHARLS, memory was assessed through an immediate word recall based on respondents’ capacity to immediately repeat, in any order, 10 Chinese nouns just read to them, followed by a delayed-recall test of respondents’ ability to repeat the same list of words 4 minutes later (24, 25). A single score of memory was calculated by averaging the immediate- and delayed-recall scores, and it ranged from 0 to 10. In addition to the memory tests, CHARLS included some components of the mental status questions of the Telephone Interview of Cognitive Status battery. Our second measure of cognitive function was based on these questions and included items on orientation, visuoconstruction, and numeric ability. Orientation was assessed by asking respondents to name the current day’s date (month, day, year) and season and to identify the correct day of the week. Visuoconstruction was assessed by asking respondents to accurately redraw a previously shown picture. Numeric ability was assessed through the serial sevens test, which asks respondents to subtract 7 from 100 (up to 5 times), and whether additional explanation or an aid such as a paper and pencil was needed to complete the task. Scores on these items were aggregated into a single score that ranged from 0 to 11 and was labeled as mental status (28). For both measures, higher scores indicate better cognitive function (27, 28). Secondhand smoke Exposure to SHS among Chinese women was assessed through several survey items that asked husbands and wives about cigarette smoking at home. Both husbands and wives were asked about the year they got married, whether they had ever smoked, if they were currently smoking cigarettes, and the age or year when they started and stopped smoking. The length of SHS exposure was calculated to reflect the total number of years that never-smoking women lived with spouses who smoked cigarettes at home. Control variables Given that cognitive function may vary across educational levels and across health and socioeconomic status, we included age in years, urban/rural residence, education (illiterate, primary education, or at least secondary education), household expenditures, and chronic diseases as control variables (27–30). To measure financial resources, we calculated the annual household expenditures. The existing literature has shown that expenditures are a better measure of economic resources available to the family in developing countries than is income (31). The summed total of annual expenditures was log transformed. Physical health conditions included hypertension and diabetes, as well as whether participants were being treated for those conditions. The measure of depressive symptoms was based on responses to the 10-item version of the Center for Epidemiologic Studies Short Depression Scale. Each of the items has 4 response options coded from 0 to 3. The total score is the sum of points for all 10 items. A dichotomous variable was created using a score of 10 or greater to indicate the presence of depressive symptoms (32). Analysis To examine the association between SHS exposure and cognitive function, we used lagged dependent-variable regression models with ordinary least squares estimation. The lagged dependent-variable model is superior for analyzing the effects of predictor variables on an outcome with 2-wave panel data while controlling for the influence of time-invariant variables (33, 34). In our models, the length of SHS exposure was the predictor variable, and memory and mental status were 2 separate outcome variables with controls for the influence of age, education, household expenditures, urban/rural residence, and chronic diseases. Prior to fitting the regression models, descriptive analyses were conducted to estimate the mean, frequency, and missing values of key study variables for the entire sample and for the different age groups. All analyses and procedures were conducted in SAS, version 9.4 (SAS Institute, Inc., Cary, North Carolina). Due to attrition from the baseline to the follow-up survey and missing data on study variables that did not occur at random, we used multiple imputation with multivariate normal distribution to replace those missing cases and adjust for this potential bias. While 2–10 imputations are recommended (35), we chose to create 10 imputed data sets to maximize the adjustment for this potential bias. The imputed data sets were analyzed separately, and the results were combined in a way that accounts for variation in the imputed values (36). RESULTS Table 1 provides a descriptive summary of the variables for all study participants and according to age group. For the entire sample, women were 57 years old on average at baseline, and the majority completed primary education. Nearly two-thirds of the respondents lived in rural areas, and the average duration that women were exposed to SHS was 30 years. The group aged 65 years or older had the highest proportion of hypertension, diabetes, and depressive symptoms, as well as the lowest education levels, compared with the other 2 groups. The mean scores of memory and mental status significantly decreased across age groups. In comparison with the group aged 45–54 years, the mean scores of memory and mental status among the group aged 55–64 years were lower. Overall, changes in the scores of memory (from 3.91 to 3.87) and mental status (from 6.82 to 6.66) between the 2 waves were significant (for memory, t = 4.30, P < 0.001; for mental status, t = 4.92, P < 0.001). Table 1. Characteristics of Female Respondents Exposed to Secondhand Smoke and Cognitive Function According to Age, China Health and Retirement Longitudinal Study, 2011 Variable  Total  Age Group  P Valuea  No.  %  45–54 years  55–64 years  ≥65 years  No.  %  No.  %  No.  %  Observations  2,037    962  47.23  863  42.37  212  10.40    Age, yearsb  56.87 (8.03)                Length of SHS exposure, yearsb  30.08 (10.52)  24.62 (7.17)  33.30 (9.35)  41.83 (12.72)  <0.001  Household expenditure, yuanb,c  20,272.09 (14,429.74)  22,747.79(14,521.27)  19,062.79(14,276.11)  13,820.29(11,775.63)  <0.001  Cognitive function in 2011b             Memory scored  3.91 (1.75)  4.24 (1.75)  3.73 (1.70)  3.17 (1.70)  <0.001   Mental status scoree  6.82 (3.24)  7.43 (2.82)  6.45 (3.08)  5.38 (3.31)  <0.001  Cognitive function in 2013b             Memory scored  3.87(1.87)  4.23 (1.79)  3.58 (1.84)  3.00 (1.76)  <0.001   Mental status scoree  6.66 (3.22)  7.20 (3.08)  6.13 (3.16)  5.28 (3.43)  <0.001  Education                  <0.001   Illiterate  717  35.20  224  23.28  375  43.46  118  55.45   Primary education  754  37.01  336  34.93  352  40.90  66  31.28   Secondary education or above  566  27.79  402  41.79  136  15.64  28  13.27  Residence                  >0.050   Urban  711  34.90  352  36.59  287  33.26  72  34.12   Rural  1,326  65.10  610  63.41  576  66.74  140  65.88  Hypertension                  >0.050   Yes, with treatment  397  19.49  132  13.71  198  22.99  67  31.68   Yes, without treatment  132  6.48  50  5.19  61  7.04  21  9.86   No  1,508  74.03  780  81.10  604  69.97  124  58.46  Diabetes                  >0.050   Yes, with treatment  209  10.26  23  2.36  46  5.28  140  6.63   Yes, without treatment  40  1.96  15  1.54  20  2.33  5  2.27   No  1,788  87.78  924  96.40  797  92.39  67  8.90  Depressive symptomsf                  <0.001   Yes  795  39.05  328  34.10  374  43.34  93  44.08   No  1,242  60.95  634  65.90  489  56.66  119  55.92  Variable  Total  Age Group  P Valuea  No.  %  45–54 years  55–64 years  ≥65 years  No.  %  No.  %  No.  %  Observations  2,037    962  47.23  863  42.37  212  10.40    Age, yearsb  56.87 (8.03)                Length of SHS exposure, yearsb  30.08 (10.52)  24.62 (7.17)  33.30 (9.35)  41.83 (12.72)  <0.001  Household expenditure, yuanb,c  20,272.09 (14,429.74)  22,747.79(14,521.27)  19,062.79(14,276.11)  13,820.29(11,775.63)  <0.001  Cognitive function in 2011b             Memory scored  3.91 (1.75)  4.24 (1.75)  3.73 (1.70)  3.17 (1.70)  <0.001   Mental status scoree  6.82 (3.24)  7.43 (2.82)  6.45 (3.08)  5.38 (3.31)  <0.001  Cognitive function in 2013b             Memory scored  3.87(1.87)  4.23 (1.79)  3.58 (1.84)  3.00 (1.76)  <0.001   Mental status scoree  6.66 (3.22)  7.20 (3.08)  6.13 (3.16)  5.28 (3.43)  <0.001  Education                  <0.001   Illiterate  717  35.20  224  23.28  375  43.46  118  55.45   Primary education  754  37.01  336  34.93  352  40.90  66  31.28   Secondary education or above  566  27.79  402  41.79  136  15.64  28  13.27  Residence                  >0.050   Urban  711  34.90  352  36.59  287  33.26  72  34.12   Rural  1,326  65.10  610  63.41  576  66.74  140  65.88  Hypertension                  >0.050   Yes, with treatment  397  19.49  132  13.71  198  22.99  67  31.68   Yes, without treatment  132  6.48  50  5.19  61  7.04  21  9.86   No  1,508  74.03  780  81.10  604  69.97  124  58.46  Diabetes                  >0.050   Yes, with treatment  209  10.26  23  2.36  46  5.28  140  6.63   Yes, without treatment  40  1.96  15  1.54  20  2.33  5  2.27   No  1,788  87.78  924  96.40  797  92.39  67  8.90  Depressive symptomsf                  <0.001   Yes  795  39.05  328  34.10  374  43.34  93  44.08   No  1,242  60.95  634  65.90  489  56.66  119  55.92  Abbreviation: SHS, secondhand smoke. aP value from a 1-way analysis of variance for continuous variables or a χ2 test for categorical variables to indicate the differences across age groups. b Values are expressed as mean (standard deviation). c 1 US dollar = 6.3 yuan. d The score range for memory was 0–10. Higher scores indicate better memory. e The score range for mental status was 0–11. Higher scores indicate better mental status. f A score of 10 or greater indicated the presence of depressive symptoms. Table 1. Characteristics of Female Respondents Exposed to Secondhand Smoke and Cognitive Function According to Age, China Health and Retirement Longitudinal Study, 2011 Variable  Total  Age Group  P Valuea  No.  %  45–54 years  55–64 years  ≥65 years  No.  %  No.  %  No.  %  Observations  2,037    962  47.23  863  42.37  212  10.40    Age, yearsb  56.87 (8.03)                Length of SHS exposure, yearsb  30.08 (10.52)  24.62 (7.17)  33.30 (9.35)  41.83 (12.72)  <0.001  Household expenditure, yuanb,c  20,272.09 (14,429.74)  22,747.79(14,521.27)  19,062.79(14,276.11)  13,820.29(11,775.63)  <0.001  Cognitive function in 2011b             Memory scored  3.91 (1.75)  4.24 (1.75)  3.73 (1.70)  3.17 (1.70)  <0.001   Mental status scoree  6.82 (3.24)  7.43 (2.82)  6.45 (3.08)  5.38 (3.31)  <0.001  Cognitive function in 2013b             Memory scored  3.87(1.87)  4.23 (1.79)  3.58 (1.84)  3.00 (1.76)  <0.001   Mental status scoree  6.66 (3.22)  7.20 (3.08)  6.13 (3.16)  5.28 (3.43)  <0.001  Education                  <0.001   Illiterate  717  35.20  224  23.28  375  43.46  118  55.45   Primary education  754  37.01  336  34.93  352  40.90  66  31.28   Secondary education or above  566  27.79  402  41.79  136  15.64  28  13.27  Residence                  >0.050   Urban  711  34.90  352  36.59  287  33.26  72  34.12   Rural  1,326  65.10  610  63.41  576  66.74  140  65.88  Hypertension                  >0.050   Yes, with treatment  397  19.49  132  13.71  198  22.99  67  31.68   Yes, without treatment  132  6.48  50  5.19  61  7.04  21  9.86   No  1,508  74.03  780  81.10  604  69.97  124  58.46  Diabetes                  >0.050   Yes, with treatment  209  10.26  23  2.36  46  5.28  140  6.63   Yes, without treatment  40  1.96  15  1.54  20  2.33  5  2.27   No  1,788  87.78  924  96.40  797  92.39  67  8.90  Depressive symptomsf                  <0.001   Yes  795  39.05  328  34.10  374  43.34  93  44.08   No  1,242  60.95  634  65.90  489  56.66  119  55.92  Variable  Total  Age Group  P Valuea  No.  %  45–54 years  55–64 years  ≥65 years  No.  %  No.  %  No.  %  Observations  2,037    962  47.23  863  42.37  212  10.40    Age, yearsb  56.87 (8.03)                Length of SHS exposure, yearsb  30.08 (10.52)  24.62 (7.17)  33.30 (9.35)  41.83 (12.72)  <0.001  Household expenditure, yuanb,c  20,272.09 (14,429.74)  22,747.79(14,521.27)  19,062.79(14,276.11)  13,820.29(11,775.63)  <0.001  Cognitive function in 2011b             Memory scored  3.91 (1.75)  4.24 (1.75)  3.73 (1.70)  3.17 (1.70)  <0.001   Mental status scoree  6.82 (3.24)  7.43 (2.82)  6.45 (3.08)  5.38 (3.31)  <0.001  Cognitive function in 2013b             Memory scored  3.87(1.87)  4.23 (1.79)  3.58 (1.84)  3.00 (1.76)  <0.001   Mental status scoree  6.66 (3.22)  7.20 (3.08)  6.13 (3.16)  5.28 (3.43)  <0.001  Education                  <0.001   Illiterate  717  35.20  224  23.28  375  43.46  118  55.45   Primary education  754  37.01  336  34.93  352  40.90  66  31.28   Secondary education or above  566  27.79  402  41.79  136  15.64  28  13.27  Residence                  >0.050   Urban  711  34.90  352  36.59  287  33.26  72  34.12   Rural  1,326  65.10  610  63.41  576  66.74  140  65.88  Hypertension                  >0.050   Yes, with treatment  397  19.49  132  13.71  198  22.99  67  31.68   Yes, without treatment  132  6.48  50  5.19  61  7.04  21  9.86   No  1,508  74.03  780  81.10  604  69.97  124  58.46  Diabetes                  >0.050   Yes, with treatment  209  10.26  23  2.36  46  5.28  140  6.63   Yes, without treatment  40  1.96  15  1.54  20  2.33  5  2.27   No  1,788  87.78  924  96.40  797  92.39  67  8.90  Depressive symptomsf                  <0.001   Yes  795  39.05  328  34.10  374  43.34  93  44.08   No  1,242  60.95  634  65.90  489  56.66  119  55.92  Abbreviation: SHS, secondhand smoke. aP value from a 1-way analysis of variance for continuous variables or a χ2 test for categorical variables to indicate the differences across age groups. b Values are expressed as mean (standard deviation). c 1 US dollar = 6.3 yuan. d The score range for memory was 0–10. Higher scores indicate better memory. e The score range for mental status was 0–11. Higher scores indicate better mental status. f A score of 10 or greater indicated the presence of depressive symptoms. Results from the regression models for the relationship between SHS exposure and each measure of cognitive function for all respondents are reported in Table 2. Scores of memory and mental status among women in 2011 were a strong predictor of their corresponding cognitive function measures 2 years later. Higher baseline scores predicted better performance in memory and mental status in 2013. After controlling for the health and demographic variables, as well as baseline scores of memory and mental status, we found that exposure to SHS significantly predicted a decline in memory but not in mental status. For each 1-year increase in the length of SHS exposure, there was an additional 0.01-point decline in scores of memory in the 2013 follow-up. Therefore, SHS exposure was associated with a greater decline in memory over a 2-year period. Table 2. Ordinary Least Squares Regression Analysis of Cognitive Function in 2013 on Length of Secondhand Smoke Exposure, Cognitive Function in 2011, and Control Variables, With Multiple Imputation Procedure, Among Women Aged 45 Years or Older, China Health and Retirement Longitudinal Study, 2011–2013 Independent Variable  Memory Score in 2013  Mental Status Score in 2013  β  95% CI  t  β  95% CI  t  Length of SHS exposure  −0.01  −0.010, −0.002  −2.07a  −0.01  −0.013, 0.004  −1.52  Memory score in 2011  0.30  0.275, 0.332  20.95b        Mental status score in 2011        0.44  0.478, 0.539  26.36c  Age  −0.03  −0.034, −0.016  −5.48c  −0.01  −0.020, 0.005  −1.40  Urban Residence  0.26  0.161, 0.363  5.08c  0.49  0.268, 0.612  5.33c  Educationd               Primary  0.62  0.507, 0.724  18.37c  1.55  1.308, 1.722  14.16c   Secondary or above  1.26  1.124, 1.393  11.12c  2.35  1.990, 2.485  17.70c  Household expendituree  0.04  −0.024, 0.099  1.21  0.32  0.069, 0.474  2.92b  Hypertensionf               Treatment  0.18  −0.307, 0.670  0.17  −0.36  −0.901, 0.746  −0.83   Without treatment  −0.21  −0.895, 0.847  −0.01  0.51  −1.008, 1.141  0.90  Diabetesg               Treatment  0.82  −0.357, 1.989  1.37  0.82  −1.486, 2.546  0.76   Without treatment  −1.30  −2.953, 0.346  −1.55  1.10  −3.705, 2.327  −0.68  Depressive symptoms  −0.21  −0.323, −0.094  −3.59c  −0.30  −0.470, −0.134  −3.29c  Independent Variable  Memory Score in 2013  Mental Status Score in 2013  β  95% CI  t  β  95% CI  t  Length of SHS exposure  −0.01  −0.010, −0.002  −2.07a  −0.01  −0.013, 0.004  −1.52  Memory score in 2011  0.30  0.275, 0.332  20.95b        Mental status score in 2011        0.44  0.478, 0.539  26.36c  Age  −0.03  −0.034, −0.016  −5.48c  −0.01  −0.020, 0.005  −1.40  Urban Residence  0.26  0.161, 0.363  5.08c  0.49  0.268, 0.612  5.33c  Educationd               Primary  0.62  0.507, 0.724  18.37c  1.55  1.308, 1.722  14.16c   Secondary or above  1.26  1.124, 1.393  11.12c  2.35  1.990, 2.485  17.70c  Household expendituree  0.04  −0.024, 0.099  1.21  0.32  0.069, 0.474  2.92b  Hypertensionf               Treatment  0.18  −0.307, 0.670  0.17  −0.36  −0.901, 0.746  −0.83   Without treatment  −0.21  −0.895, 0.847  −0.01  0.51  −1.008, 1.141  0.90  Diabetesg               Treatment  0.82  −0.357, 1.989  1.37  0.82  −1.486, 2.546  0.76   Without treatment  −1.30  −2.953, 0.346  −1.55  1.10  −3.705, 2.327  −0.68  Depressive symptoms  −0.21  −0.323, −0.094  −3.59c  −0.30  −0.470, −0.134  −3.29c  Abbreviations: CI, confidence interval; SHS, secondhand smoke. aP < 0.05. bP < 0.01. cP < 0.001. d Referent: illiterate. e Household expenditure is expressed as the natural log of the annual household expenditure. f Referent: without hypertension. g Referent: without diabetes. Table 2. Ordinary Least Squares Regression Analysis of Cognitive Function in 2013 on Length of Secondhand Smoke Exposure, Cognitive Function in 2011, and Control Variables, With Multiple Imputation Procedure, Among Women Aged 45 Years or Older, China Health and Retirement Longitudinal Study, 2011–2013 Independent Variable  Memory Score in 2013  Mental Status Score in 2013  β  95% CI  t  β  95% CI  t  Length of SHS exposure  −0.01  −0.010, −0.002  −2.07a  −0.01  −0.013, 0.004  −1.52  Memory score in 2011  0.30  0.275, 0.332  20.95b        Mental status score in 2011        0.44  0.478, 0.539  26.36c  Age  −0.03  −0.034, −0.016  −5.48c  −0.01  −0.020, 0.005  −1.40  Urban Residence  0.26  0.161, 0.363  5.08c  0.49  0.268, 0.612  5.33c  Educationd               Primary  0.62  0.507, 0.724  18.37c  1.55  1.308, 1.722  14.16c   Secondary or above  1.26  1.124, 1.393  11.12c  2.35  1.990, 2.485  17.70c  Household expendituree  0.04  −0.024, 0.099  1.21  0.32  0.069, 0.474  2.92b  Hypertensionf               Treatment  0.18  −0.307, 0.670  0.17  −0.36  −0.901, 0.746  −0.83   Without treatment  −0.21  −0.895, 0.847  −0.01  0.51  −1.008, 1.141  0.90  Diabetesg               Treatment  0.82  −0.357, 1.989  1.37  0.82  −1.486, 2.546  0.76   Without treatment  −1.30  −2.953, 0.346  −1.55  1.10  −3.705, 2.327  −0.68  Depressive symptoms  −0.21  −0.323, −0.094  −3.59c  −0.30  −0.470, −0.134  −3.29c  Independent Variable  Memory Score in 2013  Mental Status Score in 2013  β  95% CI  t  β  95% CI  t  Length of SHS exposure  −0.01  −0.010, −0.002  −2.07a  −0.01  −0.013, 0.004  −1.52  Memory score in 2011  0.30  0.275, 0.332  20.95b        Mental status score in 2011        0.44  0.478, 0.539  26.36c  Age  −0.03  −0.034, −0.016  −5.48c  −0.01  −0.020, 0.005  −1.40  Urban Residence  0.26  0.161, 0.363  5.08c  0.49  0.268, 0.612  5.33c  Educationd               Primary  0.62  0.507, 0.724  18.37c  1.55  1.308, 1.722  14.16c   Secondary or above  1.26  1.124, 1.393  11.12c  2.35  1.990, 2.485  17.70c  Household expendituree  0.04  −0.024, 0.099  1.21  0.32  0.069, 0.474  2.92b  Hypertensionf               Treatment  0.18  −0.307, 0.670  0.17  −0.36  −0.901, 0.746  −0.83   Without treatment  −0.21  −0.895, 0.847  −0.01  0.51  −1.008, 1.141  0.90  Diabetesg               Treatment  0.82  −0.357, 1.989  1.37  0.82  −1.486, 2.546  0.76   Without treatment  −1.30  −2.953, 0.346  −1.55  1.10  −3.705, 2.327  −0.68  Depressive symptoms  −0.21  −0.323, −0.094  −3.59c  −0.30  −0.470, −0.134  −3.29c  Abbreviations: CI, confidence interval; SHS, secondhand smoke. aP < 0.05. bP < 0.01. cP < 0.001. d Referent: illiterate. e Household expenditure is expressed as the natural log of the annual household expenditure. f Referent: without hypertension. g Referent: without diabetes. For each 1-year increase in age, there was an additional 0.03-point decline in scores for memory. The association between education and memory was strong, as higher levels of education protected against memory decline. Compared with illiterate respondents, the memory score in 2013 for those who finished at least primary education was 0.62 higher, and for those with secondary education or above, the score for memory was 1.26 higher. Moreover, respondents living in urban settings had higher scores in memory than respondents living in rural areas. Depressive symptoms also significantly predicted declines in memory and mental status. In our study sample, the magnitude of memory decline among those with depressive symptoms was 0.21 greater than among those without depressive symptoms, and the magnitude of the decline in mental status among those with depressive symptoms was 0.30 greater than those without depressive symptoms. Similarly, urban residence, more education, and higher household expenditures (a proxy for wealth) significantly predicted a lesser degree of decline in mental status. We further explored whether the association between SHS exposure and memory varied by age among the 3 age groups: 45–54 years, 55–64 years, and 65 years or older. As shown in Table 3, a significant relationship was found between SHS exposure and memory for the group aged 55–64 years. In this age group, for each 1-year increase in exposure to SHS, there was an additional 0.01 decline in scores of memory. Memory was better among the Chinese women who were living in urban environments than among their rural counterparts. Table 3. Ordinary Least Squares Regression of Memory in 2013 on Length of Secondhand Smoke Exposure, Memory in 2011, and Control Variables With Multiple Imputation Procedure, for Women According to Age Group, China Health and Retirement Longitudinal Study, 2011–2013 Independent Variable  Age Group  45–54 years(n = 962)  55–64 years(n = 863)  ≥65 years (n = 212)  β  95% CI  t  β  95% CI  t  β  95% CI  t  Length of SHS exposure  −0.00  −0.019, 0.004  −1.27  −0.01  −0.020, −0.001  −2.24a  −0.00  −0.014, 0.012  −0.15  Memory score in 2011  0.29  0.240, 0.332  12.23b  0.31  0.264, 0.358  13.82b  0.31  0.235, 0.382  8.30b  Age  −0.03  −0.054, 0.001  −1.91  −0.04  −0.064, −0.009  −1.99a  −0.04  −0.065, −0.009  −2.61a  Urban Residence  0.18  0.030, 0.338  2.34b  0.36  0.197, 0.523  2.94c  0.26  −0.034, 0.473  1.67  Educationd                     Primary  0.63  0.434, 0.819  6.39b  0.53  0.376, 0.688  6.60b  0.79  0.537, 1.042  6.14b   Secondary or above  1.31  1.107, 1.508  12.80b  1.12  0.894, 1.341  9.17b  1.58  1.203, 1.962  8.19b  Household expendituree  0.05  −0.041, 0.140  1.08  0.02  −0.077, 0.109  0.92  0.05  −0.067, 0.163  0.83  Hypertensionf                     Treatment  −0.10  −0.881, 0.688  −0.24  0.31  −0.313, 0.937  0.59  0.31  −1.053, 1.672  0.46   Without treatment  0.16  −0.889, 1.217  0.31  −0.36  −1.194, 0.475  −0.25  −0.44  −2.263, 1.387  −0.49  Diabetesg                     Treatment  1.26  −1.897, 4.423  0.79  0.70  −0.530, 1.938  0.41  1.13  −9.732, 10.550  0.08   Without treatment  −2.14  −6.610, 2.326  −0.94  −1.11  −2.836, 0.620  −0.61  −1.78  −14.882, 13.662  −0.09  Depressive symptoms  −0.26  −0.323, −0.094  −2.84c  −0.02  −0.338, −0.015  −2.14a  −0.22  −0.476, 0.033  −1.72  Independent Variable  Age Group  45–54 years(n = 962)  55–64 years(n = 863)  ≥65 years (n = 212)  β  95% CI  t  β  95% CI  t  β  95% CI  t  Length of SHS exposure  −0.00  −0.019, 0.004  −1.27  −0.01  −0.020, −0.001  −2.24a  −0.00  −0.014, 0.012  −0.15  Memory score in 2011  0.29  0.240, 0.332  12.23b  0.31  0.264, 0.358  13.82b  0.31  0.235, 0.382  8.30b  Age  −0.03  −0.054, 0.001  −1.91  −0.04  −0.064, −0.009  −1.99a  −0.04  −0.065, −0.009  −2.61a  Urban Residence  0.18  0.030, 0.338  2.34b  0.36  0.197, 0.523  2.94c  0.26  −0.034, 0.473  1.67  Educationd                     Primary  0.63  0.434, 0.819  6.39b  0.53  0.376, 0.688  6.60b  0.79  0.537, 1.042  6.14b   Secondary or above  1.31  1.107, 1.508  12.80b  1.12  0.894, 1.341  9.17b  1.58  1.203, 1.962  8.19b  Household expendituree  0.05  −0.041, 0.140  1.08  0.02  −0.077, 0.109  0.92  0.05  −0.067, 0.163  0.83  Hypertensionf                     Treatment  −0.10  −0.881, 0.688  −0.24  0.31  −0.313, 0.937  0.59  0.31  −1.053, 1.672  0.46   Without treatment  0.16  −0.889, 1.217  0.31  −0.36  −1.194, 0.475  −0.25  −0.44  −2.263, 1.387  −0.49  Diabetesg                     Treatment  1.26  −1.897, 4.423  0.79  0.70  −0.530, 1.938  0.41  1.13  −9.732, 10.550  0.08   Without treatment  −2.14  −6.610, 2.326  −0.94  −1.11  −2.836, 0.620  −0.61  −1.78  −14.882, 13.662  −0.09  Depressive symptoms  −0.26  −0.323, −0.094  −2.84c  −0.02  −0.338, −0.015  −2.14a  −0.22  −0.476, 0.033  −1.72  Abbreviations: CI, confidence interval; SHS, secondhand smoke. aP < 0.05. bP < 0.001. cP < 0.01. d Referent: illiterate. e Household expenditure is expressed as the natural log of the annual household expenditure (Chinese yuan). f Referent: without hypertension. g Referent: without diabetes. Table 3. Ordinary Least Squares Regression of Memory in 2013 on Length of Secondhand Smoke Exposure, Memory in 2011, and Control Variables With Multiple Imputation Procedure, for Women According to Age Group, China Health and Retirement Longitudinal Study, 2011–2013 Independent Variable  Age Group  45–54 years(n = 962)  55–64 years(n = 863)  ≥65 years (n = 212)  β  95% CI  t  β  95% CI  t  β  95% CI  t  Length of SHS exposure  −0.00  −0.019, 0.004  −1.27  −0.01  −0.020, −0.001  −2.24a  −0.00  −0.014, 0.012  −0.15  Memory score in 2011  0.29  0.240, 0.332  12.23b  0.31  0.264, 0.358  13.82b  0.31  0.235, 0.382  8.30b  Age  −0.03  −0.054, 0.001  −1.91  −0.04  −0.064, −0.009  −1.99a  −0.04  −0.065, −0.009  −2.61a  Urban Residence  0.18  0.030, 0.338  2.34b  0.36  0.197, 0.523  2.94c  0.26  −0.034, 0.473  1.67  Educationd                     Primary  0.63  0.434, 0.819  6.39b  0.53  0.376, 0.688  6.60b  0.79  0.537, 1.042  6.14b   Secondary or above  1.31  1.107, 1.508  12.80b  1.12  0.894, 1.341  9.17b  1.58  1.203, 1.962  8.19b  Household expendituree  0.05  −0.041, 0.140  1.08  0.02  −0.077, 0.109  0.92  0.05  −0.067, 0.163  0.83  Hypertensionf                     Treatment  −0.10  −0.881, 0.688  −0.24  0.31  −0.313, 0.937  0.59  0.31  −1.053, 1.672  0.46   Without treatment  0.16  −0.889, 1.217  0.31  −0.36  −1.194, 0.475  −0.25  −0.44  −2.263, 1.387  −0.49  Diabetesg                     Treatment  1.26  −1.897, 4.423  0.79  0.70  −0.530, 1.938  0.41  1.13  −9.732, 10.550  0.08   Without treatment  −2.14  −6.610, 2.326  −0.94  −1.11  −2.836, 0.620  −0.61  −1.78  −14.882, 13.662  −0.09  Depressive symptoms  −0.26  −0.323, −0.094  −2.84c  −0.02  −0.338, −0.015  −2.14a  −0.22  −0.476, 0.033  −1.72  Independent Variable  Age Group  45–54 years(n = 962)  55–64 years(n = 863)  ≥65 years (n = 212)  β  95% CI  t  β  95% CI  t  β  95% CI  t  Length of SHS exposure  −0.00  −0.019, 0.004  −1.27  −0.01  −0.020, −0.001  −2.24a  −0.00  −0.014, 0.012  −0.15  Memory score in 2011  0.29  0.240, 0.332  12.23b  0.31  0.264, 0.358  13.82b  0.31  0.235, 0.382  8.30b  Age  −0.03  −0.054, 0.001  −1.91  −0.04  −0.064, −0.009  −1.99a  −0.04  −0.065, −0.009  −2.61a  Urban Residence  0.18  0.030, 0.338  2.34b  0.36  0.197, 0.523  2.94c  0.26  −0.034, 0.473  1.67  Educationd                     Primary  0.63  0.434, 0.819  6.39b  0.53  0.376, 0.688  6.60b  0.79  0.537, 1.042  6.14b   Secondary or above  1.31  1.107, 1.508  12.80b  1.12  0.894, 1.341  9.17b  1.58  1.203, 1.962  8.19b  Household expendituree  0.05  −0.041, 0.140  1.08  0.02  −0.077, 0.109  0.92  0.05  −0.067, 0.163  0.83  Hypertensionf                     Treatment  −0.10  −0.881, 0.688  −0.24  0.31  −0.313, 0.937  0.59  0.31  −1.053, 1.672  0.46   Without treatment  0.16  −0.889, 1.217  0.31  −0.36  −1.194, 0.475  −0.25  −0.44  −2.263, 1.387  −0.49  Diabetesg                     Treatment  1.26  −1.897, 4.423  0.79  0.70  −0.530, 1.938  0.41  1.13  −9.732, 10.550  0.08   Without treatment  −2.14  −6.610, 2.326  −0.94  −1.11  −2.836, 0.620  −0.61  −1.78  −14.882, 13.662  −0.09  Depressive symptoms  −0.26  −0.323, −0.094  −2.84c  −0.02  −0.338, −0.015  −2.14a  −0.22  −0.476, 0.033  −1.72  Abbreviations: CI, confidence interval; SHS, secondhand smoke. aP < 0.05. bP < 0.001. cP < 0.01. d Referent: illiterate. e Household expenditure is expressed as the natural log of the annual household expenditure (Chinese yuan). f Referent: without hypertension. g Referent: without diabetes. DISCUSSION To our knowledge, this study is the first investigation of cognitive function in relation to household SHS exposure for middle-aged and older women in China using longitudinal data. We found that longer SHS exposure was related to greater declines over time in memory scores in this population-based sample. On average, the negative coefficient indicated that, compared with those not exposed to SHS, women with SHS exposure have a greater decline in memory scores. We also found that among women aged 55–64 years, greater exposure to SHS was related to greater declines in memory scores. No association between SHS exposure and cognitive function was observed in women aged 65 years or older, perhaps due to the relatively few cases in this age group. A larger sample may be needed to detect significant effect for this group. Despite the small regression coefficient (0.01), our findings are not trivial given that the average score of memory declined 0.04 points over the 2-year period. Additionally, our finding is quite similar in magnitude to prior research on the relationship between age and cognitive function (score declines of –0.043 for memory and –0.035 for mental status) using the 2011 CHARLS data (27), and a larger coefficient tends to be found in cross-sectional studies (37). The existing literature suggests that the “significance of effect” cannot be judged only by the P value or the regression coefficient but also by the confidence interval (37). In the present study, the 95% confidence interval for SHS on memory is small, which provides some evidence that the association between SHS and memory is high (see Tables 2 and 3). Women between the ages of 55 and 64 years who had longer exposure to household smoking in their marriage also demonstrated lower scores on memory 2 years later, after controlling for levels of education, geographic residence, household expenditures, chronic diseases, and baseline memory score. The present findings are important in strengthening the view that exposure to household SHS can hasten the onset of cognitive impairment for older Chinese women. Our study also highlighted that education protected cognitive function from decline among these older Chinese women. Educational levels for older Chinese women are generally quite low, and they are even lower within the oldest cohorts. Findings from our investigation show that even some level of education could prevent the decline of cognitive function. Given the socioeconomic development occurring in China, there have been impressive improvements over time in educational achievements, especially for women (27), which may lead to better cognitive function in several dimensions (38). In line with previous studies (27, 28, 30), we found that older women with depressive symptoms had lower cognitive function. Thus, screening and treatment for depressive symptoms presents another opportunity for interventions that could improve the cognitive function of older Chinese women. Prior research suggests that cognitive impairment tends to begin at younger ages in China than in Western countries. Existing evidence indicates that cognitive impairment is common among adults in the Chinese population starting near age 55 years (39), whereas in Western countries such cognitive impairment tends to occur approximately a decade later (around age 65 years) (9, 10). The earlier age of onset may be one reason that could explain the significant cognitive decline among the cohort of women aged 55–64 years who were exposed to SHS over a longer period of time. Women within this age group are at risk due to their older age and lower education. It is also noteworthy that the majority of respondents in our sample were from rural China. Given the combination of cohort influences and inefficient cognitive screenings in China, particularly in rural areas, it is not surprising that a significant association between SHS exposure and memory decline is found among this age group. In contrast, cognitive decline was not observed among the youngest group in our study, likely due to the combined protection of younger age and higher education. In addition, despite minimal influence, selection bias might have influenced the results. For the oldest group in our sample (those aged 65 years or older), it is possible that participants were cognitively and physically healthier than women of the same age who did not participate in CHARLS. Along these same lines, those who experienced higher health risks from environmental toxins such as SHS exposure might have suffered a premature death. The absence of a direct association between SHS exposure and mental status suggests that clinical research is needed to investigate the exact mechanism by which SHS exposure affects cognitive function pertaining to memory but not to other cognitive functioning. Perhaps future research can identify a specific gene or other biomarkers that clarify how SHS exposure affects domain-specific cognitive capacity. Limitations of the present study should also be noted. First, given the small sample of male passive smokers as well as women aged 65 years and older, it is difficult to test the interactions of sex and SHS exposure, or age and SHS exposure, on cognitive function. Thus, the connections between SHS exposure and cognitive function found in the present study should be tested more rigorously using larger samples and longer-term panel data. Second, as with any longitudinal study, selective attrition and inadequate control of all potentially related variables could have influenced the findings. Third, self-reported responses may underestimate or overestimate exposure to SHS (40). Future research could include additional observed or biological measures of SHS and cognitive functioning to improve the precision of measurement. Despite these limitations, findings from this study call for public health policy in tobacco control and smoking cessation to eliminate SHS exposure in households. In many regions of China, particularly rural communities, offering a cigarette to guests is a common and highly valued social custom (41). Policies should involve widespread public education about the harmful effects of SHS exposure to inform and change individuals’ attitudes and social customs. Sustainable efforts, including school-based anti-SHS education for children and young adults as well as community-based education for middle-aged and older adults, are needed to raise awareness of the hazards of SHS. Many rural areas in China lack the counseling, medication, and information to address tobacco addiction. As a result, it is difficult for smokers to quit smoking. SHS does not have a safe level of exposure, nor can the risk be completely mitigated through ventilation or filtration (42). The health of nonsmokers can be protected only through completely smoke-free environments (43). Therefore, education on SHS is crucially important to empower every nonsmoker to prohibit smoking indoors to reduce household SHS exposure, especially for women. Furthermore, smoke-free laws and regulations should be enforced through action by county- and city-level governments. Results of this study showed that the cognitive function of Chinese women in 2011 was a strong predictor of their cognitive function 2 years later. Therefore, cognitive health screenings for the middle-aged and younger elderly populations in urban and rural communities should be recommended, because early screening tests and behavioral interventions including SHS cessation could postpone or prevent the onset of dementia. Campaigns for SHS cessation should be required by the government and implemented in schools, hospitals, workplaces, public areas, and communities, particularly in rural areas (44). In addition, multimedia communication strategies—including television, radio, billboards, newsprint, and the Internet, as well as newer technological, smartphone-based social-networking systems—may be worthwhile to warn the Chinese people about the health hazards associated with SHS. Previous research has suggested that health warnings located directly on the cigarette packaging are an effective tool to curb individual smoking (17). More effective warning labels, such as pictorial warnings about SHS, should be located on the front of cigarette packaging in China to raise public awareness of SHS hazards, particularly among rural residents and those who are less educated. ACKNOWLEDGMENTS Author affiliations: Department of Sociology, Texas State University, San Marcos, Texas (Xi Pan); Department of Sociology, Anthropology, and Criminal Justice, Clemson University, Clemson, South Carolina (Ye Luo); and Department of Family Science and Social Work, Miami University, Oxford, Ohio (Amy Restorick Roberts). This study was not funded by any grant or financial support. We thank Dr. Cheryl Dye (Department of Public Health Sciences, Clemson University, Clemson, South Carolina) and Dr. Chad L. Smith (Department of Sociology, Texas State University, San Marcos, Texas) for excellent research assistance with manuscript revision. Conflict of interest: none declared. Abbreviations CHARLS China Health and Retirement Longitudinal Study SHS secondhand smoke REFERENCES 1 Yin P, Ma Q, Wang L, et al.  . Chronic obstructive pulmonary disease and cognitive impairment in the Chinese elderly population: a large national survey. Int J Chron Obstruct Pulmon Dis . 2016; 11: 399– 406. Google Scholar CrossRef Search ADS PubMed  2 Xiao S, Lewis M, Mellor D, et al.  . The China longitudinal ageing study: overview of the demographic, psychosocial and cognitive data of the Shanghai sample. J Ment Health . 2016; 25( 2): 131– 136. Google Scholar CrossRef Search ADS PubMed  3 Centers for Disease Control and Prevention. Cognitive impairment: the impact on health in Iowa; 2011. https://www.cdc.gov/aging/pdf/cognitive_impairment/cogimp_ia_final.pdf. Published February, 2011. Accessed June 13, 2016. 4 Zhou H, Deng J, Li J, et al.  . Study of the relationship between cigarette smoking, alcohol drinking and cognitive impairment among elderly people in China. Age Ageing . 2003; 32( 2): 205– 210. Google Scholar CrossRef Search ADS PubMed  5 Yuan L, Liu J, Ma W, et al.  . Dietary pattern and antioxidants in plasma and erythrocyte in patients with mild cognitive impairment from China. Nutrition . 2016; 32( 2): 193– 198. Google Scholar CrossRef Search ADS PubMed  6 Lin KA, Choudhury KR, Rathakrishnan BG, et al.  . Marked gender differences in progression of mild cognitive impairment over 8 years. Alzheimers Dement . 2015; 1: 103– 110. 7 Goveas JS, Rapp SR, Hogan PE, et al.  . Predictors of optimal cognitive aging in 80+ women: the Women’s Health Initiative Memory Study. J Gerontol A Biol Sci Med Sci . 2016; 71( suppl 1): S62– S71. Google Scholar CrossRef Search ADS PubMed  8 Chen R. Association of environmental tobacco smoke with dementia and Alzheimer’s disease among never smokers. Alzheimers Dement . 2012; 8( 6): 590– 595. Google Scholar CrossRef Search ADS PubMed  9 Oberg M, Jaakkola MS, Woodward A, et al.  . Worldwide burden of disease from exposure to second-hand smoke: a retrospective analysis of data from 192 countries. Lancet . 2011; 377( 9760): 139– 146. Google Scholar CrossRef Search ADS PubMed  10 Eisner MD. Passive smoking and cognitive impairment. BMJ . 2009; 338: a3070. Google Scholar CrossRef Search ADS PubMed  11 Li W, Tse LA, Au JSK, et al.  . Secondhand smoke enhances lung cancer risk in male smokers: an interaction. Nicotine Tob Res . 2016; 18( 11): 2057– 2064. Google Scholar CrossRef Search ADS PubMed  12 Barnoya J, Glantz SA. Cardiovascular effects of secondhand smoke: nearly as large as smoking. Circulation . 2005; 111( 20): 2684– 2698. Google Scholar CrossRef Search ADS PubMed  13 Anstey KJ, von Sanden C, Salim A, et al.  . Smoking as a risk factor for dementia and cognitive decline: a meta-analysis of prospective studies. Am J Epidemiol . 2007; 166( 4): 367– 378. Google Scholar CrossRef Search ADS PubMed  14 Ward A, Arrighi HM, Michels S, et al.  . Mild cognitive impairment: disparity of incidence and prevalence estimates. Alzheimers Dement . 2012; 8( 1): 14– 21. Google Scholar CrossRef Search ADS PubMed  15 Chen R, Zhang D, Chen Y, et al.  . Passive smoking and risk of cognitive impairment in women who never smoke. Arch Intern Med . 2012; 172( 3): 271– 273. Google Scholar CrossRef Search ADS PubMed  16 Barnes DE, Haight TJ, Mehta KM, et al.  . Secondhand smoke, vascular diseases, and dementia incidence: findings form the cardiovascular health cognition study. Am J Epidemiol . 2010; 171( 3): 292– 302. Google Scholar CrossRef Search ADS PubMed  17 Zhang J, Ou JX, Bai CX. Tobacco smoking in China: prevalence, disease burden, challenges and future strategies. Respirology . 2011; 16( 8): 1165– 1172. Google Scholar CrossRef Search ADS PubMed  18 Zhang DM, Hu Z, Orton S, et al.  . Socio-economic and psychosocial determinants of smoking and passive smoking in older adults. Biomed Environ Sci . 2013; 26( 6): 453– 467. Google Scholar PubMed  19 Singh-Manoux A, Kivimaki M, Glymour MM, et al.  . Timing of onset of cognitive decline: results from Whitehall II prospective cohort study. BMJ . 2012; 344: d7622. Google Scholar CrossRef Search ADS PubMed  20 Koivisto K, Reinikainen KJ, Hänninen T, et al.  . Prevalence of age-associated memory impairment in a randomly selected population from eastern Finland. Neurology . 1995; 45( 4): 741– 747. Google Scholar CrossRef Search ADS PubMed  21 Juan D, Zhou DH, Li J, et al.  . A 2-year follow-up study of cigarette smoking and risk of dementia. Eur J Neurol . 2004; 11( 4): 277– 282. Google Scholar CrossRef Search ADS PubMed  22 Li LW, Liu J, Zhang Z, et al.  . Late-life depression in Rural China: do village infrastructure and availability of community resources matter? Int J Geriatr Psychiatry . 2015; 30( 7): 729– 736. Google Scholar CrossRef Search ADS PubMed  23 Zhao Y, Strauss J, Yang G, et al.  . China Health and Retirement Longitudinal Study—2011–2012 National Baseline Users’ Guide. Beijing: Peking University; 2013. http://charls.pku.edu.cn/uploads/document/2011-charls-wave1/application/CHARLS_nationalbaseline_users_guide.pdf 24 Crimmins EM, Kim JK, Langa KM, et al.  . Assessment of cognition using surveys and neuropsychological assessment: the Health and Retirement Study and the Aging, Demographics, and Memory Study. J Gerontol B Psychol Sci Soc Sci . 2011; 66( suppl 1): i162– i171. Google Scholar CrossRef Search ADS PubMed  25 Bender AC, Austin AM, Grodstein F, et al.  . Executive function, episodic memory, and medicare expenditures. Alzheimers Dement . 2017; 13( 7): 792– 800. Google Scholar CrossRef Search ADS PubMed  26 McArdle JJ, Fisher GG, Kadlec KM. Latent variable analyses of age trends of cognition in the Health and Retirement Study, 1992–2004. Psychol Aging . 2007; 22( 3): 525– 545. Google Scholar CrossRef Search ADS PubMed  27 Lei X, Hu Y, McArdle JJ, et al.  . Gender differences in cognition among older adults in China. J Hum Resour . 2012; 47( 4): 951– 971. Google Scholar PubMed  28 Lei X, Smith JP, Sun X, et al.  . Gender differences in cognition in China and reasons for change over time: evidence from CHARLS. J Econ Ageing . 2014; 4: 46– 55. Google Scholar CrossRef Search ADS PubMed  29 Smith JP, Shen Y, Strauss J, et al.  . The effects of childhood health on adult health and SES in China. Econ Dev Cult Change . 2012; 61( 1): 127– 156. Google Scholar CrossRef Search ADS PubMed  30 Nie H, Xu Y, Liu B, et al.  . The prevalence of mild cognitive impairment about elderly population in China: a meta-analysis. Int J Geriatr Psychiatry . 2011; 26( 6): 558– 563. Google Scholar CrossRef Search ADS PubMed  31 Strauss J, Thomas D. Health over the life course. In: Schultz TP, Strauss J, eds. Handbook of Development Economics , vol. 4. Amsterdam, Holland: North Holland Press; 2008: 3375– 3474. 32 Björgvinsson T, Kertz SJ, Bigda-Peyton JS, et al.  . Psychometric properties of the CES-D-10 in a psychiatric sample. Assessment . 2013; 20( 4): 429– 436. Google Scholar CrossRef Search ADS PubMed  33 Johnson D. Two-wave panel analysis: comparing statistical methods for studying the effects of transitions. J Marriage Fam . 2005; 67( 4): 1061– 1075. Google Scholar CrossRef Search ADS   34 Allison PD. Fixed Effects Regression Methods for Longitudinal Data Using SAS . Cary, NC: SAS; 2005. 35 Rubin DB. Multiple Imputation for Survey Nonresponse . New York, NY: Wiley; 1987. Google Scholar CrossRef Search ADS   36 von Hippel PT. TEACHER’S CORNER: how many imputations are needed? A comment on Hershberger and Fisher (2003). Struct Equ Modeling . 2005; 12( 2): 334– 335. Google Scholar CrossRef Search ADS   37 Coe R. It’s the effect size, stupid: what effect size is and why it is important. Presented at the Annual Conference of the British Educational Research Association, University of Exeter, England, 2002. 38 Cagney KA, Lauderdale DS. Education, wealth, and cognitive function in later life. J Gerontol B Psychol Sci Soc Sci . 2002; 57( 2): P163– P172. Google Scholar CrossRef Search ADS PubMed  39 Huang J, Meyer JS, Zhang Z, et al.  . Progression of mild cognitive impairment to Alzheimer’s or vascular dementia versus normative aging among elderly Chinese. Curr Alzheimer Res . 2005; 2( 5): 571– 578. Google Scholar CrossRef Search ADS PubMed  40 Nondahl DM, Cruickshanks KJ, Schubert CR. A questionnaire for assessing environmental tobacco smoke exposure. Environ Res . 2005; 97( 1): 76– 82. Google Scholar CrossRef Search ADS PubMed  41 Ma S, Hoang MA, Samet JM, et al.  . Myths and attitudes that sustain smoking in China. J Health Commun . 2008; 13( 7): 654– 666. Google Scholar CrossRef Search ADS PubMed  42 Wang CP, Ma SJ, Xu XF, et al.  . The prevalence of household second-hand smoke exposure and its correlated factors in six counties of China. Tob Control . 2009; 18( 2): 121– 126. Google Scholar CrossRef Search ADS PubMed  43 American Society of Heating, Refrigerating, and Air-Conditioning Engineers. Environmental tobacco smoke. Society’s environmental tobacco smoke position, Atlanta, Georgia: American Society of Heating, Refrigerating and Air-conditioning Engineers Document Committee 2005. https://www.ashrae.org/File%20Library/docLib/About%20Us/PositionDocuments/ASHRAE_PD_Environmental_Tobacco_Smoke_2016.pdf. Accessed June 23, 2016. 44 Yang TZ, Cao C, Cottrell RR, et al.  . Second hand smoke exposure in public venues and mental disorder: a representative nationwide study of China. Tob Induc Dis . 2015; 13( 1): 18. Google Scholar CrossRef Search ADS PubMed  © The Author(s) 2018. Published by Oxford University Press on behalf of the Johns Hopkins Bloomberg School of Public Health. All rights reserved. For permissions, please e-mail: [email protected]. This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/about_us/legal/notices)

