TY - JOUR
AU - Yang,, Yong
AB - Abstract Studies of health behaviors and behavior intervention have begun to explore the potential of agent-based modeling (ABM). A review of how ABMs have been used in health behavior, behavior intervention, and corresponding insights is warranted. The goal of this study was to provide a narrative review of the applications of ABMs in health behavior change and intervention. I will focus on two perspectives: (a) the mechanism of behavior and behavior change and (b) ABMs’ use for behavior intervention. I identified and reviewed 17 ABMs applied to behaviors including physical activity, diet, alcoholic drinking, smoking, and drug use. Among these ABMs, I grouped their mechanisms of behavior change into four categories and evaluated the advantages and disadvantages of each mechanism. For behavior intervention, I evaluated the use of ABMs on levels of individual, interpersonal, and neighborhood environment. Various behavior change mechanisms and simplifications existed because of our limited knowledge of behaviors at the individual level. Utility maximization was the most frequently used mechanism. ABMs offered insights for behavior intervention including the benefits of upstream interventions and multilevel intervention, as well as balances among various factors, outcomes, and populations. ABMs have been used to model a diversity of behaviors, populations, and interventions. The use of ABMs in health behavior is at an early stage, and a major challenge is our limited knowledge of behaviors at the individual level. Implications Practice: Substitution (i.e., using population-level knowledge in ABMs when no individual-level knowledge is available) is common in ABM practice, and this may discredit the use of ABMs in the long term. Policy: ABMs confirm the benefits of upstream interventions and multilevel interventions, and ABMs can help to design interventions and policies that balance conflicting interests and avoid inadvertent effects. Research: Qualitative data and big data have the potential to provide knowledge of behaviors at the individual level that is crucial for ABM research. INTRODUCTION Behavior and health are related through multiple pathways [1]. Unhealthy behaviors account for 40% of the premature deaths in the USA [2], and many chronic diseases are attributable to suboptimal health behaviors. For example, obesity is accepted to be an outcome of complex interactions among physical activity (PA), diet, built and social environments (e.g., neighborhood, family, and social network), and genes/epigenetics [3, 4]. Although several behaviors (e.g., PA, diet, and smoking) have been studied extensively [5, 6], our overall understanding of health behaviors and behavior change mechanisms, especially the ongoing dynamic feedback between human behavior and the environment, is limited [7]. Unfortunately, this understanding is crucial in designing effective behavior interventions. Recently, studies of health behaviors and behavior intervention have begun to explore the potential of agent-based modeling (ABM) [7]. An ABM is a “bottom-up” computational model that simulates agents’ behaviors, interactions among and between agents and their environments, and among various environmental features, to achieve an understanding of the system as a whole [8]. A growing number of health behavior-related ABMs have emerged [9–12]. A recent systematic review [9] identified 22 ABMs used to study both diseases and health behaviors (e.g., walking, alcohol use, and diet). Another narrative review covered the application of ABM for chronic diseases including diabetes, cardiovascular disease, and obesity [10]. In general, these reviews have focused on why ABMs are needed and in what areas they have been applied, to illustrate ABM’s features and benefits. Given the current progress, a detailed review of how ABMs have been used in health behavior, behavior intervention, and corresponding insights is warranted. The goal of this study was to provide a narrative review of the applications of ABMs in health behavior change and intervention. I will focus on two perspectives: (a) the mechanism of behavior and behavior change and (b) ABMs’ use for behavior intervention. IDENTIFICATION OF ABMS FOR HEALTH BEHAVIORS AND INTERVENTIONS Five health behaviors were selected, including PA, diet, smoking, alcoholic drinking, and drug use. I searched PubMed (Medline), ScienceDirect, and Web of Science using a combination of keywords from three categories: (a) agent-based model, agent-based simulation, individual-based model, or individual-based simulation; (b) walking, exercise, physical activity, diet, eating, food, smoking, tobacco, drink, alcohol, or drug; and (c) policy or intervention. I excluded studies that did not focus on one of these five behaviors or those that included one of the five behaviors, but only as independent variables in studying other outcomes (e.g., ABMs that included PA and diet as independent variables to study obesity). ABMs that were not intervention oriented were excluded, as well. As Table 1 summarizes, 17 ABMs (represented by 25 articles) were identified because some appeared in multiple articles. Among them, five addressed PA, including walking and children’s active travel to school. Six were related to diet, with the majority focusing on diet disparities by income, race, or neighborhood. Two ABMs were identified for each category of alcoholic drinking, smoking, and drug use. It should be noted that this study was a narrative rather than a systematic review. It was possible that some related ABMs were neglected, and cautions should be given on some findings: for example, the comparison of numbers of ABMs among behaviors, mechanisms, and interventions. However, to my knowledge, the ABMs identified, although they may be not exhaustive, reflect the latest progress in their applications in health behavior research. Table 1 | Characteristics of the agent-based models on health behavior Aims and context Behavior mechanism Interventions and major findings Physical activity including walk and exercise Yang et al. [13, 14]. Examine the influence of built and social environments on the socioeconomic inequalities in walking, among adults in a hypothetical American city. • The choice between driving and walking is a function of distance, walking ability, and attitude toward walking. • Promote walking by (a) increasing positive attitudes toward walking and (b) improving neighborhood safety. • Previous walking experience feedbacks to shape attitude toward walking. • Findings: increasing attitudes toward walking was immediately effective but was not sustainable if the environment is not conducive to walking. Yang et al. [15]. Explain socioeconomic inequalities in utilitarian walking and explore policy to promote walking among adults in a hypothetical American city. • The choice among driving, public transit, and walking is a utility function of travel cost and attitudes toward each travel mode. • Promote walking by (a) adjusting travel cost (e.g. increase parking fee or fuel price) and (b) combining the adjustment of travel cost and the increasing of positive attitudes toward walking. • Previous travel experience feedbacks to shape attitudes toward each travel mode. • Findings: the adjustment of travel cost was effective to low socioeconomic people but not effective to high socioeconomic people. Yang et al., and Yang and Diez-Roux [16, 17]. Promote children’s walking to school in a hypothetical city. • Walking is dependent on traffic, distance to school, and attitude toward walking. • Promote walking to school by (a) improving traffic safety; (b) implementing walking school bus program; and (c) increasing attitude toward walking. • More pedestrians on the street will improve traffic safety. A child’s attitude toward walking is influenced by other children in the same school. • Findings: there is a synergistic effect between increasing attitudes toward walking and a walking school bus program. Zhang et al. [18]. Increase afterschool physical activity in children. Agents in the model represented children who enrolled in afterschool programs. • A child adjusts her/his physical activity level to the average level of her/his friends. • Compare three intervention strategies that target different children: (a) the most sedentary children, (b) the most connected children, and (c) random selection. • Findings: targeting the most sedentary children was the best at increasing their physical activity levels, but targeting the most connected children was the best at increasing physical activity for all. Lemoine et al. [19]. Assess how the bus rapid transit system (BRT) influences walking in Bogotá, Columbia. • The choice among car, taxi, bus, BRT, and walking is a utility function of travel cost, time, and attributes specific to each travel mode. • Assess by varying: (a) number of BRT lanes and (b) number of BRT and bus stations. • Findings: increasing both BRT lanes and stations can increase walking, and the increase could reach a saturation status. Diet Auchincloss et al. [20]. Examine how economic segregation generates diet disparities by income in a hypothetical city. • A household selects food stores using a utility function of price, distance, habitual behavior, and preference. • Examine by varying (a) segregation of households or stores, (b) healthy food preferences, and (c) price. • Stores with the fewest customers may close, and new stores may move in. • Finding: interventions on preferences and prices were necessary to overcome the disparities generated by segregation. Widener et al. [21]. Examine how to improve access to healthy food for