Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You and Your Team.

Learn More →

Prevalence of Visual Acuity Loss or Blindness in the US

Prevalence of Visual Acuity Loss or Blindness in the US Research JAMA Ophthalmology | Original Investigation A Bayesian Meta-analysis Abraham D. Flaxman, PhD; John S. Wittenborn, BS; Toshana Robalik, BS; Rohit Gulia, MS; Robert B. Gerzoff, MS; Elizabeth A. Lundeen, PhD, MPH; Jinan Saaddine, MD, MPH; David B. Rein, PhD, MPA; for the Vision and Eye Health Surveillance System study group Invited Commentary page 723 IMPORTANCE Globally, more than 250 million people live with visual acuity loss or blindness, Multimedia and people in the US fear losing vision more than memory, hearing, or speech. But it appears Supplemental content there are no recent empirical estimates of visual acuity loss or blindness for the US. CME Quiz at OBJECTIVE To produce estimates of visual acuity loss and blindness by age, sex, jamacmelookup.com race/ethnicity, and US state. DATA SOURCES Data from the American Community Survey (2017), National Health and Nutrition Examination Survey (1999-2008), and National Survey of Children’s Health (2017), as well as population-based studies (2000-2013), were included. STUDY SELECTION All relevant data from the US Centers for Disease Control and Prevention’s Vision and Eye Health Surveillance System were included. DATA EXTRACTION AND SYNTHESIS The prevalence of visual acuity loss or blindness was estimated, stratified when possible by factors including US state, age group, sex, race/ethnicity, and community-dwelling or group-quarters status. Data analysis occurred from March 2018 to March 2020. Author Affiliations: Institute for Health Metrics and Evaluation, MAIN OUTCOMES OR MEASURES The prevalence of visual acuity loss (defined as a University of Washington, Seattle (Flaxman, Robalik, Gulia); NORC at best-corrected visual acuity greater than or equal to 0.3 logMAR) and blindness the University of Chicago, Chicago, (defined as a logMAR of 1.0 or greater) in the better-seeing eye. Illinois (Wittenborn, Rein); Applied Statistical Consulting LLC, Atlanta, RESULTS For 2017, this meta-analysis generated an estimated US prevalence of 7.08 Georgia (Gerzoff); Division of (95% uncertainty interval, 6.32-7.89) million people living with visual acuity loss, of whom Diabetes Translation, Vision Health 1.08 (95% uncertainty interval, 0.82-1.30) million people were living with blindness. Of this, Initiative Centers for Disease Control and Prevention, Atlanta, Georgia 1.62 (95% uncertainty interval, 1.32-1.92) million persons with visual acuity loss are younger (Lundeen, Saaddine). than 40 years, and 141 000 (95% uncertainty interval, 95 000-187 000) persons with Group Information: The members blindness are younger than 40 years. of the Vision and Eye Health Surveillance System study group CONCLUSIONS AND RELEVANCE This analysis of all available data with modern methods appear in Supplement 2. produced estimates substantially higher than those previously published. Corresponding Author: Abraham D. Flaxman, PhD, Institute for Health JAMA Ophthalmol. 2021;139(7):717-723. doi:10.1001/jamaophthalmol.2021.0527 Metrics and Evaluation, University of Published online May 13, 2021. Washington, 2301 Fifth Ave, Seattle, WA 98121 (abie@uw.edu). lobally, an estimated 252.6 (95% CI, 111.4-424.5) mil- estimated national and state visual acuity loss or blindness lion people live with best-corrected visual acuity of prevalence for persons ages 40 years and older and arrived at G 20/60 or worse in the better-seeing eye. People in a similar estimate of 4.24 million cases (2.8%). Both of these 3,4 the US fear losing vision more than memory, hearing, or speech, studies are limited, since they excluded persons younger and consider visual acuity loss among the top 4 worst things than 40 years and persons living in group quarters, such as 2 3,4 that could happen to them. No existing estimates appear to nursing homes and prisons. Both studies relied on meta- have used empirical data to estimate geographic differences, analytic summaries of similar selected population-based study created estimates for persons younger than age 40 years, or data, and no other data sources, to estimate prevalence by age accounted for increased prevalence in group quarters. group, sex, and race/ethnicity and then calculated state-level Previous studies have estimated national visual acuity loss estimates by applying these summary estimates to each state’s or blindness prevalence for important age ranges. The Vision population distribution. This method may lead to inaccura- Problems in the United States (VPUS) study estimated uncor- cies because the population-based study data (while of high rectable visual impairment and blindness for persons ages 40 quality) were collected 8 to 36 years in the past from locally years and older to occur in 4.2 million individuals (2.9%) in representative samples using different methods across stud- 3 4 2010. Using similar methods and data for 2015, Varma et al ies. State-specific estimates assumed that the prevalence of jamaophthalmology.com (Reprinted) JAMA Ophthalmology July 2021 Volume 139, Number 7 717 Research Original Investigation Prevalence of Visual Acuity Loss or Blindness in the US visual acuity loss or blindness observed in the population- based studies were invariant across states, with differences be- Key Points tween states resulting only from differences in the included Question How many people in the US are living with population demographics. However, visual acuity loss or blind- uncorrectable visual acuity loss or blindness? ness prevalence may vary substantially across states because Findings This bayesian meta-analysis generated an estimate that, risk factors for visual impairment (such as diabetes, smoking, in 2017, there were 7.08 million people living with visual acuity sun exposure, nutrition, toxins, or injuries), health care ac- loss, of whom 1.08 million were living with blindness. cess (eg, health insurance, access to eye care), social determi- Meaning Per this study, uncorrectable visual acuity loss and nants of health (eg, poverty, occupational hazards), and poli- blindness are even larger drivers of health burden in the US than 5-8 cies (eg, school entry screening) vary widely across states. was previously known. The Centers for Disease Control and Prevention’s Vision and Eye Health Surveillance System (VEHSS) provides infor- (a Snellen score of 20/40 or worse) and blindness as a subset mation on diagnosed national and state-specific visual acu- ity loss or blindness prevalence based on Medicare 100% of that group, consisting of those with a logMAR of 1.0 or fee-for-service data, MarketScan private insurance claims, greater (a Snellen score of 20/200 or worse) in the better- electronic health records from the IRIS Registry (a compre- seeing eye. We first estimated all visual acuity loss of log- hensive ophthalmology eye diseases clinical registry), and MAR 0.3 or greater, and then in a second, separate calcula- self-reported response data regarding visual difficulty or tion, estimated blindness (logMAR ≥1.0). blindness from 4 national surveys (the America Community Survey [ACS], National Survey of Children’s Health [NSCH], Data Behavioral Risk Factor Surveillance System, and Na- Our model used 4 data sources: (1) data abstracted from PBS, tional Health Interview Survey) and self-reported and (2) National Health and Nutrition Examination Survey examination-based data from the National Health and (NHANES) data collected during 1999 to 2008 (the only Nutrition Examination Survey (NHANES). These sources years in which vision data were collected), (3) ACS data col- yield divergent prevalence estimates based on differences in lected in 2017, and (4) NSCH data collected in 2016. For PBS, case definitions and persons included in the data. While the we searched online sources for studies published after 1991 VEHSS provides estimates from each of these sources, to our that were representative of the target population from which knowledge, no attempt has been made to summarize these the participants were sampled, presented primary results or data into a single meta-analytic national estimate. meta-analysis of primary data, and reported age-specific, We selected all relevant VEHSS data to create new na- race/ethnicity–specific, and/or location-specific prevalence tional and state estimates of US blindness and visual acuity estimates. We identified 5 such studies for inclusion from loss for all ages for the year 2017. We used bayesian meta- (1) the Baltimore Pediatric Eye Disease Study (data collection regression to combine all relevant information from the ACS period, 2003-2007; publication date, 2008) ; (2) the (for state-to-state variation, the oldest age groups, and preva- Chinese American Eye Study (data collection period, 2010- lence in group quarters), NHANES (a primary source of infor- 2013; publication date, 2016) ; (3) the Eye Diseases Preva- mation for mean tendency, age stratification, sex, and race/ lence Research Group (EDPRG; a meta-analysis of several ethnicity variation), the NSCH (for individuals of the youngest earlier PBS; data collection period, 1985-1998; publication ages), and population-based studies (PBS), and summarized date, 2004) ; (4) the Los Angeles Latino Eye Study data col- results by point estimates and uncertainty intervals (UIs). lection period, 2000-2003; publication date, 2004) ; and (5) the Multi-Ethnic Study of Atherosclerosis Cohort (data collection period, 2000-2004; publication date, 2015). We abstracted estimated prevalence of dichotomous mea- Methods sures of visual impairment and blindness and sample size information from each study by age group, sex, and race/ Ethical Review These research activities were deemed to be not human ethnicity. Of these sources, all but the EDPRG reported pri- subjects research by the institutional review board of NORC mary data on best-corrected visual acuity, as measured by at the University of Chicago because they are based exclu- study ophthalmologists. sively on secondary analysis of existing, deidentified data For NHANES participants aged 12 years or older, we used sources. For this reason, informed consent was not required. eye examination–derived measurements of best-corrected vi- sual acuity, as measured among persons with presenting vi- Strategy sual acuity of 20/40 or worse using the Auto Refractor model We applied bayesian meta-regression methods to multiple ARK-760 (Nidek) instrument and collected as part of a visual data sources with the goal of producing estimates of the health module that was fielded from 1999 to 2008 from a na- prevalence and uncertainty interval of visual acuity loss or tionally representative sample of US individuals dwelling in blindness, stratified by age group, sex, race/ethnicity, and communities. Among those with measurements, best- state (50 US states and Washington, DC) for the year 2017. corrected visual acuity was missing for 11.51%. As described We defined visual acuity loss using US standards as a best- in the eMethods in Supplement 1, we imputed missing cat- corrected visual acuity greater than or equal to 0.3 logMAR egorical indicators of visual acuity loss and blindness using 718 JAMA Ophthalmology July 2021 Volume 139, Number 7 (Reprinted) jamaophthalmology.com Prevalence of Visual Acuity Loss or Blindness in the US Original Investigation Research Table 1. Estimated Crude Prevalence Count and Rate of People Living With Visual Acuity Loss or Blindness, Stratified by Sex and Race/Ethnicity, US, 2017 Prevalence count, millions of people Prevalence rate, % Characteristic Mean 2.5th Percentile 97.5th Percentile Mean 2.5th Percentile 97.5th Percentile Total 7.08 6.32 7.89 2.17 1.94 2.42 Female 4.16 3.62 4.69 2.52 2.19 2.84 Male 2.92 2.53 3.37 1.82 1.57 2.10 Non-Hispanic Black 1.02 0.87 1.18 2.55 2.17 2.94 White 4.27 3.68 4.87 2.16 1.86 2.47 Hispanic 1.26 1.07 1.47 2.15 1.83 2.50 Other 0.52 0.41 0.62 1.76 1.40 2.12 multiple imputation with chained equations with boot- parameters to estimate variation in prevalence as a function strapped resampling. of age, sex, race/ethnicity, and data source and assumes that TheNSCHisanationallyrepresentativesurveyofthephysi- the age-stratified prevalence rate is not changing substan- cal and emotional health of children aged 0 to 17 years that con- tially over time. Full details are provided in the eMethods tains a caregiver-reported assessment of visual difficulty, which in Supplement 1. Data analysis occurred from March 2018 reads, “Does this child have blindness or problems with seeing, to March 2020. even when wearing glasses?” The ACS is an annual nation- ally representative and state-representative survey con- ducted by the US Census Bureau to provide information on Results demographic, social, economic, and housing characteristics of the US population. Like NSCH, ACS includes a head-of- Our data abstracted from PBS consisted of 103 measure- household–reported assessment of visual difficulty, which ments of visual acuity loss and 43 measurements of blind- reads, “Is this person blind or does he/she have serious diffi- ness. The surveys used included 35 466 individuals from the culty seeing even when wearing glasses?” and for which the NHANES, 3 190 040 individuals from the ACS, and 50 212 respondent reports for all members of the household. The ACS individuals from the NSCH. also includes information on group-quartered residences, al- lowing questions to be analyzed for those in nursing homes, Visual Acuity Loss or Blindness prisons, and other institutional group quarters, separately from We estimated a US prevalence count of 7.08 (95% UI, 6.32- residents in community-dwelling households. 7.89) million people living with visual acuity loss or blind- ness (using the US standard of best-corrected visual acuity in the worse-seeing eye of a Snellen score of 20/40 or worse) in Estimation We developed 2 statistical models to assess (1) the prevalence the US in 2017, corresponding to a crude prevalence rate of rate of all visual acuity loss stratified by age group, sex, race/ 2.17% (95% UI, 1.94%-2.42%) (Table 1). The national preva- ethnicity, group-quarters status, and US state and (2) the preva- lence level of visual acuity loss or blindness increased as a func- lence rate of blindness at the same levels of stratification. The tion of age, from 0.74% (95% UI, 0.37%-1.10%) among per- model estimated the dependent variable, observed preva- sons younger than 12 years to 0.99% (95% UI, 0.80%-1.18%) lence in each stratification category, as a negative, binomially among individuals aged 50 to 54 years and 20.73% (95% UI, distributed function of the number of persons evaluated in the 17.71%-23.27%) among persons aged 85 years and older sample and independent variables measuring sex, age, race/ (Figure 1). ethnicity (non-Hispanic Black, non-Hispanic White, His- Our meta-regression also estimated that 358 000 (95% panic, and other), US state, and source of data. We applied the UI, 263 000-472 000) persons with visual acuity loss or integrative systems modeling approach developed in the Global blindness reside in group quarters, such as nursing homes Burden of Disease Study to create these estimates. Follow- and prisons. This constitutes 5.06% (95% UI, 3.78%-6.60%) ing King, our integrative systems modeling reduces to an of all persons with visual acuity loss or blindness. extension of negative binomial regression, with a piecewise We estimated 1.62 (95% UI, 1.32-1.92) million persons with linear spline to represent the nonlinear age pattern and an age- visual acuity loss or blindness are younger than 40 years. This standardizing likelihood to account for the heterogeneous re- constitutes 22.89% of all persons with visual acuity loss porting of age groups in examination study data. This al- or blindness. lowed us to include data from all 5 PBS as well as the NHANES, Crude prevalence rates of visual acuity loss or blindness NSCH, and ACS in the likelihood during parameter estima- ranged from 1.35% (95% UI, 1.02%-1.65%) in Maine to 3.59% tion. We used the DisMod-MR 1.1.1, which implements this (95% UI, 2.93%-4.26%) in West Virginia. State differences per- model in Python version 3.6 using PyMC 2, and fit the model sisted after standardization by age, sex, and race/ethnicity with 400 000 iterations of Markov chain Monte Carlo using (Figure 2). Estimated counts and prevalence rates for each US an adaptive metropolis step method. The model includes state are provided in eTable 1 in Supplement 1. jamaophthalmology.com (Reprinted) JAMA Ophthalmology July 2021 Volume 139, Number 7 719 Research Original Investigation Prevalence of Visual Acuity Loss or Blindness in the US Figure 1. Crude Prevalence of Visual Acuity Loss or Blindness by Age for All Racial/Ethnic Groups National level Hispanic Non-Hispanic Black individuals Non-Hispanic White individuals People of other races/ethnicities <12 12-17 18-24 25-29 30-34 35-39 40-44 45-49 50-54 55-59 60-64 65-69 70-74 75-79 80-84 ≥85 Age, y Increases appear starting at approximately age 60 years. ranged from 0.19% (95% UI, 0.14%-0.25%) in Utah to 0.65% Figure 2. Age-Standardized, Sex-Standardized, and Race/Ethnicity– (95% UI, 0.46%-0.83%) in West Virginia (eTable 2 in Supple- Standardized Visual Acuity Loss or Blindness Prevalence by State ment 1; Figure 3). 