Anstey, Kaarin J;Chen, Ruoling

doi: 10.1093/aje/kwx378pmid: 29370342

Abstract Pan et al. (Am J Epidemiol. 2018;187(5):911–918) reported findings that exposure to secondhand smoke (SHS) was associated with cognitive decline over the course of 2 years among middle-aged and older Chinese women who never smoked, and they also reported a dose-response relationship. SHS exposure affects vulnerable people disproportionately because they have less control or choice over their living and working environment. Smoking is an established risk factor for dementia, but recent evidence reports on dementia-risk increase have not included SHS. Many epidemiologic studies collect data on smoking but not SHS exposure. SHS may be one of the most prevalent and modifiable risk factors for cognitive decline and therefore represents a major potential target for reduction of dementia risk. Given the high prevalence of smoking in China and other parts of the world, there is an urgent need to raise awareness of SHS reduction as part of global and national strategies to reduce cognitive decline and dementia and to introduce legislation that protects nonsmokers and vulnerable children and adults from SHS. cognitive decline, dementia, prevention, risk factor The current study by Pan et al. (1) provides new evidence on the risk of secondhand smoke (SHS) for cognitive decline in mid- and late adulthood, highlighting a risk factor that has not been included as a priority in recent influential evidence reports (2, 3) on dementia prevention. This builds on previous cross-sectional research on adults aged ≥60 years from 5 provinces in China (4), reporting an association between dementia diagnosis and exposure to environmental tobacco smoke. People who develop cognitive deficits in middle age are likely to have greater risk of late-life dementia than are adults who develop cognitive impairment for the first time in late life (5). By 2030, 66 million adults globally are expected to have dementia (6). Cognitive impairment that is not severe enough to meet criteria for a diagnosis of dementia has a prevalence of 16.8% in adults aged >65 years in Canada (7) and 20% in adults aged >70 years in the United States (8). With the lack of any significant breakthrough in discovering an effective treatment or cure for dementia or Alzheimer disease, risk reduction is the mainstay of current dementia-prevention programs. Findings on SHS and cognitive decline in the Chinese Health and Retirement Longitudinal Study (CHARLS) are particularly important because the study is representative of the population, is longitudinal, and has measures comparable to those of other national surveys, including the US Health and Retirement Study (9). This enables cross-national comparison of rates of exposures for late-life disease. Within China, exposure to SHS affects approximately 50% of the population (10, 11) and within the CHARLS data set, participants had an average of 30 years of exposure. The impact of SHS on cognitive decline is likely compounded by other highly prevalent risk factors that reduce brain and cognitive reserve. Globally, the most prevalent risk factor for dementia is low education (up to lower secondary school, according to the International Standard Education Classification), which affects 40% of the world population (12) and up to 70% in China (13). Pan et al. (1) reported data for participants in middle age, which has been identified as a crucial period for focused lifestyle and vascular risk reduction interventions (14–17). The authors reported that SHS was associated with more rapid memory decline but not with decline in cognitive status as measured using a telephone-interview version of the Mini-Mental State Examination (MMSE). The MMSE is designed to detect cognitive impairment and has ceiling effects in the normal population. It usually lacks sensitivity to detect cognitive decline in normal aging, although significant change in the sample was detected over the follow-up period of 2 years. The extent of decline on the cognitive measures in CHARLS due to SHS per annum was one-third of that due to aging (1). However, it is possible that there was further cognitive decline not captured by the limited cognitive test battery in the study. Other risk factors reported by Pan et al., including low education, rural setting, and depressive symptoms, were associated with cognitive decline in the study (1) and are consistent with the wider literature. The results provide some possible explanations for observed demographic differences in cognitive impairment/dementia (18), with rural Chinese women having higher rates of dementia than rural men and urban-dwelling women. The incidence of dementia in China is similar to that in Europe and the United States (19), and prevalence rates vary in reports according to region and urban and rural setting, with generally higher rates of dementia in women (20–22). Previous research from 5 provinces in China (4) has found an increased risk of dementia and Alzheimer disease associated with environmental tobacco smoke. Findings from Pan et al. (1) provided a potential explanation for the observation of increasing prevalence of dementia in China, particularly in women (23), and suggest that without intervention, high rates of incident dementia (24) might occur. The higher rates of smoking and exposure to SHS in China compared with those in the United States and Europe may in part explain differences in incidence and prevalence rates between countries. However US smoking data (25) reveal large disparities in smoking rates that are associated with socioeconomic status and disadvantage, with the majority of US smokers now being socially disadvantaged. Given that smoking and exposure to SHS are associated with increased risk of cognitive decline and late-life dementia, these data suggest that the long-term disparities in health and health behavior of adults, and associated environmental exposures of children to factors such as SHS (26), may translate into disparities in dementia incidence in late-life in the United States and other countries. In addition to including older adults, Pan et al. (1) reported findings for middle-aged adults who had completed 2 survey interviews during 1–2 years of follow-up in CHARLS. Middle age is also when other modifiable risk factors for cognitive decline occur, such as high serum cholesterol and overweight or obese body mass index. Previous research on the impact of SHS has focused on early-life development or very old adults. Hence this report fills a gap in the life-course evidence base on the impact of SHS exposure. Other recent reports of the impact of air pollution on risk of cognitive impairment/dementia (27, 28) suggest that strategies for dementia risk reduction need to focus more broadly on environmental factors that have an impact on health. The study also raises further questions and topics for investigation. It is limited by the restriction in sex to women and by SHS measured from husband’s smoking at home. Data are needed to confirm a similar effect of SHS on men. Including SHS exposure at work and in other places would give a fuller picture of the impact of SHS on cognitive impairment. The longer-term follow-up of the participants who did not have cognitive impairment at baseline would help further clarify the causal effect of SHS on cognitive impairment. Adjustments in the data analysis including other important confounding factors, such as body mass index and stroke, would help assess the extent to which SHS increased the risk of cognitive impairment. The interaction between SHS and other known risk factors for dementia, such as the apolipoprotein ε4 genotype, and other vascular risk factors also needs exploration. Concomitant neuroimaging data would provide information on whether smoking and SHS have similar impacts on vulnerable brain regions and the accumulation of tau and amyloid pathology. Further follow-up of the sample for incident Alzheimer disease and vascular dementia would also allow for evaluation of how these results translate into risk factors for dementia. The reversibility of risk associated with SHS is not known. One study of older adults on a smoking cessation program found that successful quitters showed reduced brain atrophy compared with unsuccessful quitters, and they also had cognitive trajectories more similar to never smokers in a control group (29). However there appears to be a lack of data on removal of the risk of SHS and subsequent impact on cognitive function. SHS exposure differs from many other behavioral risk factors for chronic disease because individual behavior affects others. Infants, children, or those with low mobility may be unable to move away from environments with SHS. Adults with dementia or cognitive impairment may have a higher level of SHS exposure within the home because they may be less mobile or not recognize the harmful effects of SHS, compounding effects in this group. This raises questions about public responsibility for exposure to SHS and the role of governments in regulating exposure. In Australia, laws governing SHS exposure have been instituted to protect children. For example, adults are not allowed to smoke while driving with children. Most jurisdictions also prohibit smoking in workplaces, on public transport, on airplanes, and in restaurants. However, although Chinese legislation has called for a ban on smoking in all public places over the last few years, progress has been hindered by the high prevalence of smoking (10), and there has been limited reduction in the prevalence of SHS exposure in China (30). In the past, health concerns in relation to SHS have focused on cardiorespiratory diseases and cancer risk. However, there is now evidence that exposure to SHS is associated with an increased risk of cognitive impairment, probably also including dementia. At present, 93% of the world’s population live in countries that are not fully covered by smoke-free public health regulations, and 35% of people in the world are exposed to SHS (31). Given the prevalence and length of exposure of this potential risk factor for cognitive decline, both education and regulatory approaches appear warranted to protect the future brain health of current populations, thus reducing the epidemic of dementia in the world. ACKNOWLEDGMENTS Author affiliations: Centre for Research on Ageing, Health and Wellbeing, Research School of Population Health, College of Medicine and Health, Australian National University, Canberra, Australian Capital Territory, Australia (Kaarin J. Anstey); and Centre for Health and Social Care Improvement, Faculty of Education, Health, and Wellbeing, University of Wolverhampton, Wolverhampton, United Kingdom (Ruoling Chen). K.J.A. is currently at the School of Psychology, Faculty of Science, University of New South Wales, and Neuroscience Research Australia, in Sydney, Australia. K.J.A. and R.C. contributed equally to this work. K.J.A. is funded by National Health and Medical Research Council Fellowship (award 1002560). Conflict of interest: none declared. Abbreviations CHARLS China Health and Retirement Longitudinal Study SHS secondhand smoke REFERENCES 1 Pan X, Luo Y, Roberts AR. Secondhand smoke and women’s cognitive function in China. Am J Epidemiol . 2018; 187( 5): 911– 918. 2 Livingston G, Sommerlad A, Orgeta V, et al.  . Dementia prevention, intervention, and care. Lancet . 2017; 390( 10113): 2673– 2734. Google Scholar CrossRef Search ADS PubMed  3 Leshner AI, Landis S, Stroud C, et al.  ., eds. Preventing Cognitive Decline and Dementia: A Way Forward . Washington, DC: The National Academies Press, 2017: 180. Google Scholar CrossRef Search ADS   4 Chen R, Wilson K, Chen Y, et al.  . Association between environmental tobacco smoke exposure and dementia syndromes. Occup Environ Med . 2013; 70( 1): 63– 69. Google Scholar CrossRef Search ADS PubMed  5 Espinosa A, Alegret M, Valero S, et al.  . A longitudinal follow-up of 550 mild cognitive impairment patients: evidence for large conversion to dementia rates and detection of major risk factors involved. J Alzheimers Dis . 2013; 34( 3): 769– 780. Google Scholar PubMed  6 Alzheimer’s Disease International. The Global Impact of Dementia: an analysis of prevalence, incidence, cost and trends. World Alzheimer Reports. London, UK: Alzheimer’s Disease International; 2015: 82. 7 Graham JE, Rockwood K, Beattie BL, et al.  . Prevalence and severity of cognitive impairment with and without dementia in an elderly population. Lancet . 1997; 349( 9068): 1793– 1796. Google Scholar CrossRef Search ADS PubMed  8 Plassman BL, Langa KM, Fisher GG, et al.  . Prevalence of cognitive impairment without dementia in the United States. Ann Intern Med . 2008; 148( 6): 427– 434. Google Scholar CrossRef Search ADS PubMed  9 Chan KS, Kasper JD, Brandt J, et al.  . Measurement equivalence in ADL and IADL difficulty across international surveys of aging: findings from the HRS, SHARE, and ELSA. J Gerontol B Psychol Sci Soc Sci . 2012; 67( 1): 121– 132. Google Scholar CrossRef Search ADS PubMed  10 Yang G, Fan L, Tan J, et al.  . Smoking in China: findings of the 1996 National Prevalence Survey. JAMA . 1999; 282( 13): 1247– 1253. Google Scholar CrossRef Search ADS PubMed  11 Yang GH, Ma JM, Liu N, et al.  . Smoking and passive smoking in Chinese, 2002. Zhonghua Liu Xing Bing Xue Za Zhi . 2005; 26( 2): 77– 83. Google Scholar PubMed  12 Norton S, Matthews FE, Barnes DE, et al.  . Potential for primary prevention of Alzheimer’s disease: an analysis of population-based data. Lancet Neurol . 2014; 13( 8): 788– 794. Google Scholar CrossRef Search ADS PubMed  13 National Bureau of Statistics of China. The Sixth National Population Census of China, 2010 Population Census Data File, Beijing, China. China Statistics Press; 2010. http://www.stats.gov.cn/english/Statisticaldata/CensusData/rkpc2010/indexch.htm. Updated April 23, 2013. Accessed September 21, 2017. 14 Debette S, Seshadri S, Beiser A, et al.  . Midlife vascular risk factor exposure accelerates structural brain aging and cognitive decline. Neurology . 2011; 77( 5): 461– 468. Google Scholar CrossRef Search ADS PubMed  15 Defina LF, Willis BL, Radford NB, et al.  . The association between midlife cardiorespiratory fitness levels and later-life dementia: a cohort study. Ann Intern Med . 2013; 158( 3): 162– 168. Google Scholar CrossRef Search ADS PubMed  16 Kivipelto M, Helkala EL, Hänninen T, et al.  . Midlife vascular risk factors and late-life mild cognitive impairment: a population-based study. Neurology . 2001; 56( 12): 1683– 1689. Google Scholar CrossRef Search ADS PubMed  17 Launer LJ, Hughes T, Yu B, et al.  . Lowering midlife levels of systolic blood pressure as a public health strategy to reduce late-life dementia: perspective from the Honolulu Heart Program/Honolulu Asia Aging Study. Hypertension . 2010; 55( 6): 1352– 1359. Google Scholar CrossRef Search ADS PubMed  18 Chen R, Ma Y, Wilson K, et al.  . A multicentre community-based study of dementia cases and subcases in older people in China—the GMS-AGECAT prevalence and socio-economic correlates. Int J Geriatr Psychiatry . 2012; 27( 7): 692– 702. Google Scholar CrossRef Search ADS PubMed  19 Yuan J, Zhang Z, Wen H, et al.  . Incidence of dementia and subtypes: a cohort study in four regions in China. Alzheimers Dement . 2016; 12( 3): 262– 271. Google Scholar CrossRef Search ADS PubMed  20 Ji Y, Shi Z, Zhang Y, et al.  . Prevalence of dementia and main subtypes in rural northern China. Dement Geriatr Cogn Disord . 2015; 39( 5–6): 294– 302. Google Scholar CrossRef Search ADS PubMed  21 Wu YT, Lee HY, Norton S, et al.  . Period, birth cohort and prevalence of dementia in mainland China, Hong Kong and Taiwan: a meta-analysis. Int J Geriatr Psychiatry . 2014; 29( 12): 1212– 1220. Google Scholar CrossRef Search ADS PubMed  22 Yang L, Jin X, Yan J, et al.  . Prevalence of dementia, cognitive status and associated risk factors among elderly of Zhejiang province, China in 2014. Age Ageing . 2016; 45( 5): 708– 712. Google Scholar PubMed  23 Jia J, Wang F, Wei C, et al.  . The prevalence of dementia in urban and rural areas of China. Alzheimers Dement . 2014; 10( 1): 1– 9. Google Scholar CrossRef Search ADS PubMed  24 Chen R, Hu Z, Wei L, et al.  . Incident dementia in a defined older Chinese population. PLoS One . 2011; 6( 9): e24817. Google Scholar CrossRef Search ADS PubMed  25 Jamal A, King BA, Neff LJ, et al.  . Current cigarette smoking among adults—United States, 2005–2015. MMWR Morb Mortal Wkly Rep . 2016; 65( 44): 1205– 1211. Google Scholar CrossRef Search ADS PubMed  26 Chen R, Clifford A, Lang L, et al.  . Is exposure to secondhand smoke associated with cognitive parameters of children and adolescents?—a systematic literature review. Ann Epidemiol . 2013; 23( 10): 652– 661. Google Scholar CrossRef Search ADS PubMed  27 Calderón-Garcidueñas L, Villarreal-Ríos R. Living close to heavy traffic roads, air pollution, and dementia. Lancet . 2017; 389( 10070): 675– 677. Google Scholar CrossRef Search ADS PubMed  28 Clifford A, Lang L, Chen R, et al.  . Exposure to air pollution and cognitive functioning across the life course—a systematic literature review. Environ Res . 2016; 147: 383– 398. Google Scholar CrossRef Search ADS PubMed  29 Almeida OP, Garrido GJ, Alfonso H, et al.  . 24-month effect of smoking cessation on cognitive function and brain structure in later life. Neuroimage . 2011; 55( 4): 1480– 1489. Google Scholar CrossRef Search ADS PubMed  30 Li Z, Yao Y, Yu Y, et al.  . Prevalence and associated factors of passive smoking among women in Jilin Province, China: a cross-sectional study. Int J Environ Res Public Health . 2015; 12( 11): 13970– 13980. Google Scholar CrossRef Search ADS PubMed  31 Oberg M, Jaakkola MS, Woodward A, et al.  . Worldwide burden of disease from exposure to second-hand smoke: a retrospective analysis of data from 192 countries. Lancet . 2011; 377( 9760): 139– 146. Google Scholar CrossRef Search ADS PubMed  © The Author(s) 2018. Published by Oxford University Press on behalf of the Johns Hopkins Bloomberg School of Public Health. All rights reserved. For permissions, please e-mail: [email protected]. This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/about_us/legal/notices)

Pan, Xi;Luo, Ye;Roberts, Amy Restorick

doi: 10.1093/aje/kwx379pmid: 29370330

The commentary with in-depth and broad perspectives by Anstey and Chen (1) provides a precise assessment of the strengths and limitations of our longitudinal study on secondhand smoke (SHS) and cognitive function among women in mid- and late adulthood in China (2). We agree with their conclusions that the most important strength of our current study is that it provides new evidence on the risk of SHS for cognitive decline from a life-course perspective, using population-representative data. Most existing literature focuses either on the influence of active smoking or on very old adults. We believe ours is the first investigation providing life-course evidence for cognitive function in relation to household SHS exposure for middle-aged or older women in China. Findings of our study strengthen the view that exposure to household SHS can hasten the onset of cognitive impairment for Chinese women and affirms middle age as a crucial period for risk reduction intervention (1, 2). We also agree with Anstey and Chen’s comment that our study suggests that long-term environmental exposure to factors such as SHS, even at a minimal level, cumulatively results in cognitive decline and in disparities in the health of women and children (1–3). This life-course evidence might be replicated in the United States and other countries in order to investigate environmental risks for dementia incidence in late life (1). Anstey and Chen pointed out that our study did not find a significant association between SHS and mental status, perhaps because the Mini-Mental State Examination (MMSE) has a ceiling effect and may lack sensitivity to detect cognitive impairment in normal aging (1). A systematic review and meta-analysis conducted by Mitchell suggests that the ceiling effect means that “MMSE may not perform well in people with mild cognitive impairment and the test is only marginally impaired in the detection of mild dementia as compared to detection of moderate to severe dementia” (4, p. 44). Mitchell’s study also shows that MMSE performs best when screening, separating individuals with dementia from healthy cognitively unimpaired individuals, based on results from 108 studies (4). In the China Health and Retirement Longitudinal Study (CHARLS), the MMSE was used for dementia screening rather than diagnosis, and our study aimed to assess the association between cognitive function in memory and mental status and SHS rather than grading the severity of cognitive impairment. For this purpose, cognitive function measured by MMSE scores in our study is adequate, especially given a lower educational level among our study population. We fully agree with Anstey and Chen that the single tool of MMSE is not sufficient to test cognitive function, and a more extensive neuropsychological and clinical evaluation should be included in CHARLS as well as in future studies. As we suggested in our discussion, future research could include additional observed or biological measures of SHS and cognitive functioning to improve the precision of measurement (2). In addition, we applaud Anstey and Chen for their suggestion that further investigation should substantiate the effect of SHS on both sexes, examine SHS exposure outside households, and explore the moderation effect of smoking cessation on SHS and cognitive function over the life course (1, 5). Such studies would provide a more comprehensive understanding of the impact of SHS on cognitive impairment, to guide educational and regulatory interventions to reduce the prevalence of dementia globally. ACKNOWLEDGMENTS Author affiliations: Department of Sociology, Texas State University, San Marcos, Texas (Xi Pan); Department of Sociology, Anthropology, and Criminal Justice, Clemson University, Clemson, South Carolina (Ye Luo); and Department of Family Science and Social Work, Miami University, Oxford, Ohio (Amy Restorick Roberts). Conflict of interest: none declared. REFERENCES 1 Anstey KJ, Chen R. Invited commentary: secondhand smoke—an underrecognized risk for cognitive decline. Am J Epidemiol . 2018; 187( 5): 919– 921. 2 Pan X, Luo Y, Roberts AR. Secondhand smoke and women’s cognitive function in China. Am J Epidemiol . 2018; 187( 5): 911– 918. 3 Chen R, Clifford A, Lang L, et al.  . Is exposure to secondhand smoke associated with cognitive parameters of children and adolescents?—a systematic literature review. Ann Epidemiol . 2013; 23( 10): 652– 661. Google Scholar CrossRef Search ADS PubMed  4 Mitchell AJ. The Mini-Mental State Examination (MMSE): update on its diagnostic accuracy and clinical utility for cognitive disorders. In: Larner AJ, ed. Cognitive Screening Instruments: A Practical Approach . Cham, Switzerland: Springer; 2017: 37– 48. Google Scholar CrossRef Search ADS   5 Almeida OP, Garrido GJ, Alfonso H, et al.  . 24-month effect of smoking cessation on cognitive function and brain structure in later life. Neuroimage . 2011; 55( 4): 1480– 1489. Google Scholar CrossRef Search ADS PubMed  © The Author(s) 2018. Published by Oxford University Press on behalf of the Johns Hopkins Bloomberg School of Public Health. All rights reserved. For permissions, please e-mail: [email protected]. This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/about_us/legal/notices)

Bécares, Laia; Zhang, Nan

doi: 10.1093/aje/kwx326pmid: 29036550

Abstract Experiencing discrimination is associated with poor mental health, but how cumulative experiences of perceived interpersonal discrimination across attributes, domains, and time are associated with mental disorders is still unknown. Using data from the Study of Women’s Health Across the Nation (1996–2008), we applied latent class analysis and generalized linear models to estimate the association between cumulative exposure to perceived interpersonal discrimination and older women’s mental health. We found 4 classes of perceived interpersonal discrimination, ranging from cumulative exposure to discrimination over attributes, domains, and time to none or minimal reports of discrimination. Women who experienced cumulative perceived interpersonal discrimination over time and across attributes and domains had the highest risk of depression (Center for Epidemiologic Studies Depression Scale score ≥16) compared with women in all other classes. This was true for all women regardless of race/ethnicity, although the type and severity of perceived discrimination differed across racial/ethnic groups. Cumulative exposure to perceived interpersonal discrimination across attributes, domains, and time has an incremental negative long-term association with mental health. Studies that examine exposure to perceived discrimination due to a single attribute in 1 domain or at 1 point in time underestimate the magnitude and complexity of discrimination and its association with health. cumulative disadvantage, discrimination, mental health, race/ethnicity, women Discrimination—the treatment of someone less favorably than another person because of an identity group characteristic—has been shown to have a negative association with mental health (1–3). The majority of studies assessing whether discrimination leads to mental disorders have focused on examining how general mistreatment or race-based discrimination is adversely associated with mental health (4), although some other attributes, such as weight (5, 6), age (7, 8), sex (9), and sexual orientation (10, 11), have also received attention. Regardless of the attribution of perceived discrimination, most studies to date have examined how isolated attributes are associated with mental health. Identifying single sets of health determinants, such as unique forms and attributions of perceived discrimination, strips away the context of people’s lives and disregards the fact that individuals often embody more than 1 socially disadvantaged status and that these statuses and multiple experiences of discrimination interact to shape people’s health and life chances (12). Theories of intersectionality argue that multiple marginalizations are interlinked and operate simultaneously (13, 14) and can therefore not be understood by theoretical and empirical approaches that treat each marginalized identity as an independent subject of inquiry. Instead, examinations of the intersections between social identities (including race/ethnicity, sex, class, sexuality, indigeneity, and disability/ability, among others) and forms of systemic oppression (including racism, classism, sexism, ableism, and homophobia) provide a more nuanced understanding of the social processes that generate and reproduce poor mental health and health inequities. Excluding some exceptions (10, 15–27), the association between experiencing multiple forms of perceived discrimination due to intersecting social identities and mental health has rarely been explored. The few studies that have examined multiple types of discrimination have shown, for the most part, that experiencing numerous forms of discrimination has a greater negative association with mental health than experiencing discrimination due to 1 attribute only (10, 15–18, 20, 21). Although these studies bridge an important gap in the knowledge base about the mental health implications of experiencing discrimination due to multiple attributes, there are several important limitations to this work. For example, to our knowledge, none of these studies have examined how, in addition to the accumulation of attributions of discrimination, the accumulated experiences of discrimination over time and across multiple domains combine to affect mental health. This is important because examining single attributes, during a cross-section of people’s lifetimes or averaging across years, and exploring discrimination in 1 domain only fails to adequately capture marginalized people’s lived experiences and underestimates the harmful effect of discrimination on health. Studies of racial discrimination show a clear dose-response relationship between increasing number of domains of discrimination experienced and incremental worsening of health (28–31). The information these studies provide is vital to our understanding of the extensiveness of discrimination in people’s lives, although these dose-response studies have not taken into account how this pervasiveness is associated with health over time. In a recent study, Wallace et al. (32) explored the role of cumulative discrimination in health across time and domains, reporting that the accumulation of experienced racial discrimination over time and across several domains has an incremental negative long-term association with psychological distress. That study provided novel information on the importance of considering cumulative experiences of discrimination over time and domains, but it examined only racial discrimination and disregarded the accumulation of discrimination due to multiple attributes. We aimed to combine these 2 bodies of literature (the accumulation of experienced discrimination across attributes and across time and domains) to examine the cumulative association with mental health of the multiple forms of oppression that individuals experience over time. We did this by examining the associations that intersecting experiences of perceived interpersonal discrimination across domains, attributes, and time had with mental health in a multiethnic cohort of older women in the United States. METHODS Data and measures For this analysis, we used publicly available data from the Study of Women’s Health Across the Nation (SWAN), a community-based, multisite, longitudinal study of the menopausal transition. The study design and recruitment for SWAN have been described elsewhere (33). Briefly, SWAN recruited 3,302 women drawn from 7 cities across the United States. All sites enrolled non-Hispanic white participants, and each site also enrolled women of either African-American, Japanese-American, Chinese-American, or Hispanic racial/ethnic background. At study entry, participants were between the ages of 42 and 52 years, self-identified as a member of one of the designated racial/ethnic groups or as white, reported recent menses (<3 months prior to enrollment), were either premenopausal or early perimenopausal, and were not using hormone replacement therapy. Study questionnaires were translated into Cantonese, Japanese, and Spanish. Institutional review board approval was granted, and informed consent was obtained from each study participant. In this study, we used complete data from women who participated in SWAN from the baseline interview through wave 10 (1996–2008; n = 1,613) and belonged to the black or African-American (n = 411), Chinese-American (n = 183), Japanese-American (n = 182), or non-Hispanic white (n = 837) racial/ethnic group. Hispanic participants were excluded because of small sample sizes. Mental health Mental health was measured using the 20-item Center for Epidemiologic Studies Depression Scale (34), a measure of depressive symptomatology. Respondents were asked how often in the past week they had experienced several symptoms, including feeling depressed or feeling like everything was an effort. Response categories ranged from 0 (rarely/none of the time) to 3 (most/all of the time). Positive statements were reversed so that higher scores reflected more depressive symptoms. Center for Epidemiologic Studies Depression Scale score was modeled as a dichotomous variable following the threshold of 16 (score ≥16 = depressed), and depression was measured at baseline and wave 10. Experiences of perceived interpersonal discrimination Participants completed an adapted 10-item version of the Everyday Discrimination Scale (35), which uses a 4-point scale (1 = often, 4 = never) to assess the frequency of experiences of perceived interpersonal discrimination in respondents’ day-to-day lives. Each item starts with the following question: “In your day-to-day life, have you had the following experiences?” Example response items include “You are treated with less courtesy than other people” and “You receive poorer service than other people at restaurants or stores.” The original 9-item measure was modified in the SWAN study protocol to include one additional item, “People ignore you or act as if you are not there.” Each of the 10 domains of perceived interpersonal discrimination was dichotomized into 0 (never) or 1 (rarely, sometimes, or often). The Everyday Discrimination Scale provides options for indicating attributions of perceived interpersonal discrimination, which include race, age, sex, physical appearance, ethnicity, income level, sexual orientation, language, and other. Due to problems with small cell sizes, we combined responses regarding the attributes of race and ethnicity into “race/ethnicity” and responses regarding age, physical appearance, income level, sexual orientation, language, and other attributes into “other.” Sex was left as a single attribute. Responses to the Everyday Discrimination Scale were reverse-scored so that higher numbers indicated greater unfair treatment. Perceived interpersonal discrimination across domains and attributes was assessed over 6 waves of SWAN (baseline and waves 1, 2, 3, 7, and 10). Covariates Factors thought to be associated with both experiences of perceived interpersonal discrimination and mental health were considered in analytical models. These included age (years; continuous), marital status (single, never married; married; or separated, divorced, or widowed), nativity (born in the United States or born abroad), occupation (professional, nonmanual worker, skilled manual worker, or semiskilled/unskilled manual worker), and education (less than high school, high school graduate, some college, college graduate, or postgraduate education). Covariates were assessed at baseline. Analysis plan The accumulation of perceived interpersonal discrimination across attributes, domains, and time was captured with longitudinal latent class analysis, a person-centered approach that probabilistically assigns individuals to latent classes based upon similar patterns of observed longitudinal data. Latent class analysis was first used to evaluate the fit of a 2-class model, and we systematically increased the number of classes in subsequent models until the addition of latent classes did not further improve model fit. For each model, replication of the best log-likelihood was verified to avoid local maxima. To determine the optimal number of classes, we compared models across several model fit criteria. First, we evaluated the sample-size-adjusted Bayesian Information Criterion (36); lower relative Bayesian Information Criterion values indicate improved model fit. Given that the Bayesian Information Criterion tends to favor models with fewer latent classes (37), the Vuong-Lo-Mendell-Rubin likelihood ratio test statistic (38) was also considered. The Vuong-Lo-Mendell-Rubin likelihood ratio test statistic can be used in mixture modeling to compare the fit of the specified class solution (k-class model) with that of a model with fewer classes (k − 1 class model). A nonsignificant χ2 value suggests that a model with 1 fewer class is preferred. Entropy statistics, which measure the separation of the classes based on the posterior class membership probabilities, were also examined; entropy values approaching 1 indicate clear separation between classes (39). After identifying latent subgroups and assigning subjects to classes based on probability of membership, we used generalized linear models to examine the ways in which experiencing various forms of perceived interpersonal discrimination, experiencing discrimination over time, and experiencing discrimination across multiple domains could place women at risk for depression. Generalized linear models were fitted using the “modified Poisson” approach suggested by Zou (40), which provides relative risks and confidence intervals using robust error variances. Generalized linear models were fitted for all women combined and for each racial/ethnic group separately, to assess how classes of perceived interpersonal discrimination and depressive symptomatology were associated differently across racial/ethnic groups. All models adjusted for marital status, age, nativity, education, occupation, and mental health scores at baseline. Models that assessed the association between perceived interpersonal discrimination and depressive symptomatology for all women combined also adjusted for race/ethnicity. Statistical analyses were conducted using MPlus, version 7 (41), and Stata, version 13 (42) (StataCorp LP, College Station, Texas). RESULTS Four distinct classes of perceived interpersonal discrimination were identified in the latent class analyses (see Table 1). Class characteristics are shown in Web Table 1 (available at https://academic.oup.com/aje). The largest proportion of the sample (34%; class 3: accumulation of several domains over time; attribution due to sex and other reasons) experienced the accumulation of several domains of perceived interpersonal discrimination over time (namely being treated with less courtesy or respect; receiving poorer service; people acting as if the respondent was not smart or as if they were better than the respondent; and being ignored) and attributed their experiences of perceived interpersonal discrimination mainly to sex and other attributes. The second largest class (28%; class 4: accumulation of some domains over time; attribution due to other reasons; reduction over time) experienced accumulation of perceived interpersonal discrimination across some domains (being treated with less courtesy or respect and people acting as if they were better than the respondent), although experiences diminished over time. Attributions in class 4 were to reasons other than race/ethnicity or sex. Class 1 (21%; accumulation of perceived interpersonal discrimination over time, domains, and attributes) captured participants who had experienced the highest accumulation of perceived interpersonal discrimination over time, domains, and attributes. Finally, class 2 (17%; no experiences of perceived interpersonal discrimination) included participants who reported having no experiences or very minimal experiences of perceived interpersonal discrimination across any of the 6 time points. Table 1. Indices of the Fit of Classes of Perceived Interpersonal Discrimination Identified in Latent Class Analysis, Study of Women’s Health Across the Nation, 1996–2008 No. of Classes . Sample-Size-Adjusted BIC . Entropy . Log-Likelihood . Vuong-Lo-Mendell-Rubin Likelihood Ratio Test (Δ) . P for Δa . 2 184,538.216 0.962 −91,897.863 3 175,636.544 0.941 −87,260.206 9,260.412 0.0000 4 172,540.157 0.920 −85,525.192 3,464.453 0.0000 5 171,291.239 0.903 −84,713.913 1,619.952 0.4688 No. of Classes . Sample-Size-Adjusted BIC . Entropy . Log-Likelihood . Vuong-Lo-Mendell-Rubin Likelihood Ratio Test (Δ) . P for Δa . 2 184,538.216 0.962 −91,897.863 3 175,636.544 0.941 −87,260.206 9,260.412 0.0000 4 172,540.157 0.920 −85,525.192 3,464.453 0.0000 5 171,291.239 0.903 −84,713.913 1,619.952 0.4688 Abbreviation: BIC, Bayesian Information Criterion. aP value for the likelihood ratio test. Open in new tab Table 1. Indices of the Fit of Classes of Perceived Interpersonal Discrimination Identified in Latent Class Analysis, Study of Women’s Health Across the Nation, 1996–2008 No. of Classes . Sample-Size-Adjusted BIC . Entropy . Log-Likelihood . Vuong-Lo-Mendell-Rubin Likelihood Ratio Test (Δ) . P for Δa . 2 184,538.216 0.962 −91,897.863 3 175,636.544 0.941 −87,260.206 9,260.412 0.0000 4 172,540.157 0.920 −85,525.192 3,464.453 0.0000 5 171,291.239 0.903 −84,713.913 1,619.952 0.4688 No. of Classes . Sample-Size-Adjusted BIC . Entropy . Log-Likelihood . Vuong-Lo-Mendell-Rubin Likelihood Ratio Test (Δ) . P for Δa . 2 184,538.216 0.962 −91,897.863 3 175,636.544 0.941 −87,260.206 9,260.412 0.0000 4 172,540.157 0.920 −85,525.192 3,464.453 0.0000 5 171,291.239 0.903 −84,713.913 1,619.952 0.4688 Abbreviation: BIC, Bayesian Information Criterion. aP value for the likelihood ratio test. Open in new tab Table 2 shows the distribution of sociodemographic characteristics and outcomes of women in the SWAN study across the 4 distinct classes of perceived interpersonal discrimination. Class 1 had the highest proportions of African-American (41%) and Chinese-American (18%) women compared with any other class, whereas class 2 had the highest proportions of non-Hispanic white (64%) and Japanese-American (19%) women. The distributions of educational qualifications were similar across classes. Class 1 had the highest proportions of single and divorced women (17% and 23%, respectively), and married women were overrepresented in class 2 (76%). Class 1 had the highest mental health score at baseline and at wave 10, whereas class 2 had the lowest mental health score at both time points. Table 2. Sociodemographic Characteristics (%) of the Study Sample According to Latent Class of Perceived Interpersonal Discrimination, Study of Women’s Health Across the Nation, 1996–2008 Characteristic . Class . Class 1a (n = 366) . Class 2b (n = 203) . Class 3c (n = 561) . Class 4d (n = 483) . Age, yearse 45.9 (2.6) 46.2 (2.8) 45.8 (2.6) 45.8 (2.7) Race/ethnicity  Black or African-American 40.7 10.3 25.7 20.1  Chinese-American 17.8 5.9 12.5 7.5  Japanese-American 9.8 19.2 9.8 10.8  Non-Hispanic white 31.7 64.5 52.1 61.7 Education  Less than high school 2.7 1.5 2.9 2.5  High school graduate 12.6 13.8 12.7 14.9  Some college/technical school 33.1 30.5 31.7 32.7  College graduate 22.7 23.7 24.8 20.5  Postgraduate education 29.0 30.5 28.0 29.4 Marital status  Single or never married 16.9 9.4 13.6 14.3  Married 60.1 75.9 67.7 68.7  Separated, widowed, or divorced 23.0 14.8 18.7 17.0 Nativity  Born abroad 15.6 24.6 15.9 15.7  Born in the United States 84.4 75.4 84.1 84.3 Occupation  Professional 56.8 58.6 60.4 58.8  Nonmanual worker 6.0 7.4 5.4 6.6  Skilled manual worker 23.0 21.2 17.8 19.9  Semiskilled or unskilled manual worker 14.2 12.8 16.4 14.7 CES-D score at baselinee 0.61 (0.49) 0.33 (0.47) 0.51 (0.50) 0.46 (0.50) CES-D score at wave 10e 0.57 (0.50) 0.25 (0.43) 0.46 (0.50) 0.36 (0.48) Characteristic . Class . Class 1a (n = 366) . Class 2b (n = 203) . Class 3c (n = 561) . Class 4d (n = 483) . Age, yearse 45.9 (2.6) 46.2 (2.8) 45.8 (2.6) 45.8 (2.7) Race/ethnicity  Black or African-American 40.7 10.3 25.7 20.1  Chinese-American 17.8 5.9 12.5 7.5  Japanese-American 9.8 19.2 9.8 10.8  Non-Hispanic white 31.7 64.5 52.1 61.7 Education  Less than high school 2.7 1.5 2.9 2.5  High school graduate 12.6 13.8 12.7 14.9  Some college/technical school 33.1 30.5 31.7 32.7  College graduate 22.7 23.7 24.8 20.5  Postgraduate education 29.0 30.5 28.0 29.4 Marital status  Single or never married 16.9 9.4 13.6 14.3  Married 60.1 75.9 67.7 68.7  Separated, widowed, or divorced 23.0 14.8 18.7 17.0 Nativity  Born abroad 15.6 24.6 15.9 15.7  Born in the United States 84.4 75.4 84.1 84.3 Occupation  Professional 56.8 58.6 60.4 58.8  Nonmanual worker 6.0 7.4 5.4 6.6  Skilled manual worker 23.0 21.2 17.8 19.9  Semiskilled or unskilled manual worker 14.2 12.8 16.4 14.7 CES-D score at baselinee 0.61 (0.49) 0.33 (0.47) 0.51 (0.50) 0.46 (0.50) CES-D score at wave 10e 0.57 (0.50) 0.25 (0.43) 0.46 (0.50) 0.36 (0.48) Abbreviation: CES-D, Center for Epidemiologic Studies Depression Scale. a Accumulation of perceived discrimination over time, domains, and attributes. b No experiences of perceived interpersonal discrimination. c Accumulation of several domains over time; attribution due to sex and other reasons. d Accumulation of some domains over time; attribution due to other reasons; reduction over time. e Values are expressed as mean (standard deviation). Open in new tab Table 2. Sociodemographic Characteristics (%) of the Study Sample According to Latent Class of Perceived Interpersonal Discrimination, Study of Women’s Health Across the Nation, 1996–2008 Characteristic . Class . Class 1a (n = 366) . Class 2b (n = 203) . Class 3c (n = 561) . Class 4d (n = 483) . Age, yearse 45.9 (2.6) 46.2 (2.8) 45.8 (2.6) 45.8 (2.7) Race/ethnicity  Black or African-American 40.7 10.3 25.7 20.1  Chinese-American 17.8 5.9 12.5 7.5  Japanese-American 9.8 19.2 9.8 10.8  Non-Hispanic white 31.7 64.5 52.1 61.7 Education  Less than high school 2.7 1.5 2.9 2.5  High school graduate 12.6 13.8 12.7 14.9  Some college/technical school 33.1 30.5 31.7 32.7  College graduate 22.7 23.7 24.8 20.5  Postgraduate education 29.0 30.5 28.0 29.4 Marital status  Single or never married 16.9 9.4 13.6 14.3  Married 60.1 75.9 67.7 68.7  Separated, widowed, or divorced 23.0 14.8 18.7 17.0 Nativity  Born abroad 15.6 24.6 15.9 15.7  Born in the United States 84.4 75.4 84.1 84.3 Occupation  Professional 56.8 58.6 60.4 58.8  Nonmanual worker 6.0 7.4 5.4 6.6  Skilled manual worker 23.0 21.2 17.8 19.9  Semiskilled or unskilled manual worker 14.2 12.8 16.4 14.7 CES-D score at baselinee 0.61 (0.49) 0.33 (0.47) 0.51 (0.50) 0.46 (0.50) CES-D score at wave 10e 0.57 (0.50) 0.25 (0.43) 0.46 (0.50) 0.36 (0.48) Characteristic . Class . Class 1a (n = 366) . Class 2b (n = 203) . Class 3c (n = 561) . Class 4d (n = 483) . Age, yearse 45.9 (2.6) 46.2 (2.8) 45.8 (2.6) 45.8 (2.7) Race/ethnicity  Black or African-American 40.7 10.3 25.7 20.1  Chinese-American 17.8 5.9 12.5 7.5  Japanese-American 9.8 19.2 9.8 10.8  Non-Hispanic white 31.7 64.5 52.1 61.7 Education  Less than high school 2.7 1.5 2.9 2.5  High school graduate 12.6 13.8 12.7 14.9  Some college/technical school 33.1 30.5 31.7 32.7  College graduate 22.7 23.7 24.8 20.5  Postgraduate education 29.0 30.5 28.0 29.4 Marital status  Single or never married 16.9 9.4 13.6 14.3  Married 60.1 75.9 67.7 68.7  Separated, widowed, or divorced 23.0 14.8 18.7 17.0 Nativity  Born abroad 15.6 24.6 15.9 15.7  Born in the United States 84.4 75.4 84.1 84.3 Occupation  Professional 56.8 58.6 60.4 58.8  Nonmanual worker 6.0 7.4 5.4 6.6  Skilled manual worker 23.0 21.2 17.8 19.9  Semiskilled or unskilled manual worker 14.2 12.8 16.4 14.7 CES-D score at baselinee 0.61 (0.49) 0.33 (0.47) 0.51 (0.50) 0.46 (0.50) CES-D score at wave 10e 0.57 (0.50) 0.25 (0.43) 0.46 (0.50) 0.36 (0.48) Abbreviation: CES-D, Center for Epidemiologic Studies Depression Scale. a Accumulation of perceived discrimination over time, domains, and attributes. b No experiences of perceived interpersonal discrimination. c Accumulation of several domains over time; attribution due to sex and other reasons. d Accumulation of some domains over time; attribution due to other reasons; reduction over time. e Values are expressed as mean (standard deviation). Open in new tab Table 3 presents the descriptors of latent class membership according to relevant covariates. Older women were less likely than younger women to be in class 4 versus class 2 (odds ratio = 0.75, 95% confidence interval (CI): 0.60, 0.95). Odds of membership in any of the classes that captured different types of cumulative perceived interpersonal discrimination (class 1, 3, or 4) were much higher for African-American and Chinese-American women than for non-Hispanic white women. In the case of membership in class 1 compared with class 2, odds of membership for African-American and Chinese-American women were about 6 times those of non-Hispanic white women (see Table 3). Women with a high level of education and women born in the United States were more likely than their counterparts to be in class 1, 3, or 4 than in class 2. Women who were married were less likely than single women to be in class 1, 3, or 4 than in class 2, and so were women with semiskilled and unskilled occupations, compared with professional women. Table 3. Odds Ratios for Membership in a Latent Class Involving Perceived Interpersonal Discrimination as Compared With Class 2 (No Experiences of Perceived Discrimination), According to Sociodemographic Characteristics Relevant to Attributions of Perceived Discrimination, Study of Women’s Health Across the Nation, 1996–2008 Characteristic . Latent Class Comparison . Class 1a vs. Class 2b . Class 3c vs. Class 2 . Class 4d vs. Class 2 . OR . 95% CI . OR . 95% CI . OR . 95% CI . Age, years  40–45 1.00 Referent 1.00 Referent 1.00 Referent  ≥46 0.82 0.64, 1.05 0.86 0.69, 1.07 0.75 0.60, 0.95 Race/ethnicity  Non-Hispanic white 1.00 Referent 1.00 Referent 1.00 Referent  Black or African-American 6.29 4.37, 9.05 2.38 1.69, 3.36 1.48 1.03, 2.12  Chinese-American 5.87 3.31, 10.41 2.65 1.53, 4.61 1.39 0.77, 2.49  Japanese-American 0.87 0.57, 1.32 0.49 0.34, 0.71 0.49 0.34, 0.71 Education  Less than high school 1.00 Referent 1.00 Referent 1.00 Referent  High school graduate 4.74 2.44, 9.21 3.25 2.02, 5.24 2.55 1.60, 4.07  Some college/technical school 7.61 4.05, 14.28 3.96 2.53, 6.20 3.38 2.19, 5.22  College graduate 8.58 4.47, 16.46 4.98 3.10, 7.99 3.11 1.95, 4.97  Postgraduate education 8.37 4.38, 15.99 5.36 3.36, 8.55 3.60 2.27, 5.68 Marital status  Singe or never married 1.00 Referent 1.00 Referent 1.00 Referent  Married 0.36 0.24, 0.53 0.60 0.41, 0.88 0.59 0.40, 0.87  Separated, widowed, or divorced 0.71 0.44, 1.13 0.86 0.55, 1.36 0.72 0.45, 1.15 Nativity  Born abroad 1.00 Referent 1.00 Referent 1.00 Referent  Born in the United States 4.72 3.48, 6.39 4.42 3.41, 5.72 3.09 2.40, 3.99 Occupation  Professional 1.00 Referent 1.00 Referent 1.00 Referent  Nonmanual worker 0.64 0.37, 1.12 0.58 0.35, 0.95 0.70 0.42, 1.16  Skilled manual worker 1.07 0.75, 1.53 0.91 0.65, 1.27 0.95 0.68, 1.33  Semiskilled or unskilled worker 0.55 0.38, 0.80 0.61 0.44, 0.84 0.58 0.41, 0.81 Characteristic . Latent Class Comparison . Class 1a vs. Class 2b . Class 3c vs. Class 2 . Class 4d vs. Class 2 . OR . 95% CI . OR . 95% CI . OR . 95% CI . Age, years  40–45 1.00 Referent 1.00 Referent 1.00 Referent  ≥46 0.82 0.64, 1.05 0.86 0.69, 1.07 0.75 0.60, 0.95 Race/ethnicity  Non-Hispanic white 1.00 Referent 1.00 Referent 1.00 Referent  Black or African-American 6.29 4.37, 9.05 2.38 1.69, 3.36 1.48 1.03, 2.12  Chinese-American 5.87 3.31, 10.41 2.65 1.53, 4.61 1.39 0.77, 2.49  Japanese-American 0.87 0.57, 1.32 0.49 0.34, 0.71 0.49 0.34, 0.71 Education  Less than high school 1.00 Referent 1.00 Referent 1.00 Referent  High school graduate 4.74 2.44, 9.21 3.25 2.02, 5.24 2.55 1.60, 4.07  Some college/technical school 7.61 4.05, 14.28 3.96 2.53, 6.20 3.38 2.19, 5.22  College graduate 8.58 4.47, 16.46 4.98 3.10, 7.99 3.11 1.95, 4.97  Postgraduate education 8.37 4.38, 15.99 5.36 3.36, 8.55 3.60 2.27, 5.68 Marital status  Singe or never married 1.00 Referent 1.00 Referent 1.00 Referent  Married 0.36 0.24, 0.53 0.60 0.41, 0.88 0.59 0.40, 0.87  Separated, widowed, or divorced 0.71 0.44, 1.13 0.86 0.55, 1.36 0.72 0.45, 1.15 Nativity  Born abroad 1.00 Referent 1.00 Referent 1.00 Referent  Born in the United States 4.72 3.48, 6.39 4.42 3.41, 5.72 3.09 2.40, 3.99 Occupation  Professional 1.00 Referent 1.00 Referent 1.00 Referent  Nonmanual worker 0.64 0.37, 1.12 0.58 0.35, 0.95 0.70 0.42, 1.16  Skilled manual worker 1.07 0.75, 1.53 0.91 0.65, 1.27 0.95 0.68, 1.33  Semiskilled or unskilled worker 0.55 0.38, 0.80 0.61 0.44, 0.84 0.58 0.41, 0.81 Abbreviations: CI, confidence interval; OR, odds ratio. a Accumulation of perceived discrimination over time, domains, and attributes. b No experiences of perceived interpersonal discrimination. c Accumulation of several domains over time; attribution due to sex and other reasons. d Accumulation of some domains over time; attribution due to other reasons; reduction over time. Open in new tab Table 3. Odds Ratios for Membership in a Latent Class Involving Perceived Interpersonal Discrimination as Compared With Class 2 (No Experiences of Perceived Discrimination), According to Sociodemographic Characteristics Relevant to Attributions of Perceived Discrimination, Study of Women’s Health Across the Nation, 1996–2008 Characteristic . Latent Class Comparison . Class 1a vs. Class 2b . Class 3c vs. Class 2 . Class 4d vs. Class 2 . OR . 95% CI . OR . 95% CI . OR . 95% CI . Age, years  40–45 1.00 Referent 1.00 Referent 1.00 Referent  ≥46 0.82 0.64, 1.05 0.86 0.69, 1.07 0.75 0.60, 0.95 Race/ethnicity  Non-Hispanic white 1.00 Referent 1.00 Referent 1.00 Referent  Black or African-American 6.29 4.37, 9.05 2.38 1.69, 3.36 1.48 1.03, 2.12  Chinese-American 5.87 3.31, 10.41 2.65 1.53, 4.61 1.39 0.77, 2.49  Japanese-American 0.87 0.57, 1.32 0.49 0.34, 0.71 0.49 0.34, 0.71 Education  Less than high school 1.00 Referent 1.00 Referent 1.00 Referent  High school graduate 4.74 2.44, 9.21 3.25 2.02, 5.24 2.55 1.60, 4.07  Some college/technical school 7.61 4.05, 14.28 3.96 2.53, 6.20 3.38 2.19, 5.22  College graduate 8.58 4.47, 16.46 4.98 3.10, 7.99 3.11 1.95, 4.97  Postgraduate education 8.37 4.38, 15.99 5.36 3.36, 8.55 3.60 2.27, 5.68 Marital status  Singe or never married 1.00 Referent 1.00 Referent 1.00 Referent  Married 0.36 0.24, 0.53 0.60 0.41, 0.88 0.59 0.40, 0.87  Separated, widowed, or divorced 0.71 0.44, 1.13 0.86 0.55, 1.36 0.72 0.45, 1.15 Nativity  Born abroad 1.00 Referent 1.00 Referent 1.00 Referent  Born in the United States 4.72 3.48, 6.39 4.42 3.41, 5.72 3.09 2.40, 3.99 Occupation  Professional 1.00 Referent 1.00 Referent 1.00 Referent  Nonmanual worker 0.64 0.37, 1.12 0.58 0.35, 0.95 0.70 0.42, 1.16  Skilled manual worker 1.07 0.75, 1.53 0.91 0.65, 1.27 0.95 0.68, 1.33  Semiskilled or unskilled worker 0.55 0.38, 0.80 0.61 0.44, 0.84 0.58 0.41, 0.81 Characteristic . Latent Class Comparison . Class 1a vs. Class 2b . Class 3c vs. Class 2 . Class 4d vs. Class 2 . OR . 95% CI . OR . 95% CI . OR . 95% CI . Age, years  40–45 1.00 Referent 1.00 Referent 1.00 Referent  ≥46 0.82 0.64, 1.05 0.86 0.69, 1.07 0.75 0.60, 0.95 Race/ethnicity  Non-Hispanic white 1.00 Referent 1.00 Referent 1.00 Referent  Black or African-American 6.29 4.37, 9.05 2.38 1.69, 3.36 1.48 1.03, 2.12  Chinese-American 5.87 3.31, 10.41 2.65 1.53, 4.61 1.39 0.77, 2.49  Japanese-American 0.87 0.57, 1.32 0.49 0.34, 0.71 0.49 0.34, 0.71 Education  Less than high school 1.00 Referent 1.00 Referent 1.00 Referent  High school graduate 4.74 2.44, 9.21 3.25 2.02, 5.24 2.55 1.60, 4.07  Some college/technical school 7.61 4.05, 14.28 3.96 2.53, 6.20 3.38 2.19, 5.22  College graduate 8.58 4.47, 16.46 4.98 3.10, 7.99 3.11 1.95, 4.97  Postgraduate education 8.37 4.38, 15.99 5.36 3.36, 8.55 3.60 2.27, 5.68 Marital status  Singe or never married 1.00 Referent 1.00 Referent 1.00 Referent  Married 0.36 0.24, 0.53 0.60 0.41, 0.88 0.59 0.40, 0.87  Separated, widowed, or divorced 0.71 0.44, 1.13 0.86 0.55, 1.36 0.72 0.45, 1.15 Nativity  Born abroad 1.00 Referent 1.00 Referent 1.00 Referent  Born in the United States 4.72 3.48, 6.39 4.42 3.41, 5.72 3.09 2.40, 3.99 Occupation  Professional 1.00 Referent 1.00 Referent 1.00 Referent  Nonmanual worker 0.64 0.37, 1.12 0.58 0.35, 0.95 0.70 0.42, 1.16  Skilled manual worker 1.07 0.75, 1.53 0.91 0.65, 1.27 0.95 0.68, 1.33  Semiskilled or unskilled worker 0.55 0.38, 0.80 0.61 0.44, 0.84 0.58 0.41, 0.81 Abbreviations: CI, confidence interval; OR, odds ratio. a Accumulation of perceived discrimination over time, domains, and attributes. b No experiences of perceived interpersonal discrimination. c Accumulation of several domains over time; attribution due to sex and other reasons. d Accumulation of some domains over time; attribution due to other reasons; reduction over time. Open in new tab Compared with women who experienced the highest accumulation of perceived interpersonal discrimination (class 1), women in all other classes tended to be less likely to report depressive symptomatology (see Table 4). This was true across all racial/ethnic groups. For example, compared with African-American women in class 1, African-American women in classes 2 and 4 had 0.46 and 0.65 times the risk, respectively, of reporting depression (Table 4). Chinese-American women in classes 2, 3, and 4, who experienced less perceived interpersonal discrimination over time, domains, and attributes, were all less likely to report depression than Chinese-American women in class 1. Japanese-American women in class 3 (accumulation of several domains over time; attribution due to sex and other reasons) had 0.65 times the risk of depression, compared with women in class 1 (incidence rate ratio = 0.65, 95% CI: 0.42, 1.00). Non-Hispanic white women in class 2 or class 4 were 0.52 and 0.69 times as likely, respectively, to report depression compared with women in class 1 (see Table 4). When pooled together, women of all racial/ethnic groups in classes 2, 3, and 4 had a lower risk of depression than women in class 1. This was particularly strong for women who reported the lowest levels of perceived interpersonal discrimination (class 2), who had 0.46 times the risk of depression compared with women with the highest levels of cumulative experiences of perceived interpersonal discrimination, in class 1 (incidence rate ratio = 0.46, 95% CI: 0.35, 0.59). Table 4. Association Between Latent Class of Perceived Interpersonal Discrimination and Depressiona, Study of Women’s Health Across the Nation, 1996–2008 Class . Race/Ethnicity . Black or African-American (n = 411) . Chinese-American (n = 183) . Japanese-American (n = 182) . Non-Hispanic White (n = 837) . All Women (n = 1,613) . IRRb . 95% CI . IRRb . 95% CI . IRRb . 95% CI . IRRb . 95% CI . IRRc . 95% CI . Class 1d 1.00 Referent 1.00 Referent 1.00 Referent 1.00 Referent 1.00 Referent Class 2e 0.46 0.21, 1.00 1.72e−07 8.43e−08, 3.49e−07 0.55 0.28, 1.09 0.52 0.37, 0.72 0.46 0.35, 0.59 Class 3f 0.81 0.64, 1.04 0.50 0.30, 0.83 0.65 0.42, 1.00 0.96 0.80, 1.16 0.80 0.71, 0.91 Class 4g 0.65 0.47, 0.90 0.56 0.32, 0.98 0.76 0.50, 1.16 0.69 0.56, 0.85 0.64 0.55, 0.74 Class . Race/Ethnicity . Black or African-American (n = 411) . Chinese-American (n = 183) . Japanese-American (n = 182) . Non-Hispanic White (n = 837) . All Women (n = 1,613) . IRRb . 95% CI . IRRb . 95% CI . IRRb . 95% CI . IRRb . 95% CI . IRRc . 95% CI . Class 1d 1.00 Referent 1.00 Referent 1.00 Referent 1.00 Referent 1.00 Referent Class 2e 0.46 0.21, 1.00 1.72e−07 8.43e−08, 3.49e−07 0.55 0.28, 1.09 0.52 0.37, 0.72 0.46 0.35, 0.59 Class 3f 0.81 0.64, 1.04 0.50 0.30, 0.83 0.65 0.42, 1.00 0.96 0.80, 1.16 0.80 0.71, 0.91 Class 4g 0.65 0.47, 0.90 0.56 0.32, 0.98 0.76 0.50, 1.16 0.69 0.56, 0.85 0.64 0.55, 0.74 Abbreviations: CES-D, Center for Epidemiologic Studies Depression Scale; CI, confidence interval; IRR, incidence rate ratio. a Depression was defined as a CES-D score ≥16. b Adjusted for marital status, age, nativity, occupation, education, and CES-D score at baseline. c Adjusted for marital status, age, nativity, occupation, education, CES-D score at baseline, and race/ethnicity. d Accumulation of perceived discrimination over time, domains, and attributes. e No experiences of perceived interpersonal discrimination. f Accumulation of several domains over time; attribution due to sex and other reasons. g Accumulation of some domains over time; attribution due to other reasons; reduction over time. Open in new tab Table 4. Association Between Latent Class of Perceived Interpersonal Discrimination and Depressiona, Study of Women’s Health Across the Nation, 1996–2008 Class . Race/Ethnicity . Black or African-American (n = 411) . Chinese-American (n = 183) . Japanese-American (n = 182) . Non-Hispanic White (n = 837) . All Women (n = 1,613) . IRRb . 95% CI . IRRb . 95% CI . IRRb . 95% CI . IRRb . 95% CI . IRRc . 95% CI . Class 1d 1.00 Referent 1.00 Referent 1.00 Referent 1.00 Referent 1.00 Referent Class 2e 0.46 0.21, 1.00 1.72e−07 8.43e−08, 3.49e−07 0.55 0.28, 1.09 0.52 0.37, 0.72 0.46 0.35, 0.59 Class 3f 0.81 0.64, 1.04 0.50 0.30, 0.83 0.65 0.42, 1.00 0.96 0.80, 1.16 0.80 0.71, 0.91 Class 4g 0.65 0.47, 0.90 0.56 0.32, 0.98 0.76 0.50, 1.16 0.69 0.56, 0.85 0.64 0.55, 0.74 Class . Race/Ethnicity . Black or African-American (n = 411) . Chinese-American (n = 183) . Japanese-American (n = 182) . Non-Hispanic White (n = 837) . All Women (n = 1,613) . IRRb . 95% CI . IRRb . 95% CI . IRRb . 95% CI . IRRb . 95% CI . IRRc . 95% CI . Class 1d 1.00 Referent 1.00 Referent 1.00 Referent 1.00 Referent 1.00 Referent Class 2e 0.46 0.21, 1.00 1.72e−07 8.43e−08, 3.49e−07 0.55 0.28, 1.09 0.52 0.37, 0.72 0.46 0.35, 0.59 Class 3f 0.81 0.64, 1.04 0.50 0.30, 0.83 0.65 0.42, 1.00 0.96 0.80, 1.16 0.80 0.71, 0.91 Class 4g 0.65 0.47, 0.90 0.56 0.32, 0.98 0.76 0.50, 1.16 0.69 0.56, 0.85 0.64 0.55, 0.74 Abbreviations: CES-D, Center for Epidemiologic Studies Depression Scale; CI, confidence interval; IRR, incidence rate ratio. a Depression was defined as a CES-D score ≥16. b Adjusted for marital status, age, nativity, occupation, education, and CES-D score at baseline. c Adjusted for marital status, age, nativity, occupation, education, CES-D score at baseline, and race/ethnicity. d Accumulation of perceived discrimination over time, domains, and attributes. e No experiences of perceived interpersonal discrimination. f Accumulation of several domains over time; attribution due to sex and other reasons. g Accumulation of some domains over time; attribution due to other reasons; reduction over time. Open in new tab DISCUSSION Findings This study aimed to examine the association between cumulative exposure to perceived interpersonal discrimination over time, attributes, and domains and depression among older women. Drawing from intersectionality theory, we explored how multiple marginalizations and oppressions that women embody over time put them at increased risk of depression at older ages. We found 4 distinct classes of perceived interpersonal discrimination, ranging from increased cumulative experiences of perceived interpersonal discrimination over time, across all domains, and across attributes to none or very minimal reports of perceived discrimination. Only a minority of the sample (17%) reported having no experiences or minimal experiences of perceived discrimination, and the large majority of women experienced perceived discrimination that was attributed to multiple social identities. Through the 6 waves of data that collected information on experienced perceived discrimination, only 2 women (out of 1,613 participants who had complete data for all 6 waves) reported experiencing only racial discrimination, and only 1 reported experiencing only sex discrimination. This highlights the need to consider multiple social positions and oppressed identities when reporting the prevalence of perceived interpersonal discrimination and understanding the harmful effect of perceived interpersonal discrimination on health. Assessing 1 sole attribution of discrimination does not accurately represent the lived experiences of marginalized populations, who very often experience multiple forms of discrimination. Likewise, we found that very few participants (14 women) had experienced perceived interpersonal discrimination in 1 domain across waves or at 1 time point only (29 women). Studies that measure a single attribute, 1 domain, or 1 point in time underestimate the frequency and complexity of discrimination and its association with health. Women in class 1 (highest accumulation of perceived interpersonal discrimination) experienced the highest risk of depression compared with women in all other classes, especially compared with women in class 2 (lowest prevalence of perceived interpersonal discrimination). In previous studies, investigators have reported similar findings whereby exposure to multiple forms of discrimination is associated with significantly more depressive symptoms (15, 21), although those studies did not consider multiple domains or time points. Cumulative disadvantage theory and related models such as cumulative advantage and cumulative inequality theory (43) suggest that populations experience health outcomes and trajectories as a result of advantages or disadvantages experienced across the life course. In these analyses of the accumulation of perceived interpersonal discrimination through a period in the later stages of women’s life course, we found clear evidence of the corrosive incremental association that cumulative exposure to disadvantage, in the shape of experienced perceived interpersonal discrimination across attributes, domains, and time, has with mental health. We found that women with higher education, single women, and US-born women had greater odds of membership in class 1 compared with their noneducated, married, and foreign-born counterparts. These sociodemographic patterns in reports of discrimination have been previously reported in the literature (44–46). Patterns in the association between perceived interpersonal discrimination and mental health were similar across racial/ethnic groups, with some minor differences. This indicates that experiencing perceived discrimination is harmful for women regardless of racial/ethnic background, although the type and severity of discrimination (and therefore, the accumulated harm over the life course) differs across racial/ethnic groups. In our sample, African-American and Chinese-American women had the highest likelihood of membership in class 1 compared with non-Hispanic white women. These differences in class membership reflect that the fact that although marginalized groups, such as women, face more discrimination than privileged groups, individuals who belong to multiple stigmatized groups, such as racial/ethnic minority women and/or women with several marginalized identities, face the greatest burden of these experiences (15, 21). Japanese-American women’s mental health did not differ between participants in class 2 (lowest prevalence of perceived interpersonal discrimination) and participants in class 1 (highest accumulation of perceived interpersonal discrimination), although we found significant differences across these classes for the other racial/ethnic groups. Class 2 does not fully capture women who have never experienced any discrimination—most participants in class 2 reported very few experiences of perceived discrimination either due to 1 attribute, in 1 domain, or at 1 time point. Some studies show a J-shaped relationship between racism and health such that people who report experiencing no racism still have poor health (47), perhaps because lack of reporting does not necessarily mean that people have not experienced any discrimination but may mean that they deny these experiences as a self-defense mechanism. It is therefore possible that the lack of differences in risk of mental health found between classes 1 and 2 is due to this J-shaped curve association between perceived discrimination and health. Limitations Although this study was able to take advantage of the longitudinal and multidimensional nature of SWAN data, it was limited in some respects. First, SWAN does not ask respondents about exposure to perceived discrimination over the course of their lives, so we were unable to examine any of the processes or experiences of perceived discrimination prior to their baseline interview. We also did not have any data on vicarious exposure to discrimination or data on internalized systems of oppression, such as internalized racism and sexism, which have been shown to be detrimentally associated with poor health (29, 48). Furthermore, even though we are able to examine experiences across various domains of perceived discrimination, the domains explored do not represent the full range of places and circumstances where discrimination can be experienced. Given these measurement limitations, results presented here may underestimate the prevalence of accumulated discrimination and its association with mental health. This study focused on women only, and although these associations may be similar in men, we are not able to assert that here. In general, women are more likely than men to experience multiple forms of discrimination, since they are part of an oppressed social group and may embody other social identities that subject them to discrimination due to other attributes. For example, the accumulation of racism and sexism that racial/ethnic minority women experience in their lifetimes contributes to high levels of stress and psychiatric symptoms, but the majority of research studies focus on either race-related or sex-related stress and do not capture the ways in which racial/ethnic minority women experience both race- and sex-related stress simultaneously (49); we aimed to achieve that in this study. Applying latent class analysis to SWAN data allowed us to show how additive associations with different attributes of perceived interpersonal discrimination and multiple marginalization, over time and domains, worsen the detrimental association between discrimination and health. However, we were not able to fully examine how the interrelation and interaction between 2 attributes combined into 1 specific form of discrimination, such as gendered racism, is associated with women’s mental health. Future studies employing other statistical techniques may be able to consider in greater detail the embodied positions of multiply discriminated-against individuals and their associations with health. Conclusions This study documents, for the first time, the harm that cumulative experiences of perceived interpersonal discrimination over attributes, domains, and time have on the mental health of ethnic minority women. We found that women of all racial/ethnic minority groups who experience the highest levels of cumulative perceived discrimination are at greater risk of depression than women who experience minimal levels of perceived discrimination that do not accumulate over time, attributes, or domains. Although we found that experiencing perceived interpersonal discrimination is harmful for women regardless of racial/ethnic background, our results show that the type and severity of perceived discrimination differs across racial/ethnic groups, and it is more harmful for racial/ethnic minority women. The findings of this study highlight the need to fully capture the experiences of discrimination of marginalized populations, in order to avoid underestimating the magnitude and complexity of discrimination and its association with health. ACKNOWLEDGMENTS Author affiliations: Cathie Marsh Institute for Social Research, University of Manchester, Manchester, United Kingdom (Laia Bécares, Nan Zhang); and Department of Social Statistics, School of Social Sciences, University of Manchester, Manchester, United Kingdom (Laia Bécares). N.Z. was supported by a United Kingdom Economic and Social Research Council Global Challenge Research Fund Postdoctoral Fellowship (grant ES/P009824/1). The SWAN study was funded by the National Institutes of Health (grant NR004061), the National Institute on Aging (grants AG012495, AG012505, AG012539, AG012546, AG012553, and AG012554), the National Institute of Nursing Research (grant AG012535), and the Office of Research on Women’s Health (grant AG012531). All SWAN public-use data files are available from the website of the University of Michigan’s Inter-University Consortium for Political and Social Research (https://www.icpsr.umich.edu/icpsrweb/NACDA/series/00253). Conflict of interest: none declared. Abbreviations CI confidence interval SWAN Study of Women’s Health Across the Nation REFERENCES 1 Pascoe EA , Smart Richman L. Perceived discrimination and health: a meta-analytic review . Psychol Bull . 2009 ; 135 ( 4 ): 531 – 554 . Google Scholar Crossref Search ADS PubMed WorldCat 2 Schmitt MT , Branscombe NR, Postmes T, et al. . The consequences of perceived discrimination for psychological well-being: a meta-analytic review . Psychol Bull . 2014 ; 140 ( 4 ): 921 – 948 . Google Scholar Crossref Search ADS PubMed WorldCat 3 Williams DR , John DA, Oyserman D, et al. . Research on discrimination and health: an exploratory study of unresolved conceptual and measurement issues . Am J Public Health . 2012 ; 102 ( 5 ): 975 – 978 . Google Scholar Crossref Search ADS PubMed WorldCat 4 Lewis TT , Cogburn CD, Williams DR. Self-reported experiences of discrimination and health: scientific advances, ongoing controversies, and emerging issues . Annu Rev Clin Psychol . 2015 ; 11 : 407 – 440 . Google Scholar Crossref Search ADS PubMed WorldCat 5 Hatzenbuehler ML , Keyes KM, Hasin DS. Associations between perceived weight discrimination and the prevalence of psychiatric disorders in the general population . Obesity (Silver Spring) . 2009 ; 17 ( 11 ): 2033 – 2039 . Google Scholar Crossref Search ADS PubMed WorldCat 6 Spahlholz J , Pabst A, Riedel-Heller S, et al. . Coping with perceived weight discrimination: testing a theoretical model for examining the relationship between perceived weight discrimination and depressive symptoms in a representative sample of individuals with obesity . Int J Obes (Lond) . 2016 ; 40 ( 12 ): 1915 – 1921 . Google Scholar Crossref Search ADS PubMed WorldCat 7 Yuan ASV . Perceived age discrimination and mental health . Soc Forces . 2007 ; 86 ( 1 ): 291 – 311 . Google Scholar Crossref Search ADS WorldCat 8 Han J , Richardson VE. The relationships among perceived discrimination, self-perceptions of aging, and depressive symptoms: a longitudinal examination of age discrimination . Aging Ment Health . 2015 ; 19 ( 8 ): 747 – 755 . Google Scholar Crossref Search ADS PubMed WorldCat 9 Borrell C , Artazcoz L, Gil-González D, et al. . Determinants of perceived sexism and their role on the association of sexism with mental health . Women Health . 2011 ; 51 ( 6 ): 583 – 603 . Google Scholar Crossref Search ADS PubMed WorldCat 10 Choi KH , Paul J, Ayala G, et al. . Experiences of discrimination and their impact on the mental health among African American, Asian and Pacific Islander, and Latino men who have sex with men . Am J Public Health . 2013 ; 103 ( 5 ): 868 – 874 . Google Scholar Crossref Search ADS PubMed WorldCat 11 Mays VM , Cochran SD. Mental health correlates of perceived discrimination among lesbian, gay, and bisexual adults in the United States . Am J Public Health . 2001 ; 91 ( 11 ): 1869 – 1876 . Google Scholar Crossref Search ADS PubMed WorldCat 12 Cole ER . Intersectionality and research in psychology . Am Psychol . 2009 ; 64 ( 3 ): 170 – 180 . Google Scholar Crossref Search ADS PubMed WorldCat 13 Bauer G . Incorporating intersectionality theory into population health research methodology: challenges and the potential to advance health equity . Soc Sci Med . 2014 ; 110 : 10 – 17 . Google Scholar Crossref Search ADS PubMed WorldCat 14 Crenshaw K . Demarginalizing the intersection of race and sex: a black feminist critique of antidiscrimination doctrine, feminist theory and antiracist politics . Univ Chic Leg Forum . 1989 ; 1989 : 139 – 167 . Google Scholar OpenURL Placeholder Text WorldCat 15 Grollman EA . Multiple disadvantaged statuses and health: the role of multiple forms of discrimination . J Health Soc Behav . 2014 ; 55 ( 1 ): 3 – 19 . Google Scholar Crossref Search ADS PubMed WorldCat 16 Stuber J , Galea S, Ahern J, et al. . The association between multiple domains of discrimination and self-assessed health: a multilevel analysis of Latinos and blacks in four low-income New York City neighborhoods . Health Serv Res . 2003 ; 38 ( 6 ): 1735 – 1759 . Google Scholar Crossref Search ADS PubMed WorldCat 17 Seng JS , Lopez WD, Sperlich M, et al. . Marginalized identities, discrimination burden, and mental health: empirical exploration of an interpersonal-level approach to modeling intersectionality . Soc Sci Med . 2012 ; 75 ( 12 ): 1437 – 2446 . Google Scholar Crossref Search ADS WorldCat 18 Gayman M , Barragan J. Multiple perceived reasons for major discrimination and depression . Soc Ment Health . 2013 ; 3 ( 3 ): 203 – 220 . Google Scholar Crossref Search ADS WorldCat 19 Bécares L , Priest N. Understanding the influence of race/ethnicity, gender, and class on inequalities in academic and non-academic outcomes among eighth-grade students: findings from an intersectionality approach . PLoS One . 2015 ; 10 ( 10 ): e0141363 . Google Scholar Crossref Search ADS PubMed WorldCat 20 Garnett BR , Masyn KE, Austin SB, et al. . The intersectionality of discrimination attributes and bullying among youth: an applied latent class analysis . J Youth Adolesc . 2014 ; 43 ( 8 ): 1225 – 1239 . Google Scholar Crossref Search ADS PubMed WorldCat 21 Grollman E . Multiple forms of perceived discrimination and health among adolescents and young adults . J Health Soc Behav . 2012 ; 53 ( 2 ): 199 – 214 . Google Scholar Crossref Search ADS PubMed WorldCat 22 Balsam KF , Molina Y, Beadnell B, et al. . Measuring multiple minority stress: the LGBT People of Color Microaggressions Scale . Cultur Divers Ethnic Minor Psychol . 2011 ; 17 ( 2 ): 163 – 174 . Google Scholar Crossref Search ADS PubMed WorldCat 23 Bastos JL , Barros AJ, Celeste RK, et al. . Age, class and race discrimination: their interactions and associations with mental health among Brazilian university students . Cad Saude Publica . 2014 ; 30 ( 1 ): 175 – 186 . Google Scholar Crossref Search ADS PubMed WorldCat 24 Bucchianeri M , Eisenberg M, Wall M, et al. . Multiple types of harassment: associations with emotional well-being and unhealthy behaviors in adolescents . J Adolesc Health . 2014 ; 54 ( 6 ): 724 – 729 . Google Scholar Crossref Search ADS PubMed WorldCat 25 Carr E , Szymanski D, Taha F, et al. . Understanding the link between multiple oppressions and depression among African American women: the role of internalization . Psychol Women Q . 2014 ; 38 ( 2 ): 233 – 245 . Google Scholar Crossref Search ADS WorldCat 26 Logie C , James L, Tharao W, et al. . Associations between HIV-related stigma, racial discrimination, gender discrimination, and depression among HIV-positive African, Caribbean, and Black women in Ontario, Canada . AIDS Patient Care STDs . 2013 ; 27 ( 2 ): 114 – 122 . Google Scholar Crossref Search ADS PubMed WorldCat 27 Thompson VL , Noel JG, Campbell J. Stigmatization, discrimination, and mental health: the impact of multiple identity status . Am J Orthopsychiatry . 2004 ; 74 ( 4 ): 529 – 544 . Google Scholar Crossref Search ADS PubMed WorldCat 28 Bécares L , Atatoa-Carr P. The association between maternal and partner experienced racial discrimination and prenatal perceived stress, prenatal and postnatal depression: findings from the Growing Up in New Zealand cohort study . Int J Equity Health . 2016 ; 15 ( 1 ): 155 . Google Scholar Crossref Search ADS PubMed WorldCat 29 Bécares L , Nazroo J, Kelly Y. A longitudinal examination of maternal, family, and area-level experiences of racism on children’s socioemotional development: patterns and possible explanations . Soc Sci Med . 2015 ; 142 : 128 – 135 . Google Scholar Crossref Search ADS PubMed WorldCat 30 Harris R , Tobias M, Jeffreys M, et al. . Effects of self-reported racial discrimination and deprivation on Māori health and inequalities in New Zealand: cross-sectional study . Lancet . 2006 ; 367 ( 9527 ): 2005 – 2009 . Google Scholar Crossref Search ADS PubMed WorldCat 31 Harris R , Tobias M, Jeffreys M, et al. . Racism and health: the relationship between experience of racial discrimination and health in New Zealand . Soc Sci Med . 2006 ; 63 ( 6 ): 1428 – 1441 . Google Scholar Crossref Search ADS PubMed WorldCat 32 Wallace S , Nazroo J, Bécares L. Cumulative effect of racial discrimination on the mental health of ethnic minorities in the United Kingdom . Am J Public Health . 2016 ; 106 ( 7 ): 1294 – 1300 . Google Scholar Crossref Search ADS PubMed WorldCat 33 Sowers M , Crawford S, Sternfeld B, et al. . SWAN: a multicenter, multiethnic, community-based cohort study of women and the menopausal transition. In: Lobo R, Kelsey J, Marcus R, eds. Menopause: Biology and Pathobiology . New York, NY : Academic Press, Inc. ; 2000 : 175 – 188 . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC 34 Radloff L . The CES-D Scale: a self-report depression scale for research in the general population . Appl Psychol Meas . 1977 ; 1 ( 3 ): 385 – 401 . Google Scholar Crossref Search ADS WorldCat 35 Williams DR , Yan Yu, Jackson JS, et al. . Racial differences in physical and mental health: socio-economic status, stress and discrimination . J Health Psychol . 1997 ; 2 ( 3 ): 335 – 351 . Google Scholar Crossref Search ADS PubMed WorldCat 36 Schwarz G . Estimating the dimension of a model . Ann Stat . 1978 ; 6 ( 2 ): 461 – 464 . Google Scholar Crossref Search ADS WorldCat 37 Dayton C . Latent Class Scaling Analysis . Thousand Oaks, CA : Sage Publications ; 1998 . Google Scholar Crossref Search ADS Google Preview WorldCat COPAC 38 Lo Y , Mendell NR, Rubin DB. Testing the number of components in a normal mixture . Biometrika . 2001 ; 88 ( 3 ): 767 – 778 . Google Scholar Crossref Search ADS WorldCat 39 Celeux G , Soromenho G. An entropy criterion for assessing the number of clusters in a mixture model . J Classif . 1996 ; 13 ( 2 ): 195 – 212 . Google Scholar Crossref Search ADS WorldCat 40 Zou G . A modified Poisson regression approach to prospective studies with binary data . Am J Epidemiol . 2004 ; 159 ( 7 ): 702 – 706 . Google Scholar Crossref Search ADS PubMed WorldCat 41 Muthén LK , Muthén BO. Mplus: Statistical Analysis With Latent Variables. User’s Guide . (Version 7 (1998–2012)). Los Angeles, CA : Muthén & Muthén ; 2012 . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC 42 StataCorp LP . Stata Statistical Software. Release 13. College Station, TX : StataCorp LP ; 2013 . 43 Ferraro KF , Shippee TP. Aging and cumulative inequality: how does inequality get under the skin? Gerontologist . 2009 ; 49 ( 3 ): 333 – 343 . Google Scholar Crossref Search ADS PubMed WorldCat 44 Bécares L , Stafford M, Nazroo J. Fear of racism, employment and expected organizational racism: their association with health . Eur J Public Health . 2009 ; 19 ( 5 ): 504 – 510 . Google Scholar Crossref Search ADS PubMed WorldCat 45 Dailey AB , Kasl SV, Holford TR, et al. . Neighborhood- and individual-level socioeconomic variation in perceptions of racial discrimination . Ethn Health . 2010 ; 15 ( 2 ): 145 – 163 . Google Scholar Crossref Search ADS PubMed WorldCat 46 Hunt MO , Wise LA, Jipguep MC, et al. . Neighborhood racial composition and perceptions of racial discrimination: evidence from the Black Women’s Health Study . Soc Psychol Q . 2007 ; 70 ( 3 ): 272 – 289 . Google Scholar Crossref Search ADS WorldCat 47 Krieger N , Carney D, Lancaster K, et al. . Combining explicit and implicit measures of racial discrimination in health research . Am J Public Health . 2010 ; 100 ( 8 ): 1485 – 1492 . Google Scholar Crossref Search ADS PubMed WorldCat 48 Chae DH , Nuru-Jeter AM, Adler NE, et al. . Discrimination, racial bias, and telomere length in African-American men . Am J Prev Med . 2014 ; 46 ( 2 ): 103 – 111 . Google Scholar Crossref Search ADS PubMed WorldCat 49 Rosenthal L , Lobel M. Explaining racial disparities in adverse birth outcomes: unique sources of stress for Black American women . Soc Sci Med . 2011 ; 72 ( 6 ): 977 – 983 . Google Scholar Crossref Search ADS PubMed WorldCat © The Author(s) 2018. Published by Oxford University Press on behalf of the Johns Hopkins Bloomberg School of Public Health. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0), which permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited. © The Author(s) 2018. Published by Oxford University Press on behalf of the Johns Hopkins Bloomberg School of Public Health.