low-income households by changing food store environment in urban areas. • A household chooses the nearest food store. Supermarkets provide healthy food and convenience stores provide both healthy and unhealthy food. • Explore strategies to (a) introduce farmer’s market vendors; (b) vary the frequency of grocery shopping; (c) increase healthy food in convenience stores, and (d) introduce a mobile market distribution system. • Findings: both the spatial and nonspatial strategies should be considered to improve healthy diets among low-income households. Orr et al. [22]. Examine the influence of education quality, social network, and social norm on diet disparities between Black and White in the 100 largest metropolitan statistical areas in the USA. • Diet is a utility function of education level, neighborhood school quality, access to healthy foods, social norm, and the behaviors of others. • Explore strategies to improve (a) neighborhood school quality; (b) social influence on diet, and (c) social norm on diet. • Aging and life course events such as residential mobility were included. • Findings: improving school quality can reduce racial disparities in diet. Zhang et al. [23]. Promote healthy diet in Pasadena, CA. • Diet is a utility function of preferences, health beliefs, price, food accessibility, and demographic factors. • Examine strategies involving (a) tax on unhealthy food; (b) subsidies for healthy food; (c) promotion of social norms for a healthy diet; and (d) regulations on local food environment • Individuals’ preferences and health beliefs on diet were updated through the preceding exposures. • Findings: promoting healthy diet social norms was more effective than targeting food prices and regulating food environment. • Food outlets may switch between healthy and unhealthy food depending on their sales. Li et al. [24]. Promote fruit and vegetable consumption in New York City. • This study is based on the model of Zhang et al. [23] • Promote social norms on a healthy diet. • Findings: the promotion was less effective in neighborhoods with low education levels or a high proportion of male residents. Blok et al. [25]. Reduce diet inequality by income in Eindhoven, the Netherlands. • For a household, where to shop for food and whether to visit fast food outlets are utility functions of distance, price, and food preference. • Explore: (a) eliminating residential segregation; (b) reducing the price of healthy food, and (c) providing education. • Food outlets with few customers will close and new food outlets emerge. • Findings: eliminating residential segregation had the largest impact in reducing income inequalities in diet but negatively affected high-income households. Alcohol drinking Scott et al. [26–28]. Reduce heavy alcohol drinking among young adults in drinking venues, Melbourne, Australia. • An individual moves between drinking venues over a night. • Examine: (a) public transport extension and (b) venue lockout and last-drink policy. • An individual drinks at different rates in different settings, influenced by own status and friends. • Findings: (a) a 2-hr extension of public transport was effective, but a 24-hr extension had minimal additional benefits and (b) additional hours between the lockout and last drinks reduced transport-related harm at the cost of business revenue loss. Fitzpatrick et al. [29, 30]. Reduce college drinking through modeling a single college drink event. • Students who are similar are likely to group together. • Explore the social norms of marketing interventions: (a) variation of the influence of peer influence and (b) identify verification. • The drinking of a student is influenced by observations and interactions with other students. • Findings: social norms of marketing interventions were more effective in populations highly susceptible to social pressures of peer influence and identity verification. Smoking Haas et al. and Schaefer et al. [31, 32]. Examine the influence of friendship networks on smoking among adolescents. • An individual’s smoking is influenced by personal characteristics and friends. • Explore the strength of peer influence, smoker popularity, and their joint effects on smoking. • Friendship network changed over time depending on the similarity of smoking between friends. • Findings: (a) both peer influence and smoking-based popularity affected smoking, and their joint effects were nonlinear and (b) asymmetrical social influence: peer influence was stronger for smoking initiation than cessation. Luke et al. [33]. Examine the effect of policy to reduce tobacco retailer density in various types of town. • An individual chooses where and how much tobacco to buy with the aim to minimize the costs of tobacco and travel. • Examine retailer reduction policies: (a) random, (b) restrict certain types, (c) limit proximity to schools, and (d) limit proximity among retailers. • Findings: reducing retailer density can increase the costs of tobacco and travel, and the effect varies by town type. Drug use Moore et al., Perez et al., and Dray et al. [34–36]. Explore the impact of policies on drug use and related harms among young Australians. • A person switches among five stages of drug use: novice, occasional, regular, hardcore, and marginal, by the influence of social network, places visited, and mass media. • Examine (a) ecstasy pill-testing; (b) drug detection dogs, and (c) mass media drug prevention campaign. • Findings: (a) pill-testing was effective when adulterated ecstasy pills were widely available in a drug market; (b) the detection rate of dogs have to very high to be effective; (c) mass media campaign was effective only among moderate drug users. Bobashev et al. [37]. Evaluate the effects of opioid-related policies and interventions at the community level. • A pain patient switches among various stages including prescription-compliant, noncompliant prescription user, noncompliant opioid user, heroin user, death, under treatment, and opioid free. • Examine: (a) physician prescription drug-monitoring programs’ compliance; (b) pharmacy use of tamper-resistant pills; (c) reduce initial doses of opioids; and (d) increase naloxone availability. • A patient interacts with physicians, drug dealers, pharmacies, and emergency departments. • Findings: strong effects of naloxone use, marginal short-term effects of prescription drug-monitoring programs’ compliance, and no positive effects of tamper-resistant medications. Aims and context Behavior mechanism Interventions and major findings Physical activity including walk and exercise Yang et al. [13, 14]. Examine the influence of built and social environments on the socioeconomic inequalities in walking, among adults in a hypothetical American city. • The choice between driving and walking is a function of distance, walking ability, and attitude toward walking. • Promote walking by (a) increasing positive attitudes toward walking and (b) improving neighborhood safety. • Previous walking experience feedbacks to shape attitude toward walking. • Findings: increasing attitudes toward walking was immediately effective but was not sustainable if the environment is not conducive to walking. Yang et al. [15]. Explain socioeconomic inequalities in utilitarian walking and explore policy to promote walking among adults in a hypothetical American city. • The choice among driving, public transit, and walking is a utility function of travel cost and attitudes toward each travel mode. • Promote walking by (a) adjusting travel cost (e.g. increase parking fee or fuel price) and (b) combining the adjustment of travel cost and the increasing of positive attitudes toward walking. • Previous travel experience feedbacks to shape attitudes toward each travel mode. • Findings: the adjustment of travel cost was effective to low socioeconomic people but not effective to high socioeconomic people. Yang et al., and Yang and Diez-Roux [16, 17]. Promote children’s walking to school in a hypothetical city. • Walking is dependent on traffic, distance to school, and attitude toward walking. • Promote walking to school by (a) improving traffic safety; (b) implementing walking school bus program; and (c) increasing attitude toward walking. • More pedestrians on the street will improve traffic safety. A child’s attitude toward walking is influenced by other children in the same school. • Findings: there is a synergistic effect between increasing attitudes toward walking and a walking school bus program. Zhang et al. [18]. Increase afterschool physical activity in children. Agents in the model represented children who enrolled in afterschool programs. • A child adjusts her/his physical activity level to the average level of her/his friends. • Compare three intervention strategies that target different children: (a) the most sedentary children, (b) the most connected children, and (c) random selection. • Findings: targeting the most sedentary children was the best at increasing their physical activity levels, but targeting the most connected children was the best at increasing physical activity for all. Lemoine et al. [19]. Assess how the bus rapid transit system (BRT) influences walking in Bogotá, Columbia. • The choice among car, taxi, bus, BRT, and walking is a utility function of travel cost, time, and attributes specific to each travel mode. • Assess by varying: (a) number of BRT lanes and (b) number of BRT and bus stations. • Findings: increasing both BRT lanes and stations can increase walking, and the increase could reach a saturation status. Diet Auchincloss et al. [20]. Examine how economic segregation generates diet disparities by income in a hypothetical city. • A household