1.5 2.0 2.5 3.0 3.5 Standardized prevalence rate Discussion Our estimated number of cases of visual acuity loss or blind- ness is 68.7% higher than the previous estimate created by the VPUS study, but our estimate of blindness alone is lower. Al- though the VPUS study reported findings among people 40 years and older based on different source data, this increase in estimated visual acuity loss or blindness is largely the re- sult of our inclusion of the NHANES data in the model and our choice to use imputation instead of listwise deletion to ad- dress the missing NHANES data. We estimated higher preva- lence for Hispanic and Black individuals compared with White individuals and for women compared with men; however, at least some of these estimates are very uncertain, with a pos- terior probability distribution that crosses zero. These results Blindness are consistent with previous analyses of NHANES data, which As a subset of persons with visual acuity loss or blindness, also found a higher risk of visual acuity loss among Hispanic we estimated a 2017 US prevalence count of 1.08 (95% UI, and Black individuals compared with White individuals and 0.82-1.30) million people living with blindness, defined in women compared with men but were not able to conclude as a best-corrected visual acuity of 1.0 logMAR or greater that these higher risks were statistically significant. Other im- (corresponding to a Snellen score of 20/200 or greater) in the portant differences between our estimate and the VPUS esti- better-seeing eye. This is equal to a crude prevalence rate mate include (1) using more recent population-based study of 0.33% (95% UI, 0.25%-0.40%) (Table 2). The crude preva- data; (2) using 2017 population structure for age, sex, race/ lence rate of blindness increased substantially as a function ethnicity, and household or group quarters size; (3) account- of age, from 0.05% (95% UI, 0.02%-0.08%) among persons ing for differences in prevalence in populations in community- 12 years and younger to 0.11% (95% UI, 0.08%-0.15%) among dwelling households vs group quarters; and (4) accounting individuals aged 50 to 54 years and 5.50% (95% UI, for variations across states. 3.70%-7.30%) among persons 85 years and older (eFigure in Supplement 1). We estimated that 130 000 (95% UI, 57 000-223 000) Limitations Our analyses were limited by at least 5 factors. First, the people with blindness are living in group quarters, such as nurs- ing homes and prisons. This constitutes 11.85% (95% UI, 5.52%- NHANES data had a substantial amount (approximately 12%) of missing autorefractor examination data. Our method of ac- 18.76%) of all people living with blindness. We estimated 141 000 (95% UI, 95 000-187 000) persons with blindness are counting for missing data, multiple imputations by chained equation, resulted in a substantially higher estimate of the younger than 40 years, which constitutes 13.09% of all per- sons with blindness. Crude prevalence rates of blindness prevalence rate of visual acuity loss (2.1%) than is obtained 720 JAMA Ophthalmology July 2021 Volume 139, Number 7 (Reprinted) jamaophthalmology.com Vision loss and blindness prevalence, % Prevalence of Visual Acuity Loss or Blindness in the US Original Investigation Research Table 2. Estimated Prevalence Count of People Living With Blindness, Stratified by Sex and Race/Ethnicity, as Well as Prevalence Rates Prevalence count, millions of people Prevalence rate, % Characteristic Mean 2.5th Percentile 97.5th Percentile Mean 2.5th Percentile 97.5th Percentile Total 1.08 0.82 1.30 0.33 0.25 0.40 Female 0.64 0.48 0.79 0.38 0.29 0.48 Male 0.45 0.34 0.55 0.28 0.21 0.35 Non-Hispanic Black 0.17 0.13 0.21 0.42 0.32 0.53 White 0.74 0.56 0.92 0.37 0.28 0.47 Hispanic 0.12 0.09 0.16 0.21 0.16 0.27 Other 0.05 0.04 0.08 0.19 0.13 0.26 using the same data and listwise deletion (1.7%). While we Figure 3. Age-Standardized, Sex-Standardized, and Race/Ethnicity– believe multiple imputations by chained equation is the Standardized Blindness Prevalence Estimates by State superior method to handle missing data because it uses the strength of other information to inform the estimates, 0.2 0.3 0.4 0.5 0.6 less missing data in NHANES would have resulted in more pre- Standardized prevalence rate cise estimates. Second, our estimates may be limited by the age of some of the included data sets. The NHANES data were collected from 1999 to 2008, and data from some of the population- based examination studies included in the EDPRG meta- analyses were collected even prior to that. However, our model also included more recent PBS published after the EDPRG meta- analysis, as well as the 2016 NSCH and the 2017 ACS. Addi- tionally, time-trend analyses of the ACS did not indicate sys- tematic differences in age-stratified, sex-stratified, or race/ ethnicity–stratified vision prevalence between the years 2008 and 2017 (not shown). Third, we used survey respondent–reported values from the ACS to account for differences in visual acuity loss preva- lence at the state level and within group quarters and from the The Medicare Minimum Data Set is generated as part of a clinical assessment of all residents in Medicare-certified or NSCH for children. Since these values are not based on an ex- amination, they likely contain false-positive results at least for Medicaid-certified nursing home, and includes an assess- ment of each resident’s functional capabilities and health uncorrected refractive error. Our model corrects for system- atically higher prevalence in self-reported visual difficulty mea- needs. However, it does not collect data on visual difficulty sures. However, to estimate prevalence variation by state, in a format that we were able to integrate into our model. household status, and childhood ages, our model assumes that The Baltimore Nursing Home Eye Survey found that 47% examination data on best-corrected visual acuity, if it were col- of people living with visual acuity loss in nursing homes lected, would vary following the same pattern as these self- were blind (compared with our finding of 14.5%), and if we reported data. Furthermore, because the ACS data included generalize this 47% to the entire group-quarters population, only a single measure of severe visual difficulty or blindness, we expect an additional 118 000 (95% UI, 87 000-156 000) our model assumes that state variation is the same for both vi- people living with blindness. Finally, we estimated visual acuity loss or blindness inde- sual acuity loss and blindness together as for blindness alone and by household status. We believe this assumption is both pendently, despite the logical interdependency that every per- son living with blindness is, by definition, a person living with reasonable and currently necessary to create data-driven es- timates for state, residents in group quarters, and children, but visual acuity loss. A more complex model that estimated the 2 outcomes simultaneously could perhaps make more effi- we acknowledge that additional examination data within these strata would improve the quality of future estimates. cient use of the sparse data available and eliminate the illogi- cal possibility of estimating more people living with blind- Fourth, we have assumed that the decomposition of visual acuity loss in distinct subcategories of visual impair- ness than living with visual acuity loss. However, the model structure presented here resulted in no instances in which the ment and blindness follow the same percentage breakdown in group quarters as in households. Although it seems plau- estimated rate of blindness exceeded the estimated rate of vi- sual acuity loss at any level of stratification. sible that the fraction of visual acuity loss that is blindness is higher in group quarters than in the household population, There are also several data sources in the VEHSS that we we found no reliable, representative data source to test this were not able to include in our analysis. Medicare and hypothesis or quantify the magnitude of the difference. MarketScan claims data and IRIS registry data are both jamaophthalmology.com (Reprinted) JAMA Ophthalmology July 2021 Volume 139, Number 7 721 Research Original Investigation Prevalence of Visual Acuity Loss or Blindness in the US appealing big data sources, but we were not able to adjust for the nonrepresentative nature of the populations repre- Conclusions sented by these sources, and there is a lack of evidence on the validity of diagnosed vision loss as a measure of Visual acuity loss and blindness continue to be a substantial population-level vision. The Behavioral Risk Factor Surveil- burden to the US population, and our new analysis indicates lance System and National Health Interview Survey both that the issue is even more substantial than has previously been include respondent-reported data similar to ACS, which recognized. Efforts to collect new examination-based infor- we excluded on our assessment that the ACS sample mation on best-corrected visual acuity in the better-seeing size is substantially larger and the potential biases of eye would enhance future efforts to create more precise na- respondent-reported visual acuity loss would be com- tional and state estimates of visual acuity loss or blindness, pounded by bringing together sources from multiple instru- and this evidence base could be valuable for targeted efforts ments and surveys. to prevent or treat these conditions. ARTICLE INFORMATION Disease Control and Prevention’s scientific www.cdc.gov/visionhealth/vehss/ clearance process. index.html. Accepted for Publication: January 26, 2021. Group Information: The Vision and Eye Health 10. Flaxman AD, Vos DT, Murray CJ. An Integrative Published Online: May 13, 2021. Surveillance System Study Group members are Metaregression Framework for Descriptive doi:10.1001/jamaophthalmol.2021.0527 listed in Supplement 2. Epidemiology. University of Washington Press; 2015. Open Access: This is an open access article Disclaimer: The findings and conclusions in this 11. Dougherty M, Wittenborn J, Phillips E, distributed under the terms of the CC-BY License. report are those of the authors and do not Swenor B. Published examination-based prevalence © 2021 Flaxman AD et al. JAMA Ophthalmology. necessarily represent the official position of major eye disorders. Published 2018. Accessed Author Contributions: Dr Flaxman had full access of the US Centers for Disease Control and March 3, 2021. https://www.norc.org/PDFs/VEHSS/ to all of the data in the study and takes Prevention. EyeConditionExamLiteratureReviewVEHSS.pdf responsibility for the integrity of the data and the 12. Friedman DS, Repka MX, Katz J, et al. accuracy of the data analysis. REFERENCES Prevalence of decreased visual acuity among Concept and design: Flaxman, Wittenborn, Gulia, 1. Bourne RRA, Flaxman SR, Braithwaite T, et al; preschool-aged children in an American urban Saaddine, Rein. Vision Loss Expert Group. Magnitude, temporal population. Ophthalmology. 2008;115(10):1786-1795. Acquisition, analysis, or interpretation of data: trends, and projections of the global prevalence of doi:10.1016/j.ophtha.2008.04.006 Flaxman, Wittenborn, Robalik, Gulia, Gerzoff, blindness and distance and near vision impairment. Lundeen, Rein. 13. Varma R, Kim JS, Burkemper BS, et al; Lancet Glob Health. 2017;5(9):e888-e897. doi:10. Drafting of the manuscript: Flaxman, Wittenborn, Chinese American Eye Study Group. Prevalence and 1016/S2214-109X(17)30293-0 Robalik, Gulia, Gerzoff, Rein. causes of visual impairment and blindness in Critical revision of the manuscript for important 2. Scott AW, Bressler NM, Ffolkes S, Wittenborn JS, Chinese American adults. JAMA Ophthalmol. 2016; intellectual content: Flaxman, Wittenborn, Gulia, Jorkasky J. Public attitudes about eye and vision 134(7):785-793. doi:10.1001/jamaophthalmol. Lundeen, Saaddine, Rein. health. JAMA Ophthalmol. 2016;134(10):1111-1118. 2016.1261 Statistical analysis: Flaxman, Wittenborn, Robalik, doi:10.1001/jamaophthalmol.2016.2627 14. Congdon N, O’Colmain B, Klaver CC, et al; Eye Gulia, Gerzoff. 3. Prevent Blindness America. Vision problems in Diseases Prevalence Research Group. Causes and Obtained funding: Wittenborn, Rein. the US: prevalence of adult vision impairment and prevalence of visual impairment among adults in Administrative, technical, or material support: age-related eye disease in America. Published 2012. the United States. Arch Ophthalmol. 2004;122(4): Wittenborn, Gulia, Lundeen, Rein. Accessed September 30, 2019. http://www. 477-485. doi:10.1001/archopht.122.4.477 Supervision: Wittenborn, Lundeen, Saaddine, Rein. visionproblemsus.org/ 15. Varma R, Ying-Lai M, Klein R, Azen SP; Conflict of Interest Disclosures: Dr Flaxman 4. Varma R, Vajaranant TS, Burkemper B, et al. Los Angeles Latino Eye Study Group. Prevalence reported grants from NORC as a subcontract from a Visual impairment and blindness in adults in the and risk indicators of visual impairment and US Centers for Disease Control and Prevention United States. JAMA Ophthalmol. 2016;134(7): blindness in Latinos. Ophthalmology. 2004;111(6): during the conduct of the study; payments from 802-809. doi:10.1001/jamaophthalmol.2016.1284 1132-1140. doi:10.1016/j.ophtha.2004.02.002 Janssen, IHME, SwissRe, and Merck for Mothers for 5. Elam AR, Lee PP. High-risk populations for vision 16. Fisher DE, Shrager S, Shea SJ, et al. Visual assistance in analysis and interpretation of licensed loss and eye care underutilization. Surv Ophthalmol. impairment in White, Chinese, Black, and Hispanic data previously produced by IHME; and payment 2013;58(4):348-358. doi:10.1016/j.survophthal.2012. participants from the multi-ethnic study of from the startup Agathos Ltd for advising outside 07.005 atherosclerosis cohort. Ophthalmic Epidemiol. the submitted work. Dr Rein reported grants from 2015;22(5):321-332. doi:10.3109/09286586.2015. the US Centers for Disease Control and Prevention 6. Armstrong RA, Mousavi M. Overview of risk Vision Health Initiative during the conduct of the factors for age-related macular degeneration study. No other disclosures were reported. (AMD). J Stem Cells. 2015;10(3):171-191. 17. National Health and Nutrition Examination Survey. 2007-2008 data documentation, Funding/Support: This study was supported by 7. Pleet A, Sulewski M, Salowe RJ, et al. Risk factors codebook, and frequencies. Published 2009. funding from the US Centers for Disease Control associated with progression to blindness from Accessed October 2, 2019. https://wwwn.cdc.gov/ and Prevention Vision Health Initiative (cooperative primary open-angle glaucoma in an Nchs/Nhanes/2007-2008/VIX_E.html agreement U01DP006074, “Establish a Vision and African-American population. Ophthalmic Epidemiol. Eye Health Surveillance System for the Nation”). 2016;23(4):248-256. doi:10.1080/09286586.2016. 