Samieri, Cécilia;Morris, Martha-Clare;Bennett, David A;Berr, Claudine;Amouyel, Philippe;Dartigues, Jean-François;Tzourio, Christophe;Chasman, Daniel I;Grodstein, Francine

doi: 10.1093/aje/kwx330pmid: 29053784

Johnson, Dayna A;Hirsch, Jana A;Moore, Kari A;Redline, Susan;Roux, Ana V Diez

doi: 10.1093/aje/kwx302pmid: 29547912

Abstract Although dense neighborhood built environments support increased physical activity and lower obesity, these features may also disturb sleep. Therefore, we sought to understand the association between the built environment and objectively measured sleep. From 2010 to 2013, we analyzed data from examination 5 of the Multi-Ethnic Study of Atherosclerosis, a diverse population from 6 US cities. We fit multilevel models that assessed the association between the built environment (Street Smart Walk Score, social engagement destinations, street intersections, and population density) and sleep duration or efficiency from 1-week wrist actigraphy in 1,889 individuals. After adjustment for covariates, a 1-standard-deviation increase in Street Smart Walk Score was associated with 23% higher odds of short sleep duration (≤6 hours; odds ratio = 1.2, 95% confidence interval: 1.0, 1.4), as well as shorter average sleep duration (mean difference = −8.1 minutes, 95% confidence interval: −12.1, −4.2). Results were consistent across other built environment measures. Associations were attenuated after adjustment for survey-based measure of neighborhood noise. Dense neighborhood development may have multiple health consequence. In promoting denser neighborhoods to increase walkability, it is important to also implement strategies that reduce the adverse impacts of this development on sleep, such as noise reductions efforts. cohort, neighborhoods, noise, sleep Sleep disturbances and sleep loss are prevalent in the United States (1). It is estimated that 25% of American adults report insufficient sleep or rest at least 15 out of every 30 days, whereas 29% report sleeping less than 7 hours per night (2). Sleep problems are associated with lost productivity, motor vehicle crashes, and health outcomes that include heart disease, high blood pressure, obesity, diabetes, stroke, and all-cause mortality (3, 4). Population strategies to improve sleep require an understanding of risk factors for poor sleep, which include contextual factors such as built environments (BE). Neighborhood BE may influence sleep through a myriad of complex pathways. BE characterized by high density and more destinations could be beneficial for sleep by promoting physical activity (which can have direct beneficial effects on sleep quality), or by affecting pathways that involve adiposity or propensity for sleep apnea (5–7). Conversely, BE could have adverse effects on sleep through associations with noise, traffic, air pollution, and inopportune light exposures, which are all associated with poor sleep (8, 9). Air pollution may influence inflammation and neurotransmitters involved in sleep and sleep-related breathing disorders (9). High traffic, noise, or inopportune light exposures may lead to heightened arousal, which may result in decreased sleep efficiency and increased awakenings (10). Older populations may be more vulnerable to the effects of neighborhood environment and sleep because of underlying chronic health conditions. Understanding the link between BE and sleep is especially important given recent policies promoting high density development for health and sustainability (11). To date, there has been limited investigation of the relationships between BE and sleep, with existing work that broadly examines access to recreational facilities (12), green space (13), and urbanization (10, 14, 15). Using objectively measured BE and sleep data, we investigated the associations of BE with sleep duration and efficiency in the Multi-Ethnic Study of Atherosclerosis (MESA). In secondary analyses, we explored survey-reported measures of neighborhood noise as a factor that could partially explain the influence of BE on sleep outcomes. Because of previous research that showed sleep differences by sex (16) and the potential sex differences in coping mechanisms for environmental sleep disturbances, we investigated sex interactions in our exploratory analyses. METHODS MESA is a longitudinal study of 6,814 US adults between the ages of 45 and 84 years from 6 communities (Baltimore, Maryland; Chicago, Illinois; Forsyth County, North Carolina; Los Angeles, California; New York, New York; and Minneapolis-St. Paul, Minnesota) (17). MESA was designed to prospectively investigate risk factors for subclinical cardiovascular disease and progression to clinical disease across racial/ethnic groups including non-Hispanic white, African-American or black, Hispanic, and Asian/Chinese. Participants without cardiovascular diseases were recruited between July 2000 and August 2002, and were administered 4 follow-up examinations. Current analyses utilize data on sleep measures and neighborhood characteristics from 2 MESA ancillary studies that were collected with the fourth follow-up examination (examination 5) between April 2010 and February 2012. Institutional Review Board approval was obtained at each study site and written informed consent was obtained from all participants. Neighborhood built environment Neighborhood BE was characterized using Geographic Information Systems and linked to MESA households as part of the MESA Neighborhood Ancillary study. Examination 5 residential addresses were geocoded using the TeleAtlas EZ-Locate web-based software (Lebanon, New Hampshire). We analyzed walkability using the Street Smart (SS) Walk Score (www.walkscore.com) for each residential address, obtained from Redfin (Seattle, Washington) in August 2015. As a composite measure, SS Walk Score may be challenging for urban planners who are trying to translate research into policy. Therefore, we also analyzed 3 specific BE indicators that capture walkability (18) and that have previously been associated with walking in MESA (19–21): 1) population density, 2) street connectivity, and 3) social engagement destinations. These measures were used for 2 reasons. First, they are similar to components of the SS Walk Score, which allowed us to determine which specific features influence sleep. Second, they may be related to higher exposures to noise, light, or air pollution that may explain the association between walkability and sleep. The SS Walk Score algorithm produces scores ranging from 0 to 100 (higher scores indicate better walkability), on the basis of a distance decay to various destinations (e.g., restaurants, shopping, schools, parks, or entertainment) within 1.5 miles and was adjusted for street network characteristics (e.g. low intersection density and high block length) (22). SS Walk Score utilizes network distances by following streets to amenities and allows for multiple amenities within each category. We used standard developer SS Walk Score cutpoints, where 0–24 indicated “very car-dependent,” 25–49 indicated “car-dependent,” 50–69 indicated “somewhat walkable,” 70–89 indicated “very walkable,” and 90–100 indicated “walker’s paradise.” For population density, street connectivity, and social engagement destinations, neighborhoods were defined as half-mile Euclidean buffers. Half mile was chosen because BE elements which might influence sleep (e.g., noise, light, or traffic) are at close proximity. Sensitivity analyses were performed using 1-mile and quarter-mile Euclidean buffers; results were similar (not shown). Population density was measured as population per square mile, obtained from the 2010 US Census at block level divided by land area. When a block was not fully contained within a participant’s buffer, the population was assumed to be uniform within each block. Street network was obtained using the 2012 StreetMap Premium for ArcGIS (Esri, Redlands, California). Street connectivity was measured by intersection density, counting the number of intersections within the buffer divided by land area in hectares. Destinations that promote social engagement were purchased from the National Establishment Time-Series database (Walls & Associates, Denver, Colorado) and classified using 8-digit standard industry classification codes. From previous work, 430 standard industry classification codes were selected on the basis of likelihood of facilitating social interaction and promoting social engagement (e.g., beauty shops/barbers, entertainment venues, recreation clubs, libraries, museums, civil and political clubs, religious locations, night clubs and dining places) (23). Density (count per square mile) was created as the number of locations in business in 2010 within the buffer divided by land area. Sensitivity analyses were performed using only nighttime locations (e.g., bars or nightclubs). Results were similar for these types of locations and all other locations (Web Tables 1 and 2, available at https://academic.oup.com/aje). Exposures were transformed to z scores for ease of comparison. Neighborhood survey-based noise Survey-based noise was assessed by administering questionnaires to participants. Participants described their neighborhood (defined as a 20-minute walk, or about a mile) by responding to the question “There is a lot of noise in my neighborhood” on a 5-point scale ranging from (1) “strongly agree” to (5) “strongly disagree.” To obtain a neighborhood aggregate noise measure, the same question was administered to a random sample of residents of selected census tracts in the 6 study sites between August 2011 and May 2012. Values were reverse coded so higher scores indicated more noise. Neighborhood-level noise was measured by taking the mean of responses for all respondents living within a 1-mile buffer of the participant, excluding their own response. By averaging across individuals, a more valid measure of the objective reality of neighborhoods was obtained. This was also transformed to z score for comparison. Sleep measures Between 2010 and 2013, sleep patterns were assessed using a 1-week wrist actigraphy as part of the MESA Sleep Ancillary study. Participants wore Actiwatch Spectrum devices (Philips Respironics, Murrysville, Pennsylvania) on their nondominant wrists for 7 consecutive days, while completing a sleep diary (24). A centralized reading center at Brigham and Women’s Hospital (Boston, Massachusetts) scored all records. Details regarding measurement and scoring of actigraphic data have been previously published (25). In brief, sleep-wake status for each 30-second epoch within each rest period was computed using Actiware-Sleep scoring algorithm (version 5.59). Sleep onset was defined as 5 immobile minutes, and sleep offset was defined as 0 immobile minutes and a wake threshold of 40 counts. We examined sleep duration as the sum of all epochs scored as sleep in the main sleep period, measured in minutes. We also examined sleep efficiency, a measure of sleep continuity and disturbed sleep, which was defined as the proportion of epochs between sleep onset and offset scored as sleep. Each actigraphy-measured sleep variable was computed for each recording night and averaged across all recorded nights. Individuals with at least 5 nights of data were included, which composed 95.4% of the MESA sleep study population. Covariates Covariates that were selected a priori as potential confounders included individual sociodemographics (age, sex, race/ethnicity, education, household income), neighborhood socioeconomic status, health-related behaviors (smoking, alcohol use, physical activity), body mass index, and depressive symptoms. Age and sex were self-reported. Race/ethnicity was classified as non-Hispanic white, non-Hispanic black, Hispanic, and Asian/Chinese. Education was selected from 8 categories and a continuous measure of education years was derived using midpoints of selected categories. Similarly, combined family income was selected from 15 categories and a continuous measure was derived using category midpoints. Smoking status was self-reported and categorized as current, former, or never smoker. Alcohol use was categorized as never (0 drinks/week), moderate (≤7 drinks/week for women; <14 drinks/week for men) and heavy (>7 drinks/week for women; >15 drinks/week for men) (26). Physical activity, assessed by a questionnaire adapted from the Cross-Cultural Activity Participation Study (27), was measured as moderate to vigorous in metabolic equivalent minutes per week and categorized into quartiles. Height and weight were measured and body mass index was calculated as ratio of weight to height squared. Depressive symptomology was assessed with the Center for Epidemiologic Studies of Depression scale (after removal of the sleep item) and modeled continuously (28). Neighborhood socioeconomic status was characterized on the basis of a factor score derived from principal components analysis of US census tract-level data from the American Community Survey 5-year estimates for the years 2007–2011 (29). The factor score includes median household income, percentage of homes with interest and dividends, median value of owner-occupied housing, percentage of residents with at least a high school diploma, percentage of residents with at least a bachelor’s degree, and percentage of residents employed in managerial professions. Statistical analysis Participants with complete neighborhood and sleep data and geocodes accurate to the street or zip code +4 level were included in analyses, which accounted for 40.1% of the examination 5 sample. This resulted in a sample size of 1,889 individuals. Because of missing data, analyses including neighborhood-level noise had a sample size of 1,767 individuals. We used χ2 or analysis of variance to test for differences in covariates, BE exposures, and outcomes by 5 categories of SS Walk Score. To examine the relationship between BE and sleep, we used a series of multilevel models. Linear and logistic multi-level models, with a random intercept for each census tract (as a proxy for neighborhood), were used in the main analyses. See the Web Appendix for more details and for the intraclass correlations (Web Table 3). Sleep duration was categorized as short (≤6 hours), normal (>6 but <9 hours), and long (≥9 hours) in some analyses and analyzed continuously in minutes in other analyses. In logistic regression models, because of small sample size for long sleep (n = 33), we compared short versus normal or long sleep. Sleep efficiency was analyzed continuously. We utilized a sequential modeling approach. BE measures were model-adjusted for individual sociodemographics and neighborhood socioeconomic status (model 1), and then were further adjusted for smoking, alcohol use, body mass index, and depressive symptoms (model 2). Additionally, an adjustment for physical activity (quartiles) was included (model 3). Physical activity was examined in a separate model because it could function as a potential negative confounder of the association between greater walkability and adverse sleep outcomes, as greater walkability has been shown to be linked to more physical activity, which could in turn result in better sleep. Sensitivity analyses were performed with the addition of chronic illness, as measured by hypertension (systolic blood pressure ≥140 mmHg, diastolic blood pressure ≥90 mmHg, or use of antihypertensive medication), diabetes (fasting glucose ≥126 mg/dL or use of insulin or oral hypoglycemic medication), and emphysema/chronic obstructive pulmonary disease (self-reported). Results were similar to model 3 (not shown). Each BE measure was modeled in separate models. To test the impact of neighborhood-level survey-based noise on the association between BE and sleep measures, noise was added to all models described above. Because results are consistent across models, and results from fully adjusted models (model 3 described above) are shown. All analyses were performed using SAS software, version 9.3 (SAS Institute, Inc., Cary, North Carolina). RESULTS Just over half of all participants were women (53.8%), and also composed a higher percentage of those in “walker’s paradise” (60.3%) (Table 1). Residents of “car dependent” neighborhoods were more likely to be non-Hispanic white and non-Hispanic black and less likely to be Hispanic and Chinese than residents of other neighborhoods. Neighborhood disadvantage scores had a U-shape relationship with SS Walk Score; disadvantage was less frequent in both low- and high-walkability neighborhoods. Although all walkability categories were represented in Maryland, Minnesota, Illinois, and California, the North Carolina study site had no “very walkable” or “walker’s paradise” neighborhoods and the New York study site had almost no “car dependent” neighborhoods. More than 90% of the sample were former or never smokers, and over 50% did not currently drink alcohol. The mean depressive symptoms score was highest in the most walkable neighborhoods (6.3 (standard deviation (SD), 6.8)) compared with the least walkable neighborhoods (9.0 (SD, 8.3)). As expected, social engagement destinations, intersection density, population density, and neighborhood-level survey-based noise all increased with walkability as measured by the SS Walk Score. Table 1. Sociodemographic, Built Environment, and Sleep Characteristics by Street Smart Walk Score Walkability Category, Examination 5, Multi-Ethnic Study of Atherosclerosis, 2010–2012 Characteristic  Overall (n = 1,889)  0–24: Car Dependent (n = 452)  25–49: Car Dependent (n = 349)  50–69: Somewhat Walkable (n = 376)  70–89: Very Walkable (n = 337)  90–100: Walker’s Paradise (n = 375)  P Value  Mean (SD)  %  Mean (SD)  %  Mean (SD)  %  Mean (SD)  %  Mean (SD)  %  Mean (SD)  %  Sociodemographic                             Age, years  68.6 (9.2)    68.4 (8.8)    68.0 (9.3)    68.3 (9.1)    69.0 (9.2)    69.5 (9.6)    0.16   Women    53.8    51.1    52.1    55.8    49.5    60.3  0.03   Race/ethnicity                          <0.01    White    38.9    49.8    41.5    32.4    32.6    35.5      Chinese    12.2    8.2    12.0    18.3    20.2    4.0      African-American/black    27.0    32.1    23.8    27.4    27.0    23.5      Hispanic    21.9    10.0    22.6    21.8    20.2    37.1     Education, years  13.6 (3.9)    14.6 (2.7)    13.4 (3.6)    13.1 (4.0)    13.4 (4.5)    13.1 (4.4)    <0.01   Income (per $1,000)  55.0 (35.6)    64.2 (35.8)    54.3 (33.8)    47.8 (33.4)    53.6 (35.9)    52.9 (36.7)    <0.01   Neighborhood disadvantage score  −0.5 (1.2)    −0.3 (0.9)    −0.2 (0.8)    −0.0 (1.0)    −0.5 (1.2)    −1.5 (1.5)    <0.01   Study site                          <0.01    Forsyth County, North Carolina    15.7    49.1    16.0    5.0    0.0    0.0      New York, New York    16.1    0.7    0.6    2.4    9.8    68.8      Baltimore, Maryland    14.9    17.3    20.9    18.3    14.0    3.7      St. Paul, Minnesota    18.2    17.7    39.3    26.3    7.7    0.3      Chicago, Illinois    18.6    8.0    6.9    16.2    39.2    26.1      Los Angeles, California    16.5    7.3    16.3    31.6    29.4    1.1    Health                             Smoking status                          0.16    Never smoker    46.8    42.9    47.8    47.6    46.0    50.7      Former smoker    46.3    51.3    44.4    43.3    48.1    43.2      Current smoker    6.9    5.7    7.7    9.0    5.9    6.1     Alcohol consumption                          0.01    Not current user    56.5    52.6    55.9    56.1    58.2    60.8      Moderate    36.2    42.0    38.4    35.9    33.8    29.3      Heavy    7.3    5.3    5.7    8.0    8.0    9.9     BMIa  28.7 (5.5)    29.3 (5.9)    28.8 (5.1)    28.5 (5.3)    28.1 (5.5)    28.7 (5.6)    0.06   Depressive symptoms (CES-D (excluding sleep score))  7.4 (7.2)    6.3 (6.8)    6.9 (6.7)    7.1 (7.2)    7.6 (6.6)    9.0 (8.3)    <0.01   Moderate-vigorous physical activity, METs                              0–1,724    24.9    18.6    23.8    31.6    27.6    24.5  0.01    1,725–3,456    25.0    29.6    25.2    21.8    24.0    23.2      3,457–6,629    25.1    26.1    23.8    24.2    27.0    24.3      6,630–34,290    25.0    25.7    27.2    22.3    21.4    28.0    Built environment                             Social engagement destinations per mile2b  141.2 (222.5)    11.3 (10.3)    31.3 (23.0)    65.0 (84.4)    113.9 (55.5)    501.1 (267.3)    <0.01   Intersection density per hectareb  0.8 (0.5)    0.3 (0.2)    0.6 (0.2)    0.7 (0.2)    1.0 (0.5)    1.4 (0.5)    <0.01   Population density per mile2b  18.2 (25.6)    2.2 (1.5)    5.3 (2.6)    8.9 (6.9)    15.1 (7.7)    61.4 (27.6)    <0.01   SS Walk Score  54.4 (31.8)    —c    —c    —c    —c    —c    —c   Neighborhood survey-based noised  1.5 (0.6)    0.9 (0.4)    1.2 (0.4)    1.4 (0.4)    1.6 (0.4)    2.2 (0.3)    <0.01  Sleep outcomes                             Average sleep time, minutes  390.4 (80.6)    400.3 (76.6)    391.8 (81.1)    393.9 (75.6)    383.3 (83.8)    380.0 (85.3)    <0.01   Average sleep time, hours                              ≤6    29.8    24.1    28.6    29.3    34.1    34.4  0.03    >6 but <9    68.4    74.3    68.5    69.1    64.7    64.0      ≥9    1.7    1.5    2.9    1.6    1.2    1.6     Average sleep efficiency, %  89.9 (3.6)    90.2 (3.4)    89.6 (4.0)    90.0 (3.4)    89.7 (3.4)    89.8 (3.8)    0.17  Characteristic  Overall (n = 1,889)  0–24: Car Dependent (n = 452)  25–49: Car Dependent (n = 349)  50–69: Somewhat Walkable (n = 376)  70–89: Very Walkable (n = 337)  90–100: Walker’s Paradise (n = 375)  P Value  Mean (SD)  %  Mean (SD)  %  Mean (SD)  %  Mean (SD)  %  Mean (SD)  %  Mean (SD)  %  Sociodemographic                             Age, years  68.6 (9.2)    68.4 (8.8)    68.0 (9.3)    68.3 (9.1)    69.0 (9.2)    69.5 (9.6)    0.16   Women    53.8    51.1    52.1    55.8    49.5    60.3  0.03   Race/ethnicity                          <0.01    White    38.9    49.8    41.5    32.4    32.6    35.5      Chinese    12.2    8.2    12.0    18.3    20.2    4.0      African-American/black    27.0    32.1    23.8    27.4    27.0    23.5      Hispanic    21.9    10.0    22.6    21.8    20.2    37.1     Education, years  13.6 (3.9)    14.6 (2.7)    13.4 (3.6)    13.1 (4.0)    13.4 (4.5)    13.1 (4.4)    <0.01   Income (per $1,000)  55.0 (35.6)    64.2 (35.8)    54.3 (33.8)    47.8 (33.4)    53.6 (35.9)    52.9 (36.7)    <0.01   Neighborhood disadvantage score  −0.5 (1.2)    −0.3 (0.9)    −0.2 (0.8)    −0.0 (1.0)    −0.5 (1.2)    −1.5 (1.5)    <0.01   Study site                          <0.01    Forsyth County, North Carolina    15.7    49.1    16.0    5.0    0.0    0.0      New York, New York    16.1    0.7    0.6    2.4    9.8    68.8      Baltimore, Maryland    14.9    17.3    20.9    18.3    14.0    3.7      St. Paul, Minnesota    18.2    17.7    39.3    26.3    7.7    0.3      Chicago, Illinois    18.6    8.0    6.9    16.2    39.2    26.1      Los Angeles, California    16.5    7.3    16.3    31.6    29.4    1.1    Health                             Smoking status                          0.16    Never smoker    46.8    42.9    47.8    47.6    46.0    50.7      Former smoker    46.3    51.3    44.4    43.3    48.1    43.2      Current smoker    6.9    5.7    7.7    9.0    5.9    6.1     Alcohol consumption                          0.01    Not current user    56.5    52.6    55.9    56.1    58.2    60.8      Moderate    36.2    42.0    38.4    35.9    33.8    29.3      Heavy    7.3    5.3    5.7    8.0    8.0    9.9     BMIa  28.7 (5.5)    29.3 (5.9)    28.8 (5.1)    28.5 (5.3)    28.1 (5.5)    28.7 (5.6)    0.06   Depressive symptoms (CES-D (excluding sleep score))  7.4 (7.2)    6.3 (6.8)    6.9 (6.7)    7.1 (7.2)    7.6 (6.6)    9.0 (8.3)    <0.01   Moderate-vigorous physical activity, METs                              0–1,724    24.9    18.6    23.8    31.6    27.6    24.5  0.01    1,725–3,456    25.0    29.6    25.2    21.8    24.0    23.2      3,457–6,629    25.1    26.1    23.8    24.2    27.0    24.3      6,630–34,290    25.0    25.7    27.2    22.3    21.4    28.0    Built environment                             Social engagement destinations per mile2b  141.2 (222.5)    11.3 (10.3)    31.3 (23.0)    65.0 (84.4)    113.9 (55.5)    501.1 (267.3)    <0.01   Intersection density per hectareb  0.8 (0.5)    0.3 (0.2)    0.6 (0.2)    0.7 (0.2)    1.0 (0.5)    1.4 (0.5)    <0.01   Population density per mile2b  18.2 (25.6)    2.2 (1.5)    5.3 (2.6)    8.9 (6.9)    15.1 (7.7)    61.4 (27.6)    <0.01   SS Walk Score  54.4 (31.8)    —c    —c    —c    —c    —c    —c   Neighborhood survey-based noised  1.5 (0.6)    0.9 (0.4)    1.2 (0.4)    1.4 (0.4)    1.6 (0.4)    2.2 (0.3)    <0.01  Sleep outcomes                             Average sleep time, minutes  390.4 (80.6)    400.3 (76.6)    391.8 (81.1)    393.9 (75.6)    383.3 (83.8)    380.0 (85.3)    <0.01   Average sleep time, hours                              ≤6    29.8    24.1    28.6    29.3    34.1    34.4  0.03    >6 but <9    68.4    74.3    68.5    69.1    64.7    64.0      ≥9    1.7    1.5    2.9    1.6    1.2    1.6     Average sleep efficiency, %  89.9 (3.6)    90.2 (3.4)    89.6 (4.0)    90.0 (3.4)    89.7 (3.4)    89.8 (3.8)    0.17  Abbreviations: BMI, body mass index; CES-D, Center for Epidemiologic Studies Depression; MET, metabolic equivalent of task; SD, standard deviation; SS, Street Smart. a Weight (kg)/height (m)2. b Density measured within a half mile of the study participant’s home address. c SS Walk Score is not shown in bivariate association with itself. d Mean of respondents within 1 mile of the study participant’s home address (n = 1,767). Table 1. Sociodemographic, Built Environment, and Sleep Characteristics by Street Smart Walk Score Walkability Category, Examination 5, Multi-Ethnic Study of Atherosclerosis, 2010–2012 Characteristic  Overall (n = 1,889)  0–24: Car Dependent (n = 452)  25–49: Car Dependent (n = 349)  50–69: Somewhat Walkable (n = 376)  70–89: Very Walkable (n = 337)  90–100: Walker’s Paradise (n = 375)  P Value  Mean (SD)  %  Mean (SD)  %  Mean (SD)  %  Mean (SD)  %  Mean (SD)  %  Mean (SD)  %  Sociodemographic                             Age, years  68.6 (9.2)    68.4 (8.8)    68.0 (9.3)    68.3 (9.1)    69.0 (9.2)    69.5 (9.6)    0.16   Women    53.8    51.1    52.1    55.8    49.5    60.3  0.03   Race/ethnicity                          <0.01    White    38.9    49.8    41.5    32.4    32.6    35.5      Chinese    12.2    8.2    12.0    18.3    20.2    4.0      African-American/black    27.0    32.1    23.8    27.4    27.0    23.5      Hispanic    21.9    10.0    22.6    21.8    20.2    37.1     Education, years  13.6 (3.9)    14.6 (2.7)    13.4 (3.6)    13.1 (4.0)    13.4 (4.5)    13.1 (4.4)    <0.01   Income (per $1,000)  55.0 (35.6)    64.2 (35.8)    54.3 (33.8)    47.8 (33.4)    53.6 (35.9)    52.9 (36.7)    <0.01   Neighborhood disadvantage score  −0.5 (1.2)    −0.3 (0.9)    −0.2 (0.8)    −0.0 (1.0)    −0.5 (1.2)    −1.5 (1.5)    <0.01   Study site                          <0.01    Forsyth County, North Carolina    15.7    49.1    16.0    5.0    0.0    0.0      New York, New York    16.1    0.7    0.6    2.4    9.8    68.8      Baltimore, Maryland    14.9    17.3    20.9    18.3    14.0    3.7      St. Paul, Minnesota    18.2    17.7    39.3    26.3    7.7    0.3      Chicago, Illinois    18.6    8.0    6.9    16.2    39.2    26.1      Los Angeles, California    16.5    7.3    16.3    31.6    29.4    1.1    Health                             Smoking status                          0.16    Never smoker    46.8    42.9    47.8    47.6    46.0    50.7      Former smoker    46.3    51.3    44.4    43.3    48.1    43.2      Current smoker    6.9    5.7    7.7    9.0    5.9    6.1     Alcohol consumption                          0.01    Not current user    56.5    52.6    55.9    56.1    58.2    60.8      Moderate    36.2    42.0    38.4    35.9    33.8    29.3      Heavy    7.3    5.3    5.7    8.0    8.0    9.9     BMIa  28.7 (5.5)    29.3 (5.9)    28.8 (5.1)    28.5 (5.3)    28.1 (5.5)    28.7 (5.6)    0.06   Depressive symptoms (CES-D (excluding sleep score))  7.4 (7.2)    6.3 (6.8)    6.9 (6.7)    7.1 (7.2)    7.6 (6.6)    9.0 (8.3)    <0.01   Moderate-vigorous physical activity, METs                              0–1,724    24.9    18.6    23.8    31.6    27.6    24.5  0.01    1,725–3,456    25.0    29.6    25.2    21.8    24.0    23.2      3,457–6,629    25.1    26.1    23.8    24.2    27.0    24.3      6,630–34,290    25.0    25.7    27.2    22.3    21.4    28.0    Built environment                             Social engagement destinations per mile2b  141.2 (222.5)    11.3 (10.3)    31.3 (23.0)    65.0 (84.4)    113.9 (55.5)    501.1 (267.3)    <0.01   Intersection density per hectareb  0.8 (0.5)    0.3 (0.2)    0.6 (0.2)    0.7 (0.2)    1.0 (0.5)    1.4 (0.5)    <0.01   Population density per mile2b  18.2 (25.6)    2.2 (1.5)    5.3 (2.6)    8.9 (6.9)    15.1 (7.7)    61.4 (27.6)    <0.01   SS Walk Score  54.4 (31.8)    —c    —c    —c    —c    —c    —c   Neighborhood survey-based noised  1.5 (0.6)    0.9 (0.4)    1.2 (0.4)    1.4 (0.4)    1.6 (0.4)    2.2 (0.3)    <0.01  Sleep outcomes                             Average sleep time, minutes  390.4 (80.6)    400.3 (76.6)    391.8 (81.1)    393.9 (75.6)    383.3 (83.8)    380.0 (85.3)    <0.01   Average sleep time, hours                              ≤6    29.8    24.1    28.6    29.3    34.1    34.4  0.03    >6 but <9    68.4    74.3    68.5    69.1    64.7    64.0      ≥9    1.7    1.5    2.9    1.6    1.2    1.6     Average sleep efficiency, %  89.9 (3.6)    90.2 (3.4)    89.6 (4.0)    90.0 (3.4)    89.7 (3.4)    89.8 (3.8)    0.17  Characteristic  Overall (n = 1,889)  0–24: Car Dependent (n = 452)  25–49: Car Dependent (n = 349)  50–69: Somewhat Walkable (n = 376)  70–89: Very Walkable (n = 337)  90–100: Walker’s Paradise (n = 375)  P Value  Mean (SD)  %  Mean (SD)  %  Mean (SD)  %  Mean (SD)  %  Mean (SD)  %  Mean (SD)  %  Sociodemographic                             Age, years  68.6 (9.2)    68.4 (8.8)    68.0 (9.3)    68.3 (9.1)    69.0 (9.2)    69.5 (9.6)    0.16   Women    53.8    51.1    52.1    55.8    49.5    60.3  0.03   Race/ethnicity                          <0.01    White    38.9    49.8    41.5    32.4    32.6    35.5      Chinese    12.2    8.2    12.0    18.3    20.2    4.0      African-American/black    27.0    32.1    23.8    27.4    27.0    23.5      Hispanic    21.9    10.0    22.6    21.8    20.2    37.1     Education, years  13.6 (3.9)    14.6 (2.7)    13.4 (3.6)    13.1 (4.0)    13.4 (4.5)    13.1 (4.4)    <0.01   Income (per $1,000)  55.0 (35.6)    64.2 (35.8)    54.3 (33.8)    47.8 (33.4)    53.6 (35.9)    52.9 (36.7)    <0.01   Neighborhood disadvantage score  −0.5 (1.2)    −0.3 (0.9)    −0.2 (0.8)    −0.0 (1.0)    −0.5 (1.2)    −1.5 (1.5)    <0.01   Study site                          <0.01    Forsyth County, North Carolina    15.7    49.1    16.0    5.0    0.0    0.0      New York, New York    16.1    0.7    0.6    2.4    9.8    68.8      Baltimore, Maryland    14.9    17.3    20.9    18.3    14.0    3.7      St. Paul, Minnesota    18.2    17.7    39.3    26.3    7.7    0.3      Chicago, Illinois    18.6    8.0    6.9    16.2    39.2    26.1      Los Angeles, California    16.5    7.3    16.3    31.6    29.4    1.1    Health                             Smoking status                          0.16    Never smoker    46.8    42.9    47.8    47.6    46.0    50.7      Former smoker    46.3    51.3    44.4    43.3    48.1    43.2      Current smoker    6.9    5.7    7.7    9.0    5.9    6.1     Alcohol consumption                          0.01    Not current user    56.5    52.6    55.9    56.1    58.2    60.8      Moderate    36.2    42.0    38.4    35.9    33.8    29.3      Heavy    7.3    5.3    5.7    8.0    8.0    9.9     BMIa  28.7 (5.5)    29.3 (5.9)    28.8 (5.1)    28.5 (5.3)    28.1 (5.5)    28.7 (5.6)    0.06   Depressive symptoms (CES-D (excluding sleep score))  7.4 (7.2)    6.3 (6.8)    6.9 (6.7)    7.1 (7.2)    7.6 (6.6)    9.0 (8.3)    <0.01   Moderate-vigorous physical activity, METs                              0–1,724    24.9    18.6    23.8    31.6    27.6    24.5  0.01    1,725–3,456    25.0    29.6    25.2    21.8    24.0    23.2      3,457–6,629    25.1    26.1    23.8    24.2    27.0    24.3      6,630–34,290    25.0    25.7    27.2    22.3    21.4    28.0    Built environment                             Social engagement destinations per mile2b  141.2 (222.5)    11.3 (10.3)    31.3 (23.0)    65.0 (84.4)    113.9 (55.5)    501.1 (267.3)    <0.01   Intersection density per hectareb  0.8 (0.5)    0.3 (0.2)    0.6 (0.2)    0.7 (0.2)    1.0 (0.5)    1.4 (0.5)    <0.01   Population density per mile2b  18.2 (25.6)    2.2 (1.5)    5.3 (2.6)    8.9 (6.9)    15.1 (7.7)    61.4 (27.6)    <0.01   SS Walk Score  54.4 (31.8)    —c    —c    —c    —c    —c    —c   Neighborhood survey-based noised  1.5 (0.6)    0.9 (0.4)    1.2 (0.4)    1.4 (0.4)    1.6 (0.4)    2.2 (0.3)    <0.01  Sleep outcomes                             Average sleep time, minutes  390.4 (80.6)    400.3 (76.6)    391.8 (81.1)    393.9 (75.6)    383.3 (83.8)    380.0 (85.3)    <0.01   Average sleep time, hours                              ≤6    29.8    24.1    28.6    29.3    34.1    34.4  0.03    >6 but <9    68.4    74.3    68.5    69.1    64.7    64.0      ≥9    1.7    1.5    2.9    1.6    1.2    1.6     Average sleep efficiency, %  89.9 (3.6)    90.2 (3.4)    89.6 (4.0)    90.0 (3.4)    89.7 (3.4)    89.8 (3.8)    0.17  Abbreviations: BMI, body mass index; CES-D, Center for Epidemiologic Studies Depression; MET, metabolic equivalent of task; SD, standard deviation; SS, Street Smart. a Weight (kg)/height (m)2. b Density measured within a half mile of the study participant’s home address. c SS Walk Score is not shown in bivariate association with itself. d Mean of respondents within 1 mile of the study participant’s home address (n = 1,767). Average sleep duration was 6.5 hours, or 390.4 minutes (SD, 80.6). In unadjusted analyses, mean sleep duration was highest among participants in “car-dependent” neighborhoods (400.3 (SD, 76.6)) and decreased with increasing walkability (Table 1). Higher social engagement destination density, intersection density, population density, and SS Walk Score were associated with shorter average sleep time, higher odds of short sleep duration, and lower average sleep efficiency (Table 2). After adjustment for individual- and neighborhood-level sociodemographic characteristics, as well as individual health behaviors and outcomes, a standard deviation higher SS Walk Score was estimated to be associated with a mean 8.1 minutes of fewer sleep (95% confidence interval (CI): −12.1, −4.2). Qualitatively similar associations were observed for a standard deviation difference in social engagement destinations (−6.5 minutes, 95% CI: −11.2, −1.7), intersection density (−4.7 minutes, 95% CI: −8.7, −0.7), and population density (−7.7 minutes, 95% CI: −11.9, −3.6). Table 2. Mean Differences in Average Sleep Time, Odds Ratios of Short Sleep, and Mean Differences in Average Sleep Efficiency for a 1-Standard-Deviation Difference in Built Environment Features (n = 1,889), Multi-Ethnic Study of Atherosclerosis, 2010–2012 Sleep Parameter  Model 1a  Model 2b  Model 3c  β  95% CI  β  95% CI  β  95% CI  Average sleep time, minutes               Social engagement destinations per mile2  −6.6  −11.3, −1.8  −6.4  −11.1, −1.7  −6.5  −11.2, −1.7   Intersection density per hectare  −4.8  −8.8, −0.7  −4.7  −8.7, −0.7  −4.7  −8.7, −0.7   Population density per mile2  −8.0  −12.2, −3.9  −7.8  −11.9, −3.6  −7.7  −11.9, −3.6   SS Walk Score  −8.0  −11.9, −4.0  −8.1  −12.1, −4.2  −8.1  −12.1, −4.2  Average efficiency, %               Social engagement destinations per mile2  0.0  −0.3, 0.2  0.0  −0.3, 0.2  0.0  −0.3, 0.2   Intersection density per hectare  −0.2  −0.4, −0.1  −0.2  −0.4, −0.1  −0.2  −0.4, −0.1   Population density per mile2  −0.2  −0.4, 0.0  −0.2  −0.4, 0.0  −0.2  −0.4, 0.0   SS Walk Score  −0.1  −0.3, 0.1  −0.1  −0.3, 0.1  −0.1  −0.3, 0.1    OR  95% CI  OR  95% CI  OR  95% CI  Short sleepd               Social engagement destinations per mile2  1.2  1.1, 1.4  1.2  1.0, 1.4  1.2  1.0, 1.4   Intersection density per hectare  1.1  1.0, 1.3  1.1  1.0, 1.3  1.1  1.0, 1.3   Population density per mile2  1.2  1.1, 1.3  1.2  1.0, 1.3  1.2  1.0, 1.3   SS Walk Score  1.2  1.1, 1.4  1.2  1.0, 1.4  1.2  1.0, 1.4  Sleep Parameter  Model 1a  Model 2b  Model 3c  β  95% CI  β  95% CI  β  95% CI  Average sleep time, minutes               Social engagement destinations per mile2  −6.6  −11.3, −1.8  −6.4  −11.1, −1.7  −6.5  −11.2, −1.7   Intersection density per hectare  −4.8  −8.8, −0.7  −4.7  −8.7, −0.7  −4.7  −8.7, −0.7   Population density per mile2  −8.0  −12.2, −3.9  −7.8  −11.9, −3.6  −7.7  −11.9, −3.6   SS Walk Score  −8.0  −11.9, −4.0  −8.1  −12.1, −4.2  −8.1  −12.1, −4.2  Average efficiency, %               Social engagement destinations per mile2  0.0  −0.3, 0.2  0.0  −0.3, 0.2  0.0  −0.3, 0.2   Intersection density per hectare  −0.2  −0.4, −0.1  −0.2  −0.4, −0.1  −0.2  −0.4, −0.1   Population density per mile2  −0.2  −0.4, 0.0  −0.2  −0.4, 0.0  −0.2  −0.4, 0.0   SS Walk Score  −0.1  −0.3, 0.1  −0.1  −0.3, 0.1  −0.1  −0.3, 0.1    OR  95% CI  OR  95% CI  OR  95% CI  Short sleepd               Social engagement destinations per mile2  1.2  1.1, 1.4  1.2  1.0, 1.4  1.2  1.0, 1.4   Intersection density per hectare  1.1  1.0, 1.3  1.1  1.0, 1.3  1.1  1.0, 1.3   Population density per mile2  1.2  1.1, 1.3  1.2  1.0, 1.3  1.2  1.0, 1.3   SS Walk Score  1.2  1.1, 1.4  1.2  1.0, 1.4  1.2  1.0, 1.4  Abbreviations: CI, confidence interval; OR, odds ratio; SS, Street Smart. a Model 1 was adjusted for sex, race, age, age squared, educational level, income, and neighborhood socioeconomic status. b Model 2 was adjusted for the variables in model 1 and smoking, alcohol consumption, body mass index, and depressive symptoms, as measured using the Center for Epidemiologic Studies–Depression Scale. c Model 3 was adjusted for the variables in model 2 and physical activity level. d Short sleep was defined as ≤6 hours (vs. >6 hours). Table 2. Mean Differences in Average Sleep Time, Odds Ratios of Short Sleep, and Mean Differences in Average Sleep Efficiency for a 1-Standard-Deviation Difference in Built Environment Features (n = 1,889), Multi-Ethnic Study of Atherosclerosis, 2010–2012 Sleep Parameter  Model 1a  Model 2b  Model 3c  β  95% CI  β  95% CI  β  95% CI  Average sleep time, minutes               Social engagement destinations per mile2  −6.6  −11.3, −1.8  −6.4  −11.1, −1.7  −6.5  −11.2, −1.7   Intersection density per hectare  −4.8  −8.8, −0.7  −4.7  −8.7, −0.7  −4.7  −8.7, −0.7   Population density per mile2  −8.0  −12.2, −3.9  −7.8  −11.9, −3.6  −7.7  −11.9, −3.6   SS Walk Score  −8.0  −11.9, −4.0  −8.1  −12.1, −4.2  −8.1  −12.1, −4.2  Average efficiency, %               Social engagement destinations per mile2  0.0  −0.3, 0.2  0.0  −0.3, 0.2  0.0  −0.3, 0.2   Intersection density per hectare  −0.2  −0.4, −0.1  −0.2  −0.4, −0.1  −0.2  −0.4, −0.1   Population density per mile2  −0.2  −0.4, 0.0  −0.2  −0.4, 0.0  −0.2  −0.4, 0.0   SS Walk Score  −0.1  −0.3, 0.1  −0.1  −0.3, 0.1  −0.1  −0.3, 0.1    OR  95% CI  OR  95% CI  OR  95% CI  Short sleepd               Social engagement destinations per mile2  1.2  1.1, 1.4  1.2  1.0, 1.4  1.2  1.0, 1.4   Intersection density per hectare  1.1  1.0, 1.3  1.1  1.0, 1.3  1.1  1.0, 1.3   Population density per mile2  1.2  1.1, 1.3  1.2  1.0, 1.3  1.2  1.0, 1.3   SS Walk Score  1.2  1.1, 1.4  1.2  1.0, 1.4  1.2  1.0, 1.4  Sleep Parameter  Model 1a  Model 2b  Model 3c  β  95% CI  β  95% CI  β  95% CI  Average sleep time, minutes               Social engagement destinations per mile2  −6.6  −11.3, −1.8  −6.4  −11.1, −1.7  −6.5  −11.2, −1.7   Intersection density per hectare  −4.8  −8.8, −0.7  −4.7  −8.7, −0.7  −4.7  −8.7, −0.7   Population density per mile2  −8.0  −12.2, −3.9  −7.8  −11.9, −3.6  −7.7  −11.9, −3.6   SS Walk Score  −8.0  −11.9, −4.0  −8.1  −12.1, −4.2  −8.1  −12.1, −4.2  Average efficiency, %               Social engagement destinations per mile2  0.0  −0.3, 0.2  0.0  −0.3, 0.2  0.0  −0.3, 0.2   Intersection density per hectare  −0.2  −0.4, −0.1  −0.2  −0.4, −0.1  −0.2  −0.4, −0.1   Population density per mile2  −0.2  −0.4, 0.0  −0.2  −0.4, 0.0  −0.2  −0.4, 0.0   SS Walk Score  −0.1  −0.3, 0.1  −0.1  −0.3, 0.1  −0.1  −0.3, 0.1    OR  95% CI  OR  95% CI  OR  95% CI  Short sleepd               Social engagement destinations per mile2  1.2  1.1, 1.4  1.2  1.0, 1.4  1.2  1.0, 1.4   Intersection density per hectare  1.1  1.0, 1.3  1.1  1.0, 1.3  1.1  1.0, 1.3   Population density per mile2  1.2  1.1, 1.3  1.2  1.0, 1.3  1.2  1.0, 1.3   SS Walk Score  1.2  1.1, 1.4  1.2  1.0, 1.4  1.2  1.0, 1.4  Abbreviations: CI, confidence interval; OR, odds ratio; SS, Street Smart. a Model 1 was adjusted for sex, race, age, age squared, educational level, income, and neighborhood socioeconomic status. b Model 2 was adjusted for the variables in model 1 and smoking, alcohol consumption, body mass index, and depressive symptoms, as measured using the Center for Epidemiologic Studies–Depression Scale. c Model 3 was adjusted for the variables in model 2 and physical activity level. d Short sleep was defined as ≤6 hours (vs. >6 hours). A standard deviation higher social engagement destination, population density, and SS Walk Score was associated with 21% higher odds (adjusted odds ratio = 1.2, CI: 1.0, 1.4), 17% higher odds (adjusted odds ratio = 1.2, 95% CI: 1.0, 1.3), and 23% higher odds (adjusted odds ratio = 1.2, 95% CI: 1.0, 1.4) of short sleep duration, respectively. After adjusting for all covariates, only intersection density was statistically significantly associated with decreased average sleep efficiency (adjusted mean difference = −0.2, 95% CI: −0.4, −0.1). Higher neighborhood-level survey-based noise was associated with shorter average sleep time (adjusted mean difference = −7.7, 95% CI: −11.7, −3.7), higher odds of short sleep duration (adjusted odds ratio = 1.2, 95% CI: 1.0, 1.3) and decreased average sleep efficiency (adjusted mean difference = −0.2%, 95% CI: −0.4, 0.0). Neighborhood-level noise was correlated with BE measures (Spearman correlations: 0.7–0.8). Models that included noise resulted in attenuated associations of BE features with sleep (Table 3). For sleep duration, the association of SS Walk Score was reduced by 22.9% (−9.1 minutes (95% CI: −13.2, −5.0) versus −7.0 minutes (95% CI: −12.7, −1.3)), with a similar reduction observed in population density (−8.7 minutes (95% CI: −12.9, −4.6) versus −6.2 minutes (95% CI: −11.4, −0.9)). Larger relative reductions in the associations for social engagement destinations (−7.6 minutes (95% CI: −12.4, −2.8) versus −3.5 minutes (95% CI: −9.4, 2.3)) and intersection density (−6.0 minutes (95% CI: −10.1, −1.9) versus −2.3 minutes (95% CI: −7.3, 2.7)) were observed. The association of BE measures with odds of shorter sleep was also attenuated with the adjustment for neighborhood-level noise, with observed reductions ranging between 1% and 6%. Table 3. Mean Differences in Average Sleep Time, Odds Ratios of Short Sleep, and Mean Differences in Average Sleep Efficiency for a 1-Standard-Deviation Difference in Built Environment Features and Neighborhood Survey-Based Noise (n = 1,767), Multi-Ethnic Study of Atherosclerosis, 2010–2012 Exposure  Model 1a  Model 2b  β  95% CI  β  95% CI  Average sleep time, minutes           Social engagement destinations per mile2  −7.6  −12.4, −2.8  −3.5  −9.4, 2.3   Intersection density per hectare  −6.0  −10.1, −1.9  −2.3  −7.3, 2.7   Population density per mile2  −8.7  −12.9, −4.6  −6.2  −11.4, −0.9   SS Walk Score  −9.1  −13.2, −5.0  −7.0  −12.7, −1.3  Average efficiency, %           Social engagement destinations per mile2  −0.1  −0.3, 0.2  0.1  −0.2, 0.4   Intersection density per hectare  −0.2  −0.4, 0.0  −0.1  −0.4, 0.1   Population density per mile2  −0.2  −0.4, 0.0  −0.1  −0.4, 0.1   SS Walk Score  −0.1  −0.3, 0.1  0.1  −0.2, 0.4    OR  95% CI  OR  95% CI  Short sleepc           Social engagement destinations per mile2  1.2  1.1, 1.4  1.2  1.0, 1.4   Intersection density per hectare  1.1  1.0, 1.3  1.1  0.9, 1.2   Population density per mile2  1.2  1.0, 1.3  1.1  0.9, 1.3   SS Walk Score  1.3  1.1, 1.4  1.3  1.0, 1.5  Exposure  Model 1a  Model 2b  β  95% CI  β  95% CI  Average sleep time, minutes           Social engagement destinations per mile2  −7.6  −12.4, −2.8  −3.5  −9.4, 2.3   Intersection density per hectare  −6.0  −10.1, −1.9  −2.3  −7.3, 2.7   Population density per mile2  −8.7  −12.9, −4.6  −6.2  −11.4, −0.9   SS Walk Score  −9.1  −13.2, −5.0  −7.0  −12.7, −1.3  Average efficiency, %           Social engagement destinations per mile2  −0.1  −0.3, 0.2  0.1  −0.2, 0.4   Intersection density per hectare  −0.2  −0.4, 0.0  −0.1  −0.4, 0.1   Population density per mile2  −0.2  −0.4, 0.0  −0.1  −0.4, 0.1   SS Walk Score  −0.1  −0.3, 0.1  0.1  −0.2, 0.4    OR  95% CI  OR  95% CI  Short sleepc           Social engagement destinations per mile2  1.2  1.1, 1.4  1.2  1.0, 1.4   Intersection density per hectare  1.1  1.0, 1.3  1.1  0.9, 1.2   Population density per mile2  1.2  1.0, 1.3  1.1  0.9, 1.3   SS Walk Score  1.3  1.1, 1.4  1.3  1.0, 1.5  Abbreviations: CI, confidence interval; OR, odds ratio; SS, Street Smart. a Model 1 includes only the built environment exposure, which was adjusted for sex, race, age, age squared, educational level, income, neighborhood socioeconomic status, smoking, alcohol consumption, body mass index, depressive symptoms (as measured using the Center for Epidemiologic Studies–Depression Scale), and physical activity level. b Model 2 was adjusted for the variables in model 1 and neighborhood noise exposure. Values for neighborhood aggregate noise are as follows: average sleep time (minutes), mean difference = −7.7, 95% CI: −11.7, −3.7; for short sleep, odds ratio = 1.2, 95% CI: 1.0, 1.3; and for average efficiency (%), mean difference = −0.2, 95% CI: −0.4, 0.0. c Short sleep was defined as ≤6 hours (vs. >6 hours). Table 3. Mean Differences in Average Sleep Time, Odds Ratios of Short Sleep, and Mean Differences in Average Sleep Efficiency for a 1-Standard-Deviation Difference in Built Environment Features and Neighborhood Survey-Based Noise (n = 1,767), Multi-Ethnic Study of Atherosclerosis, 2010–2012 Exposure  Model 1a  Model 2b  β  95% CI  β  95% CI  Average sleep time, minutes           Social engagement destinations per mile2  −7.6  −12.4, −2.8  −3.5  −9.4, 2.3   Intersection density per hectare  −6.0  −10.1, −1.9  −2.3  −7.3, 2.7   Population density per mile2  −8.7  −12.9, −4.6  −6.2  −11.4, −0.9   SS Walk Score  −9.1  −13.2, −5.0  −7.0  −12.7, −1.3  Average efficiency, %           Social engagement destinations per mile2  −0.1  −0.3, 0.2  0.1  −0.2, 0.4   Intersection density per hectare  −0.2  −0.4, 0.0  −0.1  −0.4, 0.1   Population density per mile2  −0.2  −0.4, 0.0  −0.1  −0.4, 0.1   SS Walk Score  −0.1  −0.3, 0.1  0.1  −0.2, 0.4    OR  95% CI  OR  95% CI  Short sleepc           Social engagement destinations per mile2  1.2  1.1, 1.4  1.2  1.0, 1.4   Intersection density per hectare  1.1  1.0, 1.3  1.1  0.9, 1.2   Population density per mile2  1.2  1.0, 1.3  1.1  0.9, 1.3   SS Walk Score  1.3  1.1, 1.4  1.3  1.0, 1.5  Exposure  Model 1a  Model 2b  β  95% CI  β  95% CI  Average sleep time, minutes           Social engagement destinations per mile2  −7.6  −12.4, −2.8  −3.5  −9.4, 2.3   Intersection density per hectare  −6.0  −10.1, −1.9  −2.3  −7.3, 2.7   Population density per mile2  −8.7  −12.9, −4.6  −6.2  −11.4, −0.9   SS Walk Score  −9.1  −13.2, −5.0  −7.0  −12.7, −1.3  Average efficiency, %           Social engagement destinations per mile2  −0.1  −0.3, 0.2  0.1  −0.2, 0.4   Intersection density per hectare  −0.2  −0.4, 0.0  −0.1  −0.4, 0.1   Population density per mile2  −0.2  −0.4, 0.0  −0.1  −0.4, 0.1   SS Walk Score  −0.1  −0.3, 0.1  0.1  −0.2, 0.4    OR  95% CI  OR  95% CI  Short sleepc           Social engagement destinations per mile2  1.2  1.1, 1.4  1.2  1.0, 1.4   Intersection density per hectare  1.1  1.0, 1.3  1.1  0.9, 1.2   Population density per mile2  1.2  1.0, 1.3  1.1  0.9, 1.3   SS Walk Score  1.3  1.1, 1.4  1.3  1.0, 1.5  Abbreviations: CI, confidence interval; OR, odds ratio; SS, Street Smart. a Model 1 includes only the built environment exposure, which was adjusted for sex, race, age, age squared, educational level, income, neighborhood socioeconomic status, smoking, alcohol consumption, body mass index, depressive symptoms (as measured using the Center for Epidemiologic Studies–Depression Scale), and physical activity level. b Model 2 was adjusted for the variables in model 1 and neighborhood noise exposure. Values for neighborhood aggregate noise are as follows: average sleep time (minutes), mean difference = −7.7, 95% CI: −11.7, −3.7; for short sleep, odds ratio = 1.2, 95% CI: 1.0, 1.3; and for average efficiency (%), mean difference = −0.2, 95% CI: −0.4, 0.0. c Short sleep was defined as ≤6 hours (vs. >6 hours). There was no evidence of effect modification by sex in any analysis. Also, in sensitivity analyses we adjusted for site, and found that results were in the same direction but attenuated (see Web Tables 4 and 5). DISCUSSION In geographically and racially/ethnically diverse middle-aged and older adults, higher neighborhood walkability (characterized by SS Walk Score, population density, intersection density, and social destinations) was associated with shorter average sleep duration. Other measures of sleep continuity and quality were weakly associated with indices of neighborhood walkability, where the only significant adjusted association observed was for sleep efficiency and intersection density. This is among the first studies to report associations of BE with sleep. It provides evidence indicating that characteristics associated with walkability are associated with small average decreases in nightly sleep duration. The associations of more walkable environments with poorer sleep outcomes were attenuated when noise was taken into consideration. On average, a standard deviation higher walkability was associated with 23% higher odds of short sleep. A standard deviation higher walkability score was associated with an average of 8 minutes less sleep at night. Although the magnitudes of these associations are small, they are comparable to the sizes of the associations observed for other important sleep risk factors. For example, in the same data, a 5-unit higher body mass index is associated with 26% higher odds of short sleep. Short sleep duration has been linked to higher prevalence of cardiovascular risk factors, as well as increased risk of cardiovascular disease (30–35). Future research is needed to replicate our findings and determine the implications for health. Although previous work has not directly examined associations of neighborhood BE features with objectively measured sleep-wake patterns among adults, our findings are consistent with existing literature that show that residence in urban environments is associated with shorter sleep duration for both adults (10) and infants (14). Despite evidence that neighborhood social environment or physical disorder influences other sleep outcomes such as insomnia and daytime sleepiness (36, 37), we found no statistically significant associations between BE and wake after sleep onset, sleep latency, insomnia, and daytime sleepiness. This may reflect differences in the environmental measurements assessed in each study. Attention should be paid to the myriad of features that compose neighborhood environments as they may have differing impacts on sleep. The pathways through which neighborhood BE may influence sleep are complex. Various intervening mechanisms could result in opposing effects. There are numerous potential pathways through which more walkable neighborhoods could negatively influence sleep, including noise, inopportune light exposure, traffic, air pollution, and stress. Conversely, dense neighborhood BE have been shown to be associated with increased walking or physical activity, as well as decreased obesity, all of which have been associated with better sleep (7). Previous evidence has shown both noise and light exposure to be associated with shorter sleep duration across various populations (8). Places with more people and destinations could generate additional disturbances that do not exist in less populated, more isolated neighborhoods. Furthermore, mixed-use neighborhoods with retail interspersed with residential areas may have more turnover of businesses, causing intermittent disturbances such as construction. Residing in neighborhoods proximate to transportation routes may be sources of nighttime noise (8). When adjusting for neighborhood-level survey-based noise, we observed an attenuated association between the BE and sleep duration, which suggests that noise may explain some of the association. Our noise measure was limited as it was determined on the basis of a single question, and we potentially underestimated its contribution to sleep disturbance. Additional research that uses more comprehensive noise measures is needed to better quantify its contributions to sleep across neighborhoods. In the present study, intersection and population density were both associated with sleep duration. These associations deserve further exploration as they could represent the impact of traffic and air pollution. Intersection density could result in higher volume of cars and increased stop-and-go traffic, which could result in both increased noise and air pollution. Air pollution may activate pulmonary and systemic inflammation, which adversely affects sleep (9). Emerging literature has begun to examine the competing risks of walkability and air pollution (38), but future work should continue to expand on this with particular attention to their subsequent influence on sleep. Despite evidence in the MESA cohort that BE are associated with more walking (19–21) and less obesity (21, 39), it is plausible that the amounts of increase in physical activity and decrease in obesity may not be sufficient to generate better sleep outcomes. In sensitivity analyses that adjusted for physical activity, the negative association between BE measures and sleep persisted. Strengths and limitations To our knowledge, the present study is the first to quantify the association between objectively measured BE characteristics and objective measures of sleep-wake patterns by using numerous BE metrics derived from geographic information systems and 7-day wrist actigraphy. However, our metrics do not capture smaller-scale aesthetic features, such as green space or sidewalks, which may be related to sleep (13). Our measure of walkability is highly correlated with urbanicity and therefore may be capturing sleep consequences of urbanicity mediated by factors such as noise, light, and air pollution. Although we tested the association at varying buffer sizes, some misspecification of spatial context relevant to sleep is likely to be present (40, 41). We accounted for correlation in the errors of participants living in the same census tract, but some residual correlation may exist with participants in neighboring tracts. Although polysomnography is the gold standard for sleep measurement, actigraphy has been shown to be valid and correlates with sleep estimations made by polysomnography (42). There are other psychosocial measures (such as anxiety) and environmental measures (such as light) which may confound or mediate the association between the BE and sleep. The present study uses a diverse population of individuals, across numerous geographic locations in the United States. However, this sample is not a population-representative sample and results may not be generalizable to other populations that include adolescents, younger adults, or those in other geographic regions. Future work should utilize longitudinal methods to establish temporality, as our cross-sectional analyses cannot determine causality. Lastly, residual confounding by unmeasured individual and neighborhood characteristics is always a possibility in observational studies such as ours. Given strong associations between site and BE features, it was not possible to reliably estimate associations of BE features with sleep after adjusting for site, as site is essentially a rough proxy for BE. We have no reason to believe that site affects sleep through pathways not involving other factors present in the models (including BE features themselves); therefore we believe models unadjusted for site are our best approximations of the causal effects of BE within constraints of observational studies and data at our disposal. Public health implications We found evidence of associations between neighborhood walkability and shorter sleep duration, which were partially explained by noise. Although BE measures have been associated with better health outcomes (43), our results suggest the need for a deeper consideration of BE influences (including potential adverse effects on noise, light and air pollution) as policies are developed for dense, urban form. By focusing on select outcomes, previous work may have failed to capture the full realm of health impacts of BE. Potential countervailing influences of promoting higher density and street connectivity on physical activity and sleep should be further considered. Given the benefits of walkability, future studies should identify factors that may buffer the negative effects of BE characteristics and noise on sleep health. ACKNOWLEDGMENTS Author affiliations: Division of Sleep and Circadian Disorders, Harvard Medical School and Brigham and Women’s Hospital, Boston, Massachusetts (Dayna A. Johnson, Susan Redline); Department of Epidemiology and Biostatistics, Dornsife School of Public Health, Drexel University, Philadelphia, Pennsylvania (Jana A. Hirsch, Ana V. Diez Roux); and Urban Health Collaborative, Dornsife School of Public Health, Drexel University, Philadelphia, Pennsylvania (Jana A. Hirsch, Kari A. Moore, Ana V. Diez Roux). D.A.J. and J.A.H. are joint first authors and S.R. and A.V.D.R. are joint senior authors. This research was supported by the National Heart, Lung, and Blood Institute (contracts HHSN268201500003I, N01-HC-95159, N01-HC-95160, N01-HC-95161, N01-HC-95162, N01-HC-95163, N01-HC-95164, N01-HC-95165, N01-HC-95166, N01-HC-95167, N01-HC-95168 and N01-HC-95169), the National Center for Advancing Translational Sciences (grants UL1-TR-000040, UL1-TR-001079, and UL1-TR-001420), the National Heart, Lung, and Blood Institute at the National Institutes of Health (grants R01 HL071759, R01 HL1098433, R01 HL56984, T32 HL007901, R01 HL110068-03S1, and T32HL007901-19), and the National Institute of Minority Health and Health Disparities (grant P60 MD002249). We are grateful to the Carolina Population Center for training (T32 HD007168) and general (R24 HD050924) support. We thank the other investigators, staff, and participants of the MESA study for their valuable contributions. A full list of participating MESA investigators and institutions can be found at www.mesa-nhlbi.org. We thank Shannon Brines and Melissa Zagorski for creation of the geographic information systems variables and Amanda Dudley for support with license agreements and data acquisition. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institute of Health. Conflict of interest: none declared. Abbreviations BE built environment CI confidence interval MESA Multi-Ethnic Study of Atherosclerosis SD standard deviation SS Street Smart REFERENCES 1 Colten H, Altevogt BM, eds. Sleep Disorders and Sleep Deprivation: An Unmet Public Health Problem . Washington, DC: National Academies Press; 2006. 2 Centers for Disease Control and Prevention (CDC). Perceived insufficient rest or sleep among adults – United States, 2008. MMWR Morb Mortal Wkly Rep . 2009; 58( 42): 1175– 1179. PubMed  3 Centers for Disease Control and Prevention (CDC). Effect of short sleep duration on daily activites–United States, 2005–2009. MMWR Morb Mortal Wkly Rep . 2011; 60( 8): 239– 242. PubMed  4 Buxton OM, Marcelli E. Short and long sleep are positively associated with obesity, diabetes, hypertension, and cardiovascular disease among adults in the United States. Soc Sci Med . 2010; 71( 5): 1027– 1036. Google Scholar CrossRef Search ADS PubMed  5 Chambers EC, Schechter C, Tow A, et al.  . Household density and obesity in young black and white adults. Ethn Dis . 2010; 20( 4): 366– 369. Google Scholar PubMed  6 Johnson DA, Drake C, Joseph CL, et al.  . Influence of neighbourhood-level crowding on sleep-disordered breathing severity: mediation by body size. J Sleep Res . 2005; 24( 5): 559– 565. Google Scholar CrossRef Search ADS   7 Kredlow MA, Capozzoli MC, Hearon BA, et al.  . The effects of physical activity on sleep: a meta-analytic review. J Behav Med . 2015; 38( 3): 427– 449. Google Scholar CrossRef Search ADS PubMed  8 Muzet A. Environmental noise, sleep and health. Sleep Med Rev . 2007; 11( 2): 135– 142. Google Scholar CrossRef Search ADS PubMed  9 Zanobetti A, Redline S, Schwartz J, et al.  . Associations of PM10 with sleep and sleep-disordered breathing in adults from seven US urban areas. Am J Respir Crit Care Med . 2010; 182( 6): 819– 825. Google Scholar CrossRef Search ADS PubMed  10 Hale L, Do DP. Racial differences in self-reports of sleep duration in a population-based study. Sleep . 2007; 30( 9): 1096– 1103. Google Scholar CrossRef Search ADS PubMed  11 Geller AL. Smart growth: a prescription for livable cities. Am J Public Health . 2003; 93( 9): 1410– 1415. Google Scholar CrossRef Search ADS PubMed  12 Philbrook LE, El-Sheikh M. Associations between neighborhood context, physical activity, and sleep in adolescents. Sleep Health . 2016; 2( 3): 205– 210. Google Scholar CrossRef Search ADS   13 Astell-Burt T, Feng X, Kolt GS. Does access to neighbourhood green space promote a healthy duration of sleep? Novel findings from a cross-sectional study of 259 319 Australians. BMJ Open . 2013; 3( 8): e003094. Google Scholar CrossRef Search ADS PubMed  14 Bottino CJ, Rifas-Shiman SL, Kleinman KP, et al.  . The association of urbanicity with infant sleep duration. Health Place . 2012; 18( 5): 1000– 1005. Google Scholar CrossRef Search ADS PubMed  15 DeSantis A, Troxel WM, Beckman R, et al.  . Is the association between neighborhood characteristics and sleep quality mediated by psychological distress? An analysis of perceived and objective measures of 2 Pittsburgh neighborhoods. Sleep Health . 2016; 2( 4): 277– 282. Google Scholar CrossRef Search ADS PubMed  16 Krishnan V, Collop NA. Gender differences in sleep disorders. Curr Opin Pulm Med . 2006; 12( 6): 383– 389. Google Scholar CrossRef Search ADS PubMed  17 Bild DE, Bluemke DA, Burke GL, et al.  . Multi-Ethnic Study of Atherosclerosis: objectives and design. Am J Epidemiol . 2002; 156( 9): 871– 881. Google Scholar CrossRef Search ADS PubMed  18 Leslie E, Coffee N, Frank L, et al.  . Walkability of local communities: using geographic information systems to objectively assess relevant environmental attributes. Health Place . 2007; 13( 1): 111– 122. Google Scholar CrossRef Search ADS PubMed  19 Hirsch JA, Moore KA, Clarke PJ, et al.  . Changes in the built environment and changes in the amount of walking over time: longitudinal results from the Multi-Ethnic Study of Atherosclerosis. Am J Epidemiol . 2014; 180( 8): 799– 809. Google Scholar CrossRef Search ADS PubMed  20 Hirsch JA, Moore KA., Evenson KR, et al.  . Walk score® and transit score® and walking in the Multi-Ethnic Study of Atherosclerosis. Am J Prev Med . 2013; 45( 2): 158– 166. Google Scholar CrossRef Search ADS PubMed  21 Hirsch JA, Diez Roux AV, Moore KA, et al.  . Change in walking and body mass index following residential relocation: the Multi-Ethnic Study of Atherosclerosis. Am J Public Health . 2014; 104( 3): e49– e56. Google Scholar CrossRef Search ADS PubMed  22 Walk Score. Walk Score methodology. https://www.walkscore.com/methodology.shtml. Accessed February 24, 2016. 23 Hoehner CM, Schootman M. Concordance of commercial data sources for neighborhood-effects studies. J Urban Health . 2010; 87( 4): 713– 725. Google Scholar CrossRef Search ADS PubMed  24 Morgenthaler T, Alessi C, Friedman L, et al.  . Practice parameters for the use of actigraphy in the assessment of sleep and sleep disorders: an update for 2007. Sleep . 2007; 30( 4): 519– 529. Google Scholar CrossRef Search ADS PubMed  25 Chen X, Wang R, Zee P, et al.  . Racial/ethnic differences in sleep disturbances: the Multi-Ethnic Study of Atherosclerosis (MESA). Sleep . 2015; 38( 6): 877– 888. Google Scholar PubMed  26 Allen JP, Wilson V, eds. Assessing Alcohol Problems: A Guide for Clinicians and Researchers . 2nd ed. Bethesda, MD: National Institute on Alcohol Abuse and Alcoholism; 2003. 27 LaMonte MJ, Durstine JL, Addy CL, et al.  . Physical activity, physical fitness, and Framingham 10-year risk score: the cross-cultural activity participation study. J Cardiopulm Rehabil . 2001; 21( 2): 63– 70. Google Scholar CrossRef Search ADS PubMed  28 Radloff LS. The CES-D scale: a self-report depression scale for research in the general population. Appl Psychol Meas . 1977; 1( 3): 385– 401. Google Scholar CrossRef Search ADS   29 Mujahid MS, Diez Roux AV, Morenoff JD, et al.  . Assessing the measurement properties of neighborhood scales: from psychometrics to ecometrics. Am J Epidemiol . 2007; 165( 8): 858– 867. Google Scholar CrossRef Search ADS PubMed  30 Taheri S, Lin L, Austin D, et al.  . Short sleep duration is associated with reduced leptin, elevated ghrelin, and increased body mass index. PLoS Med . 2004; 1( 3): e62. Google Scholar CrossRef Search ADS PubMed  31 Vgontzas AN, Liao D, Bixler EO, et al.  . Insomnia with objective short sleep duration is associated with a high risk for hypertension. Sleep . 2009; 32( 4): 491– 497. Google Scholar CrossRef Search ADS PubMed  32 King CR, Knutson KL, Rathouz PJ, et al.  . Short sleep duration and incident coronary artery calcification. JAMA . 2008; 300( 24): 2859– 2866. Google Scholar CrossRef Search ADS PubMed  33 Knutson KL, Ryden AM, Mander BA, et al.  . Role of sleep duration and quality in the risk and severity of type 2 diabetes mellitus. Arch Intern Med . 2006; 166( 16): 1768– 1774. Google Scholar CrossRef Search ADS PubMed  34 Leng Y, Cappuccio FP, Wainwright NW, et al.  . Sleep duration and risk of fatal and nonfatal stroke: a prospective study and meta-analysis. Neurology . 2015; 84( 11): 1072– 1079. Google Scholar CrossRef Search ADS PubMed  35 Ikehara S, Iso H, Date C, et al.  . Association of sleep duration with mortality from cardiovascular disease and other causes for Japanese men and women: the JACC study. Sleep . 2009; 32( 3): 295– 301. Google Scholar CrossRef Search ADS PubMed  36 Desantis AS, Diez Roux AV, Moore K, et al.  . Associations of neighborhood characteristics with sleep timing and quality: the Multi-Ethnic Study of Atherosclerosis. Sleep . 2013; 36( 10): 1543– 1551. Google Scholar CrossRef Search ADS PubMed  37 Johnson DA, Brown DL, Morgenstern LB, et al.  . The association of neighborhood characteristics with sleep duration and daytime sleepiness. Sleep Health . 2015; 1( 3): 148– 155. Google Scholar CrossRef Search ADS PubMed  38 Marshall JD, Brauer M, Frank LD. Healthy neighborhoods: walkability and air pollution. Environ Health Perspect . 2009; 117( 11): 1752– 1759. Google Scholar CrossRef Search ADS PubMed  39 Hirsch JA, Moore KA, Barrientos-Gutierrez T, et al.  . Built environment change and change in BMI and waist circumference: Multi-Ethnic Study of Atherosclerosis. Obesity (Silver Spring) . 2014; 22( 11): 2450– 2457. Google Scholar CrossRef Search ADS PubMed  40 James P, Berrigan D, Hart JE, et al.  . Effects of buffer size and shape on associations between the built environment and energy balance. Health Place . 2014; 27: 162– 170. Google Scholar CrossRef Search ADS PubMed  41 Hirsch JA, Winters M, Clarke P, et al.  . Generating GPS activity spaces that shed light upon the mobility habits of older adults: a descriptive analysis. Int J Health Geogr . 2014; 13: 51. Google Scholar CrossRef Search ADS PubMed  42 Jean-Louis G, von Gizycki H, Zizi F, et al.  . The actigraph data analysis software: I. A novel approach to scoring and interpreting sleep-wake activity. Percept Mot Skills . 1997; 85( 1): 219– 226. Google Scholar CrossRef Search ADS PubMed  43 Leal C, Chaix B. The influence of geographic life environments on cardiometabolic risk factors: a systematic review, a methodological assessment and a research agenda. Obes Rev . 2011; 12( 3): 217– 230. Google Scholar CrossRef Search ADS PubMed  © The Author(s) 2018. Published by Oxford University Press on behalf of the Johns Hopkins Bloomberg School of Public Health. All rights reserved. For permissions, please e-mail: [email protected]. This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/about_us/legal/notices)