selects food stores using a utility function of price, distance, habitual behavior, and preference. • Examine by varying (a) segregation of households or stores, (b) healthy food preferences, and (c) price. • Stores with the fewest customers may close, and new stores may move in. • Finding: interventions on preferences and prices were necessary to overcome the disparities generated by segregation. Widener et al. [21]. Examine how to improve access to healthy food for low-income households by changing food store environment in urban areas. • A household chooses the nearest food store. Supermarkets provide healthy food and convenience stores provide both healthy and unhealthy food. • Explore strategies to (a) introduce farmer’s market vendors; (b) vary the frequency of grocery shopping; (c) increase healthy food in convenience stores, and (d) introduce a mobile market distribution system. • Findings: both the spatial and nonspatial strategies should be considered to improve healthy diets among low-income households. Orr et al. [22]. Examine the influence of education quality, social network, and social norm on diet disparities between Black and White in the 100 largest metropolitan statistical areas in the USA. • Diet is a utility function of education level, neighborhood school quality, access to healthy foods, social norm, and the behaviors of others. • Explore strategies to improve (a) neighborhood school quality; (b) social influence on diet, and (c) social norm on diet. • Aging and life course events such as residential mobility were included. • Findings: improving school quality can reduce racial disparities in diet. Zhang et al. [23]. Promote healthy diet in Pasadena, CA. • Diet is a utility function of preferences, health beliefs, price, food accessibility, and demographic factors. • Examine strategies involving (a) tax on unhealthy food; (b) subsidies for healthy food; (c) promotion of social norms for a healthy diet; and (d) regulations on local food environment • Individuals’ preferences and health beliefs on diet were updated through the preceding exposures. • Findings: promoting healthy diet social norms was more effective than targeting food prices and regulating food environment. • Food outlets may switch between healthy and unhealthy food depending on their sales. Li et al. [24]. Promote fruit and vegetable consumption in New York City. • This study is based on the model of Zhang et al. [23] • Promote social norms on a healthy diet. • Findings: the promotion was less effective in neighborhoods with low education levels or a high proportion of male residents. Blok et al. [25]. Reduce diet inequality by income in Eindhoven, the Netherlands. • For a household, where to shop for food and whether to visit fast food outlets are utility functions of distance, price, and food preference. • Explore: (a) eliminating residential segregation; (b) reducing the price of healthy food, and (c) providing education. • Food outlets with few customers will close and new food outlets emerge. • Findings: eliminating residential segregation had the largest impact in reducing income inequalities in diet but negatively affected high-income households. Alcohol drinking Scott et al. [26–28]. Reduce heavy alcohol drinking among young adults in drinking venues, Melbourne, Australia. • An individual moves between drinking venues over a night. • Examine: (a) public transport extension and (b) venue lockout and last-drink policy. • An individual drinks at different rates in different settings, influenced by own status and friends. • Findings: (a) a 2-hr extension of public transport was effective, but a 24-hr extension had minimal additional benefits and (b) additional hours between the lockout and last drinks reduced transport-related harm at the cost of business revenue loss. Fitzpatrick et al. [29, 30]. Reduce college drinking through modeling a single college drink event. • Students who are similar are likely to group together. • Explore the social norms of marketing interventions: (a) variation of the influence of peer influence and (b) identify verification. • The drinking of a student is influenced by observations and interactions with other students. • Findings: social norms of marketing interventions were more effective in populations highly susceptible to social pressures of peer influence and identity verification. Smoking Haas et al. and Schaefer et al. [31, 32]. Examine the influence of friendship networks on smoking among adolescents. • An individual’s smoking is influenced by personal characteristics and friends. • Explore the strength of peer influence, smoker popularity, and their joint effects on smoking. • Friendship network changed over time depending on the similarity of smoking between friends. • Findings: (a) both peer influence and smoking-based popularity affected smoking, and their joint effects were nonlinear and (b) asymmetrical social influence: peer influence was stronger for smoking initiation than cessation. Luke et al. [33]. Examine the effect of policy to reduce tobacco retailer density in various types of town. • An individual chooses where and how much tobacco to buy with the aim to minimize the costs of tobacco and travel. • Examine retailer reduction policies: (a) random, (b) restrict certain types, (c) limit proximity to schools, and (d) limit proximity among retailers. • Findings: reducing retailer density can increase the costs of tobacco and travel, and the effect varies by town type. Drug use Moore et al., Perez et al., and Dray et al. [34–36]. Explore the impact of policies on drug use and related harms among young Australians. • A person switches among five stages of drug use: novice, occasional, regular, hardcore, and marginal, by the influence of social network, places visited, and mass media. • Examine (a) ecstasy pill-testing; (b) drug detection dogs, and (c) mass media drug prevention campaign. • Findings: (a) pill-testing was effective when adulterated ecstasy pills were widely available in a drug market; (b) the detection rate of dogs have to very high to be effective; (c) mass media campaign was effective only among moderate drug users. Bobashev et al. [37]. Evaluate the effects of opioid-related policies and interventions at the community level. • A pain patient switches among various stages including prescription-compliant, noncompliant prescription user, noncompliant opioid user, heroin user, death, under treatment, and opioid free. • Examine: (a) physician prescription drug-monitoring programs’ compliance; (b) pharmacy use of tamper-resistant pills; (c) reduce initial doses of opioids; and (d) increase naloxone availability. • A patient interacts with physicians, drug dealers, pharmacies, and emergency departments. • Findings: strong effects of naloxone use, marginal short-term effects of prescription drug-monitoring programs’ compliance, and no positive effects of tamper-resistant medications. Open in new tab Table 1 | Characteristics of the agent-based models on health behavior Aims and context Behavior mechanism Interventions and major findings Physical activity including walk and exercise Yang et al. [13, 14]. Examine the influence of built and social environments on the socioeconomic inequalities in walking, among adults in a hypothetical American city. • The choice between driving and walking is a function of distance, walking ability, and attitude toward walking. • Promote walking by (a) increasing positive attitudes toward walking and (b) improving neighborhood safety. • Previous walking experience feedbacks to shape attitude toward walking. • Findings: increasing attitudes toward walking was immediately effective but was not sustainable if the environment is not conducive to walking. Yang et al. [15]. Explain socioeconomic inequalities in utilitarian walking and explore policy to promote walking among adults in a hypothetical American city. • The choice among driving, public transit, and walking is a utility function of travel cost and attitudes toward each travel mode. • Promote walking by (a) adjusting travel cost (e.g. increase parking fee or fuel price) and (b) combining the adjustment of travel cost and the increasing of positive attitudes toward walking. • Previous travel experience feedbacks to shape attitudes toward each travel mode. • Findings: the adjustment of travel cost was effective to low socioeconomic people but not effective to high socioeconomic people. Yang et al., and Yang and Diez-Roux [16, 17]. Promote children’s walking to school in a hypothetical city. • Walking is dependent on traffic, distance to school, and attitude toward walking. • Promote walking to school by (a) improving traffic safety; (b) implementing walking school bus program; and (c) increasing attitude toward walking. • More pedestrians on the street will improve traffic safety. A child’s attitude toward walking is influenced by other children in the same school. • Findings: there is a synergistic effect between increasing attitudes toward walking and a walking school bus program. Zhang et al. [18]. Increase afterschool physical activity in children. Agents in the model represented children who enrolled in afterschool programs. • A child adjusts her/his physical activity level to the average level of her/his friends. • Compare three intervention strategies that target different children: (a) the most sedentary children, (b) the most connected children, and (c) random selection. • Findings: targeting the most sedentary children was the best at increasing their physical activity levels, but targeting the most connected children was the best at increasing physical activity for all. Lemoine et al. [19]. Assess