18. Raghunathan TE, Lepkowski JM, Van Hoewyk J, 1193207 Solenberger P. A multivariate technique for multiply Role of the Funder/Sponsor: As coauthors, imputing missing values using a sequence of employees of the funding organization 8. Penman A, Hancock H, Papavasileiou E, et al. regression models. Survey Methodology. 2001;27 (Drs Lundeen and Saaddine) participated in the Risk factors for proliferative diabetic retinopathy in (1):85-96. design and conduct of the study; collection, African Americans with type 2 diabetes. Ophthalmic management, analysis, and interpretation of the Epidemiol. 2016;23(2):88-93. doi:10.3109/ 19. US Census Bureau. National Survey of data; preparation, review, or approval of the 09286586.2015.1119287 Children’s Health (NSCH). Published 2019. manuscript; and decision to submit the manuscript Accessed October 3, 2019. https://www.census. 9. Centers for Disease Control and Prevention. for publication. Additionally, the manuscript was gov/programs-surveys/nsch.html The Vision and Eye Health Surveillance System. reviewed and approved by the US Centers for Published 2019. Accessed October 1, 2019. https:// 20. US Census Bureau. American Community Survey (ACS). Published 2019. Accessed October 3, 722 JAMA Ophthalmology July 2021 Volume 139, Number 7 (Reprinted) jamaophthalmology.com Prevalence of Visual Acuity Loss or Blindness in the US Original Investigation Research 2019. https://www.census.gov/programs-surveys/ 23. Ko F, Vitale S, Chou C-F, Cotch MF, Saaddine J, Baltimore. N Engl J Med. 1995;332(18):1205-1209. acs Friedman DS. Prevalence of nonrefractive visual doi:10.1056/NEJM199505043321806 impairment in US adults and associated risk factors, 21. King G. Unifying Political Methodology: the 1999-2002 and 2005-2008. JAMA. 2012;308(22): Likelihood Theory of Statistical Inference. University 2361-2368. doi:10.1001/jama.2012.85685 of Michigan Press; 1998. doi:10.3998/mpub.23784 24. Tielsch JM, Javitt JC, Coleman A, Katz J, 22. Patil A, Huard D, Fonnesbeck CJ. PyMC: Sommer A. The prevalence of blindness and visual Bayesian stochastic modelling in Python. J Stat Softw. impairment among nursing home residents in 2010;35(4):1-81. doi:10.18637/jss.v035.i04 Invited Commentary Updated Numbers on the State of Visual Acuity Loss and Blindness in the US Emily Y. Chew, MD In this issue of JAMA Ophthalmology,Flaxmanetal Other Results reported the results of their study designed to estimate the The overall prevalence of visual acuity loss in the US was 7.08 rates of visual acuity loss and blindness in the US, including (95% uncertainty interval [UI], 6.32-7.89) million, translating 2,3 1 rates for each individual state. Prior published reports in to a crude prevalence rate of 2.17% (95% UI, 1.94%-2.42%). 2010 and 2015 addressing Not surprisingly, the prevalence of visual acuity loss or blind- this issue used different data ness increased with age, from 0.74% (95% UI, 0.37%-1.10%) Related article page 717 sets and statistical method- for those younger than 12 years, to 0.99% (95% UI, 0.80%- ology, resulting in rates for those 40 years and older only. 1.18%) among those aged 50 to 54 years and 20.73% (95% UI, 2,3 Both studies suggested the number of individuals older 17.71%-23.27%) among persons 85 years and older. As ex- than 40 years affected with vision impairment was approxi- pected, a range of rates of visual acuity loss was found in dif- mately 4.2 million in the US. The results of the current ferent states. Interestingly, approximately 22.89% of those with study of the population of all ages estimated the rate to be visual acuity loss or blindness were younger than 40 years. much higher, at 7.08 million people living in the US with Similar to previous analyses of the National Health and Nutri- visual acuity loss, defined as 20/40 or worse, with 1.08 mil- tion Examination Survey data, the current study estimated lion of these having blindness, defined as visual acuity of higher prevalence in Black and Hispanic individuals com- 20/200 or worse. These are important data that need to be pared with White individuals and women compared with men. explored further. However, these probabilities of differences between sex and race/ethnicity crossed zero and were not statistically significant. Methods The current study analyzed a greater number of studies, including the classic population-based studies and data Importance of These Data from the US Centers for Disease Control and Prevention’s These data underscore the burden of blindness in the US. Vision and Eye Health Surveillance, which provide data on Visual acuity loss is considered one of the most dreaded events both national and state-specific rates of visual acuity loss that individuals in the US fear compared with loss of speech, using insurance claims, registries of electronic health rec- hearing, or memory. In addition, it is important to obtain ac- ords, and self-reported data from national surveys that curate prevalence and eventually incidence data on visual im- included populations of all ages and were stratified by race/ pairment and blindness, because they have compelling pub- ethnicity and sex. Flaxman et al recognized the importance lic health implications. With this large increase in the numbers of analyzing more granular data found in state-reported of US individuals who will experience visual acuity loss or rates because of the differences found in population demo- blindness, we need to prepare to the health care systems to graphics and risk factors for visual acuity loss, including serve affected individuals. These estimates will also help to comorbidities such as diabetes, health care access, social promote potential screening and public health education for determinants of health, and lifestyle factors, such as nutri- select ocular diseases that have effective therapies that may tion and smoking. be given either as preventive therapy or active treatment to The use of the bayesian meta-analysis methods in this preserve visual acuity. study may be superior in estimating between-study hetero- Although the study results were not statistically signifi- geneity and pooled effects, especially when there is a rela- cant, the trends are similar to other studies that have found tively small number of studies, as found by Flaxman et al. women to have increased burden of blindness. Data suggest- Summarizing using various sources of data in a single ing an increased rates of visual acuity loss in women does el- meta-analysis is the innovative part of this approach. evate the alert level to consider conducting important stud- The bayesian methods also allow the researchers to inte- ies in assessing sex as a biological variable, as promoted by the grate prior knowledge and assumptions when calculating National Institutes of Health. Others have highlighted the the meta-analyses. excessive burden on Black and Hispanic individuals, again jamaophthalmology.com (Reprinted) JAMA Ophthalmology July 2021 Volume 139, Number 7 723 © 2021 American Medical Association. All rights reserved. Supplemental Online Content Flaxman AD, Wittenborn JS, Robalik T, et al . Prevalence of visual acuity loss or blindness in the US: a bayesian meta-analysis. JAMA Ophthalmol. Published , 2021. doi:10.1001/ jamaophthalmol.2021.0527 eMethods. Details, verification, and validation of our methodological approach. eFigure. Crude prevalence rate of blindness increases with age for all race and ethnic groups, starting around age 60 years. eTable 1. Estimated prevalence count of people living with visual acuity loss or blindness, stratified by state, as well as prevalence rates (in percent). eTable 2. Estimated prevalence count of people living with blindness, stratified by location, as well as prevalence rates (in percent). This supplemental material has been provided by the authors to give readers additional information about their work. © 2021 Flaxman AD et al. JAMA Ophthalmology. eMethods: details, verification, and validation of methodological approach In this supplementary appendix, we present additional details on the verification and validation of our methodological approach. To better understand the impact of modeling choices, we examined a nested sequence of intermediate models of increasing complexity. Initially we estimated the numbers of cases of visual acuity loss using only PBS data. We then added NHANES data as reference with PBS data for the estimation, and then merged NSCH data with NHANES data and PBS data in the model and generated estimates of visual acuity loss. We then added ACS data for older ages and group quarters sequentially (denoted ACS- older and ACS-gq in Table 1 below). After that we used ACS data for state-specific prevalence and for year 2017 to generate estimates (denoted ACS-state and ACS-2017 in Table 1 below). In the final model we added interaction terms along with all the datasets to generate visual acuity loss estimates. Table 1: Estimated prevalence counts (in millions) of people living with visual acuity loss or blindness, after each step of the sequence in the model. Data Sources Visual acuity loss among Visual acuity loss in all age 40+ ages (in millions) (in millions) Population Based Study data (Exam) 3.47 [2.56, 4.61] 2.92 [2.12, 3.83] © 2021 Flaxman AD et al. JAMA Ophthalmology. Exam, NHANES 4.28 [3.41, 5.25] 5.80 [4.78, 6.97] Exam, NHANES, NSCH 4.35 [3.50, 5.30] 5.83 [4.81, 6.92] Exam, NHANES, NSCH, ACS-older 4.49 [3.63, 5.47] 5.98 [5.07, 7.08] Exam, NHANES, NSCH, ACS-older, 4.63 [3.85, 5.48] 6.16 [5.29, 7.13] ACS-gc Exam, NHANES, NSCH, ACS-older, 4.57 [4.02, 5.16] 6.11 [5.38, 6.89] ACS-gc, ACS-state Exam, NHANES, NSCH, ACS-older, 5.43 [4.77, 6.16] 7.04 [6.29, 7.92] ACS-gc, ACS-state, ACS-2017 Exam, NHANES, NSCH, ACS-older, 5.46 [4.83, 6.16] 7.08 [6.34, 7.92] ACS-gc, ACS-state, ACS-2017, interaction terms Table 2 presents similar results for the subset of patients who we predicted to be blind. Table 2: Estimated prevalence counts (in millions) of people living with blindness, after each step of the sequence in the model. © 2021 Flaxman AD et al. JAMA Ophthalmology. Data Sources Blindness among age 40+ Blindness in all ages (in millions) (in millions) Population Based Study data (Exam) 0.77 [0.51, 1.15] 0.73 [0.48, 1.10] Exam, NHANES 0.71 [0.45, 1.14] 0.82 [0.56, 1.25] Exam, NHANES, NSCH 0.73 [0.48, 1.07] 0.86 [0.59, 1.20] Exam, NHANES, NSCH, ACS-older 0.69 [0.50, 0.94] 0.82 [0.60, 1.08] Exam, NHANES, NSCH, ACS-older, 0.81 [0.59, 1.09] 0.96 [0.71, 1.24] ACS-gc Exam, NHANES, NSCH, ACS-older, 0.81 [0.62, 1.04] 0.96 [0.74, 1.21] ACS-gc, ACS-state Exam, NHANES, NSCH, ACS-older, 0.91 [0.71, 1.12] 1.06 [0.84, 1.28] ACS-gc, ACS-state, ACS-2017 Exam, NHANES, NSCH, ACS-older, 0.93 [0.74, 1.17] 1.08 [0.85, 1.34] ACS-gc, ACS-state, ACS-2017, interaction terms © 2021 Flaxman AD et al. JAMA Ophthalmology. Readers may notice that our reported results when using only PBS data (row 1 of Tables 1 and 2) are lower than those reported by the Vision Problems in the U.S. (VPUS) study which was also based on population-based studies (3). That study (available at visionproblemsus.org) estimated 4.19 million people with vision impairment and blindness, of whom 1.29 million were blind. Based only on population-based studies, our model estimates 2.92 million people who had visual acuity loss or blindness, of whom approximately 730,000 were blind. This difference is driven by the PBS that were used in our model as compared to the earlier VPUS. The studies we included as PBS are described in the data section above. They can be compared to those used by VPUS by reviewing the VPUS Methods and Sources page. In general, the studies that we included reported a lower prevalence of visual impairment and blindness than the studies included in the VPUS and this lower prevalence is reflected in our model estimate. We believe specific differences between our estimates are attributable to the inclusion of older data from the Baltimore Eye Survey (collected from 1985 and 1988) (1) and the Salisbury Eye Evaluation Project (collected from 1993 and 1995) (2) in VPUS, and our inclusion of data from CHES (collected from 2010-2013), LALES (collected from 2000-2008), and MESA (collected from 2002-2004). MICE-Multiple imputation by chained equation Missing data in meta-analysis can jeopardize inferences from the study and examination data, and possibly bias the estimates. The vision examination data from NHANES is a key input in our meta-analytic estimates of the prevalence of uncorrectable visual acuity loss or blindness in the United States, but when we mapped this data to dichotomous outcomes for visual acuity loss or blindness, we were unable to determine a value for 11.51% of respondents. © 2021 Flaxman AD et al. JAMA Ophthalmology. Multiple imputation by chained equations (MICE) is a standard approach to address missing data. MICE constructs a regression model for each column with missing values, using the variables from the other columns as the predictors. If multiple columns have missing values, the MICE procedure iterates through the columns, fitting each with the previously imputed values for the other columns. Simpler approaches, such as complete-case analysis (also called list-wise deletion) and available-case analysis, may produce biased results. The mean age among individuals with incomplete eye exam was 52.8 years, substantially higher than the mean age among individuals with complete eye exam (38.6 years), and since visual acuity loss or blindness prevalence rates increase markedly as age increases, our imputation method must take age into account. Using equation 1 and 2 below, we imputed missing dichotomous indicators of visual loss and blindness using multiple imputation with chained equations (MICE) with Gaussian perturbations: (1) Visual acuity loss C(age_group) + sex + C(race_eth) + vidrva + vidlva, (2) Blindness C(age_group) + sex + C(race_eth) + vidrva + vidlva, where vidrva and vidlva are presenting visual acuity of the right eye and left eye respectively, C(age_group) respresents a “dummy coding” for age variable grouped in bins (0, 12, 18, 25, 30, 35, 40, 45, 50, 55, 60, 65, 70, 75, 80, 85, 100) and C(race_eth) represents a “dummy coding” for our categorical variable for race/ethnicity. The MICE approach requires some choices: what else should go into the regression formula for the missing variables? Should we use a Gaussian for perturbing the model parameters in the imputation model, or a non-parametric bootstrap? To identify the best © 2021 Flaxman AD et al. JAMA Ophthalmology. regression formula and perturbation method for imputing missing values of visual acuity loss or blindness, we conducted an out-of-sample validation exercise. We considered a range of regression equations for our outcomes of interest (visual acuity loss or blindness): (in the formulas below, the notation (var1 + var2)**2 means the variables var1 and var2 are included as main effects and all pair-wise interactions are also included) 1. outcome ~ age + sex + C(race) (where C(race) is a “dummy” coding of the race/ethnicity variable) 2. outcome ~ C(age_group) + sex + C(race) (where C(age_group) is a “dummy” coding of the age variable grouped in bins [0, 12, 18, 25, 30, 35, 40, 45, 50, 55, 60, 65, 70, 75, 80, 85, 100] ) 3. outcome ~ C(age_group) + sex + C(race) + vidrva (where vidrva is the presenting visual acuity in the right eye) 4. outcome ~ C(age_group) + sex + C(race) + vidlva (where vidlva is the presenting visual acuity in the left eye) 5. outcome ~ C(age_group) + sex + C(race) + vidrva + vidlva 6. outcome ~ C(age_group) + sex + C(race) + (vidrva + vidlva )**2 7. outcome ~ C(age_group) + sex + C(race) + vidrova (where vidrova is the best-corrected visual acuity in the right eye) 8. outcome ~ C(age_group) + sex + C(race) + vidlova (where vidlova is the best-corrected visual acuity in the left eye) 9. outcome ~ C(age_group) + sex + C(race) + vidrova + vidlova 10. outcome ~ C(age_group) + sex + C(race) + (vidrova + vidlova )**2 © 2021 Flaxman AD et al. JAMA Ophthalmology. 