Lopez, David S;Liu, Lydia;Rimm, Eric B;Tsilidis, Konstantinos K;de Oliveira Otto, Marcia;Wang, Run;Canfield, Steven;Giovannucci, Edward

doi: 10.1093/aje/kwx304pmid: 29020139

Abstract Coffee intake is suggested to have a positive impact on chronic diseases, yet its role in urological diseases such as erectile dysfunction (ED) remains unclear. We investigated the association of coffee intake with incidence of ED by conducting the Health Professionals Follow-Up Study, a prospective analysis of 21,403 men aged 40–75 years old. Total, regular, and decaffeinated coffee intakes were self-reported on food frequency questionnaires. ED was assessed by mean values of questionnaires in 2000, 2004 and 2008. Multivariable adjusted Cox proportional hazards models were used to compute hazard ratios for patients with incident ED (n = 7,298). No significant differences were identified for patients with incident ED after comparing highest (≥4 cups/day) with lowest (0 cups/day) categories of total (hazard ratio (HR) = 1.00, 95% confidence interval (CI): 0.90, 1.11) and regular coffee intakes (HR = 1.00, 95% CI: 0.89, 1.13). When comparing the highest category with lowest category of decaffeinated coffee intake, we found a 37% increased risk of ED (HR = 1.37, 95% CI: 1.08, 1.73), with a significant trend (P trend = 0.02). Stratified analyses also showed an association among current smokers (P trend = 0.005). Overall, long-term coffee intake was not associated with risk of ED in a prospective cohort study. coffee, decaffeinated coffee, erectile dysfunction Coffee is a rich source of caffeine, antioxidants and anti-inflammatory compounds (1–4), and has been implicated to have a potentially beneficial role against chronic diseases. Yet, the role of coffee intake in urological diseases, specifically in erectile dysfunction (ED), remains undetermined (5–9). The prevalence of ED in American men aged 20 years or older is 18.4%, which suggests that more than 18 million people are affected (10, 11). Further, in a prospective study, 17.7% of men (aged 40–75 years) reported incident ED during a 14-year follow-up (12). Among older men, these numbers substantially increase, which affects overall quality of life (11, 13, 14). An inverse association between coffee intake and improved erectile function is biologically plausible. In addition to being a major source of polyphenols, coffee has the potential to increase testosterone levels (15–17), initiate a series of pharmacologic reactions that lead to the relaxation of the cavernous smooth muscle, and improve blood supply through penile arteries (18). In a previous cross-sectional analysis using the National Health and Nutrition and Examination Survey from 2001–2004, a nationally representative sample of the US noninstitutionalized male population, the equivalent of 2–3 cups of coffee per day was associated with a lower likelihood of ED (19). A further investigation of this association was conducted among men with comorbidities such as obesity, hypertension and diabetes, which are strong risk factors for ED. The findings remained the same among obese and hypertensive men, but not among men with diabetes. Given the inherent limitations of the cross-sectional study design and a relatively small total sample size (n = 3,724) in the National Health and Nutrition and Examination Survey analyses, we decided to investigate prospectively the association of caffeinated and decaffeinated beverages with risk of ED in the Health Professionals Follow-up Study (HPFS). In the HPFS, we used an analytic sample of 21,403 men, 7,298 patients with incident ED, and a follow-up period of 10 years. Additionally, we investigated these associations among men with lifestyle factors and comorbid conditions such as obesity, hypertension, diabetes, history of smoking, and marital status. METHODS The HPFS is an ongoing prospective cohort study that started in 1986 with 51,529 middle-aged US male health professionals which included dentists, pharmacists, optometrists, osteopaths physicians, podiatrists, and veterinarians. They are followed-up with mailed questionnaires every 2 years to update their information on lifestyle and health outcomes and every 4 years to assess usual diet (94% response rate) (20). The HPFS is approved by the Human Subjects Committee at the Harvard T.H. Chan School of Public Health. Outcome assessment On the 2000 questionnaire, HPFS participants were asked to rate their ability (without treatment) to have and maintain an erection sufficient for intercourse. Each question included a time grid with year/month increments (before 1986, 1986–1989, 1990–1994, 1995 or later, and in the past 3 months) to allow participants to report historically if and when erectile function changed. Participants were again asked to report their current function (without treatment) in 2004 and 2008. Response options on the 5-point scale included very poor, poor, fair, good, and very good. Only men who at baseline in 1998 were without prior diagnosis of ED; prostate, bladder, or testicular cancer; or CVD were included in our analyses. Date of diagnosis was defined as the date of return of the 2000 questionnaire, and we censored at first report of ED. Reports of poor or very poor erectile function in any of the periods from 2000 to 2008 were considered incident cases of ED (21, 22). Coffee intake assessment Dietary data, including coffee intake, was collected from HPFS participants using food frequency questionnaires (FFQ), which calculated the cumulative dose of coffee intake. FFQ were administered in 1998 and subsequently every 4 years through 2010. In the FFQ, participants reported how frequently they consumed a specified portion of a food item over the previous year, with 9 possible answers ranging from “never or less than once a month” to “6 or more times per day.” The FFQ included questions to assess the daily intake (number of cups) of regular and decaffeinated coffee. Previously, Feskanich et al. (23) conducted a validation study and found a high correlation (r = 0.93) between participant reports of coffee intake on the FFQ compared with 2 1-week diet records separated by 6 months. Statistical analysis Each participant contributed person-time from the date of return of the 1998 questionnaire to the date of first report of ED or the end of follow-up in 2010. Coffee intake data was categorized according to levels of regular, decaffeinated and total (regular and decaffeinated) coffee intake. Because cardiovascular diseases share many similar diet risk factors as ED, we stopped updating the coffee intake and diet intake information when any cardiovascular disease was diagnosed during follow-up. In order to avoid bias due to reverse causality, we conducted analyses using cumulative average intake with a lag of 4 years to avoid using data on coffee intake from completed FFQ immediately before ED diagnosis (24) or development of cardiovascular disease and prostate, bladder or testicular cancers during follow-up. Cox proportional hazards regressions were conducted to adjust for potential confounding by ED risk factors that were previously identified in this cohort and in other studies. We used left-truncated Cox proportional hazards for time-varying covariates, with a counting process data structure using age in months as the time scale. We also stratified based on calendar year to estimate the hazard ratios for coffee intake in relation to the risk of ED by using the lowest intake of quintile as the referent group (25, 26). Covariates were updated every other year. Multivariable models were adjusted for smoking (never, past, or current 1–14, 15–24, ≥25 cigarettes/day), BMI (weight in kilograms divided by height in meters squared; <25.0, 25.0–29.9, or ≥30.0), alcohol consumption (gram/day; 0, 0.1–4.9, 5.0–14.9, 15.0–29.9, or ≥30.0), total physical activity (metabolic equivalent task/week; quintiles), history of diabetes (yes vs. no), history of hypertension (yes vs. no), history of hypercholesterolemia (yes vs. no), history of cardiovascular disease (yes vs. no), energy intake (kcal/day; quintiles), marital status (married, divorced, separated, widowed, or never married), race (white, African-American, Asian-American, or other), Alternative Healthy Eating Index score (quintiles), difficulty of falling into sleep (yes vs. no), waking up during the night (yes vs. no), not feeling rested upon waking (yes vs. no), medication for lowering cholesterol (yes vs. no), medication for lowering blood pressure (yes vs. no) and sleep medicine (yes vs. no). To test for a linear trend across categories of coffee intake, we modeled intake as a continuous variable using the median intake for each category. We conducted further stratified analyses to determine whether comorbidities and lifestyle factors such as hypertension, BMI, smoking, alcohol consumption and marital status modify the association between coffee intake and ED. Multiplicative interaction terms were incorporated into the models and likelihood-ratio tests were used to test for interaction. All analyses were conducted with SAS software, version 9 (SAS Institute, Inc., Cary, North Carolina). All P values were 2-sided and P < 0.05 indicated statistical significance. RESULTS During 10 years of follow-up among the 21,403 participants (median age of 62 years), a total of 7,298 men (34%) reported incident ED. Baseline characteristics of the participants according to categories of coffee intake in 1998 are shown in Tables 1 and 2. Overall, in 1998, 65% of cohort participants reported drinking at least 1 cup of coffee (total coffee) per day, and 11% reported consuming 4 or more cups of coffee daily. Men with higher total coffee intake had higher physical activity level, were more likely to be current smokers, and consumed more alcohol (Table 1). Similar results were observed among men with higher intake of regular coffee; in addition, they were less likely to be hypertensive and had poorer diet quality as measured by Alternative Healthy Eating Index scores (Table 2). High decaffeinated coffee drinkers had higher BMI and higher cholesterol levels, consumed more alcohol, and were more likely to be current or former smokers and hypertensive, but had slightly higher diet quality scores (Table 3). Table 1. Baseline Age-Adjusted Characteristics by Total Coffee Consumption, Health Professionals Follow-up Study, 1998–2010 Characteristic  Coffee Intake, cups/day  None (n = 2,753)  0.01–0.50 (n = 2,506)  0.50–0.99 (n = 2,208)  1.00–3.99 (n = 11,576)  ≥4.00 (n = 2,360)  P for Trend  %  Mean (SD)  %  Mean (SD)  %  Mean (SD)  %  Mean (SD)  %  Mean (SD)  Age, years    60.3 (8.1)    61.3 (8.5)    62.6 (8.7)    62.5 (8.5)    61.2 (7.6)    BMIa    25.5 (3.7)    25.3 (3.4)    25.7 (3.3)    26.1 (3.4)    26.5 (3.5)  <0.0001  Physical activity, METs/week    37.9 (41.1)    37.9 (43.1)    38.2 (40.1)    36.4 (39.2)    33.8 (36.5)  <0.0001  Currently married  91    88    90    89    90    0.71  Smoking status               Never  79    68    61    47    34    <0.0001   Past  19    30    37    49    55    <0.0001   Current  2    2    2    5    12    <0.0001  Self-reported disease               Hypertension  13    14    16    15    12    0.59   High cholesterol  40    45    46    46    43    <0.0001   Difficulty falling asleepb  21    20    21    21    19    0.99   Waking during the nightb  55    54    58    59    58    <0.0001   Not feeling rested upon wakingb  23    23    23    23    24    0.99  Medication               Low blood pressure  19    23    25    24    20    0.005   Lower cholesterol  11    11    14    13    12    0.0002   Sleep medicationb  3    3    3    4    3    0.04  Alcohol consumption, g/day    4.7 (9.7)    8.4 (12.1)    9.7 (12.4)    13.0 (14.4)    13.8 (15.7)  <0.0001  Calories, kcal/day    1,972 (604)    1,967 (394)    1,944 (601)    2.011 (602)    2.092 (692)  <0.0001  AHEI score    49.7 (11.7)    52.7 (11.4)    52.2 (11.2)    51.3 (11)    49.8 (10.7)  0.72  Regular coffee    0 (0)    0.1 (0.1)    0.4 (0.3)    1.6 (1.0)    3.6 (1.5)  <0.0001  Decaffeinated coffee    0 (0)    0.1 (0.1)    0.3 (0.3)    0.7 (0.8)    1.3 (1.5)  <0.0001  Characteristic  Coffee Intake, cups/day  None (n = 2,753)  0.01–0.50 (n = 2,506)  0.50–0.99 (n = 2,208)  1.00–3.99 (n = 11,576)  ≥4.00 (n = 2,360)  P for Trend  %  Mean (SD)  %  Mean (SD)  %  Mean (SD)  %  Mean (SD)  %  Mean (SD)  Age, years    60.3 (8.1)    61.3 (8.5)    62.6 (8.7)    62.5 (8.5)    61.2 (7.6)    BMIa    25.5 (3.7)    25.3 (3.4)    25.7 (3.3)    26.1 (3.4)    26.5 (3.5)  <0.0001  Physical activity, METs/week    37.9 (41.1)    37.9 (43.1)    38.2 (40.1)    36.4 (39.2)    33.8 (36.5)  <0.0001  Currently married  91    88    90    89    90    0.71  Smoking status               Never  79    68    61    47    34    <0.0001   Past  19    30    37    49    55    <0.0001   Current  2    2    2    5    12    <0.0001  Self-reported disease               Hypertension  13    14    16    15    12    0.59   High cholesterol  40    45    46    46    43    <0.0001   Difficulty falling asleepb  21    20    21    21    19    0.99   Waking during the nightb  55    54    58    59    58    <0.0001   Not feeling rested upon wakingb  23    23    23    23    24    0.99  Medication               Low blood pressure  19    23    25    24    20    0.005   Lower cholesterol  11    11    14    13    12    0.0002   Sleep medicationb  3    3    3    4    3    0.04  Alcohol consumption, g/day    4.7 (9.7)    8.4 (12.1)    9.7 (12.4)    13.0 (14.4)    13.8 (15.7)  <0.0001  Calories, kcal/day    1,972 (604)    1,967 (394)    1,944 (601)    2.011 (602)    2.092 (692)  <0.0001  AHEI score    49.7 (11.7)    52.7 (11.4)    52.2 (11.2)    51.3 (11)    49.8 (10.7)  0.72  Regular coffee    0 (0)    0.1 (0.1)    0.4 (0.3)    1.6 (1.0)    3.6 (1.5)  <0.0001  Decaffeinated coffee    0 (0)    0.1 (0.1)    0.3 (0.3)    0.7 (0.8)    1.3 (1.5)  <0.0001  Abbreviations: AHEI, Alternative Healthy Eating Index; BMI, body mass index; MET, metabolic equivalent of task; SD, standard deviation. a Weight (kg)/height (m)2. b Variables were ascertained in 2004 questionnaire. Table 1. Baseline Age-Adjusted Characteristics by Total Coffee Consumption, Health Professionals Follow-up Study, 1998–2010 Characteristic  Coffee Intake, cups/day  None (n = 2,753)  0.01–0.50 (n = 2,506)  0.50–0.99 (n = 2,208)  1.00–3.99 (n = 11,576)  ≥4.00 (n = 2,360)  P for Trend  %  Mean (SD)  %  Mean (SD)  %  Mean (SD)  %  Mean (SD)  %  Mean (SD)  Age, years    60.3 (8.1)    61.3 (8.5)    62.6 (8.7)    62.5 (8.5)    61.2 (7.6)    BMIa    25.5 (3.7)    25.3 (3.4)    25.7 (3.3)    26.1 (3.4)    26.5 (3.5)  <0.0001  Physical activity, METs/week    37.9 (41.1)    37.9 (43.1)    38.2 (40.1)    36.4 (39.2)    33.8 (36.5)  <0.0001  Currently married  91    88    90    89    90    0.71  Smoking status               Never  79    68    61    47    34    <0.0001   Past  19    30    37    49    55    <0.0001   Current  2    2    2    5    12    <0.0001  Self-reported disease               Hypertension  13    14    16    15    12    0.59   High cholesterol  40    45    46    46    43    <0.0001   Difficulty falling asleepb  21    20    21    21    19    0.99   Waking during the nightb  55    54    58    59    58    <0.0001   Not feeling rested upon wakingb  23    23    23    23    24    0.99  Medication               Low blood pressure  19    23    25    24    20    0.005   Lower cholesterol  11    11    14    13    12    0.0002   Sleep medicationb  3    3    3    4    3    0.04  Alcohol consumption, g/day    4.7 (9.7)    8.4 (12.1)    9.7 (12.4)    13.0 (14.4)    13.8 (15.7)  <0.0001  Calories, kcal/day    1,972 (604)    1,967 (394)    1,944 (601)    2.011 (602)    2.092 (692)  <0.0001  AHEI score    49.7 (11.7)    52.7 (11.4)    52.2 (11.2)    51.3 (11)    49.8 (10.7)  0.72  Regular coffee    0 (0)    0.1 (0.1)    0.4 (0.3)    1.6 (1.0)    3.6 (1.5)  <0.0001  Decaffeinated coffee    0 (0)    0.1 (0.1)    0.3 (0.3)    0.7 (0.8)    1.3 (1.5)  <0.0001  Characteristic  Coffee Intake, cups/day  None (n = 2,753)  0.01–0.50 (n = 2,506)  0.50–0.99 (n = 2,208)  1.00–3.99 (n = 11,576)  ≥4.00 (n = 2,360)  P for Trend  %  Mean (SD)  %  Mean (SD)  %  Mean (SD)  %  Mean (SD)  %  Mean (SD)  Age, years    60.3 (8.1)    61.3 (8.5)    62.6 (8.7)    62.5 (8.5)    61.2 (7.6)    BMIa    25.5 (3.7)    25.3 (3.4)    25.7 (3.3)    26.1 (3.4)    26.5 (3.5)  <0.0001  Physical activity, METs/week    37.9 (41.1)    37.9 (43.1)    38.2 (40.1)    36.4 (39.2)    33.8 (36.5)  <0.0001  Currently married  91    88    90    89    90    0.71  Smoking status               Never  79    68    61    47    34    <0.0001   Past  19    30    37    49    55    <0.0001   Current  2    2    2    5    12    <0.0001  Self-reported disease               Hypertension  13    14    16    15    12    0.59   High cholesterol  40    45    46    46    43    <0.0001   Difficulty falling asleepb  21    20    21    21    19    0.99   Waking during the nightb  55    54    58    59    58    <0.0001   Not feeling rested upon wakingb  23    23    23    23    24    0.99  Medication               Low blood pressure  19    23    25    24    20    0.005   Lower cholesterol  11    11    14    13    12    0.0002   Sleep medicationb  3    3    3    4    3    0.04  Alcohol consumption, g/day    4.7 (9.7)    8.4 (12.1)    9.7 (12.4)    13.0 (14.4)    13.8 (15.7)  <0.0001  Calories, kcal/day    1,972 (604)    1,967 (394)    1,944 (601)    2.011 (602)    2.092 (692)  <0.0001  AHEI score    49.7 (11.7)    52.7 (11.4)    52.2 (11.2)    51.3 (11)    49.8 (10.7)  0.72  Regular coffee    0 (0)    0.1 (0.1)    0.4 (0.3)    1.6 (1.0)    3.6 (1.5)  <0.0001  Decaffeinated coffee    0 (0)    0.1 (0.1)    0.3 (0.3)    0.7 (0.8)    1.3 (1.5)  <0.0001  Abbreviations: AHEI, Alternative Healthy Eating Index; BMI, body mass index; MET, metabolic equivalent of task; SD, standard deviation. a Weight (kg)/height (m)2. b Variables were ascertained in 2004 questionnaire. Table 2. Baseline Age-Adjusted Characteristics by Regular Coffee Consumption, Health Professionals Follow-up Study, 1998–2010 Characteristics  Coffee Intake, cups/day  None (n = 4,058)  0.01–0.50 (n = 4,112)  0.50–0.99 (n = 2,732)  1.00–3.99 (n = 9,167)  ≥4.00 (n = 1,334)  P for Trend  %  Mean (SD)  %  Mean (SD)  %  Mean (SD)  %  Mean (SD)  %  Mean (SD)  Age, years    61.7 (8.7)    62.3 (8.5)    62.5 (8.6)    62.0 (8.3)    60.3 (7.2)    BMIa    25.6 (3.5)    25.6 (3.3)    26.0 (3.5)    26.2 (3.4)    26.4 (3.4)  <0.0001  Physical activity, METs/week    38.3 (42.8)    37.4 (39.3)    36.6 (38.9)    36.2 (39.3)    32.2 (34.9)  <0.0001  Currently married  91    90    89    89    88    <0.0001  Smoking status               Never  72    61    52    45    34    <0.0001   Past  26    36    45    49    52    <0.0001   Current  2    3    4    5    14    <0.0001  Self-reported disease               Hypertension  15    16    17    14    10    0.0002   High cholesterol  43    47    47    45    42    0.47   Difficulty falling asleepb  22    22    24    20    21    0.09   Waking during the nightb  56    57    59    59    56    0.02   Not feeling rested upon wakingb  23    24    23    23    22    0.29  Medication               Low blood pressure  22    24    25    22    18    0.10   Lower cholesterol  12    13    14    13    11    0.97   Sleep medicationb  3    3    3    4    3    0.40  Alcohol consumption, g/day    6.2 (11.1)    10.0 (13.3)    10.9 (12.6)    13.5 (14.7)    13.9 (15.7)  <0.0001  Calories, kcal/day    1,944 (587)    1,976 (605)    1,980 (607)    2,033 (606)    2,100 (642)  <0.0001  AHEI score    50.7 (11.7)    52.3 (11.2)    52.2 (10.9)    50.9 (11)    48.8 (10.6)  <0.0001  Total coffee    0.3 (0.7)    0.9 (1.0)    1.5 (0.9)    2.7 (1.0)    5.0 (0.9)  <0.0001  Decaffeinated coffee    0.3 (0.7)    0.7 (1.0)    0.8 (0.9)    0.6 (0.8)    0.3 (0.6)  <0.0001  Characteristics  Coffee Intake, cups/day  None (n = 4,058)  0.01–0.50 (n = 4,112)  0.50–0.99 (n = 2,732)  1.00–3.99 (n = 9,167)  ≥4.00 (n = 1,334)  P for Trend  %  Mean (SD)  %  Mean (SD)  %  Mean (SD)  %  Mean (SD)  %  Mean (SD)  Age, years    61.7 (8.7)    62.3 (8.5)    62.5 (8.6)    62.0 (8.3)    60.3 (7.2)    BMIa    25.6 (3.5)    25.6 (3.3)    26.0 (3.5)    26.2 (3.4)    26.4 (3.4)  <0.0001  Physical activity, METs/week    38.3 (42.8)    37.4 (39.3)    36.6 (38.9)    36.2 (39.3)    32.2 (34.9)  <0.0001  Currently married  91    90    89    89    88    <0.0001  Smoking status               Never  72    61    52    45    34    <0.0001   Past  26    36    45    49    52    <0.0001   Current  2    3    4    5    14    <0.0001  Self-reported disease               Hypertension  15    16    17    14    10    0.0002   High cholesterol  43    47    47    45    42    0.47   Difficulty falling asleepb  22    22    24    20    21    0.09   Waking during the nightb  56    57    59    59    56    0.02   Not feeling rested upon wakingb  23    24    23    23    22    0.29  Medication               Low blood pressure  22    24    25    22    18    0.10   Lower cholesterol  12    13    14    13    11    0.97   Sleep medicationb  3    3    3    4    3    0.40  Alcohol consumption, g/day    6.2 (11.1)    10.0 (13.3)    10.9 (12.6)    13.5 (14.7)    13.9 (15.7)  <0.0001  Calories, kcal/day    1,944 (587)    1,976 (605)    1,980 (607)    2,033 (606)    2,100 (642)  <0.0001  AHEI score    50.7 (11.7)    52.3 (11.2)    52.2 (10.9)    50.9 (11)    48.8 (10.6)  <0.0001  Total coffee    0.3 (0.7)    0.9 (1.0)    1.5 (0.9)    2.7 (1.0)    5.0 (0.9)  <0.0001  Decaffeinated coffee    0.3 (0.7)    0.7 (1.0)    0.8 (0.9)    0.6 (0.8)    0.3 (0.6)  <0.0001  Abbreviations: AHEI, Alternative Healthy Eating Index; BMI, body mass index; MET; metabolic equivalent of task; SD, standard deviation. a Weight (kg)/height (m)2. b Variables were ascertained in the 2004 questionnaire. Table 2. Baseline Age-Adjusted Characteristics by Regular Coffee Consumption, Health Professionals Follow-up Study, 1998–2010 Characteristics  Coffee Intake, cups/day  None (n = 4,058)  0.01–0.50 (n = 4,112)  0.50–0.99 (n = 2,732)  1.00–3.99 (n = 9,167)  ≥4.00 (n = 1,334)  P for Trend  %  Mean (SD)  %  Mean (SD)  %  Mean (SD)  %  Mean (SD)  %  Mean (SD)  Age, years    61.7 (8.7)    62.3 (8.5)    62.5 (8.6)    62.0 (8.3)    60.3 (7.2)    BMIa    25.6 (3.5)    25.6 (3.3)    26.0 (3.5)    26.2 (3.4)    26.4 (3.4)  <0.0001  Physical activity, METs/week    38.3 (42.8)    37.4 (39.3)    36.6 (38.9)    36.2 (39.3)    32.2 (34.9)  <0.0001  Currently married  91    90    89    89    88    <0.0001  Smoking status               Never  72    61    52    45    34    <0.0001   Past  26    36    45    49    52    <0.0001   Current  2    3    4    5    14    <0.0001  Self-reported disease               Hypertension  15    16    17    14    10    0.0002   High cholesterol  43    47    47    45    42    0.47   Difficulty falling asleepb  22    22    24    20    21    0.09   Waking during the nightb  56    57    59    59    56    0.02   Not feeling rested upon wakingb  23    24    23    23    22    0.29  Medication               Low blood pressure  22    24    25    22    18    0.10   Lower cholesterol  12    13    14    13    11    0.97   Sleep medicationb  3    3    3    4    3    0.40  Alcohol consumption, g/day    6.2 (11.1)    10.0 (13.3)    10.9 (12.6)    13.5 (14.7)    13.9 (15.7)  <0.0001  Calories, kcal/day    1,944 (587)    1,976 (605)    1,980 (607)    2,033 (606)    2,100 (642)  <0.0001  AHEI score    50.7 (11.7)    52.3 (11.2)    52.2 (10.9)    50.9 (11)    48.8 (10.6)  <0.0001  Total coffee    0.3 (0.7)    0.9 (1.0)    1.5 (0.9)    2.7 (1.0)    5.0 (0.9)  <0.0001  Decaffeinated coffee    0.3 (0.7)    0.7 (1.0)    0.8 (0.9)    0.6 (0.8)    0.3 (0.6)  <0.0001  Characteristics  Coffee Intake, cups/day  None (n = 4,058)  0.01–0.50 (n = 4,112)  0.50–0.99 (n = 2,732)  1.00–3.99 (n = 9,167)  ≥4.00 (n = 1,334)  P for Trend  %  Mean (SD)  %  Mean (SD)  %  Mean (SD)  %  Mean (SD)  %  Mean (SD)  Age, years    61.7 (8.7)    62.3 (8.5)    62.5 (8.6)    62.0 (8.3)    60.3 (7.2)    BMIa    25.6 (3.5)    25.6 (3.3)    26.0 (3.5)    26.2 (3.4)    26.4 (3.4)  <0.0001  Physical activity, METs/week    38.3 (42.8)    37.4 (39.3)    36.6 (38.9)    36.2 (39.3)    32.2 (34.9)  <0.0001  Currently married  91    90    89    89    88    <0.0001  Smoking status               Never  72    61    52    45    34    <0.0001   Past  26    36    45    49    52    <0.0001   Current  2    3    4    5    14    <0.0001  Self-reported disease               Hypertension  15    16    17    14    10    0.0002   High cholesterol  43    47    47    45    42    0.47   Difficulty falling asleepb  22    22    24    20    21    0.09   Waking during the nightb  56    57    59    59    56    0.02   Not feeling rested upon wakingb  23    24    23    23    22    0.29  Medication               Low blood pressure  22    24    25    22    18    0.10   Lower cholesterol  12    13    14    13    11    0.97   Sleep medicationb  3    3    3    4    3    0.40  Alcohol consumption, g/day    6.2 (11.1)    10.0 (13.3)    10.9 (12.6)    13.5 (14.7)    13.9 (15.7)  <0.0001  Calories, kcal/day    1,944 (587)    1,976 (605)    1,980 (607)    2,033 (606)    2,100 (642)  <0.0001  AHEI score    50.7 (11.7)    52.3 (11.2)    52.2 (10.9)    50.9 (11)    48.8 (10.6)  <0.0001  Total coffee    0.3 (0.7)    0.9 (1.0)    1.5 (0.9)    2.7 (1.0)    5.0 (0.9)  <0.0001  Decaffeinated coffee    0.3 (0.7)    0.7 (1.0)    0.8 (0.9)    0.6 (0.8)    0.3 (0.6)  <0.0001  Abbreviations: AHEI, Alternative Healthy Eating Index; BMI, body mass index; MET; metabolic equivalent of task; SD, standard deviation. a Weight (kg)/height (m)2. b Variables were ascertained in the 2004 questionnaire. Table 3. Baseline Age-Adjusted Characteristics by Decaffeinated Coffee Consumption, Health Professionals Follow-up Study, 1998–2010 Characteristic  Coffee Intake, cups/day  None (n = 6,787)  0.01–0.50 (n = 7,280)  0.50–0.99 (n = 3,041)  1.00–3.99 (n = 4,088)  ≥4.00 (n = 207)  P for Trend  %  Mean (SD)  %  Mean (SD)  %  Mean (SD)  %  Mean (SD)  %  Mean (SD)  Age, years    61.2 (8.3)    61.5 (8.3)    62.6 (8.5)    63.4 (8.3)    63.3 (8.3)    BMIa    25.9 (3.6)    25.8 (3.3)    26.0 (3.3)    26.3 (3.4)    26.8 (3.5)  <0.0001  Physical activity, METs/week    36.1 (40)    37.2 (40.8)    36.4 (36.9)    36.6 (39.1)    34.6 (38.9)  0.60  Currently married  88    89    90    92    95    <0.0001  Smoking status               Never  59    56    51    41    31    <0.0001   Past  35    41    45    53    59    <0.0001   Current  6    3    4    5    11    0.90  Self-reported disease               Hypertension  13    13    16    17    20    <0.0001   High cholesterol  41    45    47    49    51    <0.0001   Difficulty falling asleepb  19    22    22    21    23    0.03   Waking during the nightb  53    58    62    62    62    <0.0001   Not feeling rested upon wakingb  21    23    25    24    17    0.008  Medication               Low blood pressure  20    22    26    26    25    <0.0001   Lower cholesterol  11    12    14    15    12    <0.0001   Sleep medicationb  3    3    3    4    3    0.0007  Alcohol consumption, g/day    9.8 (14.3)    11.1 (13.1)    11.8 (13.4)    12.7 (14.7)    14.6 (17.2)  <0.0001  Calories, kcal/day    1,992 (614)    2,014 (602)    2,000 (611)    1,999 (595)    2,110 (604)  0.40  AHEI score    49.3 (11.5)    52.2 (10.9)    52.0 (11)    51.8 (10.8)    51.0 (10.4)  <0.0001  Regular coffee    1.3 (1.6)    1.4 (1.3)    1.3 (1.2)    1.2 (1.2)    1.1 (1.2)  <0.0001  Decaffeinated coffee    1.3 (1.6)    1.6 (1.3)    2.0 (1.2)    3.0 (1.3)    5.6 (1.3)  <0.0001  Characteristic  Coffee Intake, cups/day  None (n = 6,787)  0.01–0.50 (n = 7,280)  0.50–0.99 (n = 3,041)  1.00–3.99 (n = 4,088)  ≥4.00 (n = 207)  P for Trend  %  Mean (SD)  %  Mean (SD)  %  Mean (SD)  %  Mean (SD)  %  Mean (SD)  Age, years    61.2 (8.3)    61.5 (8.3)    62.6 (8.5)    63.4 (8.3)    63.3 (8.3)    BMIa    25.9 (3.6)    25.8 (3.3)    26.0 (3.3)    26.3 (3.4)    26.8 (3.5)  <0.0001  Physical activity, METs/week    36.1 (40)    37.2 (40.8)    36.4 (36.9)    36.6 (39.1)    34.6 (38.9)  0.60  Currently married  88    89    90    92    95    <0.0001  Smoking status               Never  59    56    51    41    31    <0.0001   Past  35    41    45    53    59    <0.0001   Current  6    3    4    5    11    0.90  Self-reported disease               Hypertension  13    13    16    17    20    <0.0001   High cholesterol  41    45    47    49    51    <0.0001   Difficulty falling asleepb  19    22    22    21    23    0.03   Waking during the nightb  53    58    62    62    62    <0.0001   Not feeling rested upon wakingb  21    23    25    24    17    0.008  Medication               Low blood pressure  20    22    26    26    25    <0.0001   Lower cholesterol  11    12    14    15    12    <0.0001   Sleep medicationb  3    3    3    4    3    0.0007  Alcohol consumption, g/day    9.8 (14.3)    11.1 (13.1)    11.8 (13.4)    12.7 (14.7)    14.6 (17.2)  <0.0001  Calories, kcal/day    1,992 (614)    2,014 (602)    2,000 (611)    1,999 (595)    2,110 (604)  0.40  AHEI score    49.3 (11.5)    52.2 (10.9)    52.0 (11)    51.8 (10.8)    51.0 (10.4)  <0.0001  Regular coffee    1.3 (1.6)    1.4 (1.3)    1.3 (1.2)    1.2 (1.2)    1.1 (1.2)  <0.0001  Decaffeinated coffee    1.3 (1.6)    1.6 (1.3)    2.0 (1.2)    3.0 (1.3)    5.6 (1.3)  <0.0001  Abbreviations: AHEI, Alternative Healthy Eating Index; BMI, body mass index; MET; metabolic equivalent of task; SD, standard deviation. a Weight (kg)/height (m)2. b Variables were ascertained in the 2004 questionnaire. Table 3. Baseline Age-Adjusted Characteristics by Decaffeinated Coffee Consumption, Health Professionals Follow-up Study, 1998–2010 Characteristic  Coffee Intake, cups/day  None (n = 6,787)  0.01–0.50 (n = 7,280)  0.50–0.99 (n = 3,041)  1.00–3.99 (n = 4,088)  ≥4.00 (n = 207)  P for Trend  %  Mean (SD)  %  Mean (SD)  %  Mean (SD)  %  Mean (SD)  %  Mean (SD)  Age, years    61.2 (8.3)    61.5 (8.3)    62.6 (8.5)    63.4 (8.3)    63.3 (8.3)    BMIa    25.9 (3.6)    25.8 (3.3)    26.0 (3.3)    26.3 (3.4)    26.8 (3.5)  <0.0001  Physical activity, METs/week    36.1 (40)    37.2 (40.8)    36.4 (36.9)    36.6 (39.1)    34.6 (38.9)  0.60  Currently married  88    89    90    92    95    <0.0001  Smoking status               Never  59    56    51    41    31    <0.0001   Past  35    41    45    53    59    <0.0001   Current  6    3    4    5    11    0.90  Self-reported disease               Hypertension  13    13    16    17    20    <0.0001   High cholesterol  41    45    47    49    51    <0.0001   Difficulty falling asleepb  19    22    22    21    23    0.03   Waking during the nightb  53    58    62    62    62    <0.0001   Not feeling rested upon wakingb  21    23    25    24    17    0.008  Medication               Low blood pressure  20    22    26    26    25    <0.0001   Lower cholesterol  11    12    14    15    12    <0.0001   Sleep medicationb  3    3    3    4    3    0.0007  Alcohol consumption, g/day    9.8 (14.3)    11.1 (13.1)    11.8 (13.4)    12.7 (14.7)    14.6 (17.2)  <0.0001  Calories, kcal/day    1,992 (614)    2,014 (602)    2,000 (611)    1,999 (595)    2,110 (604)  0.40  AHEI score    49.3 (11.5)    52.2 (10.9)    52.0 (11)    51.8 (10.8)    51.0 (10.4)  <0.0001  Regular coffee    1.3 (1.6)    1.4 (1.3)    1.3 (1.2)    1.2 (1.2)    1.1 (1.2)  <0.0001  Decaffeinated coffee    1.3 (1.6)    1.6 (1.3)    2.0 (1.2)    3.0 (1.3)    5.6 (1.3)  <0.0001  Characteristic  Coffee Intake, cups/day  None (n = 6,787)  0.01–0.50 (n = 7,280)  0.50–0.99 (n = 3,041)  1.00–3.99 (n = 4,088)  ≥4.00 (n = 207)  P for Trend  %  Mean (SD)  %  Mean (SD)  %  Mean (SD)  %  Mean (SD)  %  Mean (SD)  Age, years    61.2 (8.3)    61.5 (8.3)    62.6 (8.5)    63.4 (8.3)    63.3 (8.3)    BMIa    25.9 (3.6)    25.8 (3.3)    26.0 (3.3)    26.3 (3.4)    26.8 (3.5)  <0.0001  Physical activity, METs/week    36.1 (40)    37.2 (40.8)    36.4 (36.9)    36.6 (39.1)    34.6 (38.9)  0.60  Currently married  88    89    90    92    95    <0.0001  Smoking status               Never  59    56    51    41    31    <0.0001   Past  35    41    45    53    59    <0.0001   Current  6    3    4    5    11    0.90  Self-reported disease               Hypertension  13    13    16    17    20    <0.0001   High cholesterol  41    45    47    49    51    <0.0001   Difficulty falling asleepb  19    22    22    21    23    0.03   Waking during the nightb  53    58    62    62    62    <0.0001   Not feeling rested upon wakingb  21    23    25    24    17    0.008  Medication               Low blood pressure  20    22    26    26    25    <0.0001   Lower cholesterol  11    12    14    15    12    <0.0001   Sleep medicationb  3    3    3    4    3    0.0007  Alcohol consumption, g/day    9.8 (14.3)    11.1 (13.1)    11.8 (13.4)    12.7 (14.7)    14.6 (17.2)  <0.0001  Calories, kcal/day    1,992 (614)    2,014 (602)    2,000 (611)    1,999 (595)    2,110 (604)  0.40  AHEI score    49.3 (11.5)    52.2 (10.9)    52.0 (11)    51.8 (10.8)    51.0 (10.4)  <0.0001  Regular coffee    1.3 (1.6)    1.4 (1.3)    1.3 (1.2)    1.2 (1.2)    1.1 (1.2)  <0.0001  Decaffeinated coffee    1.3 (1.6)    1.6 (1.3)    2.0 (1.2)    3.0 (1.3)    5.6 (1.3)  <0.0001  Abbreviations: AHEI, Alternative Healthy Eating Index; BMI, body mass index; MET; metabolic equivalent of task; SD, standard deviation. a Weight (kg)/height (m)2. b Variables were ascertained in the 2004 questionnaire. Age-adjusted associations tended to show positive trends, but after adjusting for potential confounders including history of cardiovascular disease and lifestyle risk factors, we did not find any statistically significant association between greater intakes of total coffee (P for trend = 0.37) or regular coffee (P for trend = 0.90) and ED (Table 4). No significant differences were identified after comparing the highest (≥4 cups/day) and lowest (0 cups/day) categories of total coffee (hazard ratio (HR) = 1.00, 95% confidence interval (CI): 0.90, 1.11) and regular coffee (HR = 1.00, 95% CI: 0.89, 1.13) intakes. However, in comparing highest (≥4 cups/day) with lowest (0 cups/day) categories of decaffeinated coffee, there was a 37% increased risk of ED (HR = 1.37, 95% CI: 1.08, 1.73), and the trend was significant (P for trend = 0.02). Table 4. Multivariable Associations of Cumulative Average Intakes of Total, Regular, and Decaffeinated Coffee With Erectile Dysfunction (n = 21,403), Health Professionals Follow-up Study, 1998–2010 Coffee Type and Consumption  No. of Person-Years  No. of ED Cases  Model 1a  Model 2b  HR  95% CI  P for Trend  HR  95% CI  P for Trend  Total coffee, cups/day          0.004      0.37   Nonec  24,630  829  1.00  Referent    1.00  Referent     0.01–0.50  23,167  764  0.92  0.83, 1.01    0.93  0.84, 1.03     0.50–0.99  19,705  738  0.94  0.85, 1.03    0.92  0.83, 1.02     1.00–3.99  103,629  4,242  1.04  0.96, 1.12    0.99  0.92, 1.08     ≥4.00  18,861  725  1.08  0.97, 1.19    1.00  0.90, 1.11    Regular coffee, cups/day          0.09      0.90   Nonec  34,791  1,243  1.00  Referent    1.00  Referent     0.01–0.50  36,999  1,424  1.04  0.97, 1.12    1.05  0.97, 1.13     0.50–0.99  23,978  963  1.05  0.96, 1.14    1.03  0.94, 1.12     1.00–3.99  83,857  3,279  1.06  0.99, 1.13    1.03  0.96, 1.10     ≥4.00  10,365  359  1.07  0.95, 1.20    1.00  0.89, 1.13    Decaffeinated coffee, cups/day          0.0005      0.02   Nonec  58,716  2,069  1.00  Referent    1.00  Referent     0.01–0.50  69,458  2,515  0.99  0.93, 1.05    1.00  0.94, 1.06     0.50–0.99  26,540  1,074  1.03  0.96, 1.11    1.02  0.95, 1.10     1.00–3.99  34,000  1,566  1.09  1.02, 1.17    1.06  0.99, 1.13     ≥4.00  1,278  74  1.54  1.22, 1.95    1.37  1.08, 1.73    Coffee Type and Consumption  No. of Person-Years  No. of ED Cases  Model 1a  Model 2b  HR  95% CI  P for Trend  HR  95% CI  P for Trend  Total coffee, cups/day          0.004      0.37   Nonec  24,630  829  1.00  Referent    1.00  Referent     0.01–0.50  23,167  764  0.92  0.83, 1.01    0.93  0.84, 1.03     0.50–0.99  19,705  738  0.94  0.85, 1.03    0.92  0.83, 1.02     1.00–3.99  103,629  4,242  1.04  0.96, 1.12    0.99  0.92, 1.08     ≥4.00  18,861  725  1.08  0.97, 1.19    1.00  0.90, 1.11    Regular coffee, cups/day          