how the bus rapid transit system (BRT) influences walking in Bogotá, Columbia. • The choice among car, taxi, bus, BRT, and walking is a utility function of travel cost, time, and attributes specific to each travel mode. • Assess by varying: (a) number of BRT lanes and (b) number of BRT and bus stations. • Findings: increasing both BRT lanes and stations can increase walking, and the increase could reach a saturation status. Diet Auchincloss et al. [20]. Examine how economic segregation generates diet disparities by income in a hypothetical city. • A household selects food stores using a utility function of price, distance, habitual behavior, and preference. • Examine by varying (a) segregation of households or stores, (b) healthy food preferences, and (c) price. • Stores with the fewest customers may close, and new stores may move in. • Finding: interventions on preferences and prices were necessary to overcome the disparities generated by segregation. Widener et al. [21]. Examine how to improve access to healthy food for low-income households by changing food store environment in urban areas. • A household chooses the nearest food store. Supermarkets provide healthy food and convenience stores provide both healthy and unhealthy food. • Explore strategies to (a) introduce farmer’s market vendors; (b) vary the frequency of grocery shopping; (c) increase healthy food in convenience stores, and (d) introduce a mobile market distribution system. • Findings: both the spatial and nonspatial strategies should be considered to improve healthy diets among low-income households. Orr et al. [22]. Examine the influence of education quality, social network, and social norm on diet disparities between Black and White in the 100 largest metropolitan statistical areas in the USA. • Diet is a utility function of education level, neighborhood school quality, access to healthy foods, social norm, and the behaviors of others. • Explore strategies to improve (a) neighborhood school quality; (b) social influence on diet, and (c) social norm on diet. • Aging and life course events such as residential mobility were included. • Findings: improving school quality can reduce racial disparities in diet. Zhang et al. [23]. Promote healthy diet in Pasadena, CA. • Diet is a utility function of preferences, health beliefs, price, food accessibility, and demographic factors. • Examine strategies involving (a) tax on unhealthy food; (b) subsidies for healthy food; (c) promotion of social norms for a healthy diet; and (d) regulations on local food environment • Individuals’ preferences and health beliefs on diet were updated through the preceding exposures. • Findings: promoting healthy diet social norms was more effective than targeting food prices and regulating food environment. • Food outlets may switch between healthy and unhealthy food depending on their sales. Li et al. [24]. Promote fruit and vegetable consumption in New York City. • This study is based on the model of Zhang et al. [23] • Promote social norms on a healthy diet. • Findings: the promotion was less effective in neighborhoods with low education levels or a high proportion of male residents. Blok et al. [25]. Reduce diet inequality by income in Eindhoven, the Netherlands. • For a household, where to shop for food and whether to visit fast food outlets are utility functions of distance, price, and food preference. • Explore: (a) eliminating residential segregation; (b) reducing the price of healthy food, and (c) providing education. • Food outlets with few customers will close and new food outlets emerge. • Findings: eliminating residential segregation had the largest impact in reducing income inequalities in diet but negatively affected high-income households. Alcohol drinking Scott et al. [26–28]. Reduce heavy alcohol drinking among young adults in drinking venues, Melbourne, Australia. • An individual moves between drinking venues over a night. • Examine: (a) public transport extension and (b) venue lockout and last-drink policy. • An individual drinks at different rates in different settings, influenced by own status and friends. • Findings: (a) a 2-hr extension of public transport was effective, but a 24-hr extension had minimal additional benefits and (b) additional hours between the lockout and last drinks reduced transport-related harm at the cost of business revenue loss. Fitzpatrick et al. [29, 30]. Reduce college drinking through modeling a single college drink event. • Students who are similar are likely to group together. • Explore the social norms of marketing interventions: (a) variation of the influence of peer influence and (b) identify verification. • The drinking of a student is influenced by observations and interactions with other students. • Findings: social norms of marketing interventions were more effective in populations highly susceptible to social pressures of peer influence and identity verification. Smoking Haas et al. and Schaefer et al. [31, 32]. Examine the influence of friendship networks on smoking among adolescents. • An individual’s smoking is influenced by personal characteristics and friends. • Explore the strength of peer influence, smoker popularity, and their joint effects on smoking. • Friendship network changed over time depending on the similarity of smoking between friends. • Findings: (a) both peer influence and smoking-based popularity affected smoking, and their joint effects were nonlinear and (b) asymmetrical social influence: peer influence was stronger for smoking initiation than cessation. Luke et al. [33]. Examine the effect of policy to reduce tobacco retailer density in various types of town. • An individual chooses where and how much tobacco to buy with the aim to minimize the costs of tobacco and travel. • Examine retailer reduction policies: (a) random, (b) restrict certain types, (c) limit proximity to schools, and (d) limit proximity among retailers. • Findings: reducing retailer density can increase the costs of tobacco and travel, and the effect varies by town type. Drug use Moore et al., Perez et al., and Dray et al. [34–36]. Explore the impact of policies on drug use and related harms among young Australians. • A person switches among five stages of drug use: novice, occasional, regular, hardcore, and marginal, by the influence of social network, places visited, and mass media. • Examine (a) ecstasy pill-testing; (b) drug detection dogs, and (c) mass media drug prevention campaign. • Findings: (a) pill-testing was effective when adulterated ecstasy pills were widely available in a drug market; (b) the detection rate of dogs have to very high to be effective; (c) mass media campaign was effective only among moderate drug users. Bobashev et al. [37]. Evaluate the effects of opioid-related policies and interventions at the community level. • A pain patient switches among various stages including prescription-compliant, noncompliant prescription user, noncompliant opioid user, heroin user, death, under treatment, and opioid free. • Examine: (a) physician prescription drug-monitoring programs’ compliance; (b) pharmacy use of tamper-resistant pills; (c) reduce initial doses of opioids; and (d) increase naloxone availability. • A patient interacts with physicians, drug dealers, pharmacies, and emergency departments. • Findings: strong effects of naloxone use, marginal short-term effects of prescription drug-monitoring programs’ compliance, and no positive effects of tamper-resistant medications. Aims and context Behavior mechanism Interventions and major findings Physical activity including walk and exercise Yang et al. [13, 14]. Examine the influence of built and social environments on the socioeconomic inequalities in walking, among adults in a hypothetical American city. • The choice between driving and walking is a function of distance, walking ability, and attitude toward walking. • Promote walking by (a) increasing positive attitudes toward walking and (b) improving neighborhood safety. • Previous walking experience feedbacks to shape attitude toward walking. • Findings: increasing attitudes toward walking was immediately effective but was not sustainable if the environment is not conducive to walking. Yang et al. [15]. Explain socioeconomic inequalities in utilitarian walking and explore policy to promote walking among adults in a hypothetical American city. • The choice among driving, public transit, and walking is a utility function of travel cost and attitudes toward each travel mode. • Promote walking by (a) adjusting travel cost (e.g. increase parking fee or fuel price) and (b) combining the adjustment of travel cost and the increasing of positive attitudes toward walking. • Previous travel experience feedbacks to shape attitudes toward each travel mode. • Findings: the adjustment of travel cost was effective to low socioeconomic people but not effective to high socioeconomic people. Yang et al., and Yang and Diez-Roux [16, 17]. Promote children’s walking to school in a hypothetical city. • Walking is dependent on traffic, distance to school, and attitude toward walking. • Promote walking to school by (a) improving traffic safety; (b) implementing walking school bus program; and (c) increasing attitude toward walking. • More pedestrians on the street will improve traffic safety. A child’s attitude toward walking is influenced by other children in the same school. • Findings: there is a synergistic effect between increasing attitudes toward walking and a walking school bus program. Zhang et al. [18]. Increase afterschool physical activity in children. Agents in the model represented children who enrolled in afterschool programs. • A child