11. outcome ~ C(age_group) + sex + C(race) + vidrva + vidlva + vidrova + vidlova 12. outcome ~ C(age_group) + sex + C(race) + (vidrva + vidlva + vidrova + vidlova)**2 We tested each equation for each outcome for two alternative perturbation methods: (a) the Gaussian perturbation method, which samples from a multivariate normal distribution derived from the fit of the regression equation; and (b) the bootstrap perturbation method, which samples from the conditional model fitted to a bootstrapped version of the data set. For our testing approach, we used an out-of-sample cross-validation approach from machine learning, where we withheld the presenting and best corrected visual acuity measurements for a randomly selected subset of data (our “test dataset”) and compared the model predictions for these individuals to the true values of blindness and visual acuity loss.(3) To be precise, 1. First, we selected 25% of the rows of NHANES data and redacted their vidrva, vidlva, vidrova, and vidlova values (which are all numeric values measured for presenting and best corrected visual acuity in the right and left eye; these values are always sufficient to derive the values of the visual acuity loss and blindness variables, and certain patterns of missing values among these measurements lead to missing values of the dichotomous visual acuity loss and blindness variables). For these rows, we also redacted the visual acuity loss or blindness values derived from these measurements. 2. Second, we imputed the masked values using MICE 1,000 times and took the average of these values as a probability prediction of the outcome of interest (visual acuity loss or blindness). © 2021 Flaxman AD et al. JAMA Ophthalmology. 3. Third, we compared those imputed values and actual values using the ROC curve in which the true positive rate (Sensitivity) is plotted as a function of the false positive rate (100- Specificity) for different cut-off points of a parameter. Each point on the ROC curve represents a sensitivity/specificity pair corresponding to a particular decision threshold. We measured the area under the ROC curve (AUC) as a summary metric for performance of the MICE formulation. MICE validation results Missing values were best imputed using the variables age, sex, race, vidrva and vidlva, as shown in Table 3. Table 3: Table showing the AUC score of all the models with different parameters Method / Equation VL VL Blindness Blindness Bootstrap Gaussian Bootstrap Gaussian Age + Sex + C(Race) 70.11 69.74 73.15 72.95 C(Age_group) + Sex + C(Race) 72.21 71.18 72.51 70.39 C(Age_group) + Sex + C(Race) + Vidrva 75.81 75.86 80.14 81.09 C(Age_group) + Sex + C(Race) + Vidlva 74.72 75.66 77.05 83.89 C(Age_group) + Sex + C(Race) + Vidrva + 75.51 74.66 80.75 78.53 Vidlva © 2021 Flaxman AD et al. JAMA Ophthalmology. C(Age_group) + Sex + C(Race) + (Vidrva + 74.98 75.74 78.70 78.93 Vidlva)**2 C(Age_group) + Sex + C(Race) + Vidrova 75.04 74.49 82.95 70.89 C(Age_group) + Sex + C(Race) + Vidlova 75.29 74.81 77.66 79.59 C(Age_group) + Sex + C(Race) + Vidrova 74.19 74.99 75.22 72.04 +Vidlova C(Age_group) + Sex + C(Race) + (Vidrova 74.25 72.43 69.24 75.48 + Vidlova)**@ C(Age_group) + Sex + C(Race) + Vidrva + 74.66 72.97 70.27 67.53 Vidlva + Vidrova + Vidlova C(Age_group) + Sex + C(Race) + (Vidrva + 72.60 71.26 77.00 78.62 Vidlva + Vidrova + Vidlova)**2 Our selected approach to MICE produced out-of-sample AUC of 75.51, compared with AUC of 70.79 in complete-case analysis. Bayesian meta-regression model We represent our model with a stochastic component and a systematic component, where the stochastic component is a negative binomial model of count data: (3) NegativeBinomial , , , © 2021 Flaxman AD et al. JAMA Ophthalmology. where indexes the specific measurement, is the prevalence count of those with visual acuity loss (or blindness) in measurement , is the effective sample size from which the count was is the prevalence rate typically reported in a PBS), is the prevalence taken (and so = rate predicted by the model, is the over-dispersion parameter of the negative binomial distribution (assumed to be the same for all measurements). DisMod-MR 1.1.1 uses the Python PyMC2 package to implement this Bayesian computation, and follows the formulation of the ( ) ( | ) negative binomial model provided by PyMC2, where Pr = , = , ( ) which in terms of the equation above has = , = , and = . In the systematic component of the model, we included fixed effects for sex ( ), age ( for = 0, … , ), race ( for = white, Black, Hispanic, and other races), and data source ( for =EDPRG, CHES, LALES, ACS, BPEDS, MESA, and NSCH; we coded NHANES as the reference category), as well as sex/race and race/age interaction fixed effects. We also include random effects for state (50 states and Washington, D.C.; for = 1, … ,51). Our formulation includes a piece-wise linear spline model on age, with spline knots (knot ,knot ,…) indexed by for = 1, … , , as well as intercept shifts for additional key covariates to account for differential sex ratios and age patterns by race as follows: (4) = exp{ [sex = female]} × + ( knot ) × exp [race = ] + [source = ] + [location = ] + interaction_terms , © 2021 Flaxman AD et al. JAMA Ophthalmology. where sex is the sex measured in measurement ; age and age are the start and end of the age group measured in measurement ; race is the race/ethnicity group measured in measurement ; source is the data source for measurement ; location is the state location of measurement ; and notation [variable = value] represents an indicator function which takes value 1.0 if the variable is equal to the value and 0.0 otherwise and notation (value) represents the value in the parenthesis if it is positive, and takes value zero, otherwise. We used evenly spaced spline knots at ages (0, 20, 40, 60, 80, 100). We also included interaction terms to capture the possibility that the sex and age effects were different by race/ethnicity and group quarters status: interaction_terms = [race = and sex = female] + [race = ](age ) + [race = and age > 50](age 50) + [g. quarters = inst](age ) + [g. quarters = inst][age > 50](age 50), where is the effect coefficient for race interacted with sex (for = 1,…4), is the effect coefficient for race interacted with age, is the effect coefficient for race interacted with age > 50, and age = is the midpoint of the age group measured in measurement . Together and constitute an age-dependent spline for the group quarters population with knots at 0 and 50. We used a Bayesian framework for inference with weakly informative priors for model parameters other than state to assist in regularization, which primarily allowed the data to inform the model estimates. Prior distributions for , , , , , , , and were all set to independent normal distributions with mean 0.0 and © 2021 Flaxman AD et al. JAMA Ophthalmology. standard deviation 1.0. To capture state variation in prevalence, we used informative priors for state random effects, which were informed by the state-to-state variation in age-/sex-/race- standardized endorsement rates of the respondent-reported visual acuity loss question in ACS. This prior took the form (5) Normal , , where, in short, is the log of the ratio of the standardized state prevalence to the national prevalence, and is the standard deviation of the log-ratio of the crude state prevalence to the national prevalence for all strata with at least 500 individuals, and is normal distribution with mean and standard deviation . To be precise, to estimate and , we first estimated the crude prevalence rate of respondent-reported visual acuity loss in the US in 2017 as observed in ACS data. We next estimated the prevalence rate of respondent-reported visual acuity loss for each state from the same data source, stratified by age-group, sex, race/ethnicity, and group quarters status. We then used the state-specific stratified estimates to calculate a standardized prevalence rate, standardizing with the age group, sex, race/ethnicity, and group-quarters weights at the national level, to control for demographic difference between states. We then used these data to estimate the ratio of the standardized state prevalence to the national prevalence. For each state , we set the prior on to be normally distributed with mean = log ratio , where ratio is the ratio of standardized prevalence in state to the national prevalence. We obtained an informative standard deviation for this prior as follows: analogous to the ratio use used for the mean, we constructed a ratio of the national prevalence rate to the state-specific prevalence rate for each strata (stratified by age-group, sex, race/ethnicity, and group quarters status) and calculated the standard deviation of the log-ratio for all strata with at least 500 individuals in the © 2021 Flaxman AD et al. JAMA Ophthalmology. state, age, sex, and race/ethnicity-specific strata. We put a cap on the predicted prevalence at 0.25 to ensure that we are not over estimating our results. Only data from ACS was available to inform the group-quarters effect coefficients, but unlike the state-to-state variation, we used this ACS data in the likelihood instead of constructing an informative prior. To guard against inferring with more confidence than appropriate from the ACS group quarters data, we used aggregate measurements for broad age groups (25-year intervals) and did not stratify them by sex or by race/ethnicity. We also included fixed effects for each age group, which effectively treated each comparison of the prevalence in the institutional group quarters population and the free-living population as a separate study in the meta- regression. We included all other data sources (NHANES, PBS, NSCH) in the likelihood with the assumption that they applied to a nation-level estimate, which we assumed to be constant over time. This includes data from NSCH, NHANES, and BPEDS for children under the age of 18, and also ACS data on free-living adults over the age of 85. Validation and Verification We assessed the cross-model validation of our model by setting its population parameters to those of the VPUS study (persons 40 and older in household settings), estimating the number of cases it generated when using only PBS data, and then sequentially estimating the number of cases it generated using the new data sources and model specifications that we included in our final model. See eMethods supplementary materials for full details. References © 2021 Flaxman AD et al. JAMA Ophthalmology. 1. Tielsch JM, Sommer A, Witt K, Katz J, Royall RM. Blindness and visual impairment in an American urban population: The Baltimore eye survey. Archives of ophthalmology. American Medical Association; 1990;108:286–90. 2. Munoz B, West SK, Rubin GS, Schein OD, Quigley HA, Bressler SB, et al. Causes of blindness and visual impairment in a population of older americans: The Salisbury eye evaluation study. Archives of ophthalmology. American Medical Association; 2000;118:819–25. 3. Kuhn M, Johnson K. Applied Predictive Modeling. Springer Science & Business Media; 2013. 595 p. © 2021 Flaxman AD et al. JAMA Ophthalmology. eFigure 1: Crude prevalence rate of blindness increases with age for all race and ethnic groups, starting around age 60 years. © 2021 Flaxman AD et al. JAMA Ophthalmology. eTable 1: Estimated prevalence count of people living with visual acuity loss or blindness, stratified by state, as well as prevalence rates (in percent). Prevalence Count Prevalence Rate (%) th th th th Mean 2.5 percentile 97.5 percentile Mean 2.5 percentile 97.5 percentile AK 12,300 10,600 14,400 1.66 1.43 1.95 AL 131,200 112,400 151,100 2.69 2.31 3.10 AR 160,300 135,000 184,300 2.28 1.92 2.63 AZ 89,300 75,800 104,100 2.97 2.52 3.46 CA 810,900 709,900 935,100 2.05 1.80 2.37 CO 99,000 84,200 115,600 1.76 1.50 2.06 CT 62,300 51,500 73,000 1.74 1.44 2.03 DC 22,400 19,400 25,300 3.22 2.80 3.65 DE 18,100 15,400 21,300 1.88 1.60 2.21 FL 558,600 475,700 641,500 2.66 2.27 3.06 GA 221,300 190,900 255,700 2.12 1.83 2.45 HI 30,200 24,500 35,600 2.12 1.71 2.50 IA 48,000 38,800 56,400 1.52 1.24 1.79 ID 37,000 30,600 43,500 2.15 1.78 2.53 IL 244,900 211,600 282,000 1.91 1.65 2.20 IN 141,900 120,400 166,400 2.13 1.81 2.50 © 2021 Flaxman AD et al. JAMA Ophthalmology. KS 63,000 52,600 73,200 2.16 1.80 2.51 KY 125,300 103,600 145,200 2.81 2.33 3.26 LA 123,000 104,600 141,100 2.63 2.23 3.01 MA 129,600 107,200 152,400 1.89 1.56 2.22 MD 102,800 86,900 118,900 1.70 1.44 1.97 ME 18,000 13,600 22,100 1.35 1.02 1.65 MI 200,700 168,100 234,800 2.02 1.69 2.36 MN 92,500 74,900 108,600 1.66 1.34 1.95 MO 138,000 115,500 161,200 2.26 1.89 2.64 MS 98,300 84,100 113,200 3.29 2.82 3.79 MT 19,100 15,500 23,100 1.82 1.47 2.20 NC 230,800 194,000 265,300 2.25 1.89 2.58 ND 12,100 10,000 14,500 1.60 1.32 1.92 NE 40,600 34,400 48,200 2.12 1.79 2.51 NH 23,500 18,900 28,000 1.75 1.41 2.09 NJ 169,000 143,300 192,600 1.88 1.59 2.14 NM 62,300 54,000 71,000 2.98 2.58 3.40 NV 80,300 69,300 91,700 2.68 2.31 3.06 NY 393,200 337,500 449,900 1.98 1.70 2.27 OH 251,200 214,200 296,700 2.15 1.84 2.54 © 2021 Flaxman AD et al. JAMA Ophthalmology. OK 116,400 99,900 135,200 2.96 2.54 3.44 OR 78,900 65,500 93,300 1.90 1.58 2.25 PA 302,600 250,500 352,500 2.36 1.96 2.75 RI 21,500 17,900 25,700 2.03 1.69 2.42 SC 129,800 111,100 149,800 2.58 2.21 2.98 SD 16,000 13,000 18,700 1.84 1.50 2.15 TN 172,300 143,100 199,400 2.57 2.13 2.97 TX 634,000 551,100 726,300 2.24 1.95 2.57 UT 43,000 35,900 50,000 1.39 1.16 1.61 VA 168,100 143,100 191,300 1.98 1.69 2.26 VT 13,900 11,500 16,500 2.23 1.85 2.65 WA 144,200 118,700 165,400 1.95 1.60 2.23 WI 101,600 83,500 119,200 1.75 1.44 2.06 WV 65,200 53,200 77,400 3.59 2.93 4.26 WY 10,200 8,400 12,100 1.77 1.45 2.08 © 2021 Flaxman AD et al. JAMA Ophthalmology. eTable 2: Estimated prevalence count of people living with blindness, stratified by location, as well as prevalence rates (in percent). Prevalence Count Prevalence Rate (%) th th th th Mean 2.5 percentile 97.5 percentile Mean 2.5 percentile 97.5 percentile AK 1,600 1,200 2,100 0.22 0.17 0.28 AL 21,500 16,400 27,600 0.44 0.34 0.57 AR 22,400 16,700 29,300 0.32 0.24 0.42 AZ 15,300 11,200 19,300 0.51 0.37 0.64 CA 106,200 83,600 131,000 0.27 0.21 0.33 CO 14,000 9,900 17,600 0.25 0.18 0.31 CT 10,200 7,300 13,600 0.28 0.20 0.38 DC 3,700 2,800 4,600 0.53 0.40 0.66 DE 2,900 2,100 3,600 0.30 0.22 0.38 FL 84,800 64,600 108,800 0.40 0.31 0.52 GA 33,300 25,500 41,400 0.32 0.24 0.40 HI 3,800 2,500 5,100 0.27 0.18 0.36 IA 8,100 5,300 10,900 0.26 0.17 0.35 ID 5,600 4,000 7,200 0.33 0.23 0.42 IL 38,200 27,900 48,700 0.30 0.22 0.38 IN 23,400 16,800 30,500 0.35 0.25 0.46 © 2021 Flaxman AD et al. JAMA Ophthalmology. KS 10,600 7,600 14,100 0.37 0.26 0.48 KY 21,300 15,900 27,700 0.48 0.36 0.62 LA 20,100 15,100 24,900 0.43 0.32 0.53 MA 20,600 14,400 27,100 0.30 0.21 0.40 MD 15,700 12,000 20,300 0.26 0.20 0.34 ME 3,000 2,000 4,000 0.22 0.15 0.30 MI 32,300 23,500 41,900 0.32 0.24 0.42 MN 15,400 10,300 20,300 0.28 0.18 0.36 MO 23,200 16,000 30,400 0.38 0.26 0.50 MS 16,700 13,100 20,800 0.56 0.44 0.70 MT 3,000 2,100 4,000 0.29 0.20 0.38 NC 37,000 27,800 46,400 0.36 0.27 0.45 ND 2,100 1,400 2,900 0.28 0.19 0.38 NE 7,000 5,000 9,200 0.36 0.26 0.48 NH 3,900 2,600 5,200 0.29 0.19 0.39 NJ 25,400 19,000 31,900 0.28 0.21 0.35 NM 8,300 6,500 10,600 0.40 0.31 0.51 NV 10,900 8,200 13,400 0.36 0.28 0.45 NY 59,800 45,800 75,900 0.30 0.23 0.38 OH 42,500 31,500 55,600 0.36 0.27 0.48 © 2021 Flaxman AD et al. JAMA Ophthalmology. OK 18,700 13,900 24,100 0.48 0.35 0.61 OR 11,800 8,300 15,100 0.28 0.20 0.36 PA 52,300 35,700 68,500 0.41 0.28 0.54 RI 3,400 2,400 4,500 0.33 0.22 0.43 SC 20,900 15,900 26,300 0.42 0.32 0.52 SD 2,700 1,900 3,600 0.31 0.22 0.42 TN 28,500 21,400 35,800 0.42 0.32 0.53 TX 86,100 66,200 106,200 0.30 0.23 0.38 UT 5,900 4,500 7,700 0.19 0.14 0.25 VA 25,800 19,700 32,700 0.30 0.23 0.39 VT 2,400 1,600 3,100 0.38 0.26 0.49 WA 21,300 16,000 27,200 0.29 0.22 0.37 WI 16,600 11,800 21,800 0.29 0.20 0.38 WV 11,700 8,300 15,100 0.65 0.46 0.83 WY 1,600 1,100 2,000 0.27 0.20 0.35 © 2021 Flaxman AD et al. JAMA Ophthalmology. P http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png JAMA Ophthalmology American Medical Association

Loading next page...
 
/lp/american-medical-association/prevalence-of-visual-acuity-loss-or-blindness-in-the-us-52B4YpbN06
Publisher
American Medical Association
Copyright
Copyright 2021 Flaxman AD et al. JAMA Ophthalmology.