0.09      0.90   Nonec  34,791  1,243  1.00  Referent    1.00  Referent     0.01–0.50  36,999  1,424  1.04  0.97, 1.12    1.05  0.97, 1.13     0.50–0.99  23,978  963  1.05  0.96, 1.14    1.03  0.94, 1.12     1.00–3.99  83,857  3,279  1.06  0.99, 1.13    1.03  0.96, 1.10     ≥4.00  10,365  359  1.07  0.95, 1.20    1.00  0.89, 1.13    Decaffeinated coffee, cups/day          0.0005      0.02   Nonec  58,716  2,069  1.00  Referent    1.00  Referent     0.01–0.50  69,458  2,515  0.99  0.93, 1.05    1.00  0.94, 1.06     0.50–0.99  26,540  1,074  1.03  0.96, 1.11    1.02  0.95, 1.10     1.00–3.99  34,000  1,566  1.09  1.02, 1.17    1.06  0.99, 1.13     ≥4.00  1,278  74  1.54  1.22, 1.95    1.37  1.08, 1.73    Abbreviations: CI, confidence interval; ED, erectile dysfunction; HR, hazard ratio. a Model 1 was adjusted for age in months and calendar time. b Model 2 was adjusted for the variables in model 1 and smoking (never, past, or current, categorized as 1–14, 15–24, ≥25 cigarettes/day), body mass index (weight in kilometers divided by height in meters squared; <25.0, 25.0–29.9, or ≥30.0), alcohol consumption (grams/day; 0, 0.1–4.9, 5.0–14.9, 15.0–29.9, or ≥30.0), total physical activity (metabolic equivalent of tasks per week; quintiles), history of diabetes (yes vs. no), history of hypertension (yes vs. no), history of hypercholesterolemia (yes vs. no), history of cardiovascular disease (yes vs. no), energy intake (kcal/day; quintiles), Alternative Healthy Eating Index score (quintiles), marital status (married, divorced, separated, widowed, or never married), race (white, African-American, Asian-American, or other), difficulty of falling into sleep (yes vs. no), waking during the night (yes vs. no), not feeling rested upon waking (yes vs. no), and use of cholesterol-lowering, blood pressure–lowering, or sleep-enhancing medication (yes vs. no). c Included men who drank no regular or decaffeinated coffee. Table 4. Multivariable Associations of Cumulative Average Intakes of Total, Regular, and Decaffeinated Coffee With Erectile Dysfunction (n = 21,403), Health Professionals Follow-up Study, 1998–2010 Coffee Type and Consumption  No. of Person-Years  No. of ED Cases  Model 1a  Model 2b  HR  95% CI  P for Trend  HR  95% CI  P for Trend  Total coffee, cups/day          0.004      0.37   Nonec  24,630  829  1.00  Referent    1.00  Referent     0.01–0.50  23,167  764  0.92  0.83, 1.01    0.93  0.84, 1.03     0.50–0.99  19,705  738  0.94  0.85, 1.03    0.92  0.83, 1.02     1.00–3.99  103,629  4,242  1.04  0.96, 1.12    0.99  0.92, 1.08     ≥4.00  18,861  725  1.08  0.97, 1.19    1.00  0.90, 1.11    Regular coffee, cups/day          0.09      0.90   Nonec  34,791  1,243  1.00  Referent    1.00  Referent     0.01–0.50  36,999  1,424  1.04  0.97, 1.12    1.05  0.97, 1.13     0.50–0.99  23,978  963  1.05  0.96, 1.14    1.03  0.94, 1.12     1.00–3.99  83,857  3,279  1.06  0.99, 1.13    1.03  0.96, 1.10     ≥4.00  10,365  359  1.07  0.95, 1.20    1.00  0.89, 1.13    Decaffeinated coffee, cups/day          0.0005      0.02   Nonec  58,716  2,069  1.00  Referent    1.00  Referent     0.01–0.50  69,458  2,515  0.99  0.93, 1.05    1.00  0.94, 1.06     0.50–0.99  26,540  1,074  1.03  0.96, 1.11    1.02  0.95, 1.10     1.00–3.99  34,000  1,566  1.09  1.02, 1.17    1.06  0.99, 1.13     ≥4.00  1,278  74  1.54  1.22, 1.95    1.37  1.08, 1.73    Coffee Type and Consumption  No. of Person-Years  No. of ED Cases  Model 1a  Model 2b  HR  95% CI  P for Trend  HR  95% CI  P for Trend  Total coffee, cups/day          0.004      0.37   Nonec  24,630  829  1.00  Referent    1.00  Referent     0.01–0.50  23,167  764  0.92  0.83, 1.01    0.93  0.84, 1.03     0.50–0.99  19,705  738  0.94  0.85, 1.03    0.92  0.83, 1.02     1.00–3.99  103,629  4,242  1.04  0.96, 1.12    0.99  0.92, 1.08     ≥4.00  18,861  725  1.08  0.97, 1.19    1.00  0.90, 1.11    Regular coffee, cups/day          0.09      0.90   Nonec  34,791  1,243  1.00  Referent    1.00  Referent     0.01–0.50  36,999  1,424  1.04  0.97, 1.12    1.05  0.97, 1.13     0.50–0.99  23,978  963  1.05  0.96, 1.14    1.03  0.94, 1.12     1.00–3.99  83,857  3,279  1.06  0.99, 1.13    1.03  0.96, 1.10     ≥4.00  10,365  359  1.07  0.95, 1.20    1.00  0.89, 1.13    Decaffeinated coffee, cups/day          0.0005      0.02   Nonec  58,716  2,069  1.00  Referent    1.00  Referent     0.01–0.50  69,458  2,515  0.99  0.93, 1.05    1.00  0.94, 1.06     0.50–0.99  26,540  1,074  1.03  0.96, 1.11    1.02  0.95, 1.10     1.00–3.99  34,000  1,566  1.09  1.02, 1.17    1.06  0.99, 1.13     ≥4.00  1,278  74  1.54  1.22, 1.95    1.37  1.08, 1.73    Abbreviations: CI, confidence interval; ED, erectile dysfunction; HR, hazard ratio. a Model 1 was adjusted for age in months and calendar time. b Model 2 was adjusted for the variables in model 1 and smoking (never, past, or current, categorized as 1–14, 15–24, ≥25 cigarettes/day), body mass index (weight in kilometers divided by height in meters squared; <25.0, 25.0–29.9, or ≥30.0), alcohol consumption (grams/day; 0, 0.1–4.9, 5.0–14.9, 15.0–29.9, or ≥30.0), total physical activity (metabolic equivalent of tasks per week; quintiles), history of diabetes (yes vs. no), history of hypertension (yes vs. no), history of hypercholesterolemia (yes vs. no), history of cardiovascular disease (yes vs. no), energy intake (kcal/day; quintiles), Alternative Healthy Eating Index score (quintiles), marital status (married, divorced, separated, widowed, or never married), race (white, African-American, Asian-American, or other), difficulty of falling into sleep (yes vs. no), waking during the night (yes vs. no), not feeling rested upon waking (yes vs. no), and use of cholesterol-lowering, blood pressure–lowering, or sleep-enhancing medication (yes vs. no). c Included men who drank no regular or decaffeinated coffee. Further analyses were conducted to assess possible reverse causation, that is, if development of ED led to a change in coffee habits. Thus, we examined coffee intake with a 4-year lag period between total, regular, and decaffeinated coffee intakes and ED (Table 5). We did not find any statistically significant trend associated between intakes of total (P for trend = 0.63), regular (P for trend = 0.74) and decaffeinated (P for trend = 0.64) coffee and ED. Also, there were no significant differences between the highest and lowest categories of total (HR = 0.92, 95% CI: 0.81, 1.04) and regular (HR = 0.89, 95% CI: 0.77, 1.02) coffee intakes. After comparing the highest category of decaffeinated coffee intake with the lowest ((≥4 vs. 0 cups/day), we found a 43% increased risk of ED (HR = 1.43, 95% CI: 1.10, 1.87). Table 5. Four-Year Latent Period in the Multivariable Associationsa of Total, Regular, and Decaffeinated Coffee Intakes With Erectile Dysfunction, Health Professionals Follow-up Study, 2002–2010 Type of Coffee  Coffee Intake, cups/day  No Coffeeb  0.01–0.50  0.50–0.99  1.00–3.99  ≥4.00  P for Trend  HR  95% CI  HR  95% CI  HR  95% CI  HR  95% CI  HR  95% CI  Total  1.00  Referent  0.87  0.77, 0.98  0.86  0.76, 0.97  0.93  0.85, 1.02  0.92  0.81, 1.04  0.63  Regular  1.00  Referent  0.93  0.84, 1.02  0.98  0.88, 1.09  0.98  0.91, 1.07  0.89  0.77, 1.02  0.74  Decaffeinated  1.00  Referent  0.95  0.89, 1.03  0.95  0.87, 1.05  0.99  0.91, 1.08  1.43  1.10, 1.87  0.64  Type of Coffee  Coffee Intake, cups/day  No Coffeeb  0.01–0.50  0.50–0.99  1.00–3.99  ≥4.00  P for Trend  HR  95% CI  HR  95% CI  HR  95% CI  HR  95% CI  HR  95% CI  Total  1.00  Referent  0.87  0.77, 0.98  0.86  0.76, 0.97  0.93  0.85, 1.02  0.92  0.81, 1.04  0.63  Regular  1.00  Referent  0.93  0.84, 1.02  0.98  0.88, 1.09  0.98  0.91, 1.07  0.89  0.77, 1.02  0.74  Decaffeinated  1.00  Referent  0.95  0.89, 1.03  0.95  0.87, 1.05  0.99  0.91, 1.08  1.43  1.10, 1.87  0.64  Abbreviations: CI, confidence interval; HR, hazard ratio. a Models were adjusted for age in months, calendar time, smoking (never, past, or current, categorized as 1–14, 15–24, ≥25 cigarettes/day), body mass index (weight in kilometers divided by height in meters squared; <25.0, 25.0–29.9, or ≥30.0), alcohol consumption (grams/day; 0, 0.1–4.9, 5.0–14.9, 15.0–29.9, or ≥30.0), total physical activity (metabolic equivalent of tasks per week; quintiles), history of diabetes (yes vs. no), history of hypertension (yes vs. no), history of hypercholesterolemia (yes vs. no), history of cardiovascular disease (yes vs. no), energy intake (kcal/day; quintiles), Alternative Healthy Eating Index score (quintiles), marital status (married, divorced, separated, widowed, or never married), race (white, African-American, Asian-American, or other), difficulty of falling into sleep (yes vs. no), waking during the night (yes vs. no), not feeling rested upon waking (yes vs. no), and use of cholesterol-lowering, blood pressure–lowering, or sleep-enhancing medication (yes vs. no). b Included men who drank no regular or decaffeinated coffee. Table 5. Four-Year Latent Period in the Multivariable Associationsa of Total, Regular, and Decaffeinated Coffee Intakes With Erectile Dysfunction, Health Professionals Follow-up Study, 2002–2010 Type of Coffee  Coffee Intake, cups/day  No Coffeeb  0.01–0.50  0.50–0.99  1.00–3.99  ≥4.00  P for Trend  HR  95% CI  HR  95% CI  HR  95% CI  HR  95% CI  HR  95% CI  Total  1.00  Referent  0.87  0.77, 0.98  0.86  0.76, 0.97  0.93  0.85, 1.02  0.92  0.81, 1.04  0.63  Regular  1.00  Referent  0.93  0.84, 1.02  0.98  0.88, 1.09  0.98  0.91, 1.07  0.89  0.77, 1.02  0.74  Decaffeinated  1.00  Referent  0.95  0.89, 1.03  0.95  0.87, 1.05  0.99  0.91, 1.08  1.43  1.10, 1.87  0.64  Type of Coffee  Coffee Intake, cups/day  No Coffeeb  0.01–0.50  0.50–0.99  1.00–3.99  ≥4.00  P for Trend  HR  95% CI  HR  95% CI  HR  95% CI  HR  95% CI  HR  95% CI  Total  1.00  Referent  0.87  0.77, 0.98  0.86  0.76, 0.97  0.93  0.85, 1.02  0.92  0.81, 1.04  0.63  Regular  1.00  Referent  0.93  0.84, 1.02  0.98  0.88, 1.09  0.98  0.91, 1.07  0.89  0.77, 1.02  0.74  Decaffeinated  1.00  Referent  0.95  0.89, 1.03  0.95  0.87, 1.05  0.99  0.91, 1.08  1.43  1.10, 1.87  0.64  Abbreviations: CI, confidence interval; HR, hazard ratio. a Models were adjusted for age in months, calendar time, smoking (never, past, or current, categorized as 1–14, 15–24, ≥25 cigarettes/day), body mass index (weight in kilometers divided by height in meters squared; <25.0, 25.0–29.9, or ≥30.0), alcohol consumption (grams/day; 0, 0.1–4.9, 5.0–14.9, 15.0–29.9, or ≥30.0), total physical activity (metabolic equivalent of tasks per week; quintiles), history of diabetes (yes vs. no), history of hypertension (yes vs. no), history of hypercholesterolemia (yes vs. no), history of cardiovascular disease (yes vs. no), energy intake (kcal/day; quintiles), Alternative Healthy Eating Index score (quintiles), marital status (married, divorced, separated, widowed, or never married), race (white, African-American, Asian-American, or other), difficulty of falling into sleep (yes vs. no), waking during the night (yes vs. no), not feeling rested upon waking (yes vs. no), and use of cholesterol-lowering, blood pressure–lowering, or sleep-enhancing medication (yes vs. no). b Included men who drank no regular or decaffeinated coffee. In stratified and multivariable analyses, we examined the association between coffee and ED across strata of smoking status (Table 6). We also stratified across strata of BMI categories, hypertension, marital status, and alcohol consumption (Web Tables 1–4, available at https://academic.oup.com/aje). No significant results were reported among strata of smoking status for total (P for interaction = 0.11) and regular (P for interaction = 0.80) coffee intakes, Table 4). However, among current smokers there was a positive trend associated between decaffeinated coffee intake and ED (P for trend = 0.005 and P for interaction = 0.06). Table 6. Associationsa of Cumulative Average Intakes of Total Coffee, Regular Coffee, and Decaffeinated Coffee With Erectile Dysfunction, Stratified by Smoking Status, Health Professionals Follow-up Study, 1998–2010 Type of Coffee and Smoking Status  Coffee Intake, cups/day  No Coffeeb  0.01–0.50  0.50–0.99  1.00–3.99  ≥4.00  P for Trend  P for Interaction  HR  95% CI  HR  95% CI  HR  95% CI  HR  95% CI  HR  95% CI  Total coffee              0.11   Never smoker  1.00  Referent  0.98  0.87, 1.10  0.91  0.80, 1.04  0.98  0.89, 1.08  0.99  0.84, 1.15  0.78     Past smoker  1.00  Referent  0.90  0.74, 1.09  0.97  0.80, 1.17  1.01  0.86, 1.17  1.02  0.85, 1.22  0.23     Current smoker  1.00  Referent  0.72  0.27, 1.93  0.72  0.26, 2.02  1.45  0.72, 2.95  1.23  0.59, 2.56  0.14    Regular coffee      0.80   Never smoker  1.00  Referent  1.01  0.92, 1.12  0.98  0.87, 1.10  1.00  0.92, 1.10  1.06  0.87, 1.29  0.86     Past smoker  1.00  Referent  1.11  0.97, 1.26  1.09  0.95, 1.25  1.07  0.95, 1.20  1.05  0.88, 1.25  0.76     Current smoker  1.00  Referent  1.23  0.62, 2.44  1.85  0.93, 3.70  1.50  0.84, 2.67  1.20  0.63, 2.27  0.55    Decaffeinated coffee      0.06   Never smoker  1.00  Referent  1.00  0.92, 1.09  0.97  0.87, 1.08  1.02  0.92, 1.13  1.30  0.82, 2.06  0.72     Past smoker  1.00  Referent  0.94  0.86, 1.03  1.00  0.89, 1.11  1.00  0.91, 1.10  1.31  0.98, 1.74  0.34     Current smoker  1.00  Referent  1.54  1.07, 2.21  1.54  0.98, 2.40  1.63  1.13, 2.35  2.64  0.88, 7.87  0.005    Type of Coffee and Smoking Status  Coffee Intake, cups/day  No Coffeeb  0.01–0.50  0.50–0.99  1.00–3.99  ≥4.00  P for Trend  P for Interaction  HR  95% CI  HR  95% CI  HR  95% CI  HR  95% CI  HR  95% CI  Total coffee              0.11   Never smoker  1.00  Referent  0.98  0.87, 1.10  0.91  0.80, 1.04  0.98  0.89, 1.08  0.99  0.84, 1.15  0.78     Past smoker  1.00  Referent  0.90  0.74, 1.09  0.97  0.80, 1.17  1.01  0.86, 1.17  1.02  0.85, 1.22  0.23     Current smoker  1.00  Referent  0.72  0.27, 1.93  0.72  0.26, 2.02  1.45  0.72, 2.95  1.23  0.59, 2.56  0.14    Regular coffee      0.80   Never smoker  1.00  Referent  1.01  0.92, 1.12  0.98  0.87, 1.10  1.00  0.92, 1.10  1.06  0.87, 1.29  0.86     Past smoker  1.00  Referent  1.11  0.97, 1.26  1.09  0.95, 1.25  1.07  0.95, 1.20  1.05  0.88, 1.25  0.76     Current smoker  1.00  Referent  1.23  0.62, 2.44  1.85  0.93, 3.70  1.50  0.84, 2.67  1.20  0.63, 2.27  0.55    Decaffeinated coffee      0.06   Never smoker  1.00  Referent  1.00  0.92, 1.09  0.97  0.87, 1.08  1.02  0.92, 1.13  1.30  0.82, 2.06  0.72     Past smoker  1.00  Referent  0.94  0.86, 1.03  1.00  0.89, 1.11  1.00  0.91, 1.10  1.31  0.98, 1.74  0.34     Current smoker  1.00  Referent  1.54  1.07, 2.21  1.54  0.98, 2.40  1.63  1.13, 2.35  2.64  0.88, 7.87  0.005    Abbreviations: CI, confidence interval; HR, hazard ratio. a Included men who drank no regular or decaffeinated coffee. b Models were adjusted for age in months, calendar time, smoking (never, past, or current, categorized as 1–14, 15–24, ≥25 cigarettes/day), body mass index (weight in kilometers divided by height in meters squared; <25.0, 25.0–29.9, or ≥30.0), alcohol consumption (grams/day; 0, 0.1–4.9, 5.0–14.9, 15.0–29.9, or ≥30.0), total physical activity (metabolic equivalent of tasks per week; quintiles), history of diabetes (yes vs. no), history of hypertension (yes vs. no), history of hypercholesterolemia (yes vs. no), history of cardiovascular disease (yes vs. no), energy intake (kcal/day; quintiles), Alternative Healthy Eating Index score (quintiles), marital status (married, divorced, separated, widowed, or never married), race (white, African-American, Asian-American, or other), difficulty of falling into sleep (yes vs. no), waking during the night (yes vs. no), not feeling rested upon waking (yes vs. no), and use of cholesterol-lowering, blood pressure–lowering, or sleep-enhancing medication (yes vs. no). Table 6. Associationsa of Cumulative Average Intakes of Total Coffee, Regular Coffee, and Decaffeinated Coffee With Erectile Dysfunction, Stratified by Smoking Status, Health Professionals Follow-up Study, 1998–2010 Type of Coffee and Smoking Status  Coffee Intake, cups/day  No Coffeeb  0.01–0.50  0.50–0.99  1.00–3.99  ≥4.00  P for Trend  P for Interaction  HR  95% CI  HR  95% CI  HR  95% CI  HR  95% CI  HR  95% CI  Total coffee              0.11   Never smoker  1.00  Referent  0.98  0.87, 1.10  0.91  0.80, 1.04  0.98  0.89, 1.08  0.99  0.84, 1.15  0.78     Past smoker  1.00  Referent  0.90  0.74, 1.09  0.97  0.80, 1.17  1.01  0.86, 1.17  1.02  0.85, 1.22  0.23     Current smoker  1.00  Referent  0.72  0.27, 1.93  0.72  0.26, 2.02  1.45  0.72, 2.95  1.23  0.59, 2.56  0.14    Regular coffee      0.80   Never smoker  1.00  Referent  1.01  0.92, 1.12  0.98  0.87, 1.10  1.00  0.92, 1.10  1.06  0.87, 1.29  0.86     Past smoker  1.00  Referent  1.11  0.97, 1.26  1.09  0.95, 1.25  1.07  0.95, 1.20  1.05  0.88, 1.25  0.76     Current smoker  1.00  Referent  1.23  0.62, 2.44  1.85  0.93, 3.70  1.50  0.84, 2.67  1.20  0.63, 2.27  0.55    Decaffeinated coffee      0.06   Never smoker  1.00  Referent  1.00  0.92, 1.09  0.97  0.87, 1.08  1.02  0.92, 1.13  1.30  0.82, 2.06  0.72     Past smoker  1.00  Referent  0.94  0.86, 1.03  1.00  0.89, 1.11  1.00  0.91, 1.10  1.31  0.98, 1.74  0.34     Current smoker  1.00  Referent  1.54  1.07, 2.21  1.54  0.98, 2.40  1.63  1.13, 2.35  2.64  0.88, 7.87  0.005    Type of Coffee and Smoking Status  Coffee Intake, cups/day  No Coffeeb  0.01–0.50  0.50–0.99  1.00–3.99  ≥4.00  P for Trend  P for Interaction  HR  95% CI  HR  95% CI  HR  95% CI  HR  95% CI  HR  95% CI  Total coffee              0.11   Never smoker  1.00  Referent  0.98  0.87, 1.10  0.91  0.80, 1.04  0.98  0.89, 1.08  0.99  0.84, 1.15  0.78     Past smoker  1.00  Referent  0.90  0.74, 1.09  0.97  0.80, 1.17  1.01  0.86, 1.17  1.02  0.85, 1.22  0.23     Current smoker  1.00  Referent  0.72  0.27, 1.93  0.72  0.26, 2.02  1.45  0.72, 2.95  1.23  0.59, 2.56  0.14    Regular coffee      0.80   Never smoker  1.00  Referent  1.01  0.92, 1.12  0.98  0.87, 1.10  1.00  0.92, 1.10  1.06  0.87, 1.29  0.86     Past smoker  1.00  Referent  1.11  0.97, 1.26  1.09  0.95, 1.25  1.07  0.95, 1.20  1.05  0.88, 1.25  0.76     Current smoker  1.00  Referent  1.23  0.62, 2.44  1.85  0.93, 3.70  1.50  0.84, 2.67  1.20  0.63, 2.27  0.55    Decaffeinated coffee      0.06   Never smoker  1.00  Referent  1.00  0.92, 1.09  0.97  0.87, 1.08  1.02  0.92, 1.13  1.30  0.82, 2.06  0.72     Past smoker  1.00  Referent  0.94  0.86, 1.03  1.00  0.89, 1.11  1.00  0.91, 1.10  1.31  0.98, 1.74  0.34     Current smoker  1.00  Referent  1.54  1.07, 2.21  1.54  0.98, 2.40  1.63  1.13, 2.35  2.64  0.88, 7.87  0.005    Abbreviations: CI, confidence interval; HR, hazard ratio. a Included men who drank no regular or decaffeinated coffee. b Models were adjusted for age in months, calendar time, smoking (never, past, or current, categorized as 1–14, 15–24, ≥25 cigarettes/day), body mass index (weight in kilometers divided by height in meters squared; <25.0, 25.0–29.9, or ≥30.0), alcohol consumption (grams/day; 0, 0.1–4.9, 5.0–14.9, 15.0–29.9, or ≥30.0), total physical activity (metabolic equivalent of tasks per week; quintiles), history of diabetes (yes vs. no), history of hypertension (yes vs. no), history of hypercholesterolemia (yes vs. no), history of cardiovascular disease (yes vs. no), energy intake (kcal/day; quintiles), Alternative Healthy Eating Index score (quintiles), marital status (married, divorced, separated, widowed, or never married), race (white, African-American, Asian-American, or other), difficulty of falling into sleep (yes vs. no), waking during the night (yes vs. no), not feeling rested upon waking (yes vs. no), and use of cholesterol-lowering, blood pressure–lowering, or sleep-enhancing medication (yes vs. no). Among categories of BMI (<25.0, 25.0–29.9, or ≥30.0), the associations of total, regular and decaffeinated coffee intake with ED did not reach statistical significance (P for interaction = 0.07, 0.16, and 0.88, respectively; Web Table 1). Similar findings were identified after stratifying for history of hypertension (P for interaction = 0.25, 0.25, and 0.72, respectively; Web Table 2), marital status (P for interaction = 0.32, 0.67, and 0.08, respectively, Web Table 3), and alcohol consumption (P for interaction = 0.70, 0.99, and 0.93, respectively; Web Table 4). Further stratified analyses were conducted by age (<70 or ≥70), history of diabetes (yes vs. no), medication for lowering blood pressure (yes vs. no), and difficulty of falling into sleep (yes vs. no), but no significant associations were found (data not shown). DISCUSSION In the present large observational study that investigated prospectively the association of coffee intake with ED, we observed no significant associations between intakes of total and regular coffee and ED. For decaffeinated coffee intake, there was a significant association after comparing the highest with the lowest intake and also a significant trend. In stratified analyses, none of the modifiable lifestyle factors influenced the association of coffee intake with ED, though the positive association between decaffeinated coffee and ED was observed most strongly in the stratum of current smokers, and possibly suggests residual confounding. Coffee intake has potential health benefits for various health outcomes (27–30). Yet, the present prospective investigation, which included 21,403 men followed for 10 years with a 34% rate of patients with incident ED (n = 7,298), did not support this association. Unlike in other studies in which investigators found potential associations that could potentially be the result of residual confounding (31), we controlled for a large number of confounders (e.g., sleep problems). Relatively few studies have investigated the independent association between coffee intake and ED in prospective population-based studies, including our previous study (19), in which we found an inverse association. However, in that previous study, we used cross-sectional data and a smaller sample size (n = 3,724). In a population-based prospective study of 202 patients with incident ED among Finnish men with 5 years of follow-up, Shiri et al. (9) found no association between coffee consumption and ED. No data were provided on decaffeinated coffee intake. Some aspects of our prospective study merit discussion. Although we observed an increased risk for ED in men who consumed 4 cups/day or more of decaffeinated coffee compared with those who consumed 0 cups/day, we interpret these findings cautiously for the following reasons. First, only 0.9% of men consumed 4 of cups/day or more of decaffeinated coffee. Second, there is no strong biological plausibility between decaffeinated coffee intake and ED. During the process of decaffeination, some of the polyphenols and anti-inflammatory compounds may be removed (32–36), but we found no association between regular coffee consumption and ED. It is possible that deleterious chemical compounds may be added to the decaffeinated coffee in the process of decaffeination (34–39). Third, heavy drinkers of decaffeinated coffee had an unusual lifestyle, with comorbid characteristics such as higher BMI, more current and past smoking history, as well as more hypertension, high cholesterol and higher alcohol consumption, but they also had slightly better diets (Alternative Healthy Eating Index score). Interestingly, we observed an association between decaffeinated coffee intake and ED only among current smokers, suggesting that residual confounding may have accounted for the association. Strengths of the present study include the large sample size, number of patients with incident ED, long follow-up data from participants, and detailed data on ED risk factors. These strengths increased our statistical power for the categorical comparison groups and generalizability for studies with similar study populations, and also reduced residual confounding because of our well-characterized risk factors. We used repeated measurements and assessed cumulative dietary intake to more accurately assess long-term coffee intake and reduce measurement error. Yet, the present study has some limitations as well. First, definition of ED was self-reported by participants, which will inevitably be imperfect. However, health professionals tend to report medical conditions with high accuracy, and we have previously reported on classic risk factors for ED such as BMI and physical activity (12) which have been confirmed in other populations with more detailed assessments of patients with incident ED. Although we adjusted for several potential confounders, there is still the possibility of confounding from additional unmeasured factors. However, due to our detailed and updated adjustment for confounders, it is unlikely that these would fully account for the observed findings. Second, although detailed assessment of coffee intake was conducted, this was done every 4 years, therefore, any shorter-term (<4 years) fluctuation of the intake could have been missed. However, coffee intake is consistent over time. Third, no data were available on the preparation methods of coffee, information which can provide insight on the alteration of the chemical composition of coffee. Fourth, it is possible that coffee consumption over a longer time period could have influenced risk. Yet, it is notable that coffee consumption is relatively consistent over time and drinking habits during the study period correlated with prior coffee drinking. Fifth, future studies should focus on investigating a nonlinear association between coffee intake and ED. Finally, reverse causation did not seem to play a role in our findings, but it is possible that subtle alterations in health could have influenced coffee consumption even after including a 4-year lag in our sensitivity analyses. In summary, we did not find an association, either positive or negative, between total or regular coffee intake and ED. However, decaffeinated-coffee intake seemed to be associated with ED, possibly because of residual confounding. None of the modifiable lifestyles factors influenced the association of coffee intake with ED. ACKNOWLEDGMENTS Author affiliations: Division of Epidemiology, Human Genetics and Environmental Sciences, University of Texas School of Public Health, Houston, Texas (David S. Lopez, Marcia de Oliveira Otto); Division of Urology, McGovern Medical School, University of Texas Health Science Center at Houston, Houston, Texas (David S. Lopez, Run Wang, Steven Canfield); Department of Nutrition, Harvard T.H. Chan School of Public Health, Boston, Massachusetts (Lydia Liu, Eric B. Rimm, Edward Giovannucci); Department of Epidemiology, Harvard T.H. Chan School of Public Health, Boston, Massachusetts (Lydia Liu, Eric B. Rimm, Edward Giovannucci); Channing Division of Network Medicine, Department of Medicine, Brigham and Women’s Hospital, Harvard Medical School, Boston, Massachusetts (Eric B. Rimm, Edward Giovannucci); Department of Hygiene and Epidemiology, University of Ioannina School of Medicine, Ioannina, Greece (Konstantinos K. Tsilidis); and Department of Epidemiology and Biostatistics, School of Public Health, Imperial College London, London, United Kingdom (Konstantinos K. Tsilidis). This work was supported by the National Institutes of Health and by the UTHealth McGovern Medical School-Division of Urology and Dr. Osama I. Mikhail (UTHealth) (grants UM1 CA167552 and R01 HL35464). Conflict of interest: none declared. Abbreviations BMI body mass index CI confidence interval ED erectile dysfunction FFQ food frequency questionnaire HPFS Health Professionals Follow-up Study HR hazard ratio REFERENCES 1 O’Keefe JH, Bhatti SK, Patil HR, et al.  . Effects of habitual coffee consumption on cardiometabolic disease, cardiovascular health, and all-cause mortality. J Am Coll Cardiol . 2013; 62( 12): 1043– 1051. Google Scholar CrossRef Search ADS PubMed  2 León-Carmona JR, Galano A. Is caffeine a good scavenger of oxygenated free radicals? J Phys Chem B . 2011; 115( 15): 4538– 4546. Google Scholar CrossRef Search ADS PubMed  3 Wilson KM, Kasperzyk JL, Rider JR, et al.  . Coffee consumption and prostate cancer risk and progression in the Health Professionals Follow-up Study. J Natl Cancer Inst . 2011; 103( 11): 876– 884. Google Scholar CrossRef Search ADS PubMed  4 Freedman ND, Park Y, Abnet CC, et al.  . Association of coffee drinking with total and cause-specific mortality. N Engl J Med . 2012; 366( 20): 1891– 1904. Google Scholar CrossRef Search ADS PubMed  5 Ferraro PM, Taylor EN, Gambaro G, et al.  . Caffeine intake and the risk of kidney stones. Am J Clin Nutr . 2014; 100( 6): 1596– 1603. Google Scholar CrossRef Search ADS PubMed  6 Diokno AC, Brown MB, Herzog AR. Sexual function in the elderly. Arch Intern Med . 1990; 150( 1): 197– 200. Google Scholar CrossRef Search ADS PubMed  7 Akkus E, Kadioglu A, Esen A, et al.  . Prevalence and correlates of erectile dysfunction in Turkey: a population-based study. Eur Urol . 2002; 41( 3): 298– 304. Google Scholar CrossRef Search ADS PubMed  8 Berrada S, Kadri N, Mechakra-Tahiri S, et al.  . Prevalence of erectile dysfunction and its correlates: a population-based study in Morocco. Int J Impot Res . 2003; 15( suppl 1): S3– S7. Google Scholar CrossRef Search ADS PubMed  9 Shiri R, Koskimäki J, Hakama M, et al.  . Effect of life-style factors on incidence of erectile dysfunction. Int J Impot Res . 2004; 16( 5): 389– 394. Google Scholar CrossRef Search ADS PubMed  10 Lue TF. Erectile dysfunction. N Engl J Med . 2000; 342( 24): 1802– 1813. Google Scholar CrossRef Search ADS PubMed  11 Selvin E, Burnett AL, Platz EA. Prevalence and risk factors for erectile dysfunction in the US. Am J Med . 2007; 120( 2): 151– 157. Google Scholar CrossRef Search ADS PubMed  12 Bacon CG, Mittleman MA, Kawachi I, et al.  . A prospective study of risk factors for erectile dysfunction. J Urol . 2006; 176( 1): 217– 221. Google Scholar CrossRef Search ADS PubMed  13 Francis ME, Kusek JW, Nyberg LM, et al.  . The contribution of common medical conditions and drug exposures to erectile dysfunction in adult males. J Urol . 2007; 178( 2): 591– 596. Google Scholar CrossRef Search ADS PubMed  14 Guay AT, Spark RF, Bansal S, et al.  . American Association of Clinical Endocrinologists medical guidelines for clinical practice for the evaluation and treatment of male sexual dysfunction: a couple’s problem–2003 update. Endocr Pract . 2003; 9( 1): 77– 95. Google Scholar CrossRef Search ADS PubMed  15 Sansone A, Romanelli F, Gianfrilli D, et al.  . Endocrine evaluation of erectile dysfunction. Endocrine . 2014; 46( 3): 423– 430. Google Scholar CrossRef Search ADS PubMed  16 Paton CD, Lowe T, Irvine A. Caffeinated chewing gum increases repeated sprint performance and augments increases in testosterone in competitive cyclists. Eur J Appl Physiol . 2010; 110( 6): 1243– 1250. Google Scholar CrossRef Search ADS PubMed  17 Kelly DM, Jones TH. Testosterone: a vascular hormone in health and disease. J Endocrinol . 2013; 217( 3): R47– R71. Google Scholar CrossRef Search ADS PubMed  18 Adebiyi A, Adaikan PG. Effect of caffeine on response of rabbit isolated corpus cavernosum to high K+ solution, noradrenaline and transmural electrical stimulation. Clin Exp Pharmacol Physiol . 2004; 31( 1–2): 82– 85. Google Scholar CrossRef Search ADS PubMed  19 Lopez DS, Wang R, Tsilidis KK, et al.  . Role of caffeine intake on erectile dysfunction in US men: results from NHANES 2001–2004. PLoS One  2015; 10( 4): e0123547. Google Scholar CrossRef Search ADS PubMed  20 Rimm EB, Giovannucci EL, Willett WC, et al.  . Prospective study of alcohol consumption and risk of coronary disease in men. Lancet . 1991; 338( 8765): 464– 468. Google Scholar CrossRef Search ADS PubMed  21 Bacon CG, Mittleman MA, Kawachi I, et al.  . Sexual function in men older than 50 years of age: results from the Health Professionals Follow-up Study. Ann Intern Med . 2003; 139( 3): 161– 168. Google Scholar CrossRef Search ADS PubMed  22 Gao X, Schwarzschild MA, O’Reilly EJ, et al.  . Restless legs syndrome and erectile dysfunction. Sleep . 2010; 33( 1): 75– 79. Google Scholar CrossRef Search ADS PubMed  23 Feskanich D, Rimm EB, Giovannucci EL, et al.  . Reproducibility and validity of food intake measurements from a semiquantitative food frequency questionnaire. J Am Diet Assoc . 1993; 93( 7): 790– 796. Google Scholar CrossRef Search ADS PubMed  24 Hu FB, Stampfer MJ, Rimm E, et al.  . Dietary fat and coronary heart disease: a comparison of approaches for adjusting for total energy intake and modeling repeated dietary measurements. Am J Epidemiol . 1999; 149( 6): 531– 540. Google Scholar CrossRef Search ADS PubMed  25 Cassidy A, Franz M, Rimm EB. Dietary flavonoid intake and incidence of erectile dysfunction. Am J Clin Nutr . 2016; 103( 2): 534– 541. Google Scholar CrossRef Search ADS PubMed  26 Zhang X, Zhang MJ, Fine J. A proportional hazards regression model for the subdistribution with right-censored and left-truncated competing risks data. Stat Med . 2011; 30( 16): 1933– 1951. Google Scholar CrossRef Search ADS PubMed  27 Loftfield E, Freedman ND, Graubard BI, et al.  . Association of coffee consumption with overall and cause-specific mortality in a large US prospective cohort study. Am J Epidemiol . 2015; 182( 12): 1010– 1022. Google Scholar PubMed  28 Crippa A, Discacciati A, Larsson SC, et al.  . Coffee consumption and mortality from all causes, cardiovascular disease, and cancer: a dose-response meta-analysis. Am J Epidemiol . 2014; 180( 8): 763– 775. Google Scholar CrossRef Search ADS PubMed  29 Lu Y, Zhai L, Zeng J, et al.  . Coffee consumption and prostate cancer risk: an updated meta-analysis. Cancer Causes Control . 2014; 25( 5): 591– 604. Google Scholar CrossRef Search ADS PubMed  30 Je Y, Giovannucci E. Coffee consumption and total mortality: a meta-analysis of twenty prospective cohort studies. Br J Nutr . 2014; 111( 7): 1162– 1173. Google Scholar CrossRef Search ADS PubMed  31 Lopez DS, Wang R, Tsilidis KK, et al.  . Role of caffeine intake on erectile dysfunction in US men: results from NHANES 2001-2004. PLoS One . 2015; 10( 4): e0123547. Google Scholar CrossRef Search ADS PubMed  32 Pehl C, Pfeiffer A, Wendl B, et al.  . The effect of decaffeination of coffee on gastro-oesophageal reflux in patients with reflux disease. Aliment Pharmacol Ther . 1997; 11( 3): 483– 486. Google Scholar CrossRef Search ADS PubMed  33 Spiller MA. The coffee plant and its processing. Prog Clin Biol Res . 1984; 158: 75– 89. Google Scholar PubMed  34 Borrelli RC, Visconti A, Mennella C, et al.  . Chemical characterization and antioxidant properties of coffee melanoidins. J Agric Food Chem . 2002; 50( 22): 6527– 6533. Google Scholar CrossRef Search ADS PubMed  35 Yu EK. Novel decaffeination process using cyclodextrins. Appl Microbiol Biotechnol . 1988; 28( 6): 546. Google Scholar CrossRef Search ADS   36 Chen Y, Brown PH, Hu K, et al.  . Supercritical CO2 decaffeination of unroasted coffee beans produces melanoidins with distinct NF-κB inhibitory activity. J Food Sci . 2011; 76( 7): H182– H186. Google Scholar CrossRef Search ADS PubMed  37 Superko HR, Bortz W Jr, Williams PT, et al.  . Caffeinated and decaffeinated coffee effects on plasma lipoprotein cholesterol, apolipoproteins, and lipase activity: a controlled, randomized trial. Am J Clin Nutr . 1991; 54( 3): 599– 605. Google Scholar CrossRef Search ADS PubMed  38 Centers for Disease Control and Prevention (CDC). Fatal exposure to methylene chloride among bathtub refinishers – United States, 2000–2011. MMWR Morb Mortal Wkly Rep . 2012; 61( 7): 119– 122. PubMed  39 Melnick RL, Ward JM, Huff J. War on Carcinogens: industry disputes human relevance of chemicals causing cancer in laboratory animals based on unproven hypotheses, using kidney tumors as an example. Int J Occup Environ Health . 2013; 19( 4): 255– 260. Google Scholar CrossRef Search ADS PubMed  © The Author(s) 2018. Published by Oxford University Press on behalf of the Johns Hopkins Bloomberg School of Public Health. All rights reserved. For permissions, please e-mail: [email protected]. This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/about_us/legal/notices)