adjusts her/his physical activity level to the average level of her/his friends. • Compare three intervention strategies that target different children: (a) the most sedentary children, (b) the most connected children, and (c) random selection. • Findings: targeting the most sedentary children was the best at increasing their physical activity levels, but targeting the most connected children was the best at increasing physical activity for all. Lemoine et al. [19]. Assess how the bus rapid transit system (BRT) influences walking in Bogotá, Columbia. • The choice among car, taxi, bus, BRT, and walking is a utility function of travel cost, time, and attributes specific to each travel mode. • Assess by varying: (a) number of BRT lanes and (b) number of BRT and bus stations. • Findings: increasing both BRT lanes and stations can increase walking, and the increase could reach a saturation status. Diet Auchincloss et al. [20]. Examine how economic segregation generates diet disparities by income in a hypothetical city. • A household selects food stores using a utility function of price, distance, habitual behavior, and preference. • Examine by varying (a) segregation of households or stores, (b) healthy food preferences, and (c) price. • Stores with the fewest customers may close, and new stores may move in. • Finding: interventions on preferences and prices were necessary to overcome the disparities generated by segregation. Widener et al. [21]. Examine how to improve access to healthy food for low-income households by changing food store environment in urban areas. • A household chooses the nearest food store. Supermarkets provide healthy food and convenience stores provide both healthy and unhealthy food. • Explore strategies to (a) introduce farmer’s market vendors; (b) vary the frequency of grocery shopping; (c) increase healthy food in convenience stores, and (d) introduce a mobile market distribution system. • Findings: both the spatial and nonspatial strategies should be considered to improve healthy diets among low-income households. Orr et al. [22]. Examine the influence of education quality, social network, and social norm on diet disparities between Black and White in the 100 largest metropolitan statistical areas in the USA. • Diet is a utility function of education level, neighborhood school quality, access to healthy foods, social norm, and the behaviors of others. • Explore strategies to improve (a) neighborhood school quality; (b) social influence on diet, and (c) social norm on diet. • Aging and life course events such as residential mobility were included. • Findings: improving school quality can reduce racial disparities in diet. Zhang et al. [23]. Promote healthy diet in Pasadena, CA. • Diet is a utility function of preferences, health beliefs, price, food accessibility, and demographic factors. • Examine strategies involving (a) tax on unhealthy food; (b) subsidies for healthy food; (c) promotion of social norms for a healthy diet; and (d) regulations on local food environment • Individuals’ preferences and health beliefs on diet were updated through the preceding exposures. • Findings: promoting healthy diet social norms was more effective than targeting food prices and regulating food environment. • Food outlets may switch between healthy and unhealthy food depending on their sales. Li et al. [24]. Promote fruit and vegetable consumption in New York City. • This study is based on the model of Zhang et al. [23] • Promote social norms on a healthy diet. • Findings: the promotion was less effective in neighborhoods with low education levels or a high proportion of male residents. Blok et al. [25]. Reduce diet inequality by income in Eindhoven, the Netherlands. • For a household, where to shop for food and whether to visit fast food outlets are utility functions of distance, price, and food preference. • Explore: (a) eliminating residential segregation; (b) reducing the price of healthy food, and (c) providing education. • Food outlets with few customers will close and new food outlets emerge. • Findings: eliminating residential segregation had the largest impact in reducing income inequalities in diet but negatively affected high-income households. Alcohol drinking Scott et al. [26–28]. Reduce heavy alcohol drinking among young adults in drinking venues, Melbourne, Australia. • An individual moves between drinking venues over a night. • Examine: (a) public transport extension and (b) venue lockout and last-drink policy. • An individual drinks at different rates in different settings, influenced by own status and friends. • Findings: (a) a 2-hr extension of public transport was effective, but a 24-hr extension had minimal additional benefits and (b) additional hours between the lockout and last drinks reduced transport-related harm at the cost of business revenue loss. Fitzpatrick et al. [29, 30]. Reduce college drinking through modeling a single college drink event. • Students who are similar are likely to group together. • Explore the social norms of marketing interventions: (a) variation of the influence of peer influence and (b) identify verification. • The drinking of a student is influenced by observations and interactions with other students. • Findings: social norms of marketing interventions were more effective in populations highly susceptible to social pressures of peer influence and identity verification. Smoking Haas et al. and Schaefer et al. [31, 32]. Examine the influence of friendship networks on smoking among adolescents. • An individual’s smoking is influenced by personal characteristics and friends. • Explore the strength of peer influence, smoker popularity, and their joint effects on smoking. • Friendship network changed over time depending on the similarity of smoking between friends. • Findings: (a) both peer influence and smoking-based popularity affected smoking, and their joint effects were nonlinear and (b) asymmetrical social influence: peer influence was stronger for smoking initiation than cessation. Luke et al. [33]. Examine the effect of policy to reduce tobacco retailer density in various types of town. • An individual chooses where and how much tobacco to buy with the aim to minimize the costs of tobacco and travel. • Examine retailer reduction policies: (a) random, (b) restrict certain types, (c) limit proximity to schools, and (d) limit proximity among retailers. • Findings: reducing retailer density can increase the costs of tobacco and travel, and the effect varies by town type. Drug use Moore et al., Perez et al., and Dray et al. [34–36]. Explore the impact of policies on drug use and related harms among young Australians. • A person switches among five stages of drug use: novice, occasional, regular, hardcore, and marginal, by the influence of social network, places visited, and mass media. • Examine (a) ecstasy pill-testing; (b) drug detection dogs, and (c) mass media drug prevention campaign. • Findings: (a) pill-testing was effective when adulterated ecstasy pills were widely available in a drug market; (b) the detection rate of dogs have to very high to be effective; (c) mass media campaign was effective only among moderate drug users. Bobashev et al. [37]. Evaluate the effects of opioid-related policies and interventions at the community level. • A pain patient switches among various stages including prescription-compliant, noncompliant prescription user, noncompliant opioid user, heroin user, death, under treatment, and opioid free. • Examine: (a) physician prescription drug-monitoring programs’ compliance; (b) pharmacy use of tamper-resistant pills; (c) reduce initial doses of opioids; and (d) increase naloxone availability. • A patient interacts with physicians, drug dealers, pharmacies, and emergency departments. • Findings: strong effects of naloxone use, marginal short-term effects of prescription drug-monitoring programs’ compliance, and no positive effects of tamper-resistant medications. Open in new tab MECHANISMS OF BEHAVIOR AND BEHAVIOR CHANGE Occurring within a complex adaptive system and driven by multiple determinants at multiple levels, behavior and behavior change are nonlinear, sensitive to initial conditions, vary highly, and are difficult to predict [38]. Ignoring the complexity of health behavior and behavior change may lead to resistance to intervention. One advantage of an ABM is its ability to represent behaviors at the individual level and account for dynamic interactions and feedback among and between individuals and their environments. Each individual can be characterized by a number of attributes and behavior rules that reflect a population’s heterogeneity. An ABM can describe explicitly the way in which environmental and social exposures (e.g., through a social network) may influence an individual’s internal attitude and external behavior, and the way the individual’s behavior may influence her/his environment and other individuals’ behaviors in return. Using an ABM, we can track a specific individual or obverse patterns at the population level aggregated by a specific dimension and over time. Thus, behavior, especially the mechanism of behavior change, needs to be explicit in an ABM, both qualitatively and quantitatively. For example, under what conditions does an individual choose to walk rather than use other means to travel? What is the specific mechanism that triggers an individual to switch from being a nonsmoker to a smoker? Although theoretically promising, ABM’s requirements challenge our current knowledge of behavior and behavior change. Over decades, a plethora of theories and frameworks has been proposed [39]. Generally, interventions based on