ISSN
2168-6165
eISSN
2168-6173
DOI
10.1001/jamaophthalmol.2021.0527
Publisher site
See Article on Publisher Site

Abstract

Research JAMA Ophthalmology | Original Investigation A Bayesian Meta-analysis Abraham D. Flaxman, PhD; John S. Wittenborn, BS; Toshana Robalik, BS; Rohit Gulia, MS; Robert B. Gerzoff, MS; Elizabeth A. Lundeen, PhD, MPH; Jinan Saaddine, MD, MPH; David B. Rein, PhD, MPA; for the Vision and Eye Health Surveillance System study group Invited Commentary page 723 IMPORTANCE Globally, more than 250 million people live with visual acuity loss or blindness, Multimedia and people in the US fear losing vision more than memory, hearing, or speech. But it appears Supplemental content there are no recent empirical estimates of visual acuity loss or blindness for the US. CME Quiz at OBJECTIVE To produce estimates of visual acuity loss and blindness by age, sex, jamacmelookup.com race/ethnicity, and US state. DATA SOURCES Data from the American Community Survey (2017), National Health and Nutrition Examination Survey (1999-2008), and National Survey of Children’s Health (2017), as well as population-based studies (2000-2013), were included. STUDY SELECTION All relevant data from the US Centers for Disease Control and Prevention’s Vision and Eye Health Surveillance System were included. DATA EXTRACTION AND SYNTHESIS The prevalence of visual acuity loss or blindness was estimated, stratified when possible by factors including US state, age group, sex, race/ethnicity, and community-dwelling or group-quarters status. Data analysis occurred from March 2018 to March 2020. Author Affiliations: Institute for Health Metrics and Evaluation, MAIN OUTCOMES OR MEASURES The prevalence of visual acuity loss (defined as a University of Washington, Seattle (Flaxman, Robalik, Gulia); NORC at best-corrected visual acuity greater than or equal to 0.3 logMAR) and blindness the University of Chicago, Chicago, (defined as a logMAR of 1.0 or greater) in the better-seeing eye. Illinois (Wittenborn, Rein); Applied Statistical Consulting LLC, Atlanta, RESULTS For 2017, this meta-analysis generated an estimated US prevalence of 7.08 Georgia (Gerzoff); Division of (95% uncertainty interval, 6.32-7.89) million people living with visual acuity loss, of whom Diabetes Translation, Vision Health 1.08 (95% uncertainty interval, 0.82-1.30) million people were living with blindness. Of this, Initiative Centers for Disease Control and Prevention, Atlanta, Georgia 1.62 (95% uncertainty interval, 1.32-1.92) million persons with visual acuity loss are younger (Lundeen, Saaddine). than 40 years, and 141 000 (95% uncertainty interval, 95 000-187 000) persons with Group Information: The members blindness are younger than 40 years. of the Vision and Eye Health Surveillance System study group CONCLUSIONS AND RELEVANCE This analysis of all available data with modern methods appear in Supplement 2. produced estimates substantially higher than those previously published. Corresponding Author: Abraham D. Flaxman, PhD, Institute for Health JAMA Ophthalmol. 2021;139(7):717-723. doi:10.1001/jamaophthalmol.2021.0527 Metrics and Evaluation, University of Published online May 13, 2021. Washington, 2301 Fifth Ave, Seattle, WA 98121 (abie@uw.edu). lobally, an estimated 252.6 (95% CI, 111.4-424.5) mil- estimated national and state visual acuity loss or blindness lion people live with best-corrected visual acuity of prevalence for persons ages 40 years and older and arrived at G 20/60 or worse in the better-seeing eye. People in a similar estimate of 4.24 million cases (2.8%). Both of these 3,4 the US fear losing vision more than memory, hearing, or speech, studies are limited, since they excluded persons younger and consider visual acuity loss among the top 4 worst things than 40 years and persons living in group quarters, such as 2 3,4 that could happen to them. No existing estimates appear to nursing homes and prisons. Both studies relied on meta- have used empirical data to estimate geographic differences, analytic summaries of similar selected population-based study created estimates for persons younger than age 40 years, or data, and no other data sources, to estimate prevalence by age accounted for increased prevalence in group quarters. group, sex, and race/ethnicity and then calculated state-level Previous studies have estimated national visual acuity loss estimates by applying these summary estimates to each state’s or blindness prevalence for important age ranges. The Vision population distribution. This method may lead to inaccura- Problems in the United States (VPUS) study estimated uncor- cies because the population-based study data (while of high rectable visual impairment and blindness for persons ages 40 quality) were collected 8 to 36 years in the past from locally years and older to occur in 4.2 million individuals (2.9%) in representative samples using different methods across stud- 3 4 2010. Using similar methods and data for 2015, Varma et al ies. State-specific estimates assumed that the prevalence of jamaophthalmology.com (Reprinted) JAMA Ophthalmology July 2021 Volume 139, Number 7 717 Research Original Investigation Prevalence of Visual Acuity Loss or Blindness in the US visual acuity loss or blindness observed in the population- based studies were invariant across states, with differences be- Key Points tween states resulting only from differences in the included Question How many people in the US are living with population demographics. However, visual acuity loss or blind- uncorrectable visual acuity loss or blindness? ness prevalence may vary substantially across states because Findings This bayesian meta-analysis generated an estimate that, risk factors for visual impairment (such as diabetes, smoking, in 2017, there were 7.08 million people living with visual acuity sun exposure, nutrition, toxins, or injuries), health care ac- loss, of whom 1.08 million were living with blindness. cess (eg, health insurance, access to eye care), social determi- Meaning Per this study, uncorrectable visual acuity loss and nants of health (eg, poverty, occupational hazards), and poli- blindness are even larger drivers of health burden in the US than 5-8 cies (eg, school entry screening) vary widely across states. was previously known. The Centers for Disease Control and Prevention’s Vision and Eye Health Surveillance System (VEHSS) provides infor- (a Snellen score of 20/40 or worse) and blindness as a subset mation on diagnosed national and state-specific visual acu- ity loss or blindness prevalence based on Medicare 100% of that group, consisting of those with a logMAR of 1.0 or fee-for-service data, MarketScan private insurance claims, greater (a Snellen score of 20/200 or worse) in the better- electronic health records from the IRIS Registry (a compre- seeing eye. We first estimated all visual acuity loss of log- hensive ophthalmology eye diseases clinical registry), and MAR 0.3 or greater, and then in a second, separate calcula- self-reported response data regarding visual difficulty or tion, estimated blindness (logMAR ≥1.0). blindness from 4 national surveys (the America Community Survey [ACS], National Survey of Children’s Health [NSCH], Data Behavioral Risk Factor Surveillance System, and Na- Our model used 4 data sources: (1) data abstracted from PBS, tional Health Interview Survey) and self-reported and (2) National Health and Nutrition Examination Survey examination-based data from the National Health and (NHANES) data collected during 1999 to 2008 (the only Nutrition Examination Survey (NHANES). These sources years in which vision data were collected), (3) ACS data col- yield divergent prevalence estimates based on differences in lected in 2017, and (4) NSCH data collected in 2016. For PBS, case definitions and persons included in the data. While the we searched online sources for studies published after 1991 VEHSS provides estimates from each of these sources, to our that were representative of the target population from which knowledge, no attempt has been made to summarize these the participants were sampled, presented primary results or data into a single meta-analytic national estimate. meta-analysis of primary data, and reported age-specific, We selected all relevant VEHSS data to create new na- race/ethnicity–specific, and/or location-specific prevalence tional and state estimates of US blindness and visual acuity estimates. We identified 5 such studies for inclusion from loss for all ages for the year 2017. We used bayesian meta- (1) the Baltimore Pediatric Eye Disease Study (data collection regression to combine all relevant information from the ACS period, 2003-2007; publication date, 2008) ; (2) the (for state-to-state variation, the oldest age groups, and preva- Chinese American Eye Study (data collection period, 2010- lence in group quarters), NHANES (a primary source of infor- 2013; publication date, 2016) ; (3) the Eye Diseases Preva- mation for mean tendency, age stratification, sex, and race/ lence Research Group (EDPRG; a meta-analysis of several ethnicity variation), the NSCH (for individuals of the youngest earlier PBS; data collection period, 1985-1998; publication ages), and population-based studies (PBS), and summarized date, 2004) ; (4) the Los Angeles Latino Eye Study data col- results by point estimates and uncertainty intervals (UIs). lection period, 2000-2003; publication date, 2004) ; and (5) the Multi-Ethnic Study of Atherosclerosis Cohort (data collection period, 2000-2004; publication date, 2015). We abstracted estimated prevalence of dichotomous mea- Methods sures of visual impairment and blindness and sample size information from each study by age group, sex, and race/ Ethical Review These research activities were deemed to be not human ethnicity. Of these sources, all but the EDPRG reported pri- subjects research by the institutional review board of NORC mary data on best-corrected visual acuity, as measured by at the University of Chicago because they are based exclu- study ophthalmologists. sively on secondary analysis of existing, deidentified data For NHANES participants aged 12 years or older, we used sources. For this reason, informed consent was not required. eye examination–derived measurements of best-corrected vi- sual acuity, as measured among persons with presenting vi- Strategy sual acuity of 20/40 or worse using the Auto Refractor model We applied bayesian meta-regression methods to multiple ARK-760 (Nidek) instrument and collected as part of a visual data sources with the goal of producing estimates of the health module that was fielded from 1999 to 2008 from a na- prevalence and uncertainty interval of visual acuity loss or tionally representative sample of US individuals dwelling in blindness, stratified by age group, sex, race/ethnicity, and communities. Among those with measurements, best- state (50 US states and Washington, DC) for the year 2017. corrected visual acuity was missing for 11.51%. As described We defined visual acuity loss using US standards as a best- in the eMethods in Supplement 1, we imputed missing cat- corrected visual acuity greater than or equal to 0.3 logMAR egorical indicators of visual acuity loss and blindness using 718 JAMA Ophthalmology July 2021 Volume 139, Number 7 (Reprinted) jamaophthalmology.com Prevalence of Visual Acuity Loss or Blindness in the US Original Investigation Research Table 1. Estimated Crude Prevalence Count and Rate of People Living With Visual Acuity Loss or Blindness, Stratified by Sex and Race/Ethnicity, US, 2017 Prevalence count, millions of people Prevalence rate, % Characteristic Mean 2.5th Percentile 97.5th Percentile Mean 2.5th Percentile 97.5th Percentile Total 7.08 6.32 7.89 2.17 1.94 2.42 Female 4.16 3.62 4.69 2.52 2.19 2.84 Male 2.92 2.53 3.37 1.82 1.57 2.10 Non-Hispanic Black 1.02 0.87 1.18 2.55 2.17 2.94 White 4.27 3.68 4.87 2.16 1.86 2.47 Hispanic 1.26 1.07 1.47 2.15 1.83 2.50 Other 0.52 0.41 0.62 1.76 1.40 2.12 multiple imputation with chained equations with boot- parameters to estimate variation in prevalence as a function strapped resampling. of age, sex, race/ethnicity, and data source and assumes that TheNSCHisanationallyrepresentativesurveyofthephysi- the age-stratified prevalence rate is not changing substan- cal and emotional health of children aged 0 to 17 years that con- tially over time. Full details are provided in the eMethods tains a caregiver-reported assessment of visual difficulty, which in Supplement 1. Data analysis occurred from March 2018 reads, “Does this child have blindness or problems with seeing, to March 2020. even when wearing glasses?” The ACS is an annual nation- ally representative and state-representative survey con- ducted by the US Census Bureau to provide information on Results demographic, social, economic, and housing characteristics of the US population. Like NSCH, ACS includes a head-of- Our data abstracted from PBS consisted of 103 measure- household–reported assessment of visual difficulty, which ments of visual acuity loss and 43 measurements of blind- reads, “Is this person blind or does he/she have serious diffi- ness. The surveys used included 35 466 individuals from the culty seeing even when wearing glasses?” and for which the NHANES, 3 190 040 individuals from the ACS, and 50 212 respondent reports for all members of the household. The ACS individuals from the NSCH. also includes information on group-quartered residences, al- lowing questions to be analyzed for those in nursing homes, Visual Acuity Loss or Blindness prisons, and other institutional group quarters, separately from We estimated a US prevalence count of 7.08 (95% UI, 6.32- residents in community-dwelling households. 7.89) million people living with visual acuity loss or blind- ness (using the US standard of best-corrected visual acuity in the worse-seeing eye of a Snellen score of 20/40 or worse) in Estimation We developed 2 statistical models to assess (1) the prevalence the US in 2017, corresponding to a crude prevalence rate of rate of all visual acuity loss stratified by age group, sex, race/ 2.17% (95% UI, 1.94%-2.42%) (Table 1). The national preva- ethnicity, group-quarters status, and US state and (2) the preva- lence level of visual acuity loss or blindness increased as a func- lence rate of blindness at the same levels of stratification. The tion of age, from 0.74% (95% UI, 0.37%-1.10%) among per- model estimated the dependent variable, observed preva- sons younger than 12 years to 0.99% (95% UI, 0.80%-1.18%) lence in each stratification category, as a negative, binomially among individuals aged 50 to 54 years and 20.73% (95% UI, distributed function of the number of persons evaluated in the 17.71%-23.27%) among persons aged 85 years and older sample and independent variables measuring sex, age, race/ (Figure 1). ethnicity (non-Hispanic Black, non-Hispanic White, His- Our meta-regression also estimated that 358 000 (95% panic, and other), US state, and source of data. We applied the UI, 263 000-472 000) persons with visual acuity loss or integrative systems modeling approach developed in the Global blindness reside in group quarters, such as nursing homes Burden of Disease Study to create these estimates. Follow- and prisons. This constitutes 5.06% (95% UI, 3.78%-6.60%) ing King, our integrative systems modeling reduces to an of all persons with visual acuity loss or blindness. extension of negative binomial regression, with a piecewise We estimated 1.62 (95% UI, 1.32-1.92) million persons with linear spline to represent the nonlinear age pattern and an age- visual acuity loss or blindness are younger than 40 years. This standardizing likelihood to account for the heterogeneous re- constitutes 22.89% of all persons with visual acuity loss porting of age groups in examination study data. This al- or blindness. lowed us to include data from all 5 PBS as well as the NHANES, Crude prevalence rates of visual acuity loss or blindness NSCH, and ACS in the likelihood during parameter estima- ranged from 1.35% (95% UI, 1.02%-1.65%) in Maine to 3.59% tion. We used the DisMod-MR 1.1.1, which implements this (95% UI, 2.93%-4.26%) in West Virginia. State differences per- model in Python version 3.6 using PyMC 2, and fit the model sisted after standardization by age, sex, and race/ethnicity with 400 000 iterations of Markov chain Monte Carlo using (Figure 2). Estimated counts and prevalence rates for each US an adaptive metropolis step method. The model includes state are provided in eTable 1 in Supplement 1. jamaophthalmology.com (Reprinted) JAMA Ophthalmology July 2021 Volume 139, Number 7 719 Research Original Investigation Prevalence of Visual Acuity Loss or Blindness in the US Figure 1. Crude Prevalence of Visual Acuity Loss or Blindness by Age for All Racial/Ethnic Groups National level Hispanic Non-Hispanic Black individuals Non-Hispanic White individuals People of other races/ethnicities <12 12-17 18-24 25-29 30-34 35-39 40-44 45-49 50-54 55-59 60-64 65-69 70-74 75-79 80-84 ≥85 Age, y Increases appear starting at approximately age 60 years. ranged from 0.19% (95% UI, 0.14%-0.25%) in Utah to 0.65% Figure 2. Age-Standardized, Sex-Standardized, and Race/Ethnicity– (95% UI, 0.46%-0.83%) in West Virginia (eTable 2 in Supple- Standardized Visual Acuity Loss or Blindness Prevalence by State ment 1; Figure 3). 