Krieger, Nancy;Jahn, Jaquelyn L;Waterman, Pamela D;Chen, Jarvis T

doi: 10.1093/aje/kwx312pmid: 29036268

Abstract Evidence suggests that contemporary population distributions of estrogen-receptor (ER) status among breast cancer patients may be shaped by earlier major societal events, such as the 1965 abolition of legal racial discrimination in the United States (state and local “Jim Crow” laws) and the Great Famine in China (1959–1961). We analyzed changes in ER status in relation to Jim Crow birthplace among the 46,417 black and 339,830 white US-born, non-Hispanic women in the Surveillance, Epidemiology, and End Results (SEER) 13 Registry Group who were born between 1915 and 1979 and diagnosed (ages 25–84 years, inclusive) during 1992–2012. We grouped the cases according to birth cohort and quantified the rate of change using the haldane (which scales change in relation to biological generation). The percentage of ER-positive cases rose according to birth cohort (1915–1919 to 1975–1979) only among women diagnosed before age 55. Changes according to biological generation were greater among black women than among white women, and among black women, they were greatest among those born in Jim Crow (versus non–Jim Crow) states, with this group being the only group to exhibit high haldane values (>|0.3|, indicating high rate of change). Our study’s analytical approach and findings underscore the need to consider history and societal context when analyzing ER status among breast cancer patients and racial/ethnic inequities in its distribution. black, breast cancer, estrogen receptor, generation, haldane, Jim Crow, racial segregation, secular trend Evidence indicates that expression of the estrogen receptor (ER), a key biomarker that is predictive of prognosis in breast cancer, is regulated by methylation and is modifiable (1–6). However, while the clinical significance of the temporal dynamics of breast tumor ER expression has gained attention (7, 8), research on long-term population trends in ER status of breast cancer tumors remains limited (9). The pace of change of phenotypic characteristics across generations can provide important insight into the extent to which their expression is driven by factors external to organisms and their inherited genomes (10–14). In evolutionary biology, one useful metric for quantifying this pace of change is the haldane, a “calculation of rates in phenotypic standard deviation per generation” (10, p. 453), which can be meaningfully compared within and across species, regardless of their lifespan (10–15). To date, use of the haldane to analyze contemporary rates of microevolution has focused chiefly on diverse animal species (e.g., to study the biological impact of ecosystem change induced by human activity) (10–14). By contrast, only a handful of studies have employed the haldane to quantify the pace of change in contemporary human phenotypic traits (15–18), and none have applied it to analysis of cancer. Our study took the novel step of scaling the rate of change in the prevalence of breast cancer ER status according to biological generation, as tied also to the timing of major societal changes. We specifically focus on US black and white women relative to the place and year of their birth, as framed by the system of Jim Crow laws in the United States (i.e., state and local legal racial discrimination); Jim Crow laws were abolished in 1965 (19–22). At issue is whether the racism imposed by Jim Crow laws affected health above and beyond the racism experienced in states that did not have Jim Crow laws, thereby shaping trends in both the health of black Americans and the magnitude of black versus white health inequities (19–29). Motivating our research were both: 1) persistent unresolved questions about social versus genetic causes of the US black versus white excess of ER-negative (ER−) tumors, especially at younger ages (9, 30–34); and 2) empirical evidence regarding the historical contingency of the population distribution of, and social inequalities in, breast cancer ER status (9, 35, 36). Thus, despite enormous heterogeneity in breast cancer ER status among diverse African nations (37), some researchers refer to “African” breast cancer and focus on genetic causes of black versus white disparities in ER− breast cancer (31, 34), whereas others call attention to likely social drivers of racial/ethnic disparities in breast cancer ER status and their potential malleability (30, 32, 33). Suggestively, the US black versus white ER− excess risk rose during the 1990s but then fell (35) following the 2002 publication of Women’s Health Initiative results showing that hormone therapy increased the risk of breast cancer without conferring protection for cardiovascular disease, leading to its declining use among women with access to this medication (38, 39). Two new studies have also suggested that breast cancer ER status among contemporary cases (1990s–2010s) may be shaped by major societal events that occurred decades earlier (36, 40). These studies found ER status to be associated with: 1) in the US, exposure to birthplace with Jim Crow laws among US black women (but not white women) born before 1965 (i.e., born in a state with versus without legal racial discrimination prior to passage of the US Civil Rights Act of 1964) (36); and 2) in China, exposure to the Great Famine of 1959–1961 (40). Biological pathways that could potentially link these social changes to breast cancer ER status include social adversity starting in early life, affecting nutrition, body build, and reproductive history (including age at menarche, age at first birth, parity, and breastfeeding), as well as later life access to screening and to hormone therapy (36, 40–48). METHODS Data source Our custom data set (36), provided by the National Cancer Institute (NCI) Surveillance, Epidemiology, and End Results (SEER) program, comprised women in the SEER 13 Registry Group (excluding Alaska) (49, 50) who were diagnosed with primary invasive breast cancer between January 1, 1992 (when this group was established), and December 31, 2012. The SEER data set included 47,157 US-born, black non-Hispanic women and 348,514 US-born, white non-Hispanic women who were aged 25–84 years, inclusive, at time of diagnosis, and we refer to these groups hereafter as “black” and “white.” Because of the small number of women observed who were born either before 1915 (black: 1.2%; white: 2.4%) or after 1979 (black: 0.3%; white: 0.1%), we limited the analytical data set to the 46,417 black women and 339,830 white women born between 1915 and 1979. Cancer registry data, obtained from hospital medical records, pertained to the cases’ breast cancer ER status, year of diagnosis, age at diagnosis, tumor characteristics (size, stage, histology, and grade), and vital status. The NCI added to each case record 2 additional variables: The first was birthplace, categorized as “Jim Crow” versus “non–Jim Crow,” based on data we provided to the NCI about states’ pre-1965 Jim Crow–laws status (19, 36); confidentiality policies precluded the NCI from releasing the specific state of birth. The second was the quintile ranking for the SEER socioeconomic index for each case’s residential census tract at time of diagnosis, based on the index’s annual national distribution for that year (51, 52). Use of census tract socioeconomic measures, including for cancer registries, has been validated in prior research (51–54). This work was designated by the institutional review board as exempt (Harvard School of Public Health IRB protocol #IRB13-1796). Statistical analysis To address well-known problems regarding missing birthplace data in cancer registry records (because birthplace data are obtained primarily from death certificate data (55–57)) and missing ER data (30, 58, 59), we used SAS (SAS Institute, Inc., Cary, North Carolina) PROC MI with fully conditional specification (60) to create 10 multiply imputed data sets, under the missing-at-random (MAR) assumption that the missing-data mechanism involved solely the observed data (60, 61). Variables included in the imputation model met this criterion (30, 58, 59): age at diagnosis, year of diagnosis, race/ethnicity, cancer registry, tumor characteristics (size, stage, grade, and histology), ER status, vital status, Jim Crow birthplace, and census tract socioeconomic index. As shown in Table 1, there was no missing data for race/ethnicity, age at or year of diagnosis, or vital status; missingness for census tract socioeconomic index equaled 1.2%, and it was 12.3% for ER status and 49.4% for Jim Crow birthplace. To impute ER status (as ER-positive (ER+) vs. ER−), we grouped the small percentage of cases classified as borderline (0.4%) with those classified as unknown. We conducted sensitivity analyses using logical bounds (36) and employed two extreme scenarios: 1) all missing values assigned to Jim Crow birthplace for the black women and to non–Jim Crow birthplace for the white women, and 2) vice versa. Table 1. US White and Black Non-Hispanic Women, Aged 25–84 Years and Born Between 1915–1979, Diagnosed With Invasive Primary Breast Cancer, Overall and According to Race/Ethnicity, SEER 13 Registry Group, Surveillance, Epidemiology, and End Results, United States, 1992–2012 Variable  Total (White and Black) (n = 386,247)  Black Non-Hispanic (n = 46,417)  White Non-Hispanic (n = 339,830)  No.  %  No.  %  No.  %  Registry               Atlanta  32,628  8.5  10,816  23.3  21,812  6.4   Connecticut  47,797  12.4  3,094  6.7  44,703  13.2   Detroit  53,555  13.9  12,317  26.5  41,238  12.1   Hawaii  4,207  1.1  113  0.2  4,094  1.2   Iowa  40,334  10.4  510  1.1  39,824  11.7   Los Angeles  61,281  15.9  11,618  25.0  49,663  14.6   New Mexico  14,420  3.7  265  0.6  14,155  4.2   Rural Georgia  1,596  0.4  551  1.2  1,045  0.3   San Francisco  40,614  10.5  5,020  10.8  35,594  10.5   San Jose–Monterey  18,807  4.9  526  1.1  18,281  5.4   Seattle  52,147  13.5  1,505  3.2  50,642  5.5   Utah  18,861  4.9  82  0.2  18,779  5.5  Year of diagnosisa  2002.4 (6.0)  2,003.0 (6.0)  2,002.3 (5.9)  Age at diagnosis, yearsa  60.3 (12.7)  57.2 (12.9)  60.8 (12.6)  Age group at diagnosis, years               25–44  45,989  11.9  8,242  17.8  37,747  11.1   45–54  88,358  22.9  12,175  26.2  76,183  22.4   54–84  251,900  65.2  26,000  56.0  225,900  66.5  Birth cohort               1915–1919  18,587  4.8  1,281  2.8  17,306  5.1   1920–1924  31,677  8.2  2,414  5.2  29,263  8.6   1925–1929  39,449  10.2  3,494  7.5  35,955  10.6   1930–1934  39,729  10.3  4,055  8.7  35,674  10.5   1935–1939  40,810  10.6  4,564  9.8  36,246  10.7   1940–1944  47,851  12.4  5,412  11.7  42,439  12.5   1945–1949  51,035  13.2  6,424  13.8  44,611  13.1   1950–1954  42,771  11.1  6,155  13.2  36,616  10.8   1955–1959  33,640  8.7  5,139  11.1  28,501  8.4   1960–1964  22,819  5.9  3,853  8.3  18,966  5.6   1965–1969  11,319  2.9  2,188  4.7  9,131  2.7   1970–1974  4,876  1.2  1,038  2.2  3,838  1.1   1975–1979  1,684  0.4  400  0.9  1,284  0.4  Born during the Jim Crow era               Born before 1945b  218,103  56.5  21,220  45.7  196,883  57.9   Born before 1965c  368,368  95.4  42,791  92.2  325,577  95.8  Vital status (as of 2016, when data provided)d,e               Alive  266,241  68.9  28,604  61.6  237,637  69.9   Deceased  120,006  31.1  17,813  38.4  102,193  30.1   Unknown  0  0  0  0  0  0  Tumor ER statusd               Positive  266,091  78.6  25,398  64.4  240,693  80.4   Negative  71,470  21.1  13,836  35.1  57,634  19.3   Borderline  1,207  0.4  191  0.5  1,016  0.4   Unknown  47,479  12.3  6,992  15.1  40,487  11.9   ER+ (imputed; crude), %    78.1    60.9    80.4  Contextual variables: state and census tract               Jim Crow laws in birthplaced,f                Yes  37,746  19.3  14,979  54.4  22,767  13.6    No  157,842  80.7  12,568  45.6  145,274  86.5    Unknown  190,659  49.4  18,870  40.7  171,789  50.6    Born in Jim Crow state (imputed), %    23.1    43.6    20.3   Census tract socioeconomic indexg                All years                 Quintile 1 (worst off)  49,842  13.1  18,312  39.9  31,530  9.4     Quintile 2  64,065  16.8  11,978  26.1  52,087  15.5     Quintile 3  75,334  19.8  7,624  16.6  67,710  20.2     Quintile 4  86,344  22.6  4,842  10.6  81,502  24.3     Quintile 5 (best off)  105,919  27.8  3,150  6.9  102,769  30.6    Diagnosed 1992–1995 (1990 census tract data)                 Quintile 1 (worst off)  8,431  13.7  3,195  47.5  5,236  9.5     Quintile 2  10,672  17.3  1,734  25.8  8,938  16.3     Quintile 3  11,995  19.4  864  12.9  11,131  20.2     Quintile 4  13,449  21.8  546  8.1  12,903  23.5     Quintile 5 (best off)  17,184  27.8  383  5.7  16,801  30.5    Diagnosed 1996–2005 (2000 census tract data)                 Quintile 1 (worst off)  24,467  13.2  8,911  42.3  15,556  9.4     Quintile 2  30,974  16.7  5,424  25.7  25,550  15.5     Quintile 3  36,803  19.8  3,227  15.1  33,576  20.4     Quintile 4  41,844  22.5  2,106  10.0  39,738  24.1     Quintile 5 (best off)  51,750  27.9  1,413  6.7  50,337  30.6    Diagnosed 2006–2012 (2010 census tract data)                 Quintile 1 (worst off)  16,944  12.7  6,206  34.3  10,738  9.3     Quintile 2  22,419  16.7  4,820  26.6  17,599  15.2     Quintile 3  26,536  19.8  3,533  19.5  23,003  19.9     Quintile 4  31,051  23.2  2,190  12.1  28,861  24.9     Quintile 5 (best off)  36,985  27.6  1,354  7.5  35,631  30.8     Unknown  4,743  1.2  511  1.1  4,232  1.2  Variable  Total (White and Black) (n = 386,247)  Black Non-Hispanic (n = 46,417)  White Non-Hispanic (n = 339,830)  No.  %  No.  %  No.  %  Registry               Atlanta  32,628  8.5  10,816  23.3  21,812  6.4   Connecticut  47,797  12.4  3,094  6.7  44,703  13.2   Detroit  53,555  13.9  12,317  26.5  41,238  12.1   Hawaii  4,207  1.1  113  0.2  4,094  1.2   Iowa  40,334  10.4  510  1.1  39,824  11.7   Los Angeles  61,281  15.9  11,618  25.0  49,663  14.6   New Mexico  14,420  3.7  265  0.6  14,155  4.2   Rural Georgia  1,596  0.4  551  1.2  1,045  0.3   San Francisco  40,614  10.5  5,020  10.8  35,594  10.5   San Jose–Monterey  18,807  4.9  526  1.1  18,281  5.4   Seattle  52,147  13.5  1,505  3.2  50,642  5.5   Utah  18,861  4.9  82  0.2  18,779  5.5  Year of diagnosisa  2002.4 (6.0)  2,003.0 (6.0)  2,002.3 (5.9)  Age at diagnosis, yearsa  60.3 (12.7)  57.2 (12.9)  60.8 (12.6)  Age group at diagnosis, years               25–44  45,989  11.9  8,242  17.8  37,747  11.1   45–54  88,358  22.9  12,175  26.2  76,183  22.4   54–84  251,900  65.2  26,000  56.0  225,900  66.5  Birth cohort               1915–1919  18,587  4.8  1,281  2.8  17,306  5.1   1920–1924  31,677  8.2  2,414  5.2  29,263  8.6   1925–1929  39,449  10.2  3,494  7.5  35,955  10.6   1930–1934  39,729  10.3  4,055  8.7  35,674  10.5   1935–1939  40,810  10.6  4,564  9.8  36,246  10.7   1940–1944  47,851  12.4  5,412  11.7  42,439  12.5   1945–1949  51,035  13.2  6,424  13.8  44,611  13.1   1950–1954  42,771  11.1  6,155  13.2  36,616  10.8   1955–1959  33,640  8.7  5,139  11.1  28,501  8.4   1960–1964  22,819  5.9  3,853  8.3  18,966  5.6   1965–1969  11,319  2.9  2,188  4.7  9,131  2.7   1970–1974  4,876  1.2  1,038  2.2  3,838  1.1   1975–1979  1,684  0.4  400  0.9  1,284  0.4  Born during the Jim Crow era               Born before 1945b  218,103  56.5  21,220  45.7  196,883  57.9   Born before 1965c  368,368  95.4  42,791  92.2  325,577  95.8  Vital status (as of 2016, when data provided)d,e               Alive  266,241  68.9  28,604  61.6  237,637  69.9   Deceased  120,006  31.1  17,813  38.4  102,193  30.1   Unknown  0  0  0  0  0  0  Tumor ER statusd               Positive  266,091  78.6  25,398  64.4  240,693  80.4   Negative  71,470  21.1  13,836  35.1  57,634  19.3   Borderline  1,207  0.4  191  0.5  1,016  0.4   Unknown  47,479  12.3  6,992  15.1  40,487  11.9   ER+ (imputed; crude), %    78.1    60.9    80.4  Contextual variables: state and census tract               Jim Crow laws in birthplaced,f                Yes  37,746  19.3  14,979  54.4  22,767  13.6    No  157,842  80.7  12,568  45.6  145,274  86.5    Unknown  190,659  49.4  18,870  40.7  171,789  50.6    Born in Jim Crow state (imputed), %    23.1    43.6    20.3   Census tract socioeconomic indexg                All years                 Quintile 1 (worst off)  49,842  13.1  18,312  39.9  31,530  9.4     Quintile 2  64,065  16.8  11,978  26.1  52,087  15.5     Quintile 3  75,334  19.8  7,624  16.6  67,710  20.2     Quintile 4  86,344  22.6  4,842  10.6  81,502  24.3     Quintile 5 (best off)  105,919  27.8  3,150  6.9  102,769  30.6    Diagnosed 1992–1995 (1990 census tract data)                 Quintile 1 (worst off)  8,431  13.7  3,195  47.5  5,236  9.5     Quintile 2  10,672  17.3  1,734  25.8  8,938  16.3     Quintile 3  11,995  19.4  864  12.9  11,131  20.2     Quintile 4  13,449  21.8  546  8.1  12,903  23.5     Quintile 5 (best off)  17,184  27.8  383  5.7  16,801  30.5    Diagnosed 1996–2005 (2000 census tract data)                 Quintile 1 (worst off)  24,467  13.2  8,911  42.3  15,556  9.4     Quintile 2  30,974  16.7  5,424  25.7  25,550  15.5     Quintile 3  36,803  19.8  3,227  15.1  33,576  20.4     Quintile 4  41,844  22.5  2,106  10.0  39,738  24.1     Quintile 5 (best off)  51,750  27.9  1,413  6.7  50,337  30.6    Diagnosed 2006–2012 (2010 census tract data)                 Quintile 1 (worst off)  16,944  12.7  6,206  34.3  10,738  9.3     Quintile 2  22,419  16.7  4,820  26.6  17,599  15.2     Quintile 3  26,536  19.8  3,533  19.5  23,003  19.9     Quintile 4  31,051  23.2  2,190  12.1  28,861  24.9     Quintile 5 (best off)  36,985  27.6  1,354  7.5  35,631  30.8     Unknown  4,743  1.2  511  1.1  4,232  1.2  Abbreviations: ER, estrogen receptor; SEER, Surveillance, Epidemiology, and End Results. a Values are expressed as mean (standard deviation). b Women born before 1945 lived at least the first 20 years of their lives during the Jim Crow era. c Women born before 1965 were born during the Jim Crow era. d Percentage distribution based on cases with known values. There were no cases with unknown vital status. For ER status, the percentage of tumors with unknown status ranged from 11.9% to 15.1%. For “Jim Crow” birthplace, approximately 40%–50% of cases had unknown status. e Data on vital status is included because place of birth in cancer registry records is typically obtained from the death certificate and hence is relevant to understanding missingness of place-of-birth data. f States with Jim Crow laws (n = 21): Alabama, Arizona, Arkansas, Delaware, Florida, Georgia, Kansas, Kentucky, Louisiana, Maryland, Mississippi, Missouri, New Mexico, North Carolina, Oklahoma, South Carolina, Tennessee, Texas, Virginia, West Virginia, Wyoming, and the District of Columbia; the remainder of the states did not have Jim Crow laws. g The SEER census tract socioeconomic index is the first factor yielded by a factor analysis of measures of income, poverty, unemployment, occupational class, education, and house value (51, 52). Source data comprised the 1990 and 2000 decennial census data and, starting in 2004, the 5-year annual American Community Survey data; values for intercensal years were interpolated (M. Yu, National Cancer Institute, personal communication, 2015); 1990 census tract boundaries were employed for the 1992–1995 cases, 2000 boundaries for the 1996–2005 cases, and 2010 boundaries for the 2006–2012 cases. The census tract quintile ranking was based on the national distribution for the index, and it was assigned to the case based on residential address at the time of diagnosis. Table 1. US White and Black Non-Hispanic Women, Aged 25–84 Years and Born Between 1915–1979, Diagnosed With Invasive Primary Breast Cancer, Overall and According to Race/Ethnicity, SEER 13 Registry Group, Surveillance, Epidemiology, and End Results, United States, 1992–2012 Variable  Total (White and Black) (n = 386,247)  Black Non-Hispanic (n = 46,417)  White Non-Hispanic (n = 339,830)  No.  %  No.  %  No.  %  Registry               Atlanta  32,628  8.5  10,816  23.3  21,812  6.4   Connecticut  47,797  12.4  3,094  6.7  44,703  13.2   Detroit  53,555  13.9  12,317  26.5  41,238  12.1   Hawaii  4,207  1.1  113  0.2  4,094  1.2   Iowa  40,334  10.4  510  1.1  39,824  11.7   Los Angeles  61,281  15.9  11,618  25.0  49,663  14.6   New Mexico  14,420  3.7  265  0.6  14,155  4.2   Rural Georgia  1,596  0.4  551  1.2  1,045  0.3   San Francisco  40,614  10.5  5,020  10.8  35,594  10.5   San Jose–Monterey  18,807  4.9  526  1.1  18,281  5.4   Seattle  52,147  13.5  1,505  3.2  50,642  5.5   Utah  18,861  4.9  82  0.2  18,779  5.5  Year of diagnosisa  2002.4 (6.0)  2,003.0 (6.0)  2,002.3 (5.9)  Age at diagnosis, yearsa  60.3 (12.7)  57.2 (12.9)  60.8 (12.6)  Age group at diagnosis, years               25–44  45,989  11.9  8,242  17.8  37,747  11.1   45–54  88,358  22.9  12,175  26.2  76,183  22.4   54–84  251,900  65.2  26,000  56.0  225,900  66.5  Birth cohort               1915–1919  18,587  4.8  1,281  2.8  17,306  5.1   1920–1924  31,677  8.2  2,414  5.2  29,263  8.6   1925–1929  39,449  10.2  3,494  7.5  35,955  10.6   1930–1934  39,729  10.3  4,055  8.7  35,674  10.5   1935–1939  40,810  10.6  4,564  9.8  36,246  10.7   1940–1944  47,851  12.4  5,412  11.7  42,439  12.5   1945–1949  51,035  13.2  6,424  13.8  44,611  13.1   1950–1954  42,771  11.1  6,155  13.2  36,616  10.8   1955–1959  33,640  8.7  5,139  11.1  28,501  8.4   1960–1964  22,819  5.9  3,853  8.3  18,966  5.6   1965–1969  11,319  2.9  2,188  4.7  9,131  2.7   1970–1974  4,876  1.2  1,038  2.2  3,838  1.1   1975–1979  1,684  0.4  400  0.9  1,284  0.4  Born during the Jim Crow era               Born before 1945b  218,103  56.5  21,220  45.7  196,883  57.9   Born before 1965c  368,368  95.4  42,791  92.2  325,577  95.8  Vital status (as of 2016, when data provided)d,e               Alive  266,241  68.9  28,604  61.6  237,637  69.9   Deceased  120,006  31.1  17,813  38.4  102,193  30.1   Unknown  0  0  0  0  0  0  Tumor ER statusd               Positive  266,091  78.6  25,398  64.4  240,693  80.4   Negative  71,470  21.1  13,836  35.1  57,634  19.3   Borderline  1,207  0.4  191  0.5  1,016  0.4   Unknown  47,479  12.3  6,992  15.1  40,487  11.9   ER+ (imputed; crude), %    78.1    60.9    80.4  Contextual variables: state and census tract               Jim Crow laws in birthplaced,f                Yes  37,746  19.3  14,979  54.4  22,767  13.6    No  157,842  80.7  12,568  45.6  145,274  86.5    Unknown  190,659  49.4  18,870  40.7  171,789  50.6    Born in Jim Crow state (imputed), %    23.1    43.6    20.3   Census tract socioeconomic indexg                All years                 Quintile 1 (worst off)  49,842  13.1  18,312  39.9  31,530  9.4     Quintile 2  64,065  16.8  11,978  26.1  52,087  15.5     Quintile 3  75,334  19.8  7,624  16.6  67,710  20.2     Quintile 4  86,344  22.6  4,842  10.6  81,502  24.3     Quintile 5 (best off)  105,919  27.8  3,150  6.9  102,769  30.6    Diagnosed 1992–1995 (1990 census tract data)                 Quintile 1 (worst off)  8,431  13.7  3,195  47.5  5,236  9.5     Quintile 2  10,672  17.3  1,734  25.8  8,938  16.3     Quintile 3  11,995  19.4  864  12.9  11,131  20.2     Quintile 4  13,449  21.8  546  8.1  12,903  23.5     Quintile 5 (best off)  17,184  27.8  383  5.7  16,801  30.5    Diagnosed 1996–2005 (2000 census tract data)                 Quintile 1 (worst off)  24,467  13.2  8,911  42.3  15,556  9.4     Quintile 2  30,974  16.7  5,424  25.7  25,550  15.5     Quintile 3  36,803  19.8  3,227  15.1  33,576  20.4     Quintile 4  41,844  22.5  2,106  10.0  39,738  24.1     Quintile 5 (best off)  51,750  27.9  1,413  6.7  50,337  30.6    Diagnosed 2006–2012 (2010 census tract data)                 Quintile 1 (worst off)  16,944  12.7  6,206  34.3  10,738  9.3     Quintile 2  22,419  16.7  4,820  26.6  17,599  15.2     Quintile 3  26,536  19.8  3,533  19.5  23,003  19.9     Quintile 4  31,051  23.2  2,190  12.1  28,861  24.9     Quintile 5 (best off)  36,985  27.6  1,354  7.5  35,631  30.8     Unknown  4,743  1.2  511  1.1  4,232  1.2  Variable  Total (White and Black) (n = 386,247)  Black Non-Hispanic (n = 46,417)  White Non-Hispanic (n = 339,830)  No.  %  No.  %  No.  %  Registry               Atlanta  32,628  8.5  10,816  23.3  21,812  6.4   Connecticut  47,797  12.4  3,094  6.7  44,703  13.2   Detroit  53,555  13.9  12,317  26.5  41,238  12.1   Hawaii  4,207  1.1  113  0.2  4,094  1.2   Iowa  40,334  10.4  510  1.1  39,824  11.7   Los Angeles  61,281  15.9  11,618  25.0  49,663  14.6   New Mexico  14,420  3.7  265  0.6  14,155  4.2   Rural Georgia  1,596  0.4  551  1.2  1,045  0.3   San Francisco  40,614  10.5  5,020  10.8  35,594  10.5   San Jose–Monterey  18,807  4.9  526  1.1  18,281  5.4   Seattle  52,147  13.5  1,505  3.2  50,642  5.5   Utah  18,861  4.9  82  0.2  18,779  5.5  Year of diagnosisa  2002.4 (6.0)  2,003.0 (6.0)  2,002.3 (5.9)  Age at diagnosis, yearsa  60.3 (12.7)  57.2 (12.9)  60.8 (12.6)  Age group at diagnosis, years               25–44  45,989  11.9  8,242  17.8  37,747  11.1   45–54  88,358  22.9  12,175  26.2  76,183  22.4   54–84  251,900  65.2  26,000  56.0  225,900  66.5  Birth cohort               1915–1919  18,587  4.8  1,281  2.8  17,306  5.1   1920–1924  31,677  8.2  2,414  5.2  29,263  8.6   1925–1929  39,449  10.2  3,494  7.5  35,955  10.6   1930–1934  39,729  10.3  4,055  8.7  35,674  10.5   1935–1939  40,810  10.6  4,564  9.8  36,246  10.7   1940–1944  47,851  12.4  5,412  11.7  42,439  12.5   1945–1949  51,035  13.2  6,424  13.8  44,611  13.1   1950–1954  42,771  11.1  6,155  13.2  36,616  10.8   1955–1959  33,640  8.7  5,139  11.1  28,501  8.4   1960–1964  22,819  5.9  3,853  8.3  18,966  5.6   1965–1969  11,319  2.9  2,188  4.7  9,131  2.7   1970–1974  4,876  1.2  1,038  2.2  3,838  1.1   1975–1979  1,684  0.4  400  0.9  1,284  0.4  Born during the Jim Crow era               Born before 1945b  218,103  56.5  21,220  45.7  196,883  57.9   Born before 1965c  368,368  95.4  42,791  92.2  325,577  95.8  Vital status (as of 2016, when data provided)d,e               Alive  266,241  68.9  28,604  61.6  237,637  69.9   Deceased  120,006  31.1  17,813  38.4  102,193  30.1   Unknown  0  0  0  0  0  0  Tumor ER statusd               Positive  266,091  78.6  25,398  64.4  240,693  80.4   Negative  71,470  21.1  13,836  35.1  57,634  19.3   Borderline  1,207  0.4  191  0.5  1,016  0.4   Unknown  47,479  12.3  6,992  15.1  40,487  11.9   ER+ (imputed; crude), %    78.1    60.9    80.4  Contextual variables: state and census tract               Jim Crow laws in birthplaced,f                Yes  37,746  19.3  14,979  54.4  22,767  13.6    No  157,842  80.7  12,568  45.6  145,274  86.5    Unknown  190,659  49.4  18,870  40.7  171,789  50.6    Born in Jim Crow state (imputed), %    23.1    43.6    20.3   Census tract socioeconomic indexg                All years                 Quintile 1 (worst off)  49,842  13.1  18,312  39.9  31,530  9.4     Quintile 2  64,065  16.8  11,978  26.1  52,087  15.5     Quintile 3  75,334  19.8  7,624  16.6  67,710  20.2     Quintile 4  86,344  22.6  4,842  10.6  81,502  24.3     Quintile 5 (best off)  105,919  27.8  3,150  6.9  102,769  30.6    Diagnosed 1992–1995 (1990 census tract data)                 Quintile 1 (worst off)  8,431  13.7  3,195  47.5  5,236  9.5     Quintile 2  10,672  17.3  1,734  25.8  8,938  16.3     Quintile 3  11,995  19.4  864  12.9  11,131  20.2     Quintile 4  13,449  21.8  546  8.1  12,903  23.5     Quintile 5 (best off)  17,184  27.8  383  5.7  16,801  30.5    Diagnosed 1996–2005 (2000 census tract data)                 Quintile 1 (worst off)  24,467  13.2  8,911  42.3  15,556  9.4     Quintile 2  30,974  16.7  5,424  25.7  25,550  15.5     Quintile 3  36,803  19.8  3,227  15.1  33,576  20.4     Quintile 4  41,844  22.5  2,106  10.0  39,738  24.1     Quintile 5 (best off)  51,750  27.9  1,413  6.7  50,337  30.6    Diagnosed 2006–2012 (2010 census tract data)                 Quintile 1 (worst off)  16,944  12.7  6,206  34.3  10,738  9.3     Quintile 2  22,419  16.7  4,820  26.6  17,599  15.2     Quintile 3  26,536  19.8  3,533  19.5  23,003  19.9     Quintile 4  31,051  23.2  2,190  12.1  28,861  24.9     Quintile 5 (best off)  36,985  27.6  1,354  7.5  35,631  30.8     Unknown  4,743  1.2  511  1.1  4,232  1.2  Abbreviations: ER, estrogen receptor; SEER, Surveillance, Epidemiology, and End Results. a Values are expressed as mean (standard deviation). b Women born before 1945 lived at least the first 20 years of their lives during the Jim Crow era. c Women born before 1965 were born during the Jim Crow era. d Percentage distribution based on cases with known values. There were no cases with unknown vital status. For ER status, the percentage of tumors with unknown status ranged from 11.9% to 15.1%. For “Jim Crow” birthplace, approximately 40%–50% of cases had unknown status. e Data on vital status is included because place of birth in cancer registry records is typically obtained from the death certificate and hence is relevant to understanding missingness of place-of-birth data. f States with Jim Crow laws (n = 21): Alabama, Arizona, Arkansas, Delaware, Florida, Georgia, Kansas, Kentucky, Louisiana, Maryland, Mississippi, Missouri, New Mexico, North Carolina, Oklahoma, South Carolina, Tennessee, Texas, Virginia, West Virginia, Wyoming, and the District of Columbia; the remainder of the states did not have Jim Crow laws. g The SEER census tract socioeconomic index is the first factor yielded by a factor analysis of measures of income, poverty, unemployment, occupational class, education, and house value (51, 52). Source data comprised the 1990 and 2000 decennial census data and, starting in 2004, the 5-year annual American Community Survey data; values for intercensal years were interpolated (M. Yu, National Cancer Institute, personal communication, 2015); 1990 census tract boundaries were employed for the 1992–1995 cases, 2000 boundaries for the 1996–2005 cases, and 2010 boundaries for the 2006–2012 cases. The census tract quintile ranking was based on the national distribution for the index, and it was assigned to the case based on residential address at the time of diagnosis. We first grouped the cases, stratified by race/ethnicity, into 5-year birth cohorts, and additionally stratified by age at diagnosis into 3 groups (<45 years (i.e., pre-menopausal), 45–54 years (perimenopausal), and 55 years or older (postmenopausal)), given the strong positive association between breast cancer ER+ status and age and the inverse association between age and the magnitude of the black versus white excess risk of ER− breast cancer (30–32). We then computed the percentage of ER+ tumors for each group, equal to the average percentage ER+ across the 10 imputed data sets, and employed logistic regression to evaluate the odds ratio and 95% confidence interval for linear trend across birth cohorts. For the haldane analyses, we set the reference birth cohort to be women born in 1945–1949, because it was the one birth cohort common to all 3 age groups (age at diagnosis: <45 years, 45–54 years, and 55 years or older) across the study time period. We re-expressed, as number of generations elapsed, the time difference between the reference birth cohort and each comparison birth cohort. To calculate the change by biological generation in the proportion of cases that were ER+ versus ER− we used the formula:   h=g×(ln(p2)−ln(p1))/(t2−t1),where h = haldane, p1 = proportion of cases in the reference birth cohort with the trait at time t1 (when diagnosed with breast cancer), p2 = proportion of cases in the comparison birth cohort with the trait at time t2 (when diagnosed with breast cancer), and g = number of generations elapsed between t1 and t2. This formula is a modification of the conventional haldane formula, which was developed for continuous traits (e.g., bone length) (10–13) and which calculates the difference, per pooled standard deviation, in the value of a trait at time t1 versus t2, divided by the number of generations (g) elapsed between these 2 time points. Paleontological and contemporary animal studies indicate that the haldane typically hovers around 0–0.1, with high values (indicating rapid change) defined as being >0.3 (12). We computed confidence intervals, based on log binomial regression, using the standard errors of the log risk ratio (ln(p2) − ln(p1)) for each birth cohort relative to 1947, multiplied by g/(t2 − t1). For generation length, our primary analyses set g to equal 22 years (i.e., midpoint for 20–24 years), based on historical and contemporary data about the US average age at first pregnancy (62–65). In sensitivity analyses we set g to equal 17, 27, and 32 years; because results were similar to those obtained for g = 22 years (see Web Tables 1–4, available at https://academic.oup.com/aje), we report only our primary analyses. RESULTS Table 1 presents data on the study population: US-born non-Hispanic women (46,417 black and 339,830 white) diagnosed with primary invasive breast cancer between 1992 and 2012, who were aged 25–84 years (inclusive) at diagnosis, born between 1915–1979, and included in the SEER 13 Registry Group. Fully 92.2% of the black women and 95.8% of the white women were born before 1965 (the point when Jim Crow laws were abolished), and 45.7% and 57.9%, respectively, were aged 20 years or older in 1965. Across all years, the percentages of black and white women, respectively, born in a Jim Crow state equaled 43.6% and 20.3%, and the percentages diagnosed with ER+ tumors equaled 60.9% and 80.4%. Figure 1 presents graphs, stratified by race/ethnicity and age, that simultaneously plot according to birth cohort the point estimates and 95% confidence intervals for: 1) the proportion of ER+ tumors, and 2) the haldane value (high values are indicated by the line at 0.3). The point estimates for these graphs (based on g = 22 years) and for the corresponding sensitivity analyses (g = 17, 27, and 32 years) are provided in Web Tables 1–4. Figure 1. View largeDownload slide Proportion of breast cancer cases that were estrogen receptor–positive and rate of change according to biological generation, among US-born black and white women with primary invasive breast cancer who were born 1915–1979, total and according to age at diagnosis, SEER 13 Registry Group, Surveillance, Epidemiology, and End Results, United States, 1992–2012. For estrogen receptor–positive percentage, age at diagnosis of <45 years (A), 45–54 years (B), or ≥55 years (C). For haldane values, age at diagnosis of <45 years (D), 45–54 years (E), or ≥55 years (F). Figure 1. View largeDownload slide Proportion of breast cancer cases that were estrogen receptor–positive and rate of change according to biological generation, among US-born black and white women with primary invasive breast cancer who were born 1915–1979, total and according to age at diagnosis, SEER 13 Registry Group, Surveillance, Epidemiology, and End Results, United States, 1992–2012. For estrogen receptor–positive percentage, age at diagnosis of <45 years (A), 45–54 years (B), or ≥55 years (C). For haldane values, age at diagnosis of <45 years (D), 45–54 years (E), or ≥55 years (F). Two patterns stand out. First, among both black and white women, the ER+ proportion increased over time only among women younger than age 55 years at diagnosis. Among the black women younger than age 45 years at diagnosis, the ER+ proportion rose from 43.7% (95% confidence interval (CI): 36.3, 51.0) for those born in 1945–1949 to 57.2% (95% CI: 52.2, 62.1) for those born in 1975–1979 (P for trend < 0.0001; for linear trend across birth cohorts, odds ratio (OR) = 1.13 (95% CI: 1.10, 1.17)); among those aged 45–54 years at diagnosis, the ER+ proportion rose from 50.6% (95% CI: 40.4, 60.7) for those born in 1935–1939 to 73.8% (95% CI: 68.1, 79.6) for those born in 1965–1969 (P for trend < 0.0001; for linear trend across birth cohorts, OR = 1.21 (95% CI: 1.17, 1.24)). Among black women aged 55 years or older at diagnosis, however, the ER+ proportion barely varied, and it hovered around 65% for those born in 1915–1919 through 1955–1959 (P for trend: 0.0976; for linear trend across birth cohorts, OR = 0.99 (95% CI: 0.98, 1.00)). For white women, the only marked increase in ER+ proportion occurred among those aged 45–54 years at diagnosis, for whom the percentage rose from 71.1% (95% CI: 68.4, 73.7) for those born in 1935–1939 to 83.4% (95% CI: 81.3, 85.5) for those born in 1965–1969 (P for trend < 0.0001; for linear trend across birth cohorts, OR = 1.16 (95% CI: 1.14, 1.18)). White women diagnosed at age 55 years or older were the only group to show a decline in ER+ tumors, from 84.8% (95% CI: 84.1, 85.4) for those born in 1915–1919 to 82.4% (95% CI: 80.7, 84.1) for those born in 1955–1959 (P for trend < 0.0001; for linear trend across birth cohorts, OR = 0.97 (95% CI: 0.96, 0.97)). Elevated haldane values (>|0.2|, with 95% CI excluding 0) occurred only among the black women diagnosed before age 55 years, and high haldane values (>|0.3|, with 95% CI excluding 0) occurred only among black women diagnosed between ages 45–54 years and born after 1950. By contrast, the value of the haldane consistently hovered around 0 for both black and white women diagnosed at age 55 years or older and also among white women diagnosed under age 45 years. Figure 2 presents analogous graphs, also stratified by Jim Crow birthplace (see Web Tables 1–4 for point estimates). The results indicate that an effect associated with Jim Crow laws occurred solely for black women under age 55 years, with the highest haldane values observed for those born in the states with Jim Crow laws. Additional analyses that further stratified on SEER census tract socioeconomic index did not alter these patterns (see Web Tables 1–4). Results were robust to the extreme scenarios employed in our sensitivity analyses (see Web Tables 5 and 6). Figure 2. View largeDownload slide Proportion of breast cancer cases that were estrogen receptor–positive and rate of change by biological generation, among US-born black and white women with primary invasive breast cancer who were born 1915–1979, total and according to age at diagnosis, and stratified by whether the state of birth had Jim Crow laws, SEER 13 Registry Group, Surveillance, Epidemiology, and End Results, United States, 1992–2012. For estrogen receptor–positive percentage, age at diagnosis of <45 years (A), 45–54 years (B), or ≥55 years (C). For haldane values, age at diagnosis of <45 years (D), 45–54 years (E), or ≥55 years (F). JC, states with Jim Crow laws; NJC, states without Jim Crow laws. Figure 2. View largeDownload slide Proportion of breast cancer cases that were estrogen receptor–positive and rate of change by biological generation, among US-born black and white women with primary invasive breast cancer who were born 1915–1979, total and according to age at diagnosis, and stratified by whether the state of birth had Jim Crow laws, SEER 13 Registry Group, Surveillance, Epidemiology, and End Results, United States, 1992–2012. For estrogen receptor–positive percentage, age at diagnosis of <45 years (A), 45–54 years (B), or ≥55 years (C). For haldane values, age at diagnosis of <45 years (D), 45–54 years (E), or ≥55 years (F). JC, states with Jim Crow laws; NJC, states without Jim Crow laws. DISCUSSION Our descriptive study provides novel data on the pace of change, according to biological generation, of breast cancer ER status at diagnosis among contemporary US-born black and white women. Key findings were that the percentage of ER+ cases rose, according to birth cohort (1915–1919 to 1975–1979), only among the women diagnosed before age 55 years; that the changes according to biological generation were greater for the black women than for white women; and that among the black women, they were greatest for those born in the states with Jim Crow laws (versus non–Jim Crow states), with this group the only one to exhibit high haldane values (>|0.3|). The net impact, among women under age 55 years, was to close the ER+ prevalence gap between black and white women by half. Such rapid phenotypic change can plausibly be driven only by factors exogenous to populations’ genomes (10–18), despite emphasis in the literature on genetic contributions to black versus white risk of ER− breast tumors (31–34). By contrast, among women age 55 years or older at diagnosis, no trend was evident among the black women, whereas the white women exhibited a slight decline in ER+ proportion over time. Interpretation of these results requires considering study strengths and limitations. First, we employed high-quality, NCI population-based, US cancer registry data, thus including all cases, regardless of access to care (66). Although the SEER 13 Registry Group, which in 2010 contained 13.4% of the US population (50), is not fully representative of the US population, its registry locations nevertheless were chosen to be inclusive of diverse US geographic regions and major racial/ethnic groups (66). Multiple imputation employed credible imputation models (30, 36, 58, 59), as underscored by our sensitivity analyses. Studies comparing self-reported race/ethnicity and cancer registry data, moreover, have found high congruence for black and white patients, suggesting that racial/ethnic misclassification was low (67). Our results would also not be biased by US immigration policy (e.g., the restrictive Immigration Act of 1917 and Immigration Act of 1924 or the Immigrant and Nationality Act of 1965, which expanded immigration) (68, 69), because all cases were US-born. We nevertheless lacked data on internal migration between year of birth and year of diagnosis; even without data on cumulative exposure, however, we could detect an impact of Jim Crow birthplace. We additionally lacked data on access to mammography. However, declining disparities between black and white women with respect to mammography screening rates (33, 70, 71) are unlikely to explain our results, because women younger than 45 years are not targeted for such screening, and we observed no rise in the ER+ proportion among the black women age 55 years or older. Our observational results are thus likely sufficiently sound to test our hypotheses about the pace of change, a necessary step before conducting studies to test etiological hypotheses about possible social and biological variables driving any observed trends. Consonant with our findings, the few studies on trends in the ER status of breast cancer cases provide evidence of a significant rise in ER+ proportion over time. An investigation of 11,195 US tumor specimens spanning the period of 1973–1992 found that the percentage of ER+ cases rose from 73% to 78%; unable to explain this finding by “technical improvements or changes in tumor size, age, or nodal status,” the authors inferred that “rising level of ER may reflect a change in breast cancer biology and in hormonal events that influence breast cancer genesis and growth” (72, p. 1601). A hospital-based study of 900 archival breast tumor specimens in Glasgow (United Kingdom) observed that, between 1984–1986 and 1996–1997, the proportion of ER+ cases likewise rose from 64.2% to 71.5% (73). A study of 1,290 archival specimens from the 1940s–1990s in Leeds (United Kingdom) attributed the finding that “[s]ignificantly more ER+ tumors were detected in the 1970s and 1990s cohorts compared with the 1940s cohort” to patients from the 1940s having “lived through two periods of food rationing during the First and Second World Wars” (74, pp. 272–273), given evidence that caloric restriction can affect breast cancer risk and subtype (74). A new analysis of 16,494 Chinese women diagnosed with invasive breast cancer between 1994 and 2014 in the Shanghai Cancer Center found that exposure to the Great Famine of 1959–1961 increased the risk of having an ER− tumor, especially among women diagnosed before age 50 years and who experienced the famine before menarche, leading the authors to hypothesize that “famine, malnutrition, or the associated lack of fruit and vegetable consumption in adulthood may be related to epidemiological heterogeneity within breast cancer subtypes” (40, p. 361). The handful of studies on Jim Crow laws and population health also support our findings. In this research, the strongest effects of Jim Crow abolition were observed among the US black population in Jim Crow states for decreased risk of infant mortality (23–26), premature mortality (27, 28), and ER− breast cancer (especially among black women born before 1965) (36). Likely explanations involve the myriad harms imposed by systemic, legal racial discrimination and extrajudicial violence, above and beyond forms of racial discrimination in non–Jim Crow states (19–29, 75–77). Additional evidence regarding greater change in traits according to biological generation among the US-born black versus white population comes from our haldane analyses of long-term changes in anthropometric traits and age at menarche, using nationally representative data from the National Health Examination Survey (1959–1962) and subsequent National Health and Nutrition Examination Studies (through 1999–2008) (15, 16, 18). For example, height (a trait positively associated with risk of breast cancer, and specifically hormone receptor–positive breast cancer (42)) exhibited high rates of phenotypic change (haldane > 0.3) chiefly between 1960 and 1980, especially for the black population in the highest income quintile (15). Additionally, for age at menarche (for which lower age is positively associated with risk of ER+ breast cancer (43)), haldane values of >0.3 were found solely among black women (all socioeconomic strata) and low-income white women who underwent menarche before 1960 (16). Together, these strands of evidence suggest that historical changes in socially determined early-life exposures affecting nutrition, age at menarche, and height may contribute to our study findings. Supporting this hypothesis are: 1) the persistent high and excess rates of being below the US poverty line among US black children compared with white children—notably shifting from an excess of 65.6% versus 14.4% in 1965 to 32.9% versus 12.1% in 2015 (Web Table 7) (78); and 2) steeper declines in poverty, between 1970 and 1980, in the Jim Crow versus non–Jim Crow states, especially for the black population (Web Tables 8 and 9) (79–81). Although we lacked data to explore heterogeneity in state effects, the observed Jim Crow effect was robust to inclusion of census tract socioeconomic data (Web Tables 1–6), and other research has shown that current spatial patterns of racialized impoverishment in Jim Crow states are associated with earlier spatial distributions of enslavement and greater severity of Jim Crow regimes (20–22, 82). Other socially patterned and temporally changing exposures have been associated with tumor phenotype at diagnosis (44, 45). For example, limited or no breastfeeding among parous women has been associated with increased risk of ER− breast tumors (43, 46–48). However, this association would be unlikely to explain our findings, because a black/white reversal in breastfeeding occurred in the United States in the mid-1960s, shifting from a higher to lower prevalence and duration among US black versus white women (83, 84). By contrast, the slight decline in ER+ tumors among white women diagnosed at age 55 years or older may in turn reflect their reduced use of hormone therapy following publication of the Women’s Health Initiative results (38, 39). Taken together, our novel findings open up new questions relevant to understanding changing determinants of, and health inequities in, breast cancer ER status. Empirical research linking high-quality early-life exposure data to subsequent adult cancer remains scant, including for breast cancer; the limited extant evidence, however, implicates early-life exposures (in utero and once born) to nutrition and carcinogens, as shaped by early-life socioeconomic position (41, 85, 86). Methods for augmenting cancer registry and vital statistics data on exposures by linking to online data capturing lifetime residential histories are likewise in their early stages, and they utilize data sources (such as credit-card transactions) for which electronic records typically extend only to the 1990s (87–89). Our study underscores the need for improved methods and data to pursue historically informed analyses of how societal context, past and present, shapes contemporary risk of breast cancer ER status and social disparities in their distribution. ACKNOWLEDGMENTS Author affiliations: Department of Social and Behavioral Sciences, Harvard T.H. Chan School of Public Health, Boston, Massachusetts (Nancy Krieger, Jacquelyn L. Jahn, Pamela D. Waterman, Jarvis T. Chen). This work was funded by the American Cancer Society Clinical Research Professorship awarded to N.K. Conflict of interest: none declared. Abbreviations CI confidence interval ER estrogen receptor ER+ estrogen receptor–positive ER− estrogen receptor–negative NCI National Cancer Institute OR odds ratio SEER Surveillance, Epidemiology, and End Results REFERENCES 1 Karsli-Ceppioglu S, Dagdemir A, Judes G, et al.  . Epigenetic mechanisms of breast cancer: an update of the current knowledge. Epigenomics . 2014; 6( 6): 651– 664. Google Scholar CrossRef Search ADS PubMed  2 Martinez-Arguelles DB, Papadopoulos V. Epigenetic regulation of the expression of genes involved in steroid hormone biosynthesis and action. Steroids . 2010; 75( 7): 467– 476. Google Scholar CrossRef Search ADS PubMed  3 Eyster KM. The estrogen receptors: an overview from different perspectives. Methods Mol Biol . 2016; 1366: 1– 10. Google Scholar CrossRef Search ADS PubMed  4 Hubbard WJ, Bland KI, Chaudry IH. The ERRor of our ways: estrogen-related receptors are about energy, not hormones, and are potential new targets for trauma and shock. Shock . 2015; 44( 1): 3– 15. Google Scholar CrossRef Search ADS PubMed  5 Perou CM, Sørlie T, Eisen MB, et al.  . Molecular portraits of human breast tumors. Nature . 2000; 406( 6797): 747– 752. Google Scholar CrossRef Search ADS PubMed  6 Braunstein LZ, Taghian AG. Molecular phenotype, multigene assays, and the locoregional management of breast cancer. Semin Radiat Oncol . 2016; 26( 1): 9– 16. Google Scholar CrossRef Search ADS PubMed  7 Beca F, Polyak K. Intratumor heterogeneity in breast cancer. Adv Exp Med Biol . 2016; 882: 169– 189. Google Scholar CrossRef Search ADS PubMed  8 Martelotto LG, Ng CK, Piscuoglio S, et al.  . Breast cancer intra-tumor heterogeneity. Breast Cancer Res . 2014; 16( 3): 210. Google Scholar CrossRef Search ADS PubMed  9 Krieger N. History, biology, and health inequities: emergent embodied phenotypes and the illustrative case of the breast cancer estrogen receptor. Am J Public Health . 2013; 103( 1): 22– 27. Google Scholar CrossRef Search ADS PubMed  10 Gingerich PD. Quantification and comparison of evolutionary rates. Am J Sci . 1993; 293( A): 453– 478. Google Scholar CrossRef Search ADS   11 Gingerich PD. Rates of evolution. Annu Rev Ecol Evol Syst . 2009; 40: 657– 675. Google Scholar CrossRef Search ADS   12 Hendry AP, Kinnison MT. The pace of modern life: measuring rates of contemporary microevolution. Evolution . 1999; 53( 6): 1637– 1653. Google Scholar CrossRef Search ADS PubMed  13 Hendry AP, Farrugia TJ, Kinnison MT. Human influences on rates of phenotypic change in wild animal populations. Mol Ecol . 2008; 17( 1): 20– 29. Google Scholar CrossRef Search ADS PubMed  14 DeLong JP, Forbes VE, Galic N, et al.  . How fast is fast? Eco-evolutionary dynamics and rates of change in populations and phenotypes. Ecol Evol . 2016; 6( 2): 573– 581. Google Scholar CrossRef Search ADS PubMed  15 Krieger N, Chen JT, Waterman PD, et al.  . History, haldanes and health inequities: exploring phenotypic changes in body size by generation and income level among the US-born white and black non-Hispanic populations, 1959–1962 to 2005–2008. Int J Epidemiol . 2013; 42( 1): 281– 295. Google Scholar CrossRef Search ADS PubMed  16 Krieger N, Kiang MV, Kosheleva A, et al.  . Age at menarche: 50-year socioeconomic trends among US-born black and white women. Am J Public Health . 2015; 105( 2): 388– 397. Google Scholar CrossRef Search ADS PubMed  17 Byars SG, Ewbank D, Govindaraju DR, et al.  . Natural selection in a contemporary human population. Proc Natl Acad Sci USA . 2010; 107( suppl 1): 1787– 1792. Google Scholar CrossRef Search ADS PubMed  18 Byars SG. Commentary: haldanes and trends in phenotypic change in humans. Int J Epidemiol . 2013; 42( 1): 295– 297. Google Scholar CrossRef Search ADS PubMed  19 Murray P. States’ Laws on Race and Color . Cincinnati, OH: Women’s Division of Christian Service, Board of Missions and Church Extensions, Methodist Church; 1950. 20 Wilkerson I. The Warmth of Other Suns: The Epic Story of America’s Great Migration . New York City, NY: Vintage Books; 2011. 21 Thompson-Miller R, Feagin JR, Picca LH. Jim Crow’s Legacy: The Lasting Impact of Segregation . New York City, NY: Rowman & Littlefield; 2015. 22 Cole S, Ring NJ, eds. The Folly of Jim Crow: Rethinking the Segregated South . Arlington, TX: Texas A&M University Press; 2012. 23 Chay KY, Greenstone M. The convergence in black-white infant mortality rates during the 1960’s. Am Econ Rev . 2000; 90( 2): 326– 332. Google Scholar CrossRef Search ADS   24 Almond DV, Chay KY, Greenstone M. Civil Rights, the War on Poverty, and Black-White convergence in infant mortality in the rural South and Mississippi. December 31, 2006. MIT Economics Working Paper No. 07-04. https://papers.ssrn.com/sol3/papers.cfm?abstract_id=961021. Accessed July 26, 2017. 25 Almond D, Chay KY. The long-run and intergenerational impact of poor infant health: evidence from cohorts born during the Civil Rights era. Working Paper. 2006. http://users.nber.org/~almond/chay_npc_paper.pdf. Accessed July 26, 2017. 26 Krieger N, Chen JT, Coull B, et al.  . The unique impact of abolition of Jim Crow laws on reducing inequities in infant death rates and implications for choice of comparison groups in analyzing societal determinants of health. Am J Public Health . 2013; 103( 12): 2234– 2244. Google Scholar CrossRef Search ADS PubMed  27 Krieger N, Chen JT, Coull BA, et al.  . Jim Crow and premature mortality among the US black and white population, 1960–2009: an age-period-cohort analysis. Epidemiology . 2014; 25( 4): 494– 504. Google Scholar CrossRef Search ADS PubMed  28 Kaplan G, Ranjit N, Burgard S. Lifting gates, lengthening lives: did civil rights policies improve the health of African-American women in the 1960s and 1970s? In: Schoeni RF, House JS, Kaplan G, et al.  , eds. Making Americans Healthier: Social and Economic Policy as Health Policy . New York, NY: Russell Sage Foundation; 2008: 145– 170. 29 Quadango J. Promoting civil rights through the welfare state: how Medicare integrated Southern hospitals. Soc Probl . 2000; 47( 1): 68– 89. Google Scholar CrossRef Search ADS PubMed  30 Kohler BA, Sherman RL, Howlader N, et al.  . Annual report to the Nation on the Status of Cancer, 1975–2011, featuring incidence of breast cancer subtypes by race/ethnicity, poverty, and state [published erratum appears in J Natl Cancer Inst. 2015;107(5):djv121; J Natl Cancer Inst. 2015;107(7):djv177]. J Natl Cancer Inst . 2015; 107( 6): djv048. Google Scholar CrossRef Search ADS PubMed  31 Newman LA. Disparities in breast cancer and African ancestry: a global perspective. Breast J . 2015; 21( 2): 133– 139. Google Scholar CrossRef Search ADS PubMed  32 Williams DR, Mohammed SA, Shields AE. Understanding and effectively addressing breast cancer in African American women: unpacking the social context. Cancer . 2016; 122( 14): 2138– 2149. Google Scholar CrossRef Search ADS PubMed  33 Vona-Davis L, Rose DP. The influence of socioeconomic disparities on breast cancer tumor biology and prognosis: a review. J Womens Health (Larchmt) . 2009; 18( 6): 883– 893. Google Scholar CrossRef Search ADS PubMed  34 Dietz EC, Sistrunk C, Miranda-Carboni G, et al.  . Triple-negative breast cancer in African-American women: disparities versus biology. Nat Rev Cancer . 2015; 15( 4): 248– 254. Google Scholar CrossRef Search ADS PubMed  35 Krieger N, Chen JT, Waterman PD. Temporal trends in the black/white breast cancer case ratio for estrogen receptor status: disparities are historically contingent, not innate. Cancer Causes Control . 2011; 22( 3): 511– 514. Google Scholar CrossRef Search ADS PubMed  36 Krieger N, Jahn JL, Waterman PD. Jim Crow and estrogen-receptor negative breast cancer: US-born black and white non-Hispanic women, 1992–2012. Cancer Causes Control . 2017; 28( 1): 49– 59. Google Scholar CrossRef Search ADS PubMed  37 Eng A, McCormack V, dos-Santos-Silva I. Receptor-defined subtypes of breast cancer in indigenous populations in Africa: a systematic review and meta-analysis. PLoS Med . 2014; 11( 9): e1001720. Google Scholar CrossRef Search ADS PubMed  38 Wei F, Miglioretti DL, Connelly MT, et al.  . Changes in women’s use of hormones after the Women’s Health Initiative estrogen and progestin trial by race, education, and income. J Natl Cancer Inst Monogr . 2005; 35: 106– 112. Google Scholar CrossRef Search ADS   39 Marjoribanks J, Farquhar C, Roberts H, et al.  . Long-term hormone therapy for perimenopausal and postmenopausal women. Cochrane Database Syst Rev . 2017; 1: CD004143. Google Scholar PubMed  40 Alimujiang A, Mo M, Liu Y, et al.  . The association between China’s Great famine and risk of breast cancer according to hormone receptor status: a hospital-based study. Breast Cancer Res Treat . 2016; 160( 2): 361– 369. Google Scholar CrossRef Search ADS PubMed  41 Clarke MA, Joshu CE. Early life exposures and adult cancer risk. Epidemiol Rev . 2017; 39( 1): 11– 27. Google Scholar CrossRef Search ADS PubMed  42 Zhang B, Shu XO, Delahanty RJ, et al.  . Height and breast cancer risk: evidence from prospective studies and Mendelian randomization. J Natl Cancer Inst . 2015; 107( 11): djv219. Google Scholar CrossRef Search ADS PubMed  43 Anderson KN, Schwab RB, Martinez ME. Reproductive risk factors and breast cancer subtypes: a review of the literature. Breast Cancer Res Treat . 2014; 144( 1): 1– 10. Google Scholar CrossRef Search ADS PubMed  44 Barnard ME, Boeke CE, Tamimi RM. Established breast cancer risk factors and risk of intrinsic tumor subtypes. Biochim Biophys Acta . 2015; 1856( 1): 73– 85. Google Scholar PubMed  45 Bernstein L, Lacey JV Jr. Receptors, associations, and risk factor differences by breast cancer subtypes: positive or negative? J Natl Cancer Inst . 2011; 103( 6): 451– 453. Google Scholar CrossRef Search ADS PubMed  46 Ma H, Ursin G, Xu X, et al.  . Reproductive factors and the risk of triple negative breast cancer in white women and African-American women: a pooled analysis. Breast Cancer Res . 2017; 19( 1): 6. Google Scholar CrossRef Search ADS PubMed  47 Islami F, Liu Y, Jemal A, et al.  . Breast feeding and breast cancer risk by receptor status—a systematic review and meta-analysis. Ann Oncol . 2015; 26( 12): 2398– 2407. Google Scholar PubMed  48 Lambertini M, Santoro L, Del Mastro L, et al.  . Reproductive behaviors and risk of developing breast cancer according to tumor subtype: a systematic review and meta-analysis of epidemiological studies. Cancer Treat Rev . 2016; 49: 65– 76. Google Scholar CrossRef Search ADS PubMed  49 US National Cancer Institute, Surveillance, Epidemiology, and End Results (SEER) Program. SEER Registry Groupings for Analysis. https://seer.cancer.gov/registries/terms.html. Accessed July 26, 2017. 50 US National Cancer Institute, Surveillance, Epidemiology, and End Results (SEER) Program. Number of persons by race and Hispanic ethnicity for SEER participants (2010 Census data). https://seer.cancer.gov/registries/data.html. Accessed July 26, 2017. 51 Yu M, Tatalovich Z, Gibson JT, et al.  . Using a composite index of socioeconomic status to investigate health disparities while protecting the confidentiality of cancer registry data. Cancer Causes Control . 2014; 25( 1): 81– 92. Google Scholar CrossRef Search ADS PubMed  52 Akinyemiju TF, Pisu M, Waterbor JW, et al.  . Socioeconomic status and incidence of breast cancer by hormone receptor subtype. Springerplus . 2015; 4: 508. Google Scholar CrossRef Search ADS PubMed  53 Krieger N, Chen JT, Waterman PD, et al.  . Geocoding and monitoring US socioeconomic inequalities in mortality and cancer incidence: does the choice of area-based measure and geographic level matter? The Public Health Disparities Geocoding Project. Am J Epidemiol . 2002; 156( 5): 471– 482. Google Scholar CrossRef Search ADS PubMed  54 Krieger N, Chen JT, Waterman PD, et al.  . Painting a truer picture of US socioeconomic and racial/ethnic health inequalities: the Public Health Disparities Geocoding Project. Am J Public Health . 2005; 95( 2): 312– 323. Google Scholar CrossRef Search ADS PubMed  55 Clegg LX, Reichman ME, Hankey BF, et al.  . Quality of race, Hispanic ethnicity, and immigrant status in population-based cancer registry data: implications for health disparity studies. Cancer Causes Control . 2007; 18( 2): 177– 187. Google Scholar CrossRef Search ADS PubMed  56 Montealegre JR, Zhou R, Amirian ES, et al.  . Uncovering nativity disparities in cancer patterns: multiple imputation strategy to handle missing nativity data in the Surveillance, Epidemiology, and End Results Data File. Cancer . 2014; 120( 8): 1203– 1211. Google Scholar CrossRef Search ADS PubMed  57 Pinheiro PS, Bungum TJ, Jin H. Limitations in the imputation strategy to handle missing nativity data in the Surveillance, Epidemiology, and End Results Program [letter]. Cancer . 2014; 120( 20): 3261– 3262. Google Scholar CrossRef Search ADS PubMed  58 Krieger N, Chen JT, Ware JH, et al.  . Race/ethnicity and breast cancer estrogen receptor status: impact of class, missing data, and modeling assumptions. Cancer Causes Control . 2008; 19( 10): 1305– 1318. Google Scholar CrossRef Search ADS PubMed  59 Andridge R, Noone AM, Howlader N. Imputing estrogen receptor (ER) status in a population-based cancer registry: a sensitivity analysis. Stat Med . 2017; 36( 6): 1014– 1028. Google Scholar CrossRef Search ADS PubMed  60 Lee KJ, Carlin JB. Multiple imputation for missing data: fully conditional specification versus multivariate normal imputation. Am J Epidemiol . 2010; 171( 5): 624– 632. Google Scholar CrossRef Search ADS PubMed  61 Harel O, Zhou XH. Multiple imputation: review of theory, implementation, and software. Stat Med . 2007; 26( 16): 3057– 3077. Google Scholar CrossRef Search ADS PubMed  62 US Bureau of the Census. Birth, Stillbirth, and Infant Mortality Statistics for the Birth Registration Area of the United States, 1931. Washington, DC: US Government Printing Office; 1934. (Seventeenth annual report). 63 Ventura SJ, Hamilton BE, Mathews TJ. National and State Patterns of Teen Births in the United States, 1940–2013. Hyattsville, MD: National Center for Health Statistics; 2014. (National vital statistics reports; vol 63 no 4). 64 Mathews TJ, Hamilton BE. Mean Age of Mother, 1970–2000. Hyattsville, MD: National Center for Health Statistics; 2002. (National vital statistics reports; vol 51 no 1). 65 Mathews TJ, Hamilton BE. Mean Age of Mothers Is on the Rise: United States, 2000–2014. Hyattsville, MD: National Center for Health Statistics; 2016. (NCHS data brief, no 232). 66 US National Cancer Institute, Surveillance, Epidemiology, and End Results (SEER) Program. Overview of the SEER Program. https://seer.cancer.gov/about/overview.html. Accessed July 26, 2017. 67 Gomez SL, Glaser SL. Misclassification of race/ethnicity in a population-based cancer registry (United States). Cancer Causes Control . 2006; 17( 6): 771– 781. Google Scholar CrossRef Search ADS PubMed  68 Chin GJ. The civil rights revolution comes to immigration law: A new look at the Immigration and Nationality Act of 1965. NCL Rev . 1996; 75( 1): 273– 345. 69 Ngai MM. The strange career of the illegal alien: immigration restriction and deportation policy in the United States, 1921–1965. Law Hist Rev . 2003; 21( 1): 69– 108. Google Scholar CrossRef Search ADS   70 Smith-Bindman R, Miglioretti DL, Lurie N, et al.  . Does utilization of screening mammography explain racial and ethnic differences in breast cancer? Ann Intern Med . 2016; 144( 8): 541– 553. Google Scholar CrossRef Search ADS   71 Harper S, Lynch J, Meersman SC, et al.  . Trends in area-socioeconomic and race-ethnic disparities in breast cancer incidence, stage at diagnosis, screening, mortality, and survival among women ages 50 years and over (1987–2005). Cancer Epidemiol Biomarkers Prev . 2009; 18( 1): 121– 131. Google Scholar CrossRef Search ADS PubMed  72 Pujol P, Hilsenbeck SG, Chamness GC, et al.  . Rising levels of estrogen receptor in breast cancer over 2 decades. Cancer . 1994; 74( 5): 1601– 1606. Google Scholar CrossRef Search ADS PubMed  73 Brown SB, Mallon EA, Edwards J, et al.  . Is the biology of breast cancer changing? A study of hormone receptor status 1984–1986 and 1996–1997. Br J Cancer . 2009; 100( 5): 807– 810. Google Scholar CrossRef Search ADS PubMed  74 Dowsett T, Verghese E, Pollock S, et al.  . The value of archival tissue blocks in understanding breast cancer biology. J Clin Pathol . 2014; 67( 3): 272– 275. Google Scholar CrossRef Search ADS PubMed  75 Krieger N. Discrimination and health inequities. In: Berkman LF, Kawachi I, Glymour M, eds. Social Epidemiology . 2nd ed. New York, NY: Oxford University Press; 2014: 63– 125. 76 Gee GC, Ford CL. Structural racism and health inequities: old issues, new directions. Du Bois Rev . 2011; 8( 1): 115– 132. Google Scholar CrossRef Search ADS PubMed  77 Bailey ZD, Krieger N, Agénor M, et al.  . Structural racism and health inequities: evidence and interventions. Lancet . 2017; 389( 10077): 1453– 1463. Google Scholar CrossRef Search ADS PubMed  78 US Census Bureau. Historical Poverty Tables: Peoples and Families—1959 to 2015. Table 3. Poverty status of people by age, race, and Hispanic origin. https://www.census.gov/data/tables/time-series/demo/income-poverty/historical-poverty-people.html. Accessed July 26, 2017. 79 US Census Bureau. Black persons by poverty status in 1969, 1979, 1989, and 1999—by state. CPH-L-166. https://www.census.gov/data/tables/time-series/dec/cph-series/cph-l/cph-l-166.html. Accessed July 26, 2017. 80 US Census Bureau. White persons by poverty status in 1969, 1979, 1989, and 1999—by state. CPH-L-166. https://www.census.gov/data/tables/time-series/dec/cph-series/cph-l/cph-l-165.html. Accessed July 26, 2017. 81 US Census Bureau. American Community Survey. https://www.census.gov/programs-surveys/acs/. Accessed July 26, 2017. 82 O’Connell HA. The impact of slavery on racial inequality in poverty in the contemporary US South. Soc Forces . 2012; 90( 3): 713– 734. Google Scholar CrossRef Search ADS   83 Eckhardt KW, Hendershot GE. Analysis of the reversal in breast feeding trends in the early 1970s. Public Health Rep . 1984; 99( 4): 410– 415. Google Scholar PubMed  84 Hirschman C, Hendershot GE. Trends in breast feeding among American mothers. Vital Health Stat [23] No. 3. DHEW Publication No. (PHS) 79-1979. Hyattsville, MD; 1979. https://www.cdc.gov/nchs/data/series/sr_23/sr23_003.pdf. Accessed July 26, 2017. 85 Vohra J, Marmot MG, Bauld L, et al.  . Socioeconomic position in childhood and cancer in adulthood: a rapid-review. J Epidemiol Community Health . 2016; 70( 6): 629– 634. Google Scholar CrossRef Search ADS PubMed  86 Pudrovska T, Anikputa B. The role of early-life socioeconomic status in breast cancer incidence and mortality: unraveling life course mechanisms. J Aging Health . 2012; 24( 2): 323– 344. Google Scholar CrossRef Search ADS PubMed  87 Hurley S, Hertz A, Nelson DO, et al.  . Tracing a path to the past: exploring the use of commercial credit reporting data to construct residential histories for epidemiologic studies of environmental exposures. Am J Epidemiol . 2017; 185( 3): 238– 246. Google Scholar PubMed  88 Wheeler DC, Wang A. Assessment of residential history generation using a public-record database. Int J Environ Res Public Health . 2015; 12( 9): 11670– 11682. Google Scholar CrossRef Search ADS PubMed  89 Jacquez GM, Slotnick MJ, Meliker JR, et al.  . Accuracy of commercially available residential histories for epidemiologic studies. Am J Epidemiol . 2011; 173( 2): 236– 243. Google Scholar CrossRef Search ADS PubMed  © The Author(s) 2018. Published by Oxford University Press on behalf of the Johns Hopkins Bloomberg School of Public Health. All rights reserved. For permissions, please e-mail: [email protected]. This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/about_us/legal/notices)