behavior theories are believed to be more effective than are those that are not [40]. However, many behavior change theories or frameworks are descriptive and not operational. Therefore, it is difficult to measure the constructs and test the pathways. For example, one well-established and often-cited framework, the social-ecological model [41], simply describes and demarcates boundaries between different levels and lists factors within each level but seldom elaborates on the way in which the factors interact, especially interactions across levels [42, 43]. Because of our limited knowledge of behaviors and behavior change mechanisms at the individual level, behavior change implementation in ABM requires simplification. Various behavior change mechanisms and simplifications exist and an evaluation of the advantages and disadvantages of each mechanism may guide both ABM implementation and the development of behavioral theories. Among the 17 ABMs identified, I grouped their mechanisms of behavior change into four categories including utility maximization, decision trees, social network, and stages of change (see Table 2). It should be noted that among these four mechanisms, only utility maximization and the decision trees are mutually exclusive, whereas all other pairs could be complementary. For example, the influence of social network could be a factor in a decision tree or a utility in a utility maximization function. The stages of change category, where factors and the way in which they determine jointly a switch from one stage to another, could use either utility maximization, decision trees, or social network. Table 2 | Mechanism of behavior change and level or aspect of behavior intervention in the reviewed agent-based models Mechanism of behavior change Level or aspect of policy intervention Utility maximization Decision trees Social network Stages of change Individual Financial incentives Interpersonal Social norm Neighborhood environment Yang et al. [13, 14] X X X Yang et al. [15] X X X Yang et al., and Yang and Diez- Roux [16, 17] X X X Zhang et al. [18] X X X Lemoine et al. [19] X X Auchincloss et al. [20] X X X X Widener et al. [21] X X X Orr et al. [22] X X X X Zhang et al. [23] X X X X Li et al. [24] X X Blok et al. [25] X X X X Scott et al. [26–28] X X Fitzpatrick et al. [29, 30] X X X X Haas et al. and Schaefer et al. [31, 32] X X X X Luke et al. [33]. X X Moore et al., Perez et al., and Dray et al. [34–36] X X X X Bobashev et al. [37] X X X Mechanism of behavior change Level or aspect of policy intervention Utility maximization Decision trees Social network Stages of change Individual Financial incentives Interpersonal Social norm Neighborhood environment Yang et al. [13, 14] X X X Yang et al. [15] X X X Yang et al., and Yang and Diez- Roux [16, 17] X X X Zhang et al. [18] X X X Lemoine et al. [19] X X Auchincloss et al. [20] X X X X Widener et al. [21] X X X Orr et al. [22] X X X X Zhang et al. [23] X X X X Li et al. [24] X X Blok et al. [25] X X X X Scott et al. [26–28] X X Fitzpatrick et al. [29, 30] X X X X Haas et al. and Schaefer et al. [31, 32] X X X X Luke et al. [33]. X X Moore et al., Perez et al., and Dray et al. [34–36] X X X X Bobashev et al. [37] X X X Open in new tab Table 2 | Mechanism of behavior change and level or aspect of behavior intervention in the reviewed agent-based models Mechanism of behavior change Level or aspect of policy intervention Utility maximization Decision trees Social network Stages of change Individual Financial incentives Interpersonal Social norm Neighborhood environment Yang et al. [13, 14] X X X Yang et al. [15] X X X Yang et al., and Yang and Diez- Roux [16, 17] X X X Zhang et al. [18] X X X Lemoine et al. [19] X X Auchincloss et al. [20] X X X X Widener et al. [21] X X X Orr et al. [22] X X X X Zhang et al. [23] X X X X Li et al. [24] X X Blok et al. [25] X X X X Scott et al. [26–28] X X Fitzpatrick et al. [29, 30] X X X X Haas et al. and Schaefer et al. [31, 32] X X X X Luke et al. [33]. X X Moore et al., Perez et al., and Dray et al. [34–36] X X X X Bobashev et al. [37] X X X Mechanism of behavior change Level or aspect of policy intervention Utility maximization Decision trees Social network Stages of change Individual Financial incentives Interpersonal Social norm Neighborhood environment Yang et al. [13, 14] X X X Yang et al. [15] X X X Yang et al., and Yang and Diez- Roux [16, 17] X X X Zhang et al. [18] X X X Lemoine et al. [19] X X Auchincloss et al. [20] X X X X Widener et al. [21] X X X Orr et al. [22] X X X X Zhang et al. [23] X X X X Li et al. [24] X X Blok et al. [25] X X X X Scott et al. [26–28] X X Fitzpatrick et al. [29, 30] X X X X Haas et al. and Schaefer et al. [31, 32] X X X X Luke et al. [33]. X X Moore et al., Perez et al., and Dray et al. [34–36] X X X X Bobashev et al. [37] X X X Open in new tab Utility maximization Utility maximization is based on utility theory, which states that people make rational decisions to maximize their interests [44]. In behavior change, the rationale is that a person will choose an option that maximizes benefits and minimizes costs. The general process in ABMs is to compute a score for each behavioral option using a function. The function may have a number of contributing factors and each factor may be assigned a weight to reflect the variation in their influence. Thereafter, the behavioral option with the highest score may be selected or may be more likely to be selected (i.e., weighted by the score). One example is an ABM for diet [20]. To select a store for food shopping, each store was scored by a weighted sum of factors, including food price, distance to the store, habitual behavior, and preference for healthy foods, and then the store with the highest score was selected. More than half of the 17 ABMs [15, 19, 20, 22–25, 33] used utility maximization to model behavior and behavior change mechanisms. Among these, the majority considered financial costs in the utility function, such as prices of food and cigarettes, and travel budget. Some used the utility function to examine psychological factors, such as attitudes about certain modes of travel, food habits and preferences, and general health beliefs. Accessibility (e.g., access to healthy foods or tobacco retailers) was presented in various ways, such as travel time and budget, distance, or a dichotomous indicator if certain facilities were within a threshold distance or not. Many ABMs computed the utilities using a weighted sum of related factors, whereas some used different formulas to reflect the diverse roles each factor played. For example, in an ABM for utilitarian walking [15], the utility of each travel mode was computed as the product of the attitude toward that travel mode, and the negative exponential of related travel cost. Decision trees Decision trees are represented in the shape of a “tree” and consist of a number of branches connected by decision points. Decision points are implemented using simple “if…then…else” statements that correspond to the conditions to be met. Compared with utility maximization, decision trees may not consider all factors simultaneously, and some conditions need to be satisfied before others to be considered. For example, an individual may not consider all utilities and may seek a satisfactory behavior option rather than the optimal one. Maslow’s hierarchy of needs [45], which states that a human has several levels of needs with a pyramidal structure, such that when a lower level need is realized, the next higher level need will emerge, supports the application of a decision tree to behavior change. Four ABMs [13, 14, 16, 17, 21, 26–28] used decision trees. For example, one used a decision tree to model an individual’s choice regarding utilitarian walking [13, 14]. Three conditions had to be met for a person to have a walking trip for shopping: (a) the person needed to decide to shop that day (corresponds to feasibility); (b) there had to be shops within the person’s maximum walking distance (corresponds to accessibility), and (c) a random draw with a probability equal to the person’s attitude toward walking, in which the more positive the attitude, the more likely it is that this condition will be met. In this decision process, the former conditions had to be met before the latter was evaluated. As a second example, another ABM used a decision tree to model children’s active travel to school [16, 17]. A student would walk or ride a bike to school if two conditions were met: (a) traffic safety was higher than a threshold value and (b) the attitude toward active travel was sufficiently positive to overcome the barrier of distance (i.e., the farther a child lived from school, the greater the positive attitude needed). Social networks The third behavior change mechanism is social networks, which is supported by social norm [46] and social network theories [47]. The social norm theory posits that one’s perception of the behaviors and thoughts of others in one’s social group will influence one’s behavior [46]. The theory indicates that individuals will adjust their behaviors to be similar to their peers. Social network theory emphasizes the importance of social structure, in which individuals are connected through networks and the structural characteristics of networks influence the flow of risk factors, behaviors, and health outcomes [47]. The combination of social norm and social network theories has been applied to study the diffusion of behaviors, such as smoking and alcoholic drinking [29–32]. Four ABMs [18, 29–32, 34–36] used social networks to model behaviors and behavior changes. In these ABMs, individuals adjusted their behaviors to be