1.5 2.0 2.5 3.0 3.5 Standardized prevalence rate Discussion Our estimated number of cases of visual acuity loss or blind- ness is 68.7% higher than the previous estimate created by the VPUS study, but our estimate of blindness alone is lower. Al- though the VPUS study reported findings among people 40 years and older based on different source data, this increase in estimated visual acuity loss or blindness is largely the re- sult of our inclusion of the NHANES data in the model and our choice to use imputation instead of listwise deletion to ad- dress the missing NHANES data. We estimated higher preva- lence for Hispanic and Black individuals compared with White individuals and for women compared with men; however, at least some of these estimates are very uncertain, with a pos- terior probability distribution that crosses zero. These results Blindness are consistent with previous analyses of NHANES data, which As a subset of persons with visual acuity loss or blindness, also found a higher risk of visual acuity loss among Hispanic we estimated a 2017 US prevalence count of 1.08 (95% UI, and Black individuals compared with White individuals and 0.82-1.30) million people living with blindness, defined in women compared with men but were not able to conclude as a best-corrected visual acuity of 1.0 logMAR or greater that these higher risks were statistically significant. Other im- (corresponding to a Snellen score of 20/200 or greater) in the portant differences between our estimate and the VPUS esti- better-seeing eye. This is equal to a crude prevalence rate mate include (1) using more recent population-based study of 0.33% (95% UI, 0.25%-0.40%) (Table 2). The crude preva- data; (2) using 2017 population structure for age, sex, race/ lence rate of blindness increased substantially as a function ethnicity, and household or group quarters size; (3) account- of age, from 0.05% (95% UI, 0.02%-0.08%) among persons ing for differences in prevalence in populations in community- 12 years and younger to 0.11% (95% UI, 0.08%-0.15%) among dwelling households vs group quarters; and (4) accounting individuals aged 50 to 54 years and 5.50% (95% UI, for variations across states. 3.70%-7.30%) among persons 85 years and older (eFigure in Supplement 1). We estimated that 130 000 (95% UI, 57 000-223 000) Limitations Our analyses were limited by at least 5 factors. First, the people with blindness are living in group quarters, such as nurs- ing homes and prisons. This constitutes 11.85% (95% UI, 5.52%- NHANES data had a substantial amount (approximately 12%) of missing autorefractor examination data. Our method of ac- 18.76%) of all people living with blindness. We estimated 141 000 (95% UI, 95 000-187 000) persons with blindness are counting for missing data, multiple imputations by chained equation, resulted in a substantially higher estimate of the younger than 40 years, which constitutes 13.09% of all per- sons with blindness. Crude prevalence rates of blindness prevalence rate of visual acuity loss (2.1%) than is obtained 720 JAMA Ophthalmology July 2021 Volume 139, Number 7 (Reprinted) jamaophthalmology.com Vision loss and blindness prevalence, % Prevalence of Visual Acuity Loss or Blindness in the US Original Investigation Research Table 2. Estimated Prevalence Count of People Living With Blindness, Stratified by Sex and Race/Ethnicity, as Well as Prevalence Rates Prevalence count, millions of people Prevalence rate, % Characteristic Mean 2.5th Percentile 97.5th Percentile Mean 2.5th Percentile 97.5th Percentile Total 1.08 0.82 1.30 0.33 0.25 0.40 Female 0.64 0.48 0.79 0.38 0.29 0.48 Male 0.45 0.34 0.55 0.28 0.21 0.35 Non-Hispanic Black 0.17 0.13 0.21 0.42 0.32 0.53 White 0.74 0.56 0.92 0.37 0.28 0.47 Hispanic 0.12 0.09 0.16 0.21 0.16 0.27 Other 0.05 0.04 0.08 0.19 0.13 0.26 using the same data and listwise deletion (1.7%). While we Figure 3. Age-Standardized, Sex-Standardized, and Race/Ethnicity– believe multiple imputations by chained equation is the Standardized Blindness Prevalence Estimates by State superior method to handle missing data because it uses the strength of other information to inform the estimates, 0.2 0.3 0.4 0.5 0.6 less missing data in NHANES would have resulted in more pre- Standardized prevalence rate cise estimates. Second, our estimates may be limited by the age of some of the included data sets. The NHANES data were collected from 1999 to 2008, and data from some of the population- based examination studies included in the EDPRG meta- analyses were collected even prior to that. However, our model also included more recent PBS published after the EDPRG meta- analysis, as well as the 2016 NSCH and the 2017 ACS. Addi- tionally, time-trend analyses of the ACS did not indicate sys- tematic differences in age-stratified, sex-stratified, or race/ ethnicity–stratified vision prevalence between the years 2008 and 2017 (not shown). Third, we used survey respondent–reported values from the ACS to account for differences in visual acuity loss preva- lence at the state level and within group quarters and from the The Medicare Minimum Data Set is generated as part of a clinical assessment of all residents in Medicare-certified or NSCH for children. Since these values are not based on an ex- amination, they likely contain false-positive results at least for Medicaid-certified nursing home, and includes an assess- ment of each resident’s functional capabilities and health uncorrected refractive error. Our model corrects for system- atically higher prevalence in self-reported visual difficulty mea- needs. However, it does not collect data on visual difficulty sures. However, to estimate prevalence variation by state, in a format that we were able to integrate into our model. household status, and childhood ages, our model assumes that The Baltimore Nursing Home Eye Survey found that 47% examination data on best-corrected visual acuity, if it were col- of people living with visual acuity loss in nursing homes lected, would vary following the same pattern as these self- were blind (compared with our finding of 14.5%), and if we reported data. Furthermore, because the ACS data included generalize this 47% to the entire group-quarters population, only a single measure of severe visual difficulty or blindness, we expect an additional 118 000 (95% UI, 87 000-156 000) our model assumes that state variation is the same for both vi- people living with blindness. Finally, we estimated visual acuity loss or blindness inde- sual acuity loss and blindness together as for blindness alone and by household status. We believe this assumption is both pendently, despite the logical interdependency that every per- son living with blindness is, by definition, a person living with reasonable and currently necessary to create data-driven es- timates for state, residents in group quarters, and children, but visual acuity loss. A more complex model that estimated the 2 outcomes simultaneously could perhaps make more effi- we acknowledge that additional examination data within these strata would improve the quality of future estimates. cient use of the sparse data available and eliminate the illogi- cal possibility of estimating more people living with blind- Fourth, we have assumed that the decomposition of visual acuity loss in distinct subcategories of visual impair- ness than living with visual acuity loss. However, the model structure presented here resulted in no instances in which the ment and blindness follow the same percentage breakdown in group quarters as in households. Although it seems plau- estimated rate of blindness exceeded the estimated rate of vi- sual acuity loss at any level of stratification. sible that the fraction of visual acuity loss that is blindness is higher in group quarters than in the household population, There are also several data sources in the VEHSS that we we found no reliable, representative data source to test this were not able to include in our analysis. Medicare and hypothesis or quantify the magnitude of the difference. MarketScan claims data and IRIS registry data are both jamaophthalmology.com (Reprinted) JAMA Ophthalmology July 2021 Volume 139, Number 7 721 Research Original Investigation Prevalence of Visual Acuity Loss or Blindness in the US appealing big data sources, but we were not able to adjust for the nonrepresentative nature of the populations repre- Conclusions sented by these sources, and there is a lack of evidence on the validity of diagnosed vision loss as a measure of Visual acuity loss and blindness continue to be a substantial population-level vision. The Behavioral Risk Factor Surveil- burden to the US population, and our new analysis indicates lance System and National Health Interview Survey both that the issue is even more substantial than has previously been include respondent-reported data similar to ACS, which recognized. Efforts to collect new examination-based infor- we excluded on our assessment that the ACS sample mation on best-corrected visual acuity in the better-seeing size is substantially larger and the potential biases of eye would enhance future efforts to create more precise na- respondent-reported visual acuity loss would be com- tional and state estimates of visual acuity loss or blindness, pounded by bringing together sources from multiple instru- and this evidence base could be valuable for targeted efforts ments and surveys. to prevent or treat these conditions. ARTICLE INFORMATION Disease Control and Prevention’s scientific www.cdc.gov/visionhealth/vehss/ clearance process. index.html. Accepted for Publication: January 26, 2021. Group Information: The Vision and Eye Health 10. Flaxman AD, Vos DT, Murray CJ. An Integrative Published Online: May 13, 2021. Surveillance System Study Group members are Metaregression Framework for Descriptive doi:10.1001/jamaophthalmol.2021.0527 listed in Supplement 2. Epidemiology. University of Washington Press; 2015. Open Access: This is an open access article Disclaimer: The findings and conclusions in this 11. Dougherty M, Wittenborn J, Phillips E, distributed under the terms of the CC-BY License. report are those of the authors and do not Swenor B. Published examination-based prevalence © 2021 Flaxman AD et al. JAMA Ophthalmology. necessarily represent the official position of major eye disorders. Published 2018. Accessed Author Contributions: Dr Flaxman had full access of the US Centers for Disease Control and March 3, 2021. https://www.norc.org/PDFs/VEHSS/ to all of the data in the study and takes Prevention. EyeConditionExamLiteratureReviewVEHSS.pdf responsibility for the integrity of the data and the 12. Friedman DS, Repka MX, Katz J, et al. accuracy of the data analysis. REFERENCES Prevalence of decreased visual acuity among Concept and design: Flaxman, Wittenborn, Gulia, 1. Bourne RRA, Flaxman SR, Braithwaite T, et al; preschool-aged children in an American urban Saaddine, Rein. Vision Loss Expert Group. Magnitude, temporal population. Ophthalmology. 2008;115(10):1786-1795. Acquisition, analysis, or interpretation of data: trends, and projections of the global prevalence of doi:10.1016/j.ophtha.2008.04.006 Flaxman, Wittenborn, Robalik, Gulia, Gerzoff, blindness and distance and near vision impairment. Lundeen, Rein. 13. Varma R, Kim JS, Burkemper BS, et al; Lancet Glob Health. 2017;5(9):e888-e897. doi:10. Drafting of the manuscript: Flaxman, Wittenborn, Chinese American Eye Study Group. Prevalence and 1016/S2214-109X(17)30293-0 Robalik, Gulia, Gerzoff, Rein. causes of visual impairment and blindness in Critical revision of the manuscript for important 2. Scott AW, Bressler NM, Ffolkes S, Wittenborn JS, Chinese American adults. JAMA Ophthalmol. 2016; intellectual content: Flaxman, Wittenborn, Gulia, Jorkasky J. Public attitudes about eye and vision 134(7):785-793. doi:10.1001/jamaophthalmol. Lundeen, Saaddine, Rein. health. JAMA Ophthalmol. 2016;134(10):1111-1118. 2016.1261 Statistical analysis: Flaxman, Wittenborn, Robalik, doi:10.1001/jamaophthalmol.2016.2627 14. Congdon N, O’Colmain B, Klaver CC, et al; Eye Gulia, Gerzoff. 3. Prevent Blindness America. Vision problems in Diseases Prevalence Research Group. Causes and Obtained funding: Wittenborn, Rein. the US: prevalence of adult vision impairment and prevalence of visual impairment among adults in Administrative, technical, or material support: age-related eye disease in America. Published 2012. the United States. Arch Ophthalmol. 2004;122(4): Wittenborn, Gulia, Lundeen, Rein. Accessed September 30, 2019. http://www. 477-485. doi:10.1001/archopht.122.4.477 Supervision: Wittenborn, Lundeen, Saaddine, Rein. visionproblemsus.org/ 15. Varma R, Ying-Lai M, Klein R, Azen SP; Conflict of Interest Disclosures: Dr Flaxman 4. Varma R, Vajaranant TS, Burkemper B, et al. Los Angeles Latino Eye Study Group. Prevalence reported grants from NORC as a subcontract from a Visual impairment and blindness in adults in the and risk indicators of visual impairment and US Centers for Disease Control and Prevention United States. JAMA Ophthalmol. 2016;134(7): blindness in Latinos. Ophthalmology. 2004;111(6): during the conduct of the study; payments from 802-809. doi:10.1001/jamaophthalmol.2016.1284 1132-1140. doi:10.1016/j.ophtha.2004.02.002 Janssen, IHME, SwissRe, and Merck for Mothers for 5. Elam AR, Lee PP. High-risk populations for vision 16. Fisher DE, Shrager S, Shea SJ, et al. Visual assistance in analysis and interpretation of licensed loss and eye care underutilization. Surv Ophthalmol. impairment in White, Chinese, Black, and Hispanic data previously produced by IHME; and payment 2013;58(4):348-358. doi:10.1016/j.survophthal.2012. participants from the multi-ethnic study of from the startup Agathos Ltd for advising outside 07.005 atherosclerosis cohort. Ophthalmic Epidemiol. the submitted work. Dr Rein reported grants from 2015;22(5):321-332. doi:10.3109/09286586.2015. the US Centers for Disease Control and Prevention 6. Armstrong RA, Mousavi M. Overview of risk Vision Health Initiative during the conduct of the factors for age-related macular degeneration study. No other disclosures were reported. (AMD). J Stem Cells. 2015;10(3):171-191. 17. National Health and Nutrition Examination Survey. 2007-2008 data documentation, Funding/Support: This study was supported by 7. Pleet A, Sulewski M, Salowe RJ, et al. Risk factors codebook, and frequencies. Published 2009. funding from the US Centers for Disease Control associated with progression to blindness from Accessed October 2, 2019. https://wwwn.cdc.gov/ and Prevention Vision Health Initiative (cooperative primary open-angle glaucoma in an Nchs/Nhanes/2007-2008/VIX_E.html agreement U01DP006074, “Establish a Vision and African-American population. Ophthalmic Epidemiol. Eye Health Surveillance System for the Nation”). 2016;23(4):248-256. doi:10.1080/09286586.2016. 