Hansen, Merethe S;Licaj, Idlir;Braaten, Tonje;Langhammer, Arnulf;Marchand, Loic Le;Gram, Inger T

doi: 10.1093/aje/kwx339pmid: 29087432

Abstract Whether women are more susceptible than men to smoking-related lung cancer has been a topic of controversy. To address this question, we compared risks of lung cancer associated with smoking by sex. Altogether, 585,583 participants from 3 Norwegian cohorts (Norwegian Counties Study, 40 Years Study, and Cohort of Norway (CONOR) Study) were followed until December 31, 2013, through linkage of data to national registries. We used Cox proportional hazards models and 95% confidence intervals to estimate risks. During nearly 12 million person-years of follow-up, 6,534 participants (43% women) were diagnosed with lung cancer. More men than women were heavier smokers. Compared with never smokers, male and female current smokers with ≥16 pack-years of smoking had hazard ratios for lung cancer of 27.24 (95% confidence interval (CI): 22.42, 33.09) and 23.90 (95% CI: 20.57, 27.76), respectively (P for heterogeneity = 0.30). In contrast, for current smokers, in a model with pack-years measured continuously, men had a hazard ratio of 1.43 (95% CI: 1.39, 1.48) and women a hazard ratio of 1.64 (95% CI: 1.57, 1.71) for each 10–pack-year increment of smoking (P for heterogeneity < 0.01). Our results suggest that women have an increased susceptibility to lung cancer compared with men, given the same lifetime smoking exposure. cohort studies, Cohort of Norway (CONOR) Study, histology, lung neoplasms, sex differences, smoking Tobacco smoking is the predominant cause of lung cancer, which is one of the most lethal human cancers, with a 5-year survival of only 10%–15%. Lung cancer is the most common cancer in men and the third most common cancer in women worldwide (1). In Norway, lung cancer is the second most frequent cancer among men and the third most frequent cancer among women. Lung cancer incidence in Norway increased in both sexes until 2012, when it started to plateau for men but continued to increase for women (2). The prevalence of daily smoking among Norwegian men peaked at 65% during the late 1950s, while in women the peak was at 37% around 1970 (3). In 2013, the prevalence was 15% for both sexes (4). Today, the majority of adults in Norway are ever (i.e., current or former) smokers. The continuing rising incidence of lung cancer among women globally has raised the possibility of a sex difference in the association of smoking with lung cancer. For women, a higher proportion of lung cancer patients are diagnosed at a younger age, at an earlier stage, and with adenocarcinoma, in comparison with men (1). While there are conflicting findings from case-control studies (5–7), results from 5 recent cohort studies showed similar lung cancer incidence rates among men and women with comparable smoking histories (8–12). Neither the most recent World Cancer Report (1) nor the 2014 US Surgeon General’s Report (13) discusses a possible sex difference in the risk of smoking-related cancer. The purpose of this study was to examine, by sex, the association between different measures of smoking exposure and risk of lung cancer, overall and by histological subtype, in a large Norwegian cohort. METHODS Study sample The study population included 635,840 men and women, born between 1897 and 1975, recruited from different Norwegian Health Screening Surveys conducted by the Norwegian National Health Screening Service (now included in the Norwegian Institute of Public Health): the Norwegian Counties Study, the 40 Years Study, and the Cohort of Norway (CONOR) Study. The Norwegian Counties Study took place between 1974 and 1978. Everyone aged 35–49 years and a 10% random sample of persons aged 20–34 years residing in 3 rural Norwegian counties (Finnmark, Sogn og Fjordane, and Oppland) were invited to undergo regular screening examinations for cardiovascular disease. The participation rate was 88% (14–16). In the 40 Years Study, men and women aged 40–42 years from all counties in Norway were invited to participate in a health survey during 1985–1999. In some counties, broader age groups were invited. These surveys included 420,000 Norwegians. The participation rate was 69% (17, 18). The CONOR Study consisted of 10 surveys (Tromsø Health Study IV, the second Nord-Trøndelag Health Study, the Hordaland Health Study, Oslo Study II, the Oslo Health Study, the Oppland and Hedmark Health Study, Tromsø Health Study V, the Oslo Immigrant Health Study, the Troms and Finnmark Health Study, and the second Romsås in Motion Study) from different regions in Norway, including different age groups ranging from 20 years to 103 years. These surveys were conducted in 1994–2003. The overall participation rate for the CONOR Study was 58% (14, 19, 20). In all of the studies, participants completed a baseline questionnaire including detailed assessments of smoking habits and other lifestyle factors. Height and weight were measured at the screening facility by trained nurses and were used to calculate body mass index (BMI; weight in kilograms divided by squared height in meters). We excluded participants who emigrated or died before the start of follow-up (n = 647), those with prevalent cancer (n = 11,321), and those with missing information on vital status (n = 190), measures of smoking exposure (n = 6,303), or any of the covariates (BMI, education, and physical activity; n = 31,796). Altogether, 50,257 participants were excluded, leaving 585,583 persons (52% women) in the analytical cohort. The present study was approved by the Regional Committee for Medical Research Ethics South-East, Norway, and the National Data Inspectorate. More details about our study population may be found elsewhere (21–23). Exposure information Information on current and former daily smoking, duration of smoking (in years), and number of cigarettes smoked per day was collected from the questionnaires. Former smokers were also asked about amount of time (years and/or months) since quitting. Only the CONOR Study asked participants about age at smoking initiation. In the other studies, we calculated this variable for both current (age at enrollment minus years of smoking) and former (age at enrollment minus years since quitting and duration of smoking) smokers. Among the 367,046 ever smokers, the proportion of missing values was <2% for quantity smoked (cigarettes/day) (n = 6,552), <1% for smoking duration (n = 3,051), <3% for pack-years of smoking (i.e., number of cigarettes smoked per day divided by 20, multiplied by smoking duration in years) (n = 8,280), and 21.6% for age at smoking initiation (n = 79,226). In addition, 56% (n = 77,323) of the former smokers had missing values for years since quitting smoking and age at smoking initiation. We categorized ever smokers according to age at smoking initiation (<16, 16–20, or ≥21 years), number of cigarettes smoked per day (1–10, 11–20, or >20), smoking duration (1–9, 10–19, 20–29, or ≥30 years), and pack-years of smoking (1–5, 6–15, or ≥16). Former smokers were categorized by time since smoking cessation (0–4, 5–9, or ≥10 years). We adjusted for physical activity level at study enrollment (sedentary (reading, television-watching, and other seated activity), moderate (walking, bicycling, and/or similar activities for ≥4 hours/week), or heavy (light sports or heavy gardening for ≥4 hours/week, heavy exercise, or daily competitive sports)) and BMI at study enrollment. We merged group 1 (BMI <18.5) and group 2 (BMI 18.5–24.9) in the World Health Organization’s BMI classification and retained group 3 (BMI ≥25.0) and group 4 (BMI ≥30.0) (24). We used the most recent information regarding duration of education obtained from Statistics Norway to classify subjects into 3 education categories: <10, 10–12, and ≥13 years of education. Follow-up and endpoints We used the unique 11-digit personal identification number assigned to all residents of Norway to follow all participants for 1) cancer (through linkage to the Cancer Registry of Norway) and 2) emigration or death (through linkage to the Central Population Register). These national registries are both accurate and virtually complete (25). Person-years were calculated from age at enrollment to age at lung cancer diagnosis, any incident cancer diagnosis (except basal cell carcinoma), emigration, death, or the end of follow-up (December 31, 2013), whichever occurred first. Cancer sites were identified by the anatomical sites and histological codes in the International Classification of Diseases for Oncology (26). All primary incident carcinomas of the trachea, bronchus, and lung (International Classification of Diseases, Seventh Revision, code 162 or corresponding codes from the International Classification of Diseases, Ninth Revision, or the International Classification of Diseases, Tenth Revision) were considered. Lung cancers were classified into 6 histological subtypes (squamous cell carcinoma, adenocarcinoma, large-cell carcinoma, other not specified non-small-cell carcinoma, small-cell carcinoma, and other carcinoma) according to the World Health Organization’s International Histological Classification (26). We present results on the risk of lung cancer overall and separately on risks of adenocarcinoma, squamous cell carcinoma, and small-cell carcinoma, which were the most frequent histological subtypes of lung cancer. Statistical analysis We calculated the age-standardized (2000 projected US population (27)) incidence rate of overall lung cancer by sex and smoking status. All analyses were sex-specific unless otherwise noted. We used a Cox proportional hazards model with attained age as the underlying time scale to estimate multivariate-adjusted hazard ratios and 95% confidence intervals for the associations between different measures of smoking exposure and the risk of lung cancer, overall and by histological subtype. We stratified the Cox models by cohort study and birth cohort (≤1950 and >1950) to overcome the heterogeneity for these variables. The a priori–selected covariates included in the final models were: physical activity level (sedentary, moderate, or heavy), BMI (24), and duration of education (<10, 10–12, or ≥13 years), all measured at enrollment. Never smokers were used as the reference group in all categorical smoking analyses, except for the association between years since cessation and lung cancer risk, where we used current smokers as the reference group. For former, current, and ever smokers, we estimated dose-response associations between lung cancer (overall) and the following continuous variables: smoking duration in 10-year increments, tens of cigarettes smoked per day, tens of pack-years, and age at smoking initiation. We evaluated the association between each 10-year increment of time since smoking cessation and lung cancer risk for former smokers only. In contrast to the categorical analyses, never smokers were excluded from the continuous analyses. We used fractional polynomials to determine the function of the different smoking exposures that best fitted the data (28). We entered the continuous variables into the multivariate Cox regression models via a set of defined transformations (x − 2, x − 1, x − 0.5, x0.5, x1, x2, x3, and log(x)), allowing for a maximum of 2 terms in the model. We found, as a result of these analyses, that the log-transformed model best fitted our data. We then compared the log-transformed effect of each smoking exposure for men and women and found similar sex differences. We tested for trend across categories of measures of smoking for ever smokers based on the median values in each category, with the lowest category of each smoking exposure used as the reference group. We used the Wald test to test for heterogeneity by sex for the measures of smoking exposure and the risk of lung cancer. We tested and found that the criteria for the proportional hazards assumption were met using Schoenfeld residuals (data not shown). We performed similar analyses after excluding participants who were diagnosed with lung cancer within 2 years of enrollment. Possible interactions between smoking status and education (3 categories), BMI (3 categories), and physical activity (3 categories) were assessed. When we analyzed small-cell carcinomas, we collapsed men and women due to the small number of cases among never smokers. We performed the analyses using STATA, version 14.0 (StataCorp LP, College Station, Texas). Two-sided P values less than 0.05 were considered statistically significant. RESULTS At enrollment, the proportions of never, former, and current smokers were 34%, 26%, and 40% in men and 41%, 21%, and 38% in women. During nearly 12 million person-years of follow-up, 6,534 participants (43% women) were diagnosed with lung cancer. For men, the age-standardized incidence rates of lung cancer among never, former, and current smokers were 9.2 per 100,000 person-years, 61.3 per 100,000 person-years, and 275.2 per 100,000 person-years, respectively. For women, the corresponding numbers were 17.6, 42.2, and 207.7, respectively. Adenocarcinomas were the most common tumor type for both sexes (33% in men and 41% in women). During follow-up, 14% of men and 9% of women died (from all causes). Table 1 shows that in the Norwegian Counties Study, 51% of men and 40% of women were current smokers at enrollment. In the most recent CONOR Study, these numbers were 31% for men and 32% for women. Stratified by birth cohort, 43% of men and 37% of women born in or before 1950 were current smokers. For those born after 1950, 37% of men and 39% of women were current smokers. Altogether, 18% of men and 8% of women had smoked for 30 years or more. Thirty-two percent of men and 45% of women had started to smoke after age 20 years. Age at enrollment increased from the earliest study (the Norwegian Counties Study) to the most recent study (CONOR), while years of follow-up decreased for both sexes (data not shown). The mean age at lung cancer diagnosis was 64 years for men and 63 years for women (see Web Table 1, available at https://academic.oup.com/aje). Table 1. Selected Characteristics of Study Participants at Enrollment, by Sex and Smoking Status (n = 585,583), Norwegian Health Screening Surveys, 1974–2003 Characteristic  Total No. of Participants  Men (n = 282,637)  Women (n = 302,946)  Current Smoker  Former Smoker  Never Smoker  Current Smoker  Former Smoker  Never Smoker  No.  %  No.  %  No.  %  No.  %  No.  %  No.  %  Total  585,583  113,033  40  74,639  26  94,965  34  115,350  38  64,024  21  123,572  41  Subcohort                             Norwegian Counties Study  83,500  21,416  51  9,599  23  10,898  26  16,501  40  6,054  14  19,032  46   40 Years Study  384,864  74,325  40  47,919  26  62,863  34  79,214  40  43,267  21  77,276  39   CONOR Study  117,219  17,292  31  17,121  31  21,204  38  19,635  32  14,703  24  27,264  44  Birth cohort                             ≤1950  294,280  62,008  43  42,182  29  40,698  28  55,898  37  28,262  19  65,232  44   >1950  291,303  51,025  37  32,457  24  54,267  39  59,452  39  35,762  23  58,340  38  Age at enrollment, yearsa  43 (8)  42 (7)  45 (10)  42 (8)  42 (7)  43 (8)  44 (10)  Person-years of follow-up  11,553,611  2,282,012    1,446,049    1,847,649    2,289,524    1,211,302    2,477,075    Age at diagnosis, yearsa  64 (10)  64 (10)  69 (11)  61 (12)  62 (9)  65 (12)  66 (12)  Histological typeb                             Lung cancer overall  5,514  3,147  85  458  12  109  3  2,367  84  232  8  221  8   Adenocarcinoma  1,384  957  26  172  5  63  2  900  32  130  5  119  4   Squamous cell carcinoma  1,293  811  22  99  3  9  <1  346  12  20  <1  8  <1   Small-cell carcinoma  1,161  561  15  49  1  2  <1  516  18  21  <1  12  <1  Education, years                             <10  137,691  33,627  30  16,754  22  12,735  13  36,899  32  13,081  21  24,595  20   10–12  320,642  63,463  56  41,092  55  49,477  52  66,025  57  36,648  57  63,937  52   ≥13  127,251  15,943  14  16,793  23  32,753  35  12,426  11  14,295  22  35,041  28  Physical activity levelc                             Sedentary  125,071  28,268  25  14,273  19  15,928  17  29,010  25  12,509  20  25,083  20   Moderate  297,478  52,374  46  33,423  45  38,816  41  65,408  57  35,703  56  71,754  58   High  163,034  32,391  29  26,943  36  40,221  42  20,932  18  15,812  24  26,735  22  BMIa,d  25 (4)  25 (3)  26 (3)  26 (3)  24 (4)  25 (4)  25 (4)  Duration of smoking, yearse                             1–9  60,086  5,224  5  20,745  28      7,760  7  26,357  41       10–19  104,359  19,530  17  31,031  42      28,118  24  25,680  40       20–29  174,120  75,582  67  16,810  23      72,391  63  9,337  15       ≥30  25,300  12,214  11  5,159  7      6,285  6  1,642  3      Quantity smoked, cigarettes/daye                             1–10  187,570  41,501  37  36,138  49      63,816  55  46,115  72       11–20  150,985  58,860  52  28,796  39      48,102  42  15,227  24       >20  21,816  9,758  9  7,537  10      3,028  3  1,493  2      ≥16 pack-years of smoking  103,048  50,939  45  15,678  21      31,475  27  4,956  8      Age at initiation of smoking, yearse                             ≥21  81,763  30,183  27  3,936  5      43,323  38  4,321  7       16–20  158,631  62,515  55  19,312  26      57,529  50  19,275  30       <16  47,426  19,989  18  7,283  10      14,041  12  6,113  10      ≤4 years since quitting smokingc  14,218      7,485  24          6,733  23      Characteristic  Total No. of Participants  Men (n = 282,637)  Women (n = 302,946)  Current Smoker  Former Smoker  Never Smoker  Current Smoker  Former Smoker  Never Smoker  No.  %  No.  %  No.  %  No.  %  No.  %  No.  %  Total  585,583  113,033  40  74,639  26  94,965  34  115,350  38  64,024  21  123,572  41  Subcohort                             Norwegian Counties Study  83,500  21,416  51  9,599  23  10,898  26  16,501  40  6,054  14  19,032  46   40 Years Study  384,864  74,325  40  47,919  26  62,863  34  79,214  40  43,267  21  77,276  39   CONOR Study  117,219  17,292  31  17,121  31  21,204  38  19,635  32  14,703  24  27,264  44  Birth cohort                             ≤1950  294,280  62,008  43  42,182  29  40,698  28  55,898  37  28,262  19  65,232  44   >1950  291,303  51,025  37  32,457  24  54,267  39  59,452  39  35,762  23  58,340  38  Age at enrollment, yearsa  43 (8)  42 (7)  45 (10)  42 (8)  42 (7)  43 (8)  44 (10)  Person-years of follow-up  11,553,611  2,282,012    1,446,049    1,847,649    2,289,524    1,211,302    2,477,075    Age at diagnosis, yearsa  64 (10)  64 (10)  69 (11)  61 (12)  62 (9)  65 (12)  66 (12)  Histological typeb                             Lung cancer overall  5,514  3,147  85  458  12  109  3  2,367  84  232  8  221  8   Adenocarcinoma  1,384  957  26  172  5  63  2  900  32  130  5  119  4   Squamous cell carcinoma  1,293  811  22  99  3  9  <1  346  12  20  <1  8  <1   Small-cell carcinoma  1,161  561  15  49  1  2  <1  516  18  21  <1  12  <1  Education, years                             <10  137,691  33,627  30  16,754  22  12,735  13  36,899  32  13,081  21  24,595  20   10–12  320,642  63,463  56  41,092  55  49,477  52  66,025  57  36,648  57  63,937  52   ≥13  127,251  15,943  14  16,793  23  32,753  35  12,426  11  14,295  22  35,041  28  Physical activity levelc                             Sedentary  125,071  28,268  25  14,273  19  15,928  17  29,010  25  12,509  20  25,083  20   Moderate  297,478  52,374  46  33,423  45  38,816  41  65,408  57  35,703  56  71,754  58   High  163,034  32,391  29  26,943  36  40,221  42  20,932  18  15,812  24  26,735  22  BMIa,d  25 (4)  25 (3)  26 (3)  26 (3)  24 (4)  25 (4)  25 (4)  Duration of smoking, yearse                             1–9  60,086  5,224  5  20,745  28      7,760  7  26,357  41       10–19  104,359  19,530  17  31,031  42      28,118  24  25,680  40       20–29  174,120  75,582  67  16,810  23      72,391  63  9,337  15       ≥30  25,300  12,214  11  5,159  7      6,285  6  1,642  3      Quantity smoked, cigarettes/daye                             1–10  187,570  41,501  37  36,138  49      63,816  55  46,115  72       11–20  150,985  58,860  52  28,796  39      48,102  42  15,227  24       >20  21,816  9,758  9  7,537  10      3,028  3  1,493  2      ≥16 pack-years of smoking  103,048  50,939  45  15,678  21      31,475  27  4,956  8      Age at initiation of smoking, yearse                             ≥21  81,763  30,183  27  3,936  5      43,323  38  4,321  7       16–20  158,631  62,515  55  19,312  26      57,529  50  19,275  30       <16  47,426  19,989  18  7,283  10      14,041  12  6,113  10      ≤4 years since quitting smokingc  14,218      7,485  24          6,733  23      Abbreviations: BMI, body mass index; CONOR, Cohort of Norway. a Values are expressed as mean (standard deviation). b Numbers of lung cancer subtypes do not sum to the numbers in the “overall” groups because only the main subtypes are shown in the table. c Physical activity level was defined as sedentary (reading, television-watching, and other seated activity), moderate (walking, bicycling, or similar activities for ≥4 hours/week), or heavy (light sports or heavy gardening for ≥4 hours/week, heavy exercise, or daily competitive sports). d Weight (kg)/height (m)2. e Percentages for smoking duration, quantity smoked, and age at smoking initiation do not sum to 100% because of missing values. Table 1. Selected Characteristics of Study Participants at Enrollment, by Sex and Smoking Status (n = 585,583), Norwegian Health Screening Surveys, 1974–2003 Characteristic  Total No. of Participants  Men (n = 282,637)  Women (n = 302,946)  Current Smoker  Former Smoker  Never Smoker  Current Smoker  Former Smoker  Never Smoker  No.  %  No.  %  No.  %  No.  %  No.  %  No.  %  Total  585,583  113,033  40  74,639  26  94,965  34  115,350  38  64,024  21  123,572  41  Subcohort                             Norwegian Counties Study  83,500  21,416  51  9,599  23  10,898  26  16,501  40  6,054  14  19,032  46   40 Years Study  384,864  74,325  40  47,919  26  62,863  34  79,214  40  43,267  21  77,276  39   CONOR Study  117,219  17,292  31  17,121  31  21,204  38  19,635  32  14,703  24  27,264  44  Birth cohort                             ≤1950  294,280  62,008  43  42,182  29  40,698  28  55,898  37  28,262  19  65,232  44   >1950  291,303  51,025  37  32,457  24  54,267  39  59,452  39  35,762  23  58,340  38  Age at enrollment, yearsa  43 (8)  42 (7)  45 (10)  42 (8)  42 (7)  43 (8)  44 (10)  Person-years of follow-up  11,553,611  2,282,012    1,446,049    1,847,649    2,289,524    1,211,302    2,477,075    Age at diagnosis, yearsa  64 (10)  64 (10)  69 (11)  61 (12)  62 (9)  65 (12)  66 (12)  Histological typeb                             Lung cancer overall  5,514  3,147  85  458  12  109  3  2,367  84  232  8  221  8   Adenocarcinoma  1,384  957  26  172  5  63  2  900  32  130  5  119  4   Squamous cell carcinoma  1,293  811  22  99  3  9  <1  346  12  20  <1  8  <1   Small-cell carcinoma  1,161  561  15  49  1  2  <1  516  18  21  <1  12  <1  Education, years                             <10  137,691  33,627  30  16,754  22  12,735  13  36,899  32  13,081  21  24,595  20   10–12  320,642  63,463  56  41,092  55  49,477  52  66,025  57  36,648  57  63,937  52   ≥13  127,251  15,943  14  16,793  23  32,753  35  12,426  11  14,295  22  35,041  28  Physical activity levelc                             Sedentary  125,071  28,268  25  14,273  19  15,928  17  29,010  25  12,509  20  25,083  20   Moderate  297,478  52,374  46  33,423  45  38,816  41  65,408  57  35,703  56  71,754  58   High  163,034  32,391  29  26,943  36  40,221  42  20,932  18  15,812  24  26,735  22  BMIa,d  25 (4)  25 (3)  26 (3)  26 (3)  24 (4)  25 (4)  25 (4)  Duration of smoking, yearse                             1–9  60,086  5,224  5  20,745  28      7,760  7  26,357  41       10–19  104,359  19,530  17  31,031  42      28,118  24  25,680  40       20–29  174,120  75,582  67  16,810  23      72,391  63  9,337  15       ≥30  25,300  12,214  11  5,159  7      6,285  6  1,642  3      Quantity smoked, cigarettes/daye                             1–10  187,570  41,501  37  36,138  49      63,816  55  46,115  72       11–20  150,985  58,860  52  28,796  39      48,102  42  15,227  24       >20  21,816  9,758  9  7,537  10      3,028  3  1,493  2      ≥16 pack-years of smoking  103,048  50,939  45  15,678  21      31,475  27  4,956  8      Age at initiation of smoking, yearse                             ≥21  81,763  30,183  27  3,936  5      43,323  38  4,321  7       16–20  158,631  62,515  55  19,312  26      57,529  50  19,275  30       <16  47,426  19,989  18  7,283  10      14,041  12  6,113  10      ≤4 years since quitting smokingc  14,218      7,485  24          6,733  23      Characteristic  Total No. of Participants  Men (n = 282,637)  Women (n = 302,946)  Current Smoker  Former Smoker  Never Smoker  Current Smoker  Former Smoker  Never Smoker  No.  %  No.  %  No.  %  No.  %  No.  %  No.  %  Total  585,583  113,033  40  74,639  26  94,965  34  115,350  38  64,024  21  123,572  41  Subcohort                             Norwegian Counties Study  83,500  21,416  51  9,599  23  10,898  26  16,501  40  6,054  14  19,032  46   40 Years Study  384,864  74,325  40  47,919  26  62,863  34  79,214  40  43,267  21  77,276  39   CONOR Study  117,219  17,292  31  17,121  31  21,204  38  19,635  32  14,703  24  27,264  44  Birth cohort                             ≤1950  294,280  62,008  43  42,182  29  40,698  28  55,898  37  28,262  19  65,232  44   >1950  291,303  51,025  37  32,457  24  54,267  39  59,452  39  35,762  23  58,340  38  Age at enrollment, yearsa  43 (8)  42 (7)  45 (10)  42 (8)  42 (7)  43 (8)  44 (10)  Person-years of follow-up  11,553,611  2,282,012    1,446,049    1,847,649    2,289,524    1,211,302    2,477,075    Age at diagnosis, yearsa  64 (10)  64 (10)  69 (11)  61 (12)  62 (9)  65 (12)  66 (12)  Histological typeb                             Lung cancer overall  5,514  3,147  85  458  12  109  3  2,367  84  232  8  221  8   Adenocarcinoma  1,384  957  26  172  5  63  2  900  32  130  5  119  4   Squamous cell carcinoma  1,293  811  22  99  3  9  <1  346  12  20  <1  8  <1   Small-cell carcinoma  1,161  561  15  49  1  2  <1  516  18  21  <1  12  <1  Education, years                             <10  137,691  33,627  30  16,754  22  12,735  13  36,899  32  13,081  21  24,595  20   10–12  320,642  63,463  56  41,092  55  49,477  52  66,025  57  36,648  57  63,937  52   ≥13  127,251  15,943  14  16,793  23  32,753  35  12,426  11  14,295  22  35,041  28  Physical activity levelc                             Sedentary  125,071  28,268  25  14,273  19  15,928  17  29,010  25  12,509  20  25,083  20   Moderate  297,478  52,374  46  33,423  45  38,816  41  65,408  57  35,703  56  71,754  58   High  163,034  32,391  29  26,943  36  40,221  42  20,932  18  15,812  24  26,735  22  BMIa,d  25 (4)  25 (3)  26 (3)  26 (3)  24 (4)  25 (4)  25 (4)  Duration of smoking, yearse                             1–9  60,086  5,224  5  20,745  28      7,760  7  26,357  41       10–19  104,359  19,530  17  31,031  42      28,118  24  25,680  40       20–29  174,120  75,582  67  16,810  23      72,391  63  9,337  15       ≥30  25,300  12,214  11  5,159  7      6,285  6  1,642  3      Quantity smoked, cigarettes/daye                             1–10  187,570  41,501  37  36,138  49      63,816  55  46,115  72       11–20  150,985  58,860  52  28,796  39      48,102  42  15,227  24       >20  21,816  9,758  9  7,537  10      3,028  3  1,493  2      ≥16 pack-years of smoking  103,048  50,939  45  15,678  21      31,475  27  4,956  8      Age at initiation of smoking, yearse                             ≥21  81,763  30,183  27  3,936  5      43,323  38  4,321  7       16–20  158,631  62,515  55  19,312  26      57,529  50  19,275  30       <16  47,426  19,989  18  7,283  10      14,041  12  6,113  10      ≤4 years since quitting smokingc  14,218      7,485  24          6,733  23      Abbreviations: BMI, body mass index; CONOR, Cohort of Norway. a Values are expressed as mean (standard deviation). b Numbers of lung cancer subtypes do not sum to the numbers in the “overall” groups because only the main subtypes are shown in the table. c Physical activity level was defined as sedentary (reading, television-watching, and other seated activity), moderate (walking, bicycling, or similar activities for ≥4 hours/week), or heavy (light sports or heavy gardening for ≥4 hours/week, heavy exercise, or daily competitive sports). d Weight (kg)/height (m)2. e Percentages for smoking duration, quantity smoked, and age at smoking initiation do not sum to 100% because of missing values. Table 2 shows that the overall incidence rates of lung cancer for men in the Norwegian Counties Study, the 40 Years Study, and the CONOR Study were 102.4 per 100,000 person-years, 50.4 per 100,000 person-years, and 83.2 per 100,000 person-years, respectively. Corresponding incidence rates for women were 59.4, 42.0, and 51.5, respectively. Table 2. Incidence Ratesa of Lung Cancer by Subcohort, Sex, and Smoking Status (n = 585,583), Norwegian Health Screening Surveys, 1974–2003 Smoking Status  Norwegian Counties Study (1974–1978)  40 Years Study (1985–1989)  CONOR Study (1994–2003)  Men  Women  Men  Women  Men  Women  Overallb  102.4  59.4  50.4  42.0  83.2  51.5  Never smoker  7.7  7.5  5.7  8.2  4.5  14.0  Former smoker  28.3  25.3  21.5  13.6  76.1  35.1  Current smoker  188.7  135.2  107.0  90.8  189.7  114.4  Smoking Status  Norwegian Counties Study (1974–1978)  40 Years Study (1985–1989)  CONOR Study (1994–2003)  Men  Women  Men  Women  Men  Women  Overallb  102.4  59.4  50.4  42.0  83.2  51.5  Never smoker  7.7  7.5  5.7  8.2  4.5  14.0  Former smoker  28.3  25.3  21.5  13.6  76.1  35.1  Current smoker  188.7  135.2  107.0  90.8  189.7  114.4  Abbreviation: CONOR, Cohort of Norway. a Incidence rates per 100,000 person-years. b Never, former, and current smokers combined. Table 2. Incidence Ratesa of Lung Cancer by Subcohort, Sex, and Smoking Status (n = 585,583), Norwegian Health Screening Surveys, 1974–2003 Smoking Status  Norwegian Counties Study (1974–1978)  40 Years Study (1985–1989)  CONOR Study (1994–2003)  Men  Women  Men  Women  Men  Women  Overallb  102.4  59.4  50.4  42.0  83.2  51.5  Never smoker  7.7  7.5  5.7  8.2  4.5  14.0  Former smoker  28.3  25.3  21.5  13.6  76.1  35.1  Current smoker  188.7  135.2  107.0  90.8  189.7  114.4  Smoking Status  Norwegian Counties Study (1974–1978)  40 Years Study (1985–1989)  CONOR Study (1994–2003)  Men  Women  Men  Women  Men  Women  Overallb  102.4  59.4  50.4  42.0  83.2  51.5  Never smoker  7.7  7.5  5.7  8.2  4.5  14.0  Former smoker  28.3  25.3  21.5  13.6  76.1  35.1  Current smoker  188.7  135.2  107.0  90.8  189.7  114.4  Abbreviation: CONOR, Cohort of Norway. a Incidence rates per 100,000 person-years. b Never, former, and current smokers combined. Table 3 shows that in comparison with their never-smoker counterparts, both men (hazard ratio = 19.12, 95% confidence interval (CI): 15.78, 23.18) and women (hazard ratio = 13.63, 95% CI: 11.83, 15.70) who were current smokers at baseline had significantly increased risks of lung cancer overall. In both sexes, former smoking and current smoking showed significant associations with lung cancer risk for smoking duration, number of cigarettes smoked daily, pack-years of smoking, and age at smoking initiation (all P’s for trend < 0.01). The heterogeneity test for both former and current smokers and overall lung cancer risk showed that these associations were stronger for men than for women (both P’s for heterogeneity = 0.01). Table 3. Multivariate-Adjusteda Hazard Ratios for Lung Cancer According to Sex and Measures of Smoking (n = 585,583), Norwegian Health Screening Surveys, 1974–2003 Smoking Measure  Men  Women  Heterogeneity Test for Women vs. Men  No.  HR  95% CI  No.  HR  95% CI  Smoking status                 Never smoker  109  1.00  Referent  221  1.00  Referent     Former smoker  458  3.64  2.95, 4.49  232  2.51  2.08, 3.02  0.01   Current smoker  3,147  19.12  15.78, 23.18  2,367  13.63  11.83, 15.70  0.01  Duration of former smoking, years                 1–9  23  0.96  0.61, 1.51  42  1.21  0.86, 1.69  0.43   10–19  85  1.95  1.46, 2.60  77  2.11  1.62, 2.75  0.69   20–29  167  5.89  4.59, 7.57  65  4.26  3.22, 5.64  0.09   ≥30  177  11.41  8.48, 15.36  43  7.19  5.01, 10.32  0.05    P for trendb    <0.01    <0.01     Linear variablec    1.83  1.67, 2.00    1.76  1.54, 2.02  0.68  Duration of current smoking, years                 1–9  34  6.23  4.12, 9.43  43  3.02  2.12, 4.31  0.01   10–19  274  11.35  9.06, 14.21  351  8.03  6.75, 9.55  0.02   20–29  1,971  19.89  16.35, 24.20  1,589  15.02  12.94, 17.42  0.02   ≥30  858  23.19  18.79, 28.61  368  21.92  18.08, 26.56  0.70    P for trendb    <0.01    <0.01     Linear variablec    1.37  1.30, 1.45    1.69  1.58, 1.81  <0.01  Quantity formerly smoked, cigarettes/day                 1–10  150  2.23  1.73, 2.87  139  2.00  1.62, 2.4  0.53   11–20  206  4.55  3.60, 5.76  80  4.29  3.30, 5.58  0.74   >20  77  6.14  4.56, 8.26  10  5.67  3.00, 10.73  0.83    P for trendb    0.05    0.43     Linear variablec    1.34  1.23, 1.46    1.54  1.33, 1.80  0.11  Quantity currently smoked, cigarettes/day                 1–10  763  11.53  9.42, 14.11  956  9.57  8.24, 11.11  0.15   11–20  1,783  22.95  18.88, 27.89  1,262  19.73  16.99, 22.90  0.23   >20  507  39.84  32.32, 49.12  140  34.46  27.74, 42.81  0.35    P for trendb    <0.01    <0.01     Linear variablec    1.54  1.49, 1.60    1.76  1.67, 1.86  <0.01  Pack-years of former smokingd                 1–5  33  0.95  0.64, 1.40  79  1.49  1.15, 1.93  0.06   6–15  145  2.97  2.30, 3.82  86  3.05  2.37, 3.93  0.88   ≥16  253  7.98  6.29, 10.13  60  7.43  5.54, 9.97  0.71    P for trendb    <0.01    0.01     Linear variablec    1.52  1.43, 1.62    1.85  1.65, 2.07  <0.01  Pack-years of current smokingd                 1–5  88  5.73  4.31, 7.61  142  3.84  3.09, 4.76  0.03   6–15  906  13.54  11.08, 16.55  1,133  12.67  10.92, 14.69  0.60   ≥16  2,052  27.24  22.42, 33.09  1,072  23.90  20.57, 27.76  0.30    P for trendb    <0.01    <0.01     Linear variablec    1.43  1.39, 1.48    1.64  1.57, 1.71  <0.01  Age at initiation of former smoking, years                 ≥21  22  2.57  1.50, 4.39  29  2.78  1.80, 4.30  0.82   16–20  124  4.26  2.94, 6.18  54  2.95  2.04, 4.25  0.17   <16  46  5.36  3.47, 8.27  11  3.38  1.73, 6.59  0.26    P for trendb    0.01    0.14     Linear variablee    0.92  0.88, 0.96    0.97  0.93, 1.01  0.12  Age at initiation of current smoking, years                 ≥21  641  12.01  9.79, 14.73  857  9.70  8.34, 11.28  0.10   16–20  1,819  20.92  17.22, 25.43  1,239  18.83  16.19, 21.89  0.40   <16  682  29.28  23.85, 35.93  261  28.50  23.54, 34.51  0.85    P for trendb    <0.01    <0.01     Linear variablee    0.92  0.91, 0.93    0.91  0.91, 0.92  0.20  Years since cessation of former smokingf                 Current smoker    1.00  Referent    1.00  Referent     0–4  61  0.44  0.34, 0.57  31  0.36  0.25, 0.51  0.38   5–9  34  0.35  0.25, 0.49  14  0.25  0.14, 0.42  0.28   ≥10  92  0.16  0.13, 0.20  44  0.17  0.12, 0.23  0.77    P for trendb    <0.01    <0.01     Linear variablec    0.57  0.49, 0.66    0.66  0.53, 0.83  0.26  Smoking Measure  Men  Women  Heterogeneity Test for Women vs. Men  No.  HR  95% CI  No.  HR  95% CI  Smoking status                 Never smoker  109  1.00  Referent  221  1.00  Referent     Former smoker  458  3.64  2.95, 4.49  232  2.51  2.08, 3.02  0.01   Current smoker  3,147  19.12  15.78, 23.18  2,367  13.63  11.83, 15.70  0.01  Duration of former smoking, years                 1–9  23  0.96  0.61, 1.51  42  1.21  0.86, 1.69  0.43   10–19  85  1.95  1.46, 2.60  77  2.11  1.62, 2.75  0.69   20–29  167  5.89  4.59, 7.57  65  4.26  3.22, 5.64  0.09   ≥30  177  11.41  8.48, 15.36  43  7.19  5.01, 10.32  0.05    P for trendb    <0.01    <0.01     Linear variablec    1.83  1.67, 2.00    1.76  1.54, 2.02  0.68  Duration of current smoking, years                 1–9  34  6.23  4.12, 9.43  43  3.02  2.12, 4.31  0.01   10–19  274  11.35  9.06, 14.21  351  8.03  6.75, 9.55  0.02   20–29  1,971  19.89  16.35, 24.20  1,589  15.02  12.94, 17.42  0.02   ≥30  858  23.19  18.79, 28.61  368  21.92  18.08, 26.56  0.70    P for trendb    <0.01    <0.01     Linear variablec    1.37  1.30, 1.45    1.69  1.58, 1.81  <0.01  Quantity formerly smoked, cigarettes/day                 1–10  150  2.23  1.73, 2.87  139  2.00  1.62, 2.4  0.53   11–20  206  4.55  3.60, 5.76  80  4.29  3.30, 5.58  0.74   >20  77  6.14  4.56, 8.26  10  5.67  3.00, 10.73  0.83    P for trendb    0.05    0.43     Linear variablec    1.34  1.23, 1.46    1.54  1.33, 1.80  0.11  Quantity currently smoked, cigarettes/day                 1–10  763  11.53  9.42, 14.11  956  9.57  8.24, 11.11  0.15   11–20  1,783  22.95  18.88, 27.89  1,262  19.73  16.99, 22.90  0.23   >20  507  39.84  32.32, 49.12  140  34.46  27.74, 42.81  0.35    P for trendb    <0.01    <0.01     Linear variablec    1.54  1.49, 1.60    1.76  1.67, 1.86  <0.01  Pack-years of former smokingd                 1–5  33  0.95  0.64, 1.40  79  1.49  1.15, 1.93  0.06   6–15  145  2.97  2.30, 3.82  86  3.05  2.37, 3.93  0.88   ≥16  253  7.98  6.29, 10.13  60  7.43  5.54, 9.97  0.71    P for trendb    <0.01    0.01     Linear variablec    1.52  1.43, 1.62    1.85  1.65, 2.07  <0.01  Pack-years of current smokingd                 1–5  88  5.73  4.31, 7.61  142  3.84  3.09, 4.76  0.03   6–15  906  13.54  11.08, 16.55  1,133  12.67  10.92, 14.69  0.60   ≥16  2,052  27.24  22.42, 33.09  1,072  23.90  20.57, 27.76  0.30    P for trendb    <0.01    <0.01     Linear variablec    1.43  1.39, 1.48    1.64  1.57, 1.71  <0.01  Age at initiation of former smoking, years                 ≥21  22  2.57  1.50, 4.39  29  2.78  1.80, 4.30  0.82   16–20  124  4.26  2.94, 6.18  54  2.95  2.04, 4.25  0.17   <16  46  5.36  3.47, 8.27  11  3.38  1.73, 6.59  0.26    P for trendb    0.01    0.14     Linear variablee    0.92  0.88, 0.96    0.97  0.93, 1.01  0.12  Age at initiation of current smoking, years                 ≥21  641  12.01  9.79, 14.73  857  9.70  8.34, 11.28  0.10   16–20  1,819  20.92  17.22, 25.43  1,239  18.83  16.19, 21.89  0.40   <16  682  29.28  23.85, 35.93  261  28.50  23.54, 34.51  0.85    P for trendb    <0.01    <0.01     Linear variablee    0.92  0.91, 0.93    0.91  0.91, 0.92  0.20  Years since cessation of former smokingf                 Current smoker    1.00  Referent    1.00  Referent     0–4  61  0.44  0.34, 0.57  31  0.36  0.25, 0.51  0.38   5–9  34  0.35  0.25, 0.49  14  0.25  0.14, 0.42  0.28   ≥10  92  0.16  0.13, 0.20  44  0.17  0.12, 0.23  0.77    P for trendb    <0.01    <0.01     Linear variablec    0.57  0.49, 0.66    0.66  0.53, 0.83  0.26  Abbreviations: CI, confidence interval; HR, hazard ratio. a Adjusted for age, body mass index, physical activity level (all at enrollment), and duration of education. b Trend test without never smokers included. c Per 10-year increment for smoking duration, per 10 cigarettes/day for quantity smoked, per 10 pack-years for pack-years in former and current smokers, and per 10-year increment since smoking cessation in former smokers, for lung cancer overall. d Pack-years were calculated as numbers of cigarettes smoked per day, divided by 20 and multiplied by smoking duration in years. e Per year for age at smoking initiation in former and current smokers, for lung cancer overall. f For smoking cessation, current smokers were the reference group. Table 3. Multivariate-Adjusteda Hazard Ratios for Lung Cancer According to Sex and Measures of Smoking (n = 585,583), Norwegian Health Screening Surveys, 1974–2003 Smoking Measure  Men  Women  Heterogeneity Test for Women vs. Men  No.  HR  95% CI  No.  HR  95% CI  Smoking status                 Never smoker  109  1.00  Referent  221  1.00  Referent     Former smoker  458  3.64  2.95, 4.49  232  2.51  2.08, 3.02  0.01   Current smoker  3,147  19.12  15.78, 23.18  2,367  13.63  11.83, 15.70  0.01  Duration of former smoking, years                 1–9  23  0.96  0.61, 1.51  42  1.21  0.86, 1.69  0.43   10–19  85  1.95  1.46, 2.60  77  2.11  1.62, 2.75  0.69   20–29  167  5.89  4.59, 7.57  65  4.26  3.22, 5.64  0.09   ≥30  177  11.41  8.48, 15.36  43  7.19  5.01, 10.32  0.05    P for trendb    <0.01    <0.01     Linear variablec    1.83  1.67, 2.00    1.76  1.54, 2.02  0.68  Duration of current smoking, years                 1–9  34  6.23  4.12, 9.43  43  3.02  2.12, 4.31  0.01   10–19  274  11.35  9.06, 14.21  351  8.03  6.75, 9.55  0.02   20–29  1,971  19.89  16.35, 24.20  1,589  15.02  12.94, 17.42  0.02   ≥30  858  23.19  18.79, 28.61  368  21.92  18.08, 26.56  0.70    P for trendb    <0.01    <0.01     Linear variablec    1.37  1.30, 1.45    1.69  1.58, 1.81  <0.01  Quantity formerly smoked, cigarettes/day                 1–10  150  2.23  1.73, 2.87  139  2.00  1.62, 2.4  0.53   11–20  206  4.55  3.60, 5.76  80  4.29  3.30, 5.58  0.74   >20  77  6.14  4.56, 8.26  10  5.67  3.00, 10.73  0.83    P for trendb    0.05    0.43     Linear variablec    1.34  1.23, 1.46    1.54  1.33, 1.80  0.11  Quantity currently smoked, cigarettes/day                 1–10  763  11.53  9.42, 14.11  956  9.57  8.24, 11.11  0.15   11–20  1,783  22.95  18.88, 27.89  1,262  19.73  16.99, 22.90  0.23   >20  507  39.84  32.32, 49.12  140  34.46  27.74, 42.81  0.35    P for trendb    <0.01    <0.01     Linear variablec    1.54  1.49, 1.60    1.76  1.67, 1.86  <0.01  Pack-years of former smokingd                 1–5  33  0.95  0.64, 1.40  79  1.49  1.15, 1.93  0.06   6–15  145  2.97  2.30, 3.82  86  3.05  2.37, 3.93  0.88   ≥16  253  7.98  6.29, 10.13  60  7.43  5.54, 9.97  0.71    P for trendb    <0.01    0.01     Linear variablec    1.52  1.43, 1.62    1.85  1.65, 2.07  <0.01  Pack-years of current smokingd                 1–5  88  5.73  4.31, 7.61  142  3.84  3.09, 4.76  0.03   6–15  906  13.54  11.08, 16.55  1,133  12.67  10.92, 14.69  0.60   ≥16  2,052  27.24  22.42, 33.09  1,072  23.90  20.57, 27.76  0.30    P for trendb    <0.01    <0.01     Linear variablec    1.43  1.39, 1.48    1.64  1.57, 1.71  <0.01  Age at initiation of former smoking, years                 ≥21  22  2.57  1.50, 4.39  29  2.78  1.80, 4.30  0.82   16–20  124  4.26  2.94, 6.18  54  2.95  2.04, 4.25  0.17   <16  46  5.36  3.47, 8.27  11  3.38  1.73, 6.59  0.26    P for trendb    0.01    0.14     Linear variablee    0.92  0.88, 0.96    0.97  0.93, 1.01  0.12  Age at initiation of current smoking, years                 ≥21  641  12.01  9.79, 14.73  857  9.70  8.34, 11.28  0.10   16–20  1,819  20.92  17.22, 25.43  1,239  18.83  16.19, 21.89  0.40   <16  682  29.28  23.85, 35.93  261  28.50  23.54, 34.51  0.85    P for trendb    <0.01    <0.01     Linear variablee    0.92  0.91, 0.93    0.91  0.91, 0.92  0.20  Years since cessation of former smokingf                 Current smoker    1.00  Referent    1.00  Referent     0–4  61  0.44  0.34, 0.57  31  0.36  0.25, 0.51  0.38   5–9  34  0.35  0.25, 0.49  14  0.25  0.14, 0.42  0.28   ≥10  92  0.16  0.13, 0.20  44  0.17  0.12, 0.23  0.77    P for trendb    <0.01    <0.01     Linear variablec    0.57  0.49, 0.66    0.66  0.53, 0.83  0.26  Smoking Measure  Men  Women  Heterogeneity Test for Women vs. Men  No.  HR  95% CI  No.  