similar to other individuals in their social networks. For example, one examined the way social networks influenced college drinking [29, 30]. In this model, students were assumed to be (a) more likely to associate with other students with similar characteristics and (b) adjust their drinking according to observations and interactions with other students. The model examined the effect of misperceptions of college drinking, that is, a student may have an inaccurate idea of how much other students drink, and this misperception affected her/his drinking, which then creates a cycle, that affects other students’ drinking. Another ABM [31, 32] explored the complexity further and assumed that not only did friends influence a person’s smoking, but also individuals’ similarity in smoking status influenced the friendship networks. Stages of change The fourth behavior change mechanism, which is similar to the transtheoretical model [48], is called stages of change. Rather than simply dichotomizing certain behaviors as “present” or “absent” (e.g., smoking vs. not smoking) or “healthy” and “unhealthy” (e.g., healthy vs. unhealthy diet), stages of change assumes that health behavior change is a process consisting of multiple stages. Although the transtheoretical model posits five stages of change including precontemplation, contemplation, preparation, action, and maintenance, the stages of behavior changes in an ABM with an “implementation” perspective. An ABM may combine stages of precontemplation, contemplation, and preparation together because these stages tend to occur internally and an ABM tends to focus on the explicit progression and regression of a specific behavior that can be observed externally. These behaviors include those within an individual and interactions between the individual with other individuals, and environment. It should be noted that the division of stages in an ABM can be either generic for most health behaviors or specific to a certain behavior. For example, tobacco regulatory research has recognized the importance of an ABM to represent the key processes underlying tobacco use including initiation, progression, cessation, and relapse [49], and some processes such as initiation and cessation may be further divided into several subprocesses. However, the above stage division may be applied to other health behaviors such as alcohol drinking and drug use as well. We found that two ABMs [34–37] used stages of change. The first one of drug use [34–36] used “generic division,” that is, it assumed that a person switches among five stages of drug use: novice, occasional, regular, hardcore, and marginal. The stages of change were triggered largely by the social network, places visited, and mass media. The second ABM of opioid use [37] was “specific to the behavior,” that is, it modeled a pain patient’s transition among several stages, such as compliant prescription user, noncompliant prescription user, noncompliant opioid user, and heroin user. The stages of change were triggered through the patient’s interactions with physicians, drug dealers, pharmacies, and emergency departments. EVALUATION OF THE FOUR BEHAVIOR CHANGE MECHANISMS Among the four mechanisms, utility maximization was used most commonly. This is not surprising given that utility maximization functions have a long history of use in choice modeling [50]. However, utility maximization has been criticized because of its inability to capture the complexity of human decision making, which is subject to limitations in computing, information availability, and rationality [51, 52]. The wide use of utility maximization is also attributable to another pragmatic reason: it is convenient for researchers to use knowledge at the population level as a substitute when knowledge at the individual level is unavailable or limited. For many health behaviors, the empirical evidence amassed to date provides a knowledge base only at the population level. Typically, an epidemiological study uses statistical analyses to examine how and to what extent each of a number of selected covariates is associated with an outcome (i.e., a health behavior) by adjusting a number of other variables. The statistics provide quantitative estimates of the association between each covariate and the outcome. Thus, the evidence accumulated on the associations can facilitate the use of utility maximization in ABMs by choosing factors and assigning weights to each factor. However, this is problematic because the associations between a certain health behavior and covariates at the population level are unlikely to be the same as the way in which these covariates determine health behavior at the individual level. Thus, this pragmatic substitution is a suboptimal choice at best. The mechanism of social network is used commonly among addictive behaviors including smoking, drinking alcohol, and drug use. This reflects the significance of the social norm theory in research on addictive behaviors. One benefit of adopting social network mechanism is to take advantage of well-established network analysis technology and availability of real-word data [47]. The stages of change mechanism have a unique advantage by dividing one behavior change into a number of stages. Because each stage has different determinants that may influence to the behavior differently, an ABM including explicit stages of change could be used to examine and pinpoint the effect of interventions to each stage and to design tailored interventions accordingly. Unfortunately, this mechanism is used least frequently, which may be attributable to the dearth of knowledge about fine-grained health behavior and behavior change. Thus, dichotomizing a health behavior, rather than separating it into several stages, is another pragmatic simplification. Four ABMs used decision trees. Decision trees are a more straightforward way to mimic a human’s decision process (e.g., human’s heuristics, bounded rationality, and other cognitive biases) and are easier to interpret compared with utility maximization. Particularly, decision trees may benefit from the recent development of big data and data mining [53, 54]. As illustrated in several studies on travel mode choice [51, 55, 56], decision tree learning, a data-mining technique, is a powerful tool to apply to datasets to identify the patterns and connections between/among explanatory variables and behavior choice. The results of decision trees mining will result in a predictive model in the format of a “tree” that can be translated easily to ABMs’ mechanisms of behavior and behavior change. ABMS IN BEHAVIOR INTERVENTION One advantage of an ABM is that it can serve as a “virtual testbed” to examine an intervention’s effects through a series of “what-if” simulations, especially for certain interventions that are unethical and uneconomical. ABMs can inform interventions both prospectively (i.e., by showing and predicting their potential effect) and retrospectively (i.e., to understand the success or failure of a program or policy that was in place) [57]. According to the socioecological model [41], health behavior interventions may be implemented at multiple levels; however, studies of the assessment of multilevel interventions are limited [58, 59]. ABM has been recognized as a promising method to examine the way in which factors from multiple levels interact and influence health behavior and behavior change jointly [7, 60]. A review of the way ABMs used for behavior intervention may provide insights into practical intervention. As Table 1 shows, all 17 ABMs aimed to promote healthy behavior, reduce unhealthy behaviors, or examine the mechanisms that drive behavior patterns at the population level (e.g., disparities in health behaviors). Several ABMs addressed health disparities at the individual level, such as socioeconomic disparities in walking [13–15] and diet [20, 21, 25], and racial disparities in diet [22]. One ABM examined socioeconomic disparities in vegetable consumption at the neighborhood level [24]. Some ABMs were applied to hypothetical cases, whereas others were based on empirical data, for example, those for a specific school [18] or specific cities [19, 21, 23–28]. Special programs also have been investigated, such as walking school bus [16], mobile market distribution system [21], as well as ecstasy pill-testing and drug detection dogs [34–36]. As shown in Table 2, I evaluated the use of ABMs for behavior intervention on three levels: individual, interpersonal, and neighborhood environment. For the levels of individual and interpersonal, I singled out the aspects of financial incentives and social norms, respectively. Individual level and financial incentive aspect The aim of individual-level interventions on health behaviors is to encourage people to adopt healthy behaviors, especially lifestyle behaviors, by providing knowledge or changing their attitudes about certain behaviors. At this level, I focused on the aspect of financial incentive, one potentially effective method to promote healthy behavior change [61]. It is believed that people give more value to outcomes that occur sooner and more certainly than those occurring later and less certainly [62, 63]. Therefore, a financial incentive intervention has an advantage because the financial benefits and costs are concrete and immediate, compared with the delay and uncertainty of expected health benefits and costs [64]. A majority of the 17 ABMs implemented interventions at the individual level, such as promoting positive attitudes about walking and active travel to school [13, 14, 16, 17] and increasing the preference for healthy foods [20]. Four ABMs examined