18. Raghunathan TE, Lepkowski JM, Van Hoewyk J, 1193207 Solenberger P. A multivariate technique for multiply Role of the Funder/Sponsor: As coauthors, imputing missing values using a sequence of employees of the funding organization 8. Penman A, Hancock H, Papavasileiou E, et al. regression models. Survey Methodology. 2001;27 (Drs Lundeen and Saaddine) participated in the Risk factors for proliferative diabetic retinopathy in (1):85-96. design and conduct of the study; collection, African Americans with type 2 diabetes. Ophthalmic management, analysis, and interpretation of the Epidemiol. 2016;23(2):88-93. doi:10.3109/ 19. US Census Bureau. National Survey of data; preparation, review, or approval of the 09286586.2015.1119287 Children’s Health (NSCH). Published 2019. manuscript; and decision to submit the manuscript Accessed October 3, 2019. https://www.census. 9. Centers for Disease Control and Prevention. for publication. Additionally, the manuscript was gov/programs-surveys/nsch.html The Vision and Eye Health Surveillance System. reviewed and approved by the US Centers for Published 2019. Accessed October 1, 2019. https:// 20. US Census Bureau. American Community Survey (ACS). Published 2019. Accessed October 3, 722 JAMA Ophthalmology July 2021 Volume 139, Number 7 (Reprinted) jamaophthalmology.com Prevalence of Visual Acuity Loss or Blindness in the US Original Investigation Research 2019. https://www.census.gov/programs-surveys/ 23. Ko F, Vitale S, Chou C-F, Cotch MF, Saaddine J, Baltimore. N Engl J Med. 1995;332(18):1205-1209. acs Friedman DS. Prevalence of nonrefractive visual doi:10.1056/NEJM199505043321806 impairment in US adults and associated risk factors, 21. King G. Unifying Political Methodology: the 1999-2002 and 2005-2008. JAMA. 2012;308(22): Likelihood Theory of Statistical Inference. University 2361-2368. doi:10.1001/jama.2012.85685 of Michigan Press; 1998. doi:10.3998/mpub.23784 24. Tielsch JM, Javitt JC, Coleman A, Katz J, 22. Patil A, Huard D, Fonnesbeck CJ. PyMC: Sommer A. The prevalence of blindness and visual Bayesian stochastic modelling in Python. J Stat Softw. impairment among nursing home residents in 2010;35(4):1-81. doi:10.18637/jss.v035.i04 Invited Commentary Updated Numbers on the State of Visual Acuity Loss and Blindness in the US Emily Y. Chew, MD In this issue of JAMA Ophthalmology,Flaxmanetal Other Results reported the results of their study designed to estimate the The overall prevalence of visual acuity loss in the US was 7.08 rates of visual acuity loss and blindness in the US, including (95% uncertainty interval [UI], 6.32-7.89) million, translating 2,3 1 rates for each individual state. Prior published reports in to a crude prevalence rate of 2.17% (95% UI, 1.94%-2.42%). 2010 and 2015 addressing Not surprisingly, the prevalence of visual acuity loss or blind- this issue used different data ness increased with age, from 0.74% (95% UI, 0.37%-1.10%) Related article page 717 sets and statistical method- for those younger than 12 years, to 0.99% (95% UI, 0.80%- ology, resulting in rates for those 40 years and older only. 1.18%) among those aged 50 to 54 years and 20.73% (95% UI, 2,3 Both studies suggested the number of individuals older 17.71%-23.27%) among persons 85 years and older. As ex- than 40 years affected with vision impairment was approxi- pected, a range of rates of visual acuity loss was found in dif- mately 4.2 million in the US. The results of the current ferent states. Interestingly, approximately 22.89% of those with study of the population of all ages estimated the rate to be visual acuity loss or blindness were younger than 40 years. much higher, at 7.08 million people living in the US with Similar to previous analyses of the National Health and Nutri- visual acuity loss, defined as 20/40 or worse, with 1.08 mil- tion Examination Survey data, the current study estimated lion of these having blindness, defined as visual acuity of higher prevalence in Black and Hispanic individuals com- 20/200 or worse. These are important data that need to be pared with White individuals and women compared with men. explored further. However, these probabilities of differences between sex and race/ethnicity crossed zero and were not statistically significant. Methods The current study analyzed a greater number of studies, including the classic population-based studies and data Importance of These Data from the US Centers for Disease Control and Prevention’s These data underscore the burden of blindness in the US. Vision and Eye Health Surveillance, which provide data on Visual acuity loss is considered one of the most dreaded events both national and state-specific rates of visual acuity loss that individuals in the US fear compared with loss of speech, using insurance claims, registries of electronic health rec- hearing, or memory. In addition, it is important to obtain ac- ords, and self-reported data from national surveys that curate prevalence and eventually incidence data on visual im- included populations of all ages and were stratified by race/ pairment and blindness, because they have compelling pub- ethnicity and sex. Flaxman et al recognized the importance lic health implications. With this large increase in the numbers of analyzing more granular data found in state-reported of US individuals who will experience visual acuity loss or rates because of the differences found in population demo- blindness, we need to prepare to the health care systems to graphics and risk factors for visual acuity loss, including serve affected individuals. These estimates will also help to comorbidities such as diabetes, health care access, social promote potential screening and public health education for determinants of health, and lifestyle factors, such as nutri- select ocular diseases that have effective therapies that may tion and smoking. be given either as preventive therapy or active treatment to The use of the bayesian meta-analysis methods in this preserve visual acuity. study may be superior in estimating between-study hetero- Although the study results were not statistically signifi- geneity and pooled effects, especially when there is a rela- cant, the trends are similar to other studies that have found tively small number of studies, as found by Flaxman et al. women to have increased burden of blindness. Data suggest- Summarizing using various sources of data in a single ing an increased rates of visual acuity loss in women does el- meta-analysis is the innovative part of this approach. evate the alert level to consider conducting important stud- The bayesian methods also allow the researchers to inte- ies in assessing sex as a biological variable, as promoted by the grate prior knowledge and assumptions when calculating National Institutes of Health. Others have highlighted the the meta-analyses. excessive burden on Black and Hispanic individuals, again jamaophthalmology.com (Reprinted) JAMA Ophthalmology July 2021 Volume 139, Number 7 723 © 2021 American Medical Association. All rights reserved. Supplemental Online Content Flaxman AD, Wittenborn JS, Robalik T, et al . Prevalence of visual acuity loss or blindness in the US: a bayesian meta-analysis. JAMA Ophthalmol. Published , 2021. doi:10.1001/ jamaophthalmol.2021.0527 eMethods. Details, verification, and validation of our methodological approach. eFigure. Crude prevalence rate of blindness increases with age for all race and ethnic groups, starting around age 60 years. eTable 1. Estimated prevalence count of people living with visual acuity loss or blindness, stratified by state, as well as prevalence rates (in percent). eTable 2. Estimated prevalence count of people living with blindness, stratified by location, as well as prevalence rates (in percent). This supplemental material has been provided by the authors to give readers additional information about their work. © 2021 Flaxman AD et al. JAMA Ophthalmology. eMethods: details, verification, and validation of methodological approach In this supplementary appendix, we present additional details on the verification and validation of our methodological approach. To better understand the impact of modeling choices, we examined a nested sequence of intermediate models of increasing complexity. Initially we estimated the numbers of cases of visual acuity loss using only PBS data. We then added NHANES data as reference with PBS data for the estimation, and then merged NSCH data with NHANES data and PBS data in the model and generated estimates of visual acuity loss. We then added ACS data for older ages and group quarters sequentially (denoted ACS- older and ACS-gq in Table 1 below). After that we used ACS data for state-specific prevalence and for year 2017 to generate estimates (denoted ACS-state and ACS-2017 in Table 1 below). In the final model we added interaction terms along with all the datasets to generate visual acuity loss estimates. Table 1: Estimated prevalence counts (in millions) of people living with visual acuity loss or blindness, after each step of the sequence in the model. Data Sources Visual acuity loss among Visual acuity loss in all age 40+ ages (in millions) (in millions) Population Based Study data (Exam) 3.47 [2.56, 4.61] 2.92 [2.12, 3.83] © 2021 Flaxman AD et al. JAMA Ophthalmology. Exam, NHANES 4.28 [3.41, 5.25] 5.80 [4.78, 6.97] Exam, NHANES, NSCH 4.35 [3.50, 5.30] 5.83 [4.81, 6.92] Exam, NHANES, NSCH, ACS-older 4.49 [3.63, 5.47] 5.98 [5.07, 7.08] Exam, NHANES, NSCH, ACS-older, 4.63 [3.85, 5.48] 6.16 [5.29, 7.13] ACS-gc Exam, NHANES, NSCH, ACS-older, 4.57 [4.02, 5.16] 6.11 [5.38, 6.89] ACS-gc, ACS-state Exam, NHANES, NSCH, ACS-older, 5.43 [4.77, 6.16] 7.04 [6.29, 7.92] ACS-gc, ACS-state, ACS-2017 Exam, NHANES, NSCH, ACS-older, 5.46 [4.83, 6.16] 7.08 [6.34, 7.92] ACS-gc, ACS-state, ACS-2017, interaction terms Table 2 presents similar results for the subset of patients who we predicted to be blind. Table 2: Estimated prevalence counts (in millions) of people living with blindness, after each step of the sequence in the model. © 2021 Flaxman AD et al. JAMA Ophthalmology. Data Sources Blindness among age 40+ Blindness in all ages (in millions) (in millions) Population Based Study data (Exam) 0.77 [0.51, 1.15] 0.73 [0.48, 1.10] Exam, NHANES 0.71 [0.45, 1.14] 0.82 [0.56, 1.25] Exam, NHANES, NSCH 0.73 [0.48, 1.07] 0.86 [0.59, 1.20] Exam, NHANES, NSCH, ACS-older 0.69 [0.50, 0.94] 0.82 [0.60, 1.08] Exam, NHANES, NSCH, ACS-older, 0.81 [0.59, 1.09] 0.96 [0.71, 1.24] ACS-gc Exam, NHANES, NSCH, ACS-older, 0.81 [0.62, 1.04] 0.96 [0.74, 1.21] ACS-gc, ACS-state Exam, NHANES, NSCH, ACS-older, 0.91 [0.71, 1.12] 1.06 [0.84, 1.28] ACS-gc, ACS-state, ACS-2017 Exam, NHANES, NSCH, ACS-older, 0.93 [0.74, 1.17] 1.08 [0.85, 1.34] ACS-gc, ACS-state, ACS-2017, interaction terms © 2021 Flaxman AD et al. JAMA Ophthalmology. Readers may notice that our reported results when using only PBS data (row 1 of Tables 1 and 2) are lower than those reported by the Vision Problems in the U.S. (VPUS) study which was also based on population-based studies (3). That study (available at visionproblemsus.org) estimated 4.19 million people with vision impairment and blindness, of whom 1.29 million were blind. Based only on population-based studies, our model estimates 2.92 million people who had visual acuity loss or blindness, of whom approximately 730,000 were blind. This difference is driven by the PBS that were used in our model as compared to the earlier VPUS. The studies we included as PBS are described in the data section above. They can be compared to those used by VPUS by reviewing the VPUS Methods and Sources page. In general, the studies that we included reported a lower prevalence of visual impairment and blindness than the studies included in the VPUS and this lower prevalence is reflected in our model estimate. We believe specific differences between our estimates are attributable to the inclusion of older data from the Baltimore Eye Survey (collected from 1985 and 1988) (1) and the Salisbury Eye Evaluation Project (collected from 1993 and 1995) (2) in VPUS, and our inclusion of data from CHES (collected from 2010-2013), LALES (collected from 2000-2008), and MESA (collected from 2002-2004). MICE-Multiple imputation by chained equation Missing data in meta-analysis can jeopardize inferences from the study and examination data, and possibly bias the estimates. The vision examination data from NHANES is a key input in our meta-analytic estimates of the prevalence of uncorrectable visual acuity loss or blindness in the United States, but when we mapped this data to dichotomous outcomes for visual acuity loss or blindness, we were unable to determine a value for 11.51% of respondents. © 2021 Flaxman AD et al. JAMA Ophthalmology. Multiple imputation by chained equations (MICE) is a standard approach to address missing data. MICE constructs a regression model for each column with missing values, using the variables from the other columns as the predictors. If multiple columns have missing values, the MICE procedure iterates through the columns, fitting each with the previously imputed values for the other columns. Simpler approaches, such as complete-case analysis (also called list-wise deletion) and available-case analysis, may produce biased results. The mean age among individuals with incomplete eye exam was 52.8 years, substantially higher than the mean age among individuals with complete eye exam (38.6 years), and since visual acuity loss or blindness prevalence rates increase markedly as age increases, our imputation method must take age into account. Using equation 1 and 2 below, we imputed missing dichotomous indicators of visual loss and blindness using multiple imputation with chained equations (MICE) with Gaussian perturbations: (1) Visual acuity loss C(age_group) + sex + C(race_eth) + vidrva + vidlva, (2) Blindness C(age_group) + sex + C(race_eth) + vidrva + vidlva, where vidrva and vidlva are presenting visual acuity of the right eye and left eye respectively, C(age_group) respresents a “dummy coding” for age variable grouped in bins (0, 12, 18, 25, 30, 35, 40, 45, 50, 55, 60, 65, 70, 75, 80, 85, 100) and C(race_eth) represents a “dummy coding” for our categorical variable for race/ethnicity. The MICE approach requires some choices: what else should go into the regression formula for the missing variables? Should we use a Gaussian for perturbing the model parameters in the imputation model, or a non-parametric bootstrap? To identify the best © 2021 Flaxman AD et al. JAMA Ophthalmology. regression formula and perturbation method for imputing missing values of visual acuity loss or blindness, we conducted an out-of-sample validation exercise. We considered a range of regression equations for our outcomes of interest (visual acuity loss or blindness): (in the formulas below, the notation (var1 + var2)**2 means the variables var1 and var2 are included as main effects and all pair-wise interactions are also included) 1. outcome ~ age + sex + C(race) (where C(race) is a “dummy” coding of the race/ethnicity variable) 2. outcome ~ C(age_group) + sex + C(race) (where C(age_group) is a “dummy” coding of the age variable grouped in bins [0, 12, 18, 25, 30, 35, 40, 45, 50, 55, 60, 65, 70, 75, 80, 85, 100] ) 3. outcome ~ C(age_group) + sex + C(race) + vidrva (where vidrva is the presenting visual acuity in the right eye) 4. outcome ~ C(age_group) + sex + C(race) + vidlva (where vidlva is the presenting visual acuity in the left eye) 5. outcome ~ C(age_group) + sex + C(race) + vidrva + vidlva 6. outcome ~ C(age_group) + sex + C(race) + (vidrva + vidlva )**2 7. outcome ~ C(age_group) + sex + C(race) + vidrova (where vidrova is the best-corrected visual acuity in the right eye) 8. outcome ~ C(age_group) + sex + C(race) + vidlova (where vidlova is the best-corrected visual acuity in the left eye) 9. outcome ~ C(age_group) + sex + C(race) + vidrova + vidlova 10. outcome ~ C(age_group) + sex + C(race) + (vidrova + vidlova )**2 © 2021 Flaxman AD et al. JAMA Ophthalmology. 