HR  95% CI  Smoking status                 Never smoker  109  1.00  Referent  221  1.00  Referent     Former smoker  458  3.64  2.95, 4.49  232  2.51  2.08, 3.02  0.01   Current smoker  3,147  19.12  15.78, 23.18  2,367  13.63  11.83, 15.70  0.01  Duration of former smoking, years                 1–9  23  0.96  0.61, 1.51  42  1.21  0.86, 1.69  0.43   10–19  85  1.95  1.46, 2.60  77  2.11  1.62, 2.75  0.69   20–29  167  5.89  4.59, 7.57  65  4.26  3.22, 5.64  0.09   ≥30  177  11.41  8.48, 15.36  43  7.19  5.01, 10.32  0.05    P for trendb    <0.01    <0.01     Linear variablec    1.83  1.67, 2.00    1.76  1.54, 2.02  0.68  Duration of current smoking, years                 1–9  34  6.23  4.12, 9.43  43  3.02  2.12, 4.31  0.01   10–19  274  11.35  9.06, 14.21  351  8.03  6.75, 9.55  0.02   20–29  1,971  19.89  16.35, 24.20  1,589  15.02  12.94, 17.42  0.02   ≥30  858  23.19  18.79, 28.61  368  21.92  18.08, 26.56  0.70    P for trendb    <0.01    <0.01     Linear variablec    1.37  1.30, 1.45    1.69  1.58, 1.81  <0.01  Quantity formerly smoked, cigarettes/day                 1–10  150  2.23  1.73, 2.87  139  2.00  1.62, 2.4  0.53   11–20  206  4.55  3.60, 5.76  80  4.29  3.30, 5.58  0.74   >20  77  6.14  4.56, 8.26  10  5.67  3.00, 10.73  0.83    P for trendb    0.05    0.43     Linear variablec    1.34  1.23, 1.46    1.54  1.33, 1.80  0.11  Quantity currently smoked, cigarettes/day                 1–10  763  11.53  9.42, 14.11  956  9.57  8.24, 11.11  0.15   11–20  1,783  22.95  18.88, 27.89  1,262  19.73  16.99, 22.90  0.23   >20  507  39.84  32.32, 49.12  140  34.46  27.74, 42.81  0.35    P for trendb    <0.01    <0.01     Linear variablec    1.54  1.49, 1.60    1.76  1.67, 1.86  <0.01  Pack-years of former smokingd                 1–5  33  0.95  0.64, 1.40  79  1.49  1.15, 1.93  0.06   6–15  145  2.97  2.30, 3.82  86  3.05  2.37, 3.93  0.88   ≥16  253  7.98  6.29, 10.13  60  7.43  5.54, 9.97  0.71    P for trendb    <0.01    0.01     Linear variablec    1.52  1.43, 1.62    1.85  1.65, 2.07  <0.01  Pack-years of current smokingd                 1–5  88  5.73  4.31, 7.61  142  3.84  3.09, 4.76  0.03   6–15  906  13.54  11.08, 16.55  1,133  12.67  10.92, 14.69  0.60   ≥16  2,052  27.24  22.42, 33.09  1,072  23.90  20.57, 27.76  0.30    P for trendb    <0.01    <0.01     Linear variablec    1.43  1.39, 1.48    1.64  1.57, 1.71  <0.01  Age at initiation of former smoking, years                 ≥21  22  2.57  1.50, 4.39  29  2.78  1.80, 4.30  0.82   16–20  124  4.26  2.94, 6.18  54  2.95  2.04, 4.25  0.17   <16  46  5.36  3.47, 8.27  11  3.38  1.73, 6.59  0.26    P for trendb    0.01    0.14     Linear variablee    0.92  0.88, 0.96    0.97  0.93, 1.01  0.12  Age at initiation of current smoking, years                 ≥21  641  12.01  9.79, 14.73  857  9.70  8.34, 11.28  0.10   16–20  1,819  20.92  17.22, 25.43  1,239  18.83  16.19, 21.89  0.40   <16  682  29.28  23.85, 35.93  261  28.50  23.54, 34.51  0.85    P for trendb    <0.01    <0.01     Linear variablee    0.92  0.91, 0.93    0.91  0.91, 0.92  0.20  Years since cessation of former smokingf                 Current smoker    1.00  Referent    1.00  Referent     0–4  61  0.44  0.34, 0.57  31  0.36  0.25, 0.51  0.38   5–9  34  0.35  0.25, 0.49  14  0.25  0.14, 0.42  0.28   ≥10  92  0.16  0.13, 0.20  44  0.17  0.12, 0.23  0.77    P for trendb    <0.01    <0.01     Linear variablec    0.57  0.49, 0.66    0.66  0.53, 0.83  0.26  Abbreviations: CI, confidence interval; HR, hazard ratio. a Adjusted for age, body mass index, physical activity level (all at enrollment), and duration of education. b Trend test without never smokers included. c Per 10-year increment for smoking duration, per 10 cigarettes/day for quantity smoked, per 10 pack-years for pack-years in former and current smokers, and per 10-year increment since smoking cessation in former smokers, for lung cancer overall. d Pack-years were calculated as numbers of cigarettes smoked per day, divided by 20 and multiplied by smoking duration in years. e Per year for age at smoking initiation in former and current smokers, for lung cancer overall. f For smoking cessation, current smokers were the reference group. Table 3 shows that, compared with never smokers, male current smokers with ≥16 pack-years of smoking had a hazard ratio for lung cancer of 27.24 (95% CI: 22.42, 33.09), and female current smokers had a hazard ratio of 23.90 (95% CI: 20.57, 27.76) (P for heterogeneity = 0.30). Table 3 further shows that, for current smokers, the test for heterogeneity by sex for each variable category compared with never smokers was statistically significant for duration of smoking (all P’s for heterogeneity < 0.05) but was not significant for the upper category (≥30 years of smoking) or for the other variables (number of cigarettes smoked per day, age at smoking initiation, and pack-years (except for the lowest category (1–5 pack-years))). Additionally, for former smokers, the test for heterogeneity by sex was not significant for any of the smoking variables. For current smokers, the increase in lung cancer risk was significantly greater in women than in men when we examined the various measures of smoking exposure as continuous variables; for each 10–pack-year increment, the hazard ratio was 1.43 (95% CI: 1.39, 1.48) in men and 1.64 (95% CI: 1.57, 1.71) in women (P for heterogeneity < 0.01) (Table 3). The test for heterogeneity was significant for increments (as continuous measures) of 10 years of smoking duration and 10 cigarettes/day, with a higher risk for women who were current smokers than for men (both P’s for heterogeneity < 0.01). When we examined the various measures of smoking exposure as continuous variables in former smokers, the heterogeneity test was significant by sex for 10 pack-years, with a higher increased risk of lung cancer per increment for women compared with men (P for heterogeneity < 0.01). For ever smokers, the increase in risk of lung cancer overall differed significantly by sex, with a greater increased risk in women for increments of 10 years of smoking, 10 cigarettes/day, and 10 pack-years (all P’s for heterogeneity < 0.01) (Table 4). We observed similar significant differences by sex for squamous cell carcinoma. For adenocarcinoma, we observed significant differences for increments of 10 cigarettes/day and 10 pack-years (Table 4). The log-transformed models in ever smokers and by cell type showed similar differences by sex (data not shown). Due to few cases of small-cell carcinoma, especially among never smokers, we did not stratify by sex when we examined this subtype. We found a significant dose-response association between smoking duration and risk of small-cell carcinoma (results not shown). Table 4. Multivariate-Adjusteda Hazard Ratios for Lung Cancer According to Sex, Measures of Smoking, and Histological Subtype (n = 585,583), Norwegian Health Screening Surveys, 1974–2003 Smoking Measure  Men  Women  Heterogeneity Test for Men vs. Women  No.  HR  95% CI  No.  HR  95% CI  Lung Cancer Overall  Smoking status                 Never smoker  109  1.00  Referent  221  1.00  Referent     Former smoker  458  3.64  2.95, 4.49  232  2.51  2.08, 3.02  0.01   Current smoker  3,147  19.12  15.78, 23.18  2,367  13.63  11.83, 15.70  0.01  Smoking duration, years                 1–9  57  1.78  1.29, 2.47  85  1.63  1.26, 2.11  0.68   10–19  359  5.12  4.12, 6.34  428  5.09  4.32, 5.99  0.97   20–29  2,138  16.04  13.21, 19.48  1,654  13.05  11.29, 15.08  0.10   ≥30  1,035  24.50  19.92, 30.13  411  21.47  17.84, 25.84  0.35    P for trendb    <0.01    <0.01     Linear variablec    1.90  1.82, 1.98    2.11  2.00, 2.23  <0.01  Quantity smoked, cigarettes/day                 1–10  913  6.93  5.68, 8.46  1,095  6.25  5.40, 7.24  0.42   11–20  1,989  15.52  12.78, 18.83  1,342  15.53  13.41, 17.97  1.00   >20  584  23.03  18.75, 28.29  150  24.85  20.13, 30.69  0.61    P for trendb    <0.01    <0.01     Linear variablec    1.44  1.40, 1.49    1.82  1.74, 1.91  <0.01  Pack-years of smokingd                 1–5  121  2.34  1.80, 3.03  221  2.40  1.99, 2.90  0.87   6–15  1,051  8.86  7.27, 10.81  1,219  10.03  8.67, 11.61  0.32   ≥16  2,305  22.12  18.22, 26.84  1,132  21.27  18.34, 24.66  0.75    P for trendb    <0.01    <0.01     Linear variablec    1.57  1.53, 1.61    1.84  1.78, 1.91  <0.01  Age at smoking initiation                 ≥21  663  10.70  8.73, 13.11  886  8.76  7.54, 0.17  0.12   16–20  1,943  18.59  15.30, 22.58  1,293  16.23  13.99, 18.83  0.28   <16  728  26.86  21.10, 31.71  278  24.38  20.19, 29.44  0.68    P for trendb    <0.01    <0.01     Linear variablee    0.92  0.91, 0.93    0.92  0.91, 0.93  0.51  Adenocarcinoma  Smoking status                 Never smoker  63  1.00  Referent  119  1.00  Referent     Former smoker  172  2.65  1.98, 3.55  130  2.45  1.91, 3.15  0.93   Current smoker  957  10.33  7.98, 13.37  900  8.52  7.00, 10.36  0.23  Smoking duration                 1–9  26  1.41  0.89, 2.25  44  1.48  1.04, 2.11  0.88   10–19  114  2.94  2.16, 4.01  190  3.99  3.17, 5.04  0.12   20–29  730  9.60  7.39, 12.46  666  8.96  7.32, 10.96  0.68   ≥30  253  13.67  10.13, 18.44  128  12.51  9.37, 16.71  0.68    P for trendb    <0.01    <0.01     Linear variablec    1.83  1.69, 1.98    1.91  1.75, 2.08  0.47  Quantity smoked, cigarettes/day                 1–10  272  3.92  2.97, 5.16  448  4.42  3.60, 5.42  0.49   11–20  652  9.07  6.99, 11.77  521  9.84  8.02, 12.09  0.63   >20  166  11.83  8.83, 15.86  60  16.51  12.04, 22.63  0.13    P for trendb    <0.01    <0.01     Linear variablec    1.38  1.30, 1.47    1.69  1.57, 1.83  <0.01  Pack-years of smokingd                 1–5  49  1.68  1.15, 2.44  118  2.23  1.73, 2.89  0.22   6–15  344  5.32  4.05, 6.97  490  6.83  5.57, 8.37  0.15   ≥16  695  12.41  9.55, 16.12  419  13.17  10.67, 16.25  0.73    P for trendb    <0.01    <0.01     Linear variablec    1.51  1.44, 1.58    1.76  1.66, 1.87  <0.01  Age at smoking initiation                 ≥21  186  5.52  4.14, 7.36  336  5.83  4.72, 7.20  0.76   16–20  582  10.18  7.82, 13.25  484  9.70  7.87, 11.96  0.78   <16  248  16.18  12.20, 21.45  123  16.38  12.51, 21.45  0.95    P for trendb    <0.01    <0.01     Linear variablee    0.91  0.90, 0.93    0.92  0.90, 0.93  0.61  Squamous Cell Carcinoma  Smoking status                 Never smoker  9  1.00  Referent  8  1.00  Referent     Former smoker  99  9.01  4.55, 17.85  20  6.64  2.91, 15.13  0.87   Current smoker  811  57.49  29.76, 111.05  346  61.63  30.10, 126.20  1.00  Smoking duration                 1–9  14  5.39  2.31, 12.57  7  3.86  1.37, 10.92  0.63   10–19  94  15.50  7.81, 30.74  53  17.55  8.32, 37.02  0.81   20–29  512  44.95  23.21, 87.05  234  54.72  26.87, 111.42  0.62   ≥30  289  70.28  35.83, 137.84  67  110.35  51.17, 237.97  0.39    P for trendb    <0.01    <0.01     Linear variablec    1.90  1.75, 2.07    2.45  2.11, 2.84  <0.01  Quantity smoked, cigarettes/day                 1–10  215  18.46  9.47, 35.99  149  24.35  11.93, 49.70  0.58   11–20  480  44.01  22.74, 85.17  191  69.06  33.84, 140.96  0.36   >20  184  85.98  43.98, 168.09  25  131.74  58.98, 294.25  0.42    P for trendb    <0.01    <0.01     Linear variablec    1.57  1.48, 1.67    2.02  1.79, 2.27  <0.01  Pack-years of smokingd                 1–5  27  6.14  2.89, 13.08  18  5.59  2.42, 12.93  0.87   6–15  250  24.07  12.37, 46.86  175  42.50  20.83, 86.69  0.25   ≥16  601  66.72  34.49, 129.07  168  102.56  50.08, 210.02  0.39    P for trendb    <0.01    <0.01     Linear variablec    1.69  1.61, 1.77    1.99  1.82, 2.19  <0.01  Age at smoking initiation                 ≥21  181  33.58  17.18, 65.63  137  37.54  18.36, 76.80  0.82   16–20  486  55.06  28.42, 106.65  183  71.94  35.09, 147.49  0.59   <16  187  79.03  40.37, 154.69  32  101.15  45.75, 223.64  0.64    P for trendb    <0.01    <0.01     Linear variablee    0.93  0.91, 0.94    0.91  0.89, 0.94  0.28  Smoking Measure  Men  Women  Heterogeneity Test for Men vs. Women  No.  HR  95% CI  No.  HR  95% CI  Lung Cancer Overall  Smoking status                 Never smoker  109  1.00  Referent  221  1.00  Referent     Former smoker  458  3.64  2.95, 4.49  232  2.51  2.08, 3.02  0.01   Current smoker  3,147  19.12  15.78, 23.18  2,367  13.63  11.83, 15.70  0.01  Smoking duration, years                 1–9  57  1.78  1.29, 2.47  85  1.63  1.26, 2.11  0.68   10–19  359  5.12  4.12, 6.34  428  5.09  4.32, 5.99  0.97   20–29  2,138  16.04  13.21, 19.48  1,654  13.05  11.29, 15.08  0.10   ≥30  1,035  24.50  19.92, 30.13  411  21.47  17.84, 25.84  0.35    P for trendb    <0.01    <0.01     Linear variablec    1.90  1.82, 1.98    2.11  2.00, 2.23  <0.01  Quantity smoked, cigarettes/day                 1–10  913  6.93  5.68, 8.46  1,095  6.25  5.40, 7.24  0.42   11–20  1,989  15.52  12.78, 18.83  1,342  15.53  13.41, 17.97  1.00   >20  584  23.03  18.75, 28.29  150  24.85  20.13, 30.69  0.61    P for trendb    <0.01    <0.01     Linear variablec    1.44  1.40, 1.49    1.82  1.74, 1.91  <0.01  Pack-years of smokingd                 1–5  121  2.34  1.80, 3.03  221  2.40  1.99, 2.90  0.87   6–15  1,051  8.86  7.27, 10.81  1,219  10.03  8.67, 11.61  0.32   ≥16  2,305  22.12  18.22, 26.84  1,132  21.27  18.34, 24.66  0.75    P for trendb    <0.01    <0.01     Linear variablec    1.57  1.53, 1.61    1.84  1.78, 1.91  <0.01  Age at smoking initiation                 ≥21  663  10.70  8.73, 13.11  886  8.76  7.54, 0.17  0.12   16–20  1,943  18.59  15.30, 22.58  1,293  16.23  13.99, 18.83  0.28   <16  728  26.86  21.10, 31.71  278  24.38  20.19, 29.44  0.68    P for trendb    <0.01    <0.01     Linear variablee    0.92  0.91, 0.93    0.92  0.91, 0.93  0.51  Adenocarcinoma  Smoking status                 Never smoker  63  1.00  Referent  119  1.00  Referent     Former smoker  172  2.65  1.98, 3.55  130  2.45  1.91, 3.15  0.93   Current smoker  957  10.33  7.98, 13.37  900  8.52  7.00, 10.36  0.23  Smoking duration                 1–9  26  1.41  0.89, 2.25  44  1.48  1.04, 2.11  0.88   10–19  114  2.94  2.16, 4.01  190  3.99  3.17, 5.04  0.12   20–29  730  9.60  7.39, 12.46  666  8.96  7.32, 10.96  0.68   ≥30  253  13.67  10.13, 18.44  128  12.51  9.37, 16.71  0.68    P for trendb    <0.01    <0.01     Linear variablec    1.83  1.69, 1.98    1.91  1.75, 2.08  0.47  Quantity smoked, cigarettes/day                 1–10  272  3.92  2.97, 5.16  448  4.42  3.60, 5.42  0.49   11–20  652  9.07  6.99, 11.77  521  9.84  8.02, 12.09  0.63   >20  166  11.83  8.83, 15.86  60  16.51  12.04, 22.63  0.13    P for trendb    <0.01    <0.01     Linear variablec    1.38  1.30, 1.47    1.69  1.57, 1.83  <0.01  Pack-years of smokingd                 1–5  49  1.68  1.15, 2.44  118  2.23  1.73, 2.89  0.22   6–15  344  5.32  4.05, 6.97  490  6.83  5.57, 8.37  0.15   ≥16  695  12.41  9.55, 16.12  419  13.17  10.67, 16.25  0.73    P for trendb    <0.01    <0.01     Linear variablec    1.51  1.44, 1.58    1.76  1.66, 1.87  <0.01  Age at smoking initiation                 ≥21  186  5.52  4.14, 7.36  336  5.83  4.72, 7.20  0.76   16–20  582  10.18  7.82, 13.25  484  9.70  7.87, 11.96  0.78   <16  248  16.18  12.20, 21.45  123  16.38  12.51, 21.45  0.95    P for trendb    <0.01    <0.01     Linear variablee    0.91  0.90, 0.93    0.92  0.90, 0.93  0.61  Squamous Cell Carcinoma  Smoking status                 Never smoker  9  1.00  Referent  8  1.00  Referent     Former smoker  99  9.01  4.55, 17.85  20  6.64  2.91, 15.13  0.87   Current smoker  811  57.49  29.76, 111.05  346  61.63  30.10, 126.20  1.00  Smoking duration                 1–9  14  5.39  2.31, 12.57  7  3.86  1.37, 10.92  0.63   10–19  94  15.50  7.81, 30.74  53  17.55  8.32, 37.02  0.81   20–29  512  44.95  23.21, 87.05  234  54.72  26.87, 111.42  0.62   ≥30  289  70.28  35.83, 137.84  67  110.35  51.17, 237.97  0.39    P for trendb    <0.01    <0.01     Linear variablec    1.90  1.75, 2.07    2.45  2.11, 2.84  <0.01  Quantity smoked, cigarettes/day                 1–10  215  18.46  9.47, 35.99  149  24.35  11.93, 49.70  0.58   11–20  480  44.01  22.74, 85.17  191  69.06  33.84, 140.96  0.36   >20  184  85.98  43.98, 168.09  25  131.74  58.98, 294.25  0.42    P for trendb    <0.01    <0.01     Linear variablec    1.57  1.48, 1.67    2.02  1.79, 2.27  <0.01  Pack-years of smokingd                 1–5  27  6.14  2.89, 13.08  18  5.59  2.42, 12.93  0.87   6–15  250  24.07  12.37, 46.86  175  42.50  20.83, 86.69  0.25   ≥16  601  66.72  34.49, 129.07  168  102.56  50.08, 210.02  0.39    P for trendb    <0.01    <0.01     Linear variablec    1.69  1.61, 1.77    1.99  1.82, 2.19  <0.01  Age at smoking initiation                 ≥21  181  33.58  17.18, 65.63  137  37.54  18.36, 76.80  0.82   16–20  486  55.06  28.42, 106.65  183  71.94  35.09, 147.49  0.59   <16  187  79.03  40.37, 154.69  32  101.15  45.75, 223.64  0.64    P for trendb    <0.01    <0.01     Linear variablee    0.93  0.91, 0.94    0.91  0.89, 0.94  0.28  Abbreviations: CI, confidence interval; HR, hazard ratio. a Adjusted for age, body mass index, physical activity level (all at enrollment), and duration of education. b Trend test without never smokers included. c Per 10-year increment for smoking duration, per 10 cigarettes/day for quantity smoked, per 10 pack-years for pack-years among ever smokers, for lung cancer overall, adenocarcinoma, and squamous cell carcinoma. d Pack-years were calculated as numbers of cigarettes smoked per day, divided by 20 and multiplied by smoking duration in years. e Per year for age at smoking initiation among ever smokers, for lung cancer overall, adenocarcinoma, and squamous cell carcinoma. Table 4. Multivariate-Adjusteda Hazard Ratios for Lung Cancer According to Sex, Measures of Smoking, and Histological Subtype (n = 585,583), Norwegian Health Screening Surveys, 1974–2003 Smoking Measure  Men  Women  Heterogeneity Test for Men vs. Women  No.  HR  95% CI  No.  HR  95% CI  Lung Cancer Overall  Smoking status                 Never smoker  109  1.00  Referent  221  1.00  Referent     Former smoker  458  3.64  2.95, 4.49  232  2.51  2.08, 3.02  0.01   Current smoker  3,147  19.12  15.78, 23.18  2,367  13.63  11.83, 15.70  0.01  Smoking duration, years                 1–9  57  1.78  1.29, 2.47  85  1.63  1.26, 2.11  0.68   10–19  359  5.12  4.12, 6.34  428  5.09  4.32, 5.99  0.97   20–29  2,138  16.04  13.21, 19.48  1,654  13.05  11.29, 15.08  0.10   ≥30  1,035  24.50  19.92, 30.13  411  21.47  17.84, 25.84  0.35    P for trendb    <0.01    <0.01     Linear variablec    1.90  1.82, 1.98    2.11  2.00, 2.23  <0.01  Quantity smoked, cigarettes/day                 1–10  913  6.93  5.68, 8.46  1,095  6.25  5.40, 7.24  0.42   11–20  1,989  15.52  12.78, 18.83  1,342  15.53  13.41, 17.97  1.00   >20  584  23.03  18.75, 28.29  150  24.85  20.13, 30.69  0.61    P for trendb    <0.01    <0.01     Linear variablec    1.44  1.40, 1.49    1.82  1.74, 1.91  <0.01  Pack-years of smokingd                 1–5  121  2.34  1.80, 3.03  221  2.40  1.99, 2.90  0.87   6–15  1,051  8.86  7.27, 10.81  1,219  10.03  8.67, 11.61  0.32   ≥16  2,305  22.12  18.22, 26.84  1,132  21.27  18.34, 24.66  0.75    P for trendb    <0.01    <0.01     Linear variablec    1.57  1.53, 1.61    1.84  1.78, 1.91  <0.01  Age at smoking initiation                 ≥21  663  10.70  8.73, 13.11  886  8.76  7.54, 0.17  0.12   16–20  1,943  18.59  15.30, 22.58  1,293  16.23  13.99, 18.83  0.28   <16  728  26.86  21.10, 31.71  278  24.38  20.19, 29.44  0.68    P for trendb    <0.01    <0.01     Linear variablee    0.92  0.91, 0.93    0.92  0.91, 0.93  0.51  Adenocarcinoma  Smoking status                 Never smoker  63  1.00  Referent  119  1.00  Referent     Former smoker  172  2.65  1.98, 3.55  130  2.45  1.91, 3.15  0.93   Current smoker  957  10.33  7.98, 13.37  900  8.52  7.00, 10.36  0.23  Smoking duration                 1–9  26  1.41  0.89, 2.25  44  1.48  1.04, 2.11  0.88   10–19  114  2.94  2.16, 4.01  190  3.99  3.17, 5.04  0.12   20–29  730  9.60  7.39, 12.46  666  8.96  7.32, 10.96  0.68   ≥30  253  13.67  10.13, 18.44  128  12.51  9.37, 16.71  0.68    P for trendb    <0.01    <0.01     Linear variablec    1.83  1.69, 1.98    1.91  1.75, 2.08  0.47  Quantity smoked, cigarettes/day                 1–10  272  3.92  2.97, 5.16  448  4.42  3.60, 5.42  0.49   11–20  652  9.07  6.99, 11.77  521  9.84  8.02, 12.09  0.63   >20  166  11.83  8.83, 15.86  60  16.51  12.04, 22.63  0.13    P for trendb    <0.01    <0.01     Linear variablec    1.38  1.30, 1.47    1.69  1.57, 1.83  <0.01  Pack-years of smokingd                 1–5  49  1.68  1.15, 2.44  118  2.23  1.73, 2.89  0.22   6–15  344  5.32  4.05, 6.97  490  6.83  5.57, 8.37  0.15   ≥16  695  12.41  9.55, 16.12  419  13.17  10.67, 16.25  0.73    P for trendb    <0.01    <0.01     Linear variablec    1.51  1.44, 1.58    1.76  1.66, 1.87  <0.01  Age at smoking initiation                 ≥21  186  5.52  4.14, 7.36  336  5.83  4.72, 7.20  0.76   16–20  582  10.18  7.82, 13.25  484  9.70  7.87, 11.96  0.78   <16  248  16.18  12.20, 21.45  123  16.38  12.51, 21.45  0.95    P for trendb    <0.01    <0.01     Linear variablee    0.91  0.90, 0.93    0.92  0.90, 0.93  0.61  Squamous Cell Carcinoma  Smoking status                 Never smoker  9  1.00  Referent  8  1.00  Referent     Former smoker  99  9.01  4.55, 17.85  20  6.64  2.91, 15.13  0.87   Current smoker  811  57.49  29.76, 111.05  346  61.63  30.10, 126.20  1.00  Smoking duration                 1–9  14  5.39  2.31, 12.57  7  3.86  1.37, 10.92  0.63   10–19  94  15.50  7.81, 30.74  53  17.55  8.32, 37.02  0.81   20–29  512  44.95  23.21, 87.05  234  54.72  26.87, 111.42  0.62   ≥30  289  70.28  35.83, 137.84  67  110.35  51.17, 237.97  0.39    P for trendb    <0.01    <0.01     Linear variablec    1.90  1.75, 2.07    2.45  2.11, 2.84  <0.01  Quantity smoked, cigarettes/day                 1–10  215  18.46  9.47, 35.99  149  24.35  11.93, 49.70  0.58   11–20  480  44.01  22.74, 85.17  191  69.06  33.84, 140.96  0.36   >20  184  85.98  43.98, 168.09  25  131.74  58.98, 294.25  0.42    P for trendb    <0.01    <0.01     Linear variablec    1.57  1.48, 1.67    2.02  1.79, 2.27  <0.01  Pack-years of smokingd                 1–5  27  6.14  2.89, 13.08  18  5.59  2.42, 12.93  0.87   6–15  250  24.07  12.37, 46.86  175  42.50  20.83, 86.69  0.25   ≥16  601  66.72  34.49, 129.07  168  102.56  50.08, 210.02  0.39    P for trendb    <0.01    <0.01     Linear variablec    1.69  1.61, 1.77    1.99  1.82, 2.19  <0.01  Age at smoking initiation                 ≥21  181  33.58  17.18, 65.63  137  37.54  18.36, 76.80  0.82   16–20  486  55.06  28.42, 106.65  183  71.94  35.09, 147.49  0.59   <16  187  79.03  40.37, 154.69  32  101.15  45.75, 223.64  0.64    P for trendb    <0.01    <0.01     Linear variablee    0.93  0.91, 0.94    0.91  0.89, 0.94  0.28  Smoking Measure  Men  Women  Heterogeneity Test for Men vs. Women  No.  HR  95% CI  No.  HR  95% CI  Lung Cancer Overall  Smoking status                 Never smoker  109  1.00  Referent  221  1.00  Referent     Former smoker  458  3.64  2.95, 4.49  232  2.51  2.08, 3.02  0.01   Current smoker  3,147  19.12  15.78, 23.18  2,367  13.63  11.83, 15.70  0.01  Smoking duration, years                 1–9  57  1.78  1.29, 2.47  85  1.63  1.26, 2.11  0.68   10–19  359  5.12  4.12, 6.34  428  5.09  4.32, 5.99  0.97   20–29  2,138  16.04  13.21, 19.48  1,654  13.05  11.29, 15.08  0.10   ≥30  1,035  24.50  19.92, 30.13  411  21.47  17.84, 25.84  0.35    P for trendb    <0.01    <0.01     Linear variablec    1.90  1.82, 1.98    2.11  2.00, 2.23  <0.01  Quantity smoked, cigarettes/day                 1–10  913  6.93  5.68, 8.46  1,095  6.25  5.40, 7.24  0.42   11–20  1,989  15.52  12.78, 18.83  1,342  15.53  13.41, 17.97  1.00   >20  584  23.03  18.75, 28.29  150  24.85  20.13, 30.69  0.61    P for trendb    <0.01    <0.01     Linear variablec    1.44  1.40, 1.49    1.82  1.74, 1.91  <0.01  Pack-years of smokingd                 1–5  121  2.34  1.80, 3.03  221  2.40  1.99, 2.90  0.87   6–15  1,051  8.86  7.27, 10.81  1,219  10.03  8.67, 11.61  0.32   ≥16  2,305  22.12  18.22, 26.84  1,132  21.27  18.34, 24.66  0.75    P for trendb    <0.01    <0.01     Linear variablec    1.57  1.53, 1.61    1.84  1.78, 1.91  <0.01  Age at smoking initiation                 ≥21  663  10.70  8.73, 13.11  886  8.76  7.54, 0.17  0.12   16–20  1,943  18.59  15.30, 22.58  1,293  16.23  13.99, 18.83  0.28   <16  728  26.86  21.10, 31.71  278  24.38  20.19, 29.44  0.68    P for trendb    <0.01    <0.01     Linear variablee    0.92  0.91, 0.93    0.92  0.91, 0.93  0.51  Adenocarcinoma  Smoking status                 Never smoker  63  1.00  Referent  119  1.00  Referent     Former smoker  172  2.65  1.98, 3.55  130  2.45  1.91, 3.15  0.93   Current smoker  957  10.33  7.98, 13.37  900  8.52  7.00, 10.36  0.23  Smoking duration                 1–9  26  1.41  0.89, 2.25  44  1.48  1.04, 2.11  0.88   10–19  114  2.94  2.16, 4.01  190  3.99  3.17, 5.04  0.12   20–29  730  9.60  7.39, 12.46  666  8.96  7.32, 10.96  0.68   ≥30  253  13.67  10.13, 18.44  128  12.51  9.37, 16.71  0.68    P for trendb    <0.01    <0.01     Linear variablec    1.83  1.69, 1.98    1.91  1.75, 2.08  0.47  Quantity smoked, cigarettes/day                 1–10  272  3.92  2.97, 5.16  448  4.42  3.60, 5.42  0.49   11–20  652  9.07  6.99, 11.77  521  9.84  8.02, 12.09  0.63   >20  166  11.83  8.83, 15.86  60  16.51  12.04, 22.63  0.13    P for trendb    <0.01    <0.01     Linear variablec    1.38  1.30, 1.47    1.69  1.57, 1.83  <0.01  Pack-years of smokingd                 1–5  49  1.68  1.15, 2.44  118  2.23  1.73, 2.89  0.22   6–15  344  5.32  4.05, 6.97  490  6.83  5.57, 8.37  0.15   ≥16  695  12.41  9.55, 16.12  419  13.17  10.67, 16.25  0.73    P for trendb    <0.01    <0.01     Linear variablec    1.51  1.44, 1.58    1.76  1.66, 1.87  <0.01  Age at smoking initiation                 ≥21  186  5.52  4.14, 7.36  336  5.83  4.72, 7.20  0.76   16–20  582  10.18  7.82, 13.25  484  9.70  7.87, 11.96  0.78   <16  248  16.18  12.20, 21.45  123  16.38  12.51, 21.45  0.95    P for trendb    <0.01    <0.01     Linear variablee    0.91  0.90, 0.93    0.92  0.90, 0.93  0.61  Squamous Cell Carcinoma  Smoking status                 Never smoker  9  1.00  Referent  8  1.00  Referent     Former smoker  99  9.01  4.55, 17.85  20  6.64  2.91, 15.13  0.87   Current smoker  811  57.49  29.76, 111.05  346  61.63  30.10, 126.20  1.00  Smoking duration                 1–9  14  5.39  2.31, 12.57  7  3.86  1.37, 10.92  0.63   10–19  94  15.50  7.81, 30.74  53  17.55  8.32, 37.02  0.81   20–29  512  44.95  23.21, 87.05  234  54.72  26.87, 111.42  0.62   ≥30  289  70.28  35.83, 137.84  67  110.35  51.17, 237.97  0.39    P for trendb    <0.01    <0.01     Linear variablec    1.90  1.75, 2.07    2.45  2.11, 2.84  <0.01  Quantity smoked, cigarettes/day                 1–10  215  18.46  9.47, 35.99  149  24.35  11.93, 49.70  0.58   11–20  480  44.01  22.74, 85.17  191  69.06  33.84, 140.96  0.36   >20  184  85.98  43.98, 168.09  25  131.74  58.98, 294.25  0.42    P for trendb    <0.01    <0.01     Linear variablec    1.57  1.48, 1.67    2.02  1.79, 2.27  <0.01  Pack-years of smokingd                 1–5  27  6.14  2.89, 13.08  18  5.59  2.42, 12.93  0.87   6–15  250  24.07  12.37, 46.86  175  42.50  20.83, 86.69  0.25   ≥16  601  66.72  34.49, 129.07  168  102.56  50.08, 210.02  0.39    P for trendb    <0.01    <0.01     Linear variablec    1.69  1.61, 1.77    1.99  1.82, 2.19  <0.01  Age at smoking initiation                 ≥21  181  33.58  17.18, 65.63  137  37.54  18.36, 76.80  0.82   16–20  486  55.06  28.42, 106.65  183  71.94  35.09, 147.49  0.59   <16  187  79.03  40.37, 154.69  32  101.15  45.75, 223.64  0.64    P for trendb    <0.01    <0.01     Linear variablee    0.93  0.91, 0.94    0.91  0.89, 0.94  0.28  Abbreviations: CI, confidence interval; HR, hazard ratio. a Adjusted for age, body mass index, physical activity level (all at enrollment), and duration of education. b Trend test without never smokers included. c Per 10-year increment for smoking duration, per 10 cigarettes/day for quantity smoked, per 10 pack-years for pack-years among ever smokers, for lung cancer overall, adenocarcinoma, and squamous cell carcinoma. d Pack-years were calculated as numbers of cigarettes smoked per day, divided by 20 and multiplied by smoking duration in years. e Per year for age at smoking initiation among ever smokers, for lung cancer overall, adenocarcinoma, and squamous cell carcinoma. Of the interactions tested, neither BMI nor physical activity nor education was statistically significant for any of the outcomes investigated (data not shown). The overall results stayed materially the same when we excluded subjects with lung cancer diagnosed within 2 years of enrollment (data not shown). Web Table 2 shows that for male current smokers, the overall lung cancer risk differed significantly between the 40 Years Study and the CONOR Study, with a greater risk in the more recent study (CONOR). For female current smokers, the risks of lung cancer differed significantly between all 3 studies, with the highest risk being seen in the earliest study (the Norwegian Counties Study). Web Table 3 shows that male current smokers born in or before 1950 had a hazard ratio of 23.11 (95% CI: 18.30, 29.20), and the corresponding hazard ratio for men born after 1950 was 10.75 (95% CI: 7.62, 15.16) (P for heterogeneity < 0.01). Among women, the risk of lung cancer was greatest in the older birth cohort (born ≤1950) as well. DISCUSSION In this large prospective study, we found that compared with women, more men were ever and heavier smokers. More female than male never smokers were diagnosed with lung cancer during follow-up. The age-standardized incidence rate of lung cancer in men was more than 6-fold greater among former smokers and 30-fold greater among current smokers, compared with never smokers. The corresponding rates for lung cancer in women were more than doubled in former smokers and more than 10-fold greater in current smokers, compared with never smokers. When we analyzed smoking exposure according to categorical groups (smoking status), we did not detect a difference between men and women. However, when we analyzed smoking exposure as a continuous variable, female current smokers had a significantly higher risk of lung cancer than male current smokers for increments of pack-years, cigarettes per day, and smoking duration. The pattern of a greater risk of lung cancer for women compared with men remained after exclusion of subjects diagnosed with lung cancer within the first 2 years after enrollment. Five cohort studies published between 2004 and 2015, including 470–17,670 lung cancer cases, did not find a sex difference in susceptibility to the carcinogenic effects of cigarette smoke (8, 9, 11, 12, 29). These cohort studies analyzed the risk of lung cancer according to fixed categories of smoking exposure. Our results are in accordance with theirs when we analyze the data this way. The increased risk of lung cancer among women that we found when we analyzed the data continuously is most likely concealed when the smoking exposure data are categorized. Since men are heavier smokers than women, within each category they are likely to be more heavily exposed than female smokers. Furthermore, the reference group for women comprises more lung cancer cases than that for men. This will also inflate the lung cancer risk for male smokers and attenuate that for women. The higher incidence rates of lung cancer for female never smokers compared with male never smokers is probably explained by more women than men being exposed to passive smoking. This is also what was found in a recent review (30). In Norway, men in every age group have a higher rate of death from cardiovascular disease than women (31); this was also found in the current study and could support the explanation for our finding of a sex difference in risk of lung cancer. A recent study from the European Prospective Investigation Into Cancer and Nutrition revealed lower relative risks of lung cancer in female current smokers than in male current smokers (32). However, this could be explained by a much heavier smoking burden among men compared with women in countries included in the European Prospective Investigation Into Cancer and Nutrition, such as Italy, Spain, and Greece (32). Our results suggesting lower relative risks of lung cancer in the most recent birth cohorts are not in accordance with findings observed in the United States (11) and the United Kingdom (33), respectively, where changes were towards higher relative risks. The difference observed by birth cohort between our study and the US study is mainly due to the fact that we stratified birth cohorts by the year 1950 (in or before 1950 or after 1950), while the 2 most recent time periods (1982–1988 and 2000–2010) from the US study were closer in time and had shorter durations of follow-up (6–10 years). Likewise, a possible explanation for the change towards a lower relative risk in the most recent birth cohort in our study is that the proportion of lung cancer cases among persons under age 50 years is only 10% in Norway (2). Our study had several major strengths. It was based on a large, prospectively assembled Norwegian cohort comprising a high proportion of male and female ever smokers, with long, virtually complete follow-up due to the national registries in Norway. Another major strength is that the questions about duration of smoking (in years) and number of cigarettes smoked per day were open instead of presented with fixed categories. Moreover, we had more than 6,500 lung cancer cases, giving us more stable risk estimates and results that were less prone to chance. We were also able to examine the association between smoking and lung cancer according to histological subtype and according to different measures of smoking exposure. A limitation of our study was the lack of updated information on smoking status during follow-up. In Norway, the proportion of daily smokers has decreased steadily, with a steeper reduction in men, and protection from passive smoking has increased, especially during the last decade of our follow-up period. This may explain some of the heterogeneity across birth cohorts and study cohorts. Because more men than women have quit smoking, this could explain some of the differences in lung cancer risk by sex that we found. We lacked information about passive smoking from the majority of the participants. Our reference group was therefore most likely contaminated with passive smokers. Since more men than women were smokers in our population, it is likely that more female never smokers than male never smokers were exposed to passive smoking. For women, this will have attenuated our observed risk of lung cancer among ever smokers; for men, it will have increased our observed risk of lung cancer among ever smokers. Approximately 10% of the Norwegian population reported being occasional smokers during follow-up (34). Some of these occasional smokers may have been excluded from our analytical sample due to insufficient smoking information, whereas others may have been included in the reference group, together with women exposed to passive smoking. This misclassification would most likely have attenuated the revealed associations between smoking and lung cancer. We also lacked information on possible confounders such as radon exposure and other exposures to air pollutants (1). We cannot rule out the possibility of residual confounding due to the above factors, or to other factors we did not measure. In conclusion, our results suggest that women are more susceptible than men to lung cancer given the same smoking exposure. Efforts to eliminate smoking in both sexes should continue. ACKNOWLEDGMENTS Author affiliations: Department of Community Medicine, Faculty of Health Sciences, UiT The Arctic University of Norway, Tromsø, Norway (Merethe S. Hansen, Idlir Licaj, Tonje Braaten, Inger T. Gram); Clinical Research Department, Centre François Baclesse, Caen, France (Idlir Licaj); General Practice Research Unit, Department of Public Health and Nursing, Faculty of Medicine and Health Sciences, Norwegian University of Science and Technology, Trondheim, Norway (Arnulf Langhammer); and Epidemiology Program, University of Hawaii Cancer Center, Honolulu, Hawaii (Loic Le Marchand). All authors contributed equally to the work. M.S.H., a doctoral student, was supported by the Northern Norway Health Authority (grant SFP1227-15). I.L., a postdoctoral researcher, was supported by grant agreement 4510766-2013 from the Norwegian Cancer Society. We acknowledge the services of the Cohort of Norway (CONOR) Study investigators and the contributing research centers delivering data to CONOR. Conflict of interest: none declared. Abbreviations BMI body mass index CI confidence interval CONOR Cohort of Norway REFERENCES 1 Steward BW, Wild CP, eds. World Cancer Report 2014. Lyon, France: International Agency for Research on Cancer; 2014. 2 Cancer Registry of Norway. Cancer in Norway 2013. https://www.kreftregisteret.no/Generelt/Publikasjoner/Cancer-in-Norway/Cancer-in-Norway-2013/. Published 2013. Updated March 2015. Accessed March 10, 2016. 3 Lund I, Lund KE. Lifetime smoking habits among Norwegian men and women born between 1890 and 1994: a cohort analysis using cross-sectional data. BMJ Open . 2014; 4( 10): e005539. Google Scholar CrossRef Search ADS PubMed  4 Statistics Norway. Smoking habits, 2013. https://www.ssb.no/en/helse/statistikker/royk/aar/2014-02-05. Published 2014. Accessed March 10, 2016. 5 Powell HA, Iyen-Omofoman B, Hubbard RB, et al.  . The association between smoking quantity and lung cancer in men and women. Chest . 2013; 143( 1): 123– 129. Google Scholar CrossRef Search ADS PubMed  6 De Matteis S, Consonni D, Pesatori AC, et al.  . Are women who smoke at higher risk for lung cancer than men who smoke? Am J Epidemiol . 2013; 177( 7): 601– 612. Google Scholar CrossRef Search ADS PubMed  7 Risch HA, Howe GR, Jain M, et al.  . Are female smokers at higher risk for lung cancer than male smokers? A case-control analysis by histologic type. Am J Epidemiol . 1993; 138( 5): 281– 293. Google Scholar CrossRef Search ADS PubMed  8 Bain C, Feskanich D, Speizer FE, et al.  . Lung cancer rates in men and women with comparable histories of smoking. J Natl Cancer Inst . 2004; 96( 11): 826– 834. Google Scholar CrossRef Search ADS PubMed  9 Freedman ND, Leitzmann MF, Hollenbeck AR, et al.  . Cigarette smoking and subsequent risk of lung cancer in men and women: analysis of a prospective cohort study. Lancet Oncol . 2008; 9( 7): 649– 656. Google Scholar CrossRef Search ADS PubMed  10 Freedman ND, Abnet CC, Caporaso NE, et al.  . Impact of changing US cigarette smoking patterns on incident cancer: risks of 20 smoking-related cancers among the women and men of the NIH-AARP cohort. Int J Epidemiol . 2016; 45( 3): 846– 856. Google Scholar CrossRef Search ADS PubMed  11 Thun MJ, Carter BD, Feskanich D, et al.  . 50-year trends in smoking-related mortality in the United States. N Engl J Med . 2013; 368( 4): 351– 364. Google Scholar CrossRef Search ADS PubMed  12 Prizment AE, Yatsuya H, Lutsey PL, et al.  . Smoking behavior and lung cancer in a biracial cohort: the Atherosclerosis Risk in Communities Study. Am J Prev Med . 2014; 46( 6): 624– 632. Google Scholar CrossRef Search ADS PubMed  13 Office of the Surgeon General, US Public Health Service. The Health Consequences of Smoking—50 Years of Progress . Atlanta, GA: US Department of Health and Human Services; 2014. 14 Stocks T, Borena W, Strohmaier S, et al.  . Cohort profile: the Metabolic syndrome and Cancer project (Me-Can). Int J Epidemiol . 2010; 39( 3): 660– 667. Google Scholar CrossRef Search ADS PubMed  15 Bjartveit K, Foss OP, Gjervig T. The cardiovascular disease study in Norwegian counties. Results from first screening. Acta Med Scand Suppl . 1983; 675: 1– 184. Google Scholar PubMed  16 Tverdal A, Bjartveit K. Health consequences of reduced daily cigarette consumption. Tob Control . 2006; 15( 6): 472– 480. Google Scholar CrossRef Search ADS PubMed  17 Aires N, Selmer R, Thelle D. The validity of self-reported leisure time physical activity, and its relationship to serum cholesterol, blood pressure and body mass index. A population based study of 332,182 men and women aged 40–42 years. Eur J Epidemiol . 2003; 18( 6): 479– 485. Google Scholar CrossRef Search ADS PubMed  18 Bjartveit K, Stensvold I, Lund-Larsen PG, et al.  . Cardiovascular screenings in Norwegian counties. Trends in risk pattern during the period 1985–90 among persons aged 40–42 in 4 counties [in Norwegian]. Tidsskr Nor Laegeforen . 1991; 111( 17): 2072– 2076. Google Scholar PubMed  19 Aamodt G, Søgaard AJ, Naess Ø, et al.  . The CONOR database—a little piece of Norway [in Norwegian]. Tidsskr Nor Laegeforen . 2010; 130( 3): 264– 265. Google Scholar CrossRef Search ADS PubMed  20 Naess O, Søgaard AJ, Arnesen E, et al.  . Cohort profile: Cohort of Norway (CONOR). Int J Epidemiol . 2008; 37( 3): 481– 485. Google Scholar CrossRef Search ADS PubMed  21 Parajuli R, Bjerkaas E, Tverdal A, et al.  . The increased risk of colon cancer due to cigarette smoking may be greater in women than men. Cancer Epidemiol Biomarkers Prev . 2013; 22( 5): 862– 871. Google Scholar CrossRef Search ADS PubMed  22 Bjerkaas E, Parajuli R, Weiderpass E, et al.  . Smoking duration before first childbirth: an emerging risk factor for breast cancer? Results from 302,865 Norwegian women. Cancer Causes Control . 2013; 24( 7): 1347– 1356. Google Scholar CrossRef Search ADS PubMed  23 Licaj I, Jacobsen BK, Selmer RM, et al.  . Smoking and risk of ovarian cancer by histological subtypes: an analysis among 300 000 Norwegian women. Br J Cancer . 2017; 116( 2): 270– 276. Google Scholar CrossRef Search ADS PubMed  24 World Health Organization. Global Database on Body Mass Index. BMI classification. http://www.assessmentpsychology.com/icbmi.htm. Published 2006. Updated February 15, 2018. Accessed March 10, 2016. 25 Larsen IK, Småstuen M, Johannesen TB, et al.  . Data quality at the Cancer Registry of Norway: an overview of comparability, completeness, validity and timeliness. Eur J Cancer . 2009; 45( 7): 1218– 1231. Google Scholar CrossRef Search ADS PubMed  26 World Health Organization. International Classification of Diseases for Oncology (ICD-O-3) . Geneva, Switzerland: World Health Organization; 2000. 27 Klein RJ, Schoenborn CA. Age adjustment using the 2000 projected U.S. population. Healthy People 2000 Stat Notes . 2001;( 20): 1– 9. 28 Royston P, Ambler G, Sauerbrei W. The use of fractional polynomials to model continuous risk variables in epidemiology. Int J Epidemiol . 1999; 28( 5): 964– 974. Google Scholar CrossRef Search ADS PubMed  29 Freedman ND, Abnet CC, Caporaso NE, et al.  . Impact of changing US cigarette smoking patterns on incident cancer: risks of 20 smoking-related cancers among the women and men of the NIH-AARP cohort. Int J Epidemiol . 2016; 45( 3): 846– 856. Google Scholar CrossRef Search ADS PubMed  30 Sun S, Schiller JH, Gazdar AF. Lung cancer in never smokers—a different disease. Nat Rev Cancer . 2007; 7( 10): 778– 790. Google Scholar CrossRef Search ADS PubMed  31 Norwegian Institute of Public Health. Cardiovascular Disease in Norway. Oslo, Norway: Norwegian Institute of Public Health; 2014. 32 Agudo A, Bonet C, Travier N, et al.  . Impact of cigarette smoking on cancer risk in the European Prospective Investigation into Cancer and Nutrition study. J Clin Oncol . 2012; 30( 36): 4550– 4557. Google Scholar CrossRef Search ADS PubMed  33 Pirie K, Peto R, Reeves GK, et al.  . The 21st century hazards of smoking and benefits of stopping: a prospective study of one million women in the UK. Lancet . 2013; 381( 9861): 133– 141. Google Scholar CrossRef Search ADS PubMed  34 Norwegian Institute of Public Health. Health and Ageing in Norway—Public Health Report 2014. Oslo, Norway: Norwegian Institute of Public Health; 2014. https://www.fhi.no/en/op/hin/helse-i-ulike-befolkningsgrupper/health-and-ageing-in-norway/. Updated April 12, 2016. Accessed March 10, 2016. © The Author(s) 2018. Published by Oxford University Press on behalf of the Johns Hopkins Bloomberg School of Public Health. All rights reserved. For permissions, please e-mail: [email protected]. This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/about_us/legal/notices)