11. outcome ~ C(age_group) + sex + C(race) + vidrva + vidlva + vidrova + vidlova 12. outcome ~ C(age_group) + sex + C(race) + (vidrva + vidlva + vidrova + vidlova)**2 We tested each equation for each outcome for two alternative perturbation methods: (a) the Gaussian perturbation method, which samples from a multivariate normal distribution derived from the fit of the regression equation; and (b) the bootstrap perturbation method, which samples from the conditional model fitted to a bootstrapped version of the data set. For our testing approach, we used an out-of-sample cross-validation approach from machine learning, where we withheld the presenting and best corrected visual acuity measurements for a randomly selected subset of data (our “test dataset”) and compared the model predictions for these individuals to the true values of blindness and visual acuity loss.(3) To be precise, 1. First, we selected 25% of the rows of NHANES data and redacted their vidrva, vidlva, vidrova, and vidlova values (which are all numeric values measured for presenting and best corrected visual acuity in the right and left eye; these values are always sufficient to derive the values of the visual acuity loss and blindness variables, and certain patterns of missing values among these measurements lead to missing values of the dichotomous visual acuity loss and blindness variables). For these rows, we also redacted the visual acuity loss or blindness values derived from these measurements. 2. Second, we imputed the masked values using MICE 1,000 times and took the average of these values as a probability prediction of the outcome of interest (visual acuity loss or blindness). © 2021 Flaxman AD et al. JAMA Ophthalmology. 3. Third, we compared those imputed values and actual values using the ROC curve in which the true positive rate (Sensitivity) is plotted as a function of the false positive rate (100- Specificity) for different cut-off points of a parameter. Each point on the ROC curve represents a sensitivity/specificity pair corresponding to a particular decision threshold. We measured the area under the ROC curve (AUC) as a summary metric for performance of the MICE formulation. MICE validation results Missing values were best imputed using the variables age, sex, race, vidrva and vidlva, as shown in Table 3. Table 3: Table showing the AUC score of all the models with different parameters Method / Equation VL VL Blindness Blindness Bootstrap Gaussian Bootstrap Gaussian Age + Sex + C(Race) 70.11 69.74 73.15 72.95 C(Age_group) + Sex + C(Race) 72.21 71.18 72.51 70.39 C(Age_group) + Sex + C(Race) + Vidrva 75.81 75.86 80.14 81.09 C(Age_group) + Sex + C(Race) + Vidlva 74.72 75.66 77.05 83.89 C(Age_group) + Sex + C(Race) + Vidrva + 75.51 74.66 80.75 78.53 Vidlva © 2021 Flaxman AD et al. JAMA Ophthalmology. C(Age_group) + Sex + C(Race) + (Vidrva + 74.98 75.74 78.70 78.93 Vidlva)**2 C(Age_group) + Sex + C(Race) + Vidrova 75.04 74.49 82.95 70.89 C(Age_group) + Sex + C(Race) + Vidlova 75.29 74.81 77.66 79.59 C(Age_group) + Sex + C(Race) + Vidrova 74.19 74.99 75.22 72.04 +Vidlova C(Age_group) + Sex + C(Race) + (Vidrova 74.25 72.43 69.24 75.48 + Vidlova)**@ C(Age_group) + Sex + C(Race) + Vidrva + 74.66 72.97 70.27 67.53 Vidlva + Vidrova + Vidlova C(Age_group) + Sex + C(Race) + (Vidrva + 72.60 71.26 77.00 78.62 Vidlva + Vidrova + Vidlova)**2 Our selected approach to MICE produced out-of-sample AUC of 75.51, compared with AUC of 70.79 in complete-case analysis. Bayesian meta-regression model We represent our model with a stochastic component and a systematic component, where the stochastic component is a negative binomial model of count data: (3) NegativeBinomial , , , © 2021 Flaxman AD et al. JAMA Ophthalmology. where indexes the specific measurement, is the prevalence count of those with visual acuity loss (or blindness) in measurement , is the effective sample size from which the count was is the prevalence rate typically reported in a PBS), is the prevalence taken (and so = rate predicted by the model, is the over-dispersion parameter of the negative binomial distribution (assumed to be the same for all measurements). DisMod-MR 1.1.1 uses the Python PyMC2 package to implement this Bayesian computation, and follows the formulation of the ( ) ( | ) negative binomial model provided by PyMC2, where Pr = , = , ( ) which in terms of the equation above has = , = , and = . In the systematic component of the model, we included fixed effects for sex ( ), age ( for = 0, … , ), race ( for = white, Black, Hispanic, and other races), and data source ( for =EDPRG, CHES, LALES, ACS, BPEDS, MESA, and NSCH; we coded NHANES as the reference category), as well as sex/race and race/age interaction fixed effects. We also include random effects for state (50 states and Washington, D.C.; for = 1, … ,51). Our formulation includes a piece-wise linear spline model on age, with spline knots (knot ,knot ,…) indexed by for = 1, … , , as well as intercept shifts for additional key covariates to account for differential sex ratios and age patterns by race as follows: (4) = exp{ [sex = female]} × + ( knot ) × exp [race = ] + [source = ] + [location = ] + interaction_terms , © 2021 Flaxman AD et al. JAMA Ophthalmology. where sex is the sex measured in measurement ; age and age are the start and end of the age group measured in measurement ; race is the race/ethnicity group measured in measurement ; source is the data source for measurement ; location is the state location of measurement ; and notation [variable = value] represents an indicator function which takes value 1.0 if the variable is equal to the value and 0.0 otherwise and notation (value) represents the value in the parenthesis if it is positive, and takes value zero, otherwise. We used evenly spaced spline knots at ages (0, 20, 40, 60, 80, 100). We also included interaction terms to capture the possibility that the sex and age effects were different by race/ethnicity and group quarters status: interaction_terms = [race = and sex = female] + [race = ](age ) + [race = and age > 50](age 50) + [g. quarters = inst](age ) + [g. quarters = inst][age > 50](age 50), where is the effect coefficient for race interacted with sex (for = 1,…4), is the effect coefficient for race interacted with age, is the effect coefficient for race interacted with age > 50, and age = is the midpoint of the age group measured in measurement . Together and constitute an age-dependent spline for the group quarters population with knots at 0 and 50. We used a Bayesian framework for inference with weakly informative priors for model parameters other than state to assist in regularization, which primarily allowed the data to inform the model estimates. Prior distributions for , , , , , , , and were all set to independent normal distributions with mean 0.0 and © 2021 Flaxman AD et al. JAMA Ophthalmology. standard deviation 1.0. To capture state variation in prevalence, we used informative priors for state random effects, which were informed by the state-to-state variation in age-/sex-/race- standardized endorsement rates of the respondent-reported visual acuity loss question in ACS. This prior took the form (5) Normal , , where, in short, is the log of the ratio of the standardized state prevalence to the national prevalence, and is the standard deviation of the log-ratio of the crude state prevalence to the national prevalence for all strata with at least 500 individuals, and is normal distribution with mean and standard deviation . To be precise, to estimate and , we first estimated the crude prevalence rate of respondent-reported visual acuity loss in the US in 2017 as observed in ACS data. We next estimated the prevalence rate of respondent-reported visual acuity loss for each state from the same data source, stratified by age-group, sex, race/ethnicity, and group quarters status. We then used the state-specific stratified estimates to calculate a standardized prevalence rate, standardizing with the age group, sex, race/ethnicity, and group-quarters weights at the national level, to control for demographic difference between states. We then used these data to estimate the ratio of the standardized state prevalence to the national prevalence. For each state , we set the prior on to be normally distributed with mean = log ratio , where ratio is the ratio of standardized prevalence in state to the national prevalence. We obtained an informative standard deviation for this prior as follows: analogous to the ratio use used for the mean, we constructed a ratio of the national prevalence rate to the state-specific prevalence rate for each strata (stratified by age-group, sex, race/ethnicity, and group quarters status) and calculated the standard deviation of the log-ratio for all strata with at least 500 individuals in the © 2021 Flaxman AD et al. JAMA Ophthalmology. state, age, sex, and race/ethnicity-specific strata. We put a cap on the predicted prevalence at 0.25 to ensure that we are not over estimating our results. Only data from ACS was available to inform the group-quarters effect coefficients, but unlike the state-to-state variation, we used this ACS data in the likelihood instead of constructing an informative prior. To guard against inferring with more confidence than appropriate from the ACS group quarters data, we used aggregate measurements for broad age groups (25-year intervals) and did not stratify them by sex or by race/ethnicity. We also included fixed effects for each age group, which effectively treated each comparison of the prevalence in the institutional group quarters population and the free-living population as a separate study in the meta- regression. We included all other data sources (NHANES, PBS, NSCH) in the likelihood with the assumption that they applied to a nation-level estimate, which we assumed to be constant over time. This includes data from NSCH, NHANES, and BPEDS for children under the age of 18, and also ACS data on free-living adults over the age of 85. Validation and Verification We assessed the cross-model validation of our model by setting its population parameters to those of the VPUS study (persons 40 and older in household settings), estimating the number of cases it generated when using only PBS data, and then sequentially estimating the number of cases it generated using the new data sources and model specifications that we included in our final model. See eMethods supplementary materials for full details. References © 2021 Flaxman AD et al. JAMA Ophthalmology. 1. Tielsch JM, Sommer A, Witt K, Katz J, Royall RM. Blindness and visual impairment in an American urban population: The Baltimore eye survey. Archives of ophthalmology. American Medical Association; 1990;108:286–90. 2. Munoz B, West SK, Rubin GS, Schein OD, Quigley HA, Bressler SB, et al. Causes of blindness and visual impairment in a population of older americans: The Salisbury eye evaluation study. Archives of ophthalmology. American Medical Association; 2000;118:819–25. 3. Kuhn M, Johnson K. Applied Predictive Modeling. Springer Science & Business Media; 2013. 595 p. © 2021 Flaxman AD et al. JAMA Ophthalmology. eFigure 1: Crude prevalence rate of blindness increases with age for all race and ethnic groups, starting around age 60 years. © 2021 Flaxman AD et al. JAMA Ophthalmology. eTable 1: Estimated prevalence count of people living with visual acuity loss or blindness, stratified by state, as well as prevalence rates (in percent). Prevalence Count Prevalence Rate (%) th th th th Mean 2.5 percentile 97.5 percentile Mean 2.5 percentile 97.5 percentile AK 12,300 10,600 14,400 1.66 1.43 1.95 AL 131,200 112,400 151,100 2.69 2.31 3.10 AR 160,300 135,000 184,300 2.28 1.92 2.63 AZ 89,300 75,800 104,100 2.97 2.52 3.46 CA 810,900 709,900 935,100 2.05 1.80 2.37 CO 99,000 84,200 115,600 1.76 1.50 2.06 CT 62,300 51,500 73,000 1.74 1.44 2.03 DC 22,400 19,400 25,300 3.22 2.80 3.65 DE 18,100 15,400 21,300 1.88 1.60 2.21 FL 558,600 475,700 641,500 2.66 2.27 3.06 GA 221,300 190,900 255,700 2.12 1.83 2.45 HI 30,200 24,500 35,600 2.12 1.71 2.50 IA 48,000 38,800 56,400 1.52 1.24 1.79 ID 37,000 30,600 43,500 2.15 1.78 2.53 IL 244,900 211,600 282,000 1.91 1.65 2.20 IN 141,900 120,400 166,400 2.13 1.81 2.50 © 2021 Flaxman AD et al. JAMA Ophthalmology. KS 63,000 52,600 73,200 2.16 1.80 2.51 KY 125,300 103,600 145,200 2.81 2.33 3.26 LA 123,000 104,600 141,100 2.63 2.23 3.01 MA 129,600 107,200 152,400 1.89 1.56 2.22 MD 102,800 86,900 118,900 1.70 1.44 1.97 ME 18,000 13,600 22,100 1.35 1.02 1.65 MI 200,700 168,100 234,800 2.02 1.69 2.36 MN 92,500 74,900 108,600 1.66 1.34 1.95 MO 138,000 115,500 161,200 2.26 1.89 2.64 MS 98,300 84,100 113,200 3.29 2.82 3.79 MT 19,100 15,500 23,100 1.82 1.47 2.20 NC 230,800 194,000 265,300 2.25 1.89 2.58 ND 12,100 10,000 14,500 1.60 1.32 1.92 NE 40,600 34,400 48,200 2.12 1.79 2.51 NH 23,500 18,900 28,000 1.75 1.41 2.09 NJ 169,000 143,300 192,600 1.88 1.59 2.14 NM 62,300 54,000 71,000 2.98 2.58 3.40 NV 80,300 69,300 91,700 2.68 2.31 3.06 NY 393,200 337,500 449,900 1.98 1.70 2.27 OH 251,200 214,200 296,700 2.15 1.84 2.54 © 2021 Flaxman AD et al. JAMA Ophthalmology. OK 116,400 99,900 135,200 2.96 2.54 3.44 OR 78,900 65,500 93,300 1.90 1.58 2.25 PA 302,600 250,500 352,500 2.36 1.96 2.75 RI 21,500 17,900 25,700 2.03 1.69 2.42 SC 129,800 111,100 149,800 2.58 2.21 2.98 SD 16,000 13,000 18,700 1.84 1.50 2.15 TN 172,300 143,100 199,400 2.57 2.13 2.97 TX 634,000 551,100 726,300 2.24 1.95 2.57 UT 43,000 35,900 50,000 1.39 1.16 1.61 VA 168,100 143,100 191,300 1.98 1.69 2.26 VT 13,900 11,500 16,500 2.23 1.85 2.65 WA 144,200 118,700 165,400 1.95 1.60 2.23 WI 101,600 83,500 119,200 1.75 1.44 2.06 WV 65,200 53,200 77,400 3.59 2.93 4.26 WY 10,200 8,400 12,100 1.77 1.45 2.08 © 2021 Flaxman AD et al. JAMA Ophthalmology. eTable 2: Estimated prevalence count of people living with blindness, stratified by location, as well as prevalence rates (in percent). Prevalence Count Prevalence Rate (%) th th th th Mean 2.5 percentile 97.5 percentile Mean 2.5 percentile 97.5 percentile AK 1,600 1,200 2,100 0.22 0.17 0.28 AL 21,500 16,400 27,600 0.44 0.34 0.57 AR 22,400 16,700 29,300 0.32 0.24 0.42 AZ 15,300 11,200 19,300 0.51 0.37 0.64 CA 106,200 83,600 131,000 0.27 0.21 0.33 CO 14,000 9,900 17,600 0.25 0.18 0.31 CT 10,200 7,300 13,600 0.28 0.20 0.38 DC 3,700 2,800 4,600 0.53 0.40 0.66 DE 2,900 2,100 3,600 0.30 0.22 0.38 FL 84,800 64,600 108,800 0.40 0.31 0.52 GA 33,300 25,500 41,400 0.32 0.24 0.40 HI 3,800 2,500 5,100 0.27 0.18 0.36 IA 8,100 5,300 10,900 0.26 0.17 0.35 ID 5,600 4,000 7,200 0.33 0.23 0.42 IL 38,200 27,900 48,700 0.30 0.22 0.38 IN 23,400 16,800 30,500 0.35 0.25 0.46 © 2021 Flaxman AD et al. JAMA Ophthalmology. KS 10,600 7,600 14,100 0.37 0.26 0.48 KY 21,300 15,900 27,700 0.48 0.36 0.62 LA 20,100 15,100 24,900 0.43 0.32 0.53 MA 20,600 14,400 27,100 0.30 0.21 0.40 MD 15,700 12,000 20,300 0.26 0.20 0.34 ME 3,000 2,000 4,000 0.22 0.15 0.30 MI 32,300 23,500 41,900 0.32 0.24 0.42 MN 15,400 10,300 20,300 0.28 0.18 0.36 MO 23,200 16,000 30,400 0.38 0.26 0.50 MS 16,700 13,100 20,800 0.56 0.44 0.70 MT 3,000 2,100 4,000 0.29 0.20 0.38 NC 37,000 27,800 46,400 0.36 0.27 0.45 ND 2,100 1,400 2,900 0.28 0.19 0.38 NE 7,000 5,000 9,200 0.36 0.26 0.48 NH 3,900 2,600 5,200 0.29 0.19 0.39 NJ 25,400 19,000 31,900 0.28 0.21 0.35 NM 8,300 6,500 10,600 0.40 0.31 0.51 NV 10,900 8,200 13,400 0.36 0.28 0.45 NY 59,800 45,800 75,900 0.30 0.23 0.38 OH 42,500 31,500 55,600 0.36 0.27 0.48 © 2021 Flaxman AD et al. JAMA Ophthalmology. OK 18,700 13,900 24,100 0.48 0.35 0.61 OR 11,800 8,300 15,100 0.28 0.20 0.36 PA 52,300 35,700 68,500 0.41 0.28 0.54 RI 3,400 2,400 4,500 0.33 0.22 0.43 SC 20,900 15,900 26,300 0.42 0.32 0.52 SD 2,700 1,900 3,600 0.31 0.22 0.42 TN 28,500 21,400 35,800 0.42 0.32 0.53 TX 86,100 66,200 106,200 0.30 0.23 0.38 UT 5,900 4,500 7,700 0.19 0.14 0.25 VA 25,800 19,700 32,700 0.30 0.23 0.39 VT 2,400 1,600 3,100 0.38 0.26 0.49 WA 21,300 16,000 27,200 0.29 0.22 0.37 WI 16,600 11,800 21,800 0.29 0.20 0.38 WV 11,700 8,300 15,100 0.65 0.46 0.83 WY 1,600 1,100 2,000 0.27 0.20 0.35 © 2021 Flaxman AD et al. JAMA Ophthalmology. P

Journal

JAMA OphthalmologyAmerican Medical Association

Published: Jul 13, 2021

References