financial incentive interventions, such as adjusting parking fees and fuel prices to discourage driving and promote active travel [15], decreasing the price of healthy foods [20, 25], and increasing tax on unhealthy foods [23]. One ABM [20] found that both the increase in preferences for healthy foods and the reduction in their prices were necessary to address income disparities in a healthy diet. Interpersonal level and social norm aspect At the interpersonal level, given that a person’s behavior is likely to be influenced by her/his family members, friends, and other persons in her/his social network, interventions can take advantage of the network to promote the transmission of healthy behaviors. Here, I singled out the aspect of the social norm. Social norms are believed to be associated with health behavior [65] and are rooted in the fundamental idea that people believe in common standards for behaviors and that violation of these standards will incur penalties [66]. Thus, changing social norms positively (e.g., glorifying healthy behaviors and stigmatizing unhealthy ones) may shape behavior at the population level accordingly [65, 66]. Six ABMs modeled interventions at the interpersonal level. For example, one [18] found that targeting the most sedentary children was the best intervention to increase PA among sedentary children, whereas targeting the most connected children was the best intervention to increase PA among all children. Six ABMs examined the effect of social norm changes. For example, one ABM found that interventions that emphasized healthy eating norms were more effective than those that targeted food prices or regulated local food outlets directly [23]. Neighborhood environment level Recently, there has been a growing interest in behavior interventions that focus on the neighborhood environment, because of the recognition that individuals are products of their environment, and behavior change may not be sustainable if the environment is not conducive to healthy behavior [67]. Therefore, improving the neighborhood environment could promote multiple healthy behaviors and outcomes, and the benefits may be sustainable and throughout the entire population. Ten ABMs examined the effect of either environmental improvement or increased access to supporting facilities on certain behaviors. For example, Lemoine et al.’s ABM [19] assessed the way in which the rapid transit bus system influenced walking by varying the number of lanes and densities of bus stations. Another ABM found that interventions that focused on attitude toward walking did not work well if neighborhood environments were not conducive to walking [13]. INSIGHTS FROM THE USE OF ABMS IN BEHAVIOR INTERVENTION The 17 ABMs offered several insights. First, these models confirmed the benefits of upstream interventions. Generally, upstream determinants are defined as features of the social environment, such as socioeconomic status, education, housing, and neighborhood conditions, that influence health outcomes and behaviors fundamentally, but indirectly through long and complex pathways [68, 69]. Several upstream determinants have been covered by the reviewed ABMs, including neighborhood environment, social norms, and education. For example, one ABM [13, 14] found that positive attitudes about walking diminished over time if the neighborhood environment was not conducive to walking. Zhang et al. [23] found that interventions that emphasize healthy eating norms were more effective than were those that targeted food prices or regulated local food outlets directly. Orr et al. [22] found that improving school quality can reduce racial disparities in diet. ABMs’ advantage in addressing complex pathways among multiple intervening factors makes it a uniquely promising method to examine upstream interventions on health behavior. Second, a majority of ABMs that applied interventions from multiple levels or aspects found additional benefits of the interventions. For example, a synergistic effect was identified between the implementation of a walking school bus program and improvement in attitudes about walking [16, 17]. In contrast, to promote a healthy diet, interventions in both food preferences and prices were necessary to overcome the diet disparity generated by residential segregation [20]. Another example is that both spatial (e.g., use spatial optimization model to locate mobile food markets across food desert areas) and nonspatial strategies should be considered in efforts to improve diets among low-income households [21]. The third insight was the trade-off effect. One ABM [18] found that targeting the most sedentary children was the best way to increase their PA levels, but targeting the most connected children was the best to increase PA among all children. In another example, Blok et al. [25] found that eliminating residential segregation had the greatest effect in reducing income inequalities in food consumption, but may negatively influence high-income households by decreasing their consumption of healthy foods. Similarly, Scott et al. [26–28] found that additional hours between an establishment’s business closure and last call for drinks not only reduced harm for drinkers, but also minimized business revenues simultaneously. The fourth insight was the nonlinear and asymmetrical effect. Lemoine et al. [19] found that, although an increase in the number of public transit lines and stations increased walking, the increase reached a saturation point, such that, above a certain point, further increases in public transit lines and stations no longer increased walking. Another example is the asymmetrical social influence on smoking, where peer influence was found to be stronger in initiating smoking than in smoking cessation [31, 32]. The four insights above illustrated and highlighted the promise of ABMs. ABM can help us see the “bigger picture” and take advantage of the complex pathways among various determinants and outcomes to better design interventions. As discussed earlier, the first (upstream factors) and second (multiple levels or aspects) insights focused on the interactions among determinants of health behaviors, whereas the third insight addressed the interactions among health behaviors. ABM can be used to study the complex pathways from upstream to downstream determinants, and to study multiple outcomes simultaneously [70]. The fourth insight of the nonlinear and asymmetrical effect may guide interventions more precisely by determining which factors should be targeted, and how great an effect should be spent considering the balance in cost-effectiveness. CONCLUSIONS The review confirmed a growing interest in the use of ABM in health behavior and intervention, reflected by its application to a diversity of behaviors, populations, and interventions. Among the four behavior change mechanisms, utility maximization was used most frequently, and social network was used commonly among addictive behaviors. The mechanisms of decision trees and stages of change, despite their advantages in theory, were used less frequently in practice. ABMs have been used to examine interventions at the individual level, interpersonal level, neighborhood environment levels, as well as interventions involving financial incentives, social norms, and special programs and policies. Although ABM’s use in health behavior study is still in its early stages and ABM’s advantages have not been explored fully, important insights have emerged from its application in behavior intervention. ABM’ advantages in addressing complex pathways make it a uniquely promising method to examine the effect of multiple intervening factors on health behaviors, particularly, upstream and multilevel interventions, as well as balances among various factors, outcomes, and populations. A major challenge to the use of ABM in health behavior study is our limited knowledge of behaviors at the individual level. This is not surprising because of ABM’s flexibility in representing the real-world from the perspective of the modeler. However, ABM’s flexibility also demands a comprehensive knowledge base that may not exist yet. Most reviewed ABMs addressed this issue by either simplification or substitution. For example, the complicated human decision process (likely to be much more complicated than the decision trees mechanism) was modeled in ABMs as a process of computing and comparing utilities for each option. Also, the dynamic and ongoing behavior change was modeled in ABMs as a dichotomous process. Substitution, that is, using population-level knowledge in ABMs when no individual-level knowledge is available, has been applied in many ABMs. Substitution is problematic in theory and may discredit the use of ABMs in the long term. However, the potential problem and implications of using population-level data to model individual behavioral choices are not well understood. To address the above challenge, it is crucial for ABMs to integrate “rules” from various fields and synthesizes various data sources. One important but relatively neglected data source is qualitative data. Qualitative data usually are rich and detailed with respect to relationships, individual decisions, and other dynamic processes that can inform ABMs’ basic mechanisms [71]. However, qualitative data’s degree of involvement in ABMs is limited, although, for a decade, it has been advocated [71–74]. Recently, health researchers have recognized the potentials of big data, for example, the rapidly accumulating availability of smartphones, sensors, and other types of digital, time-stamped, and contextualized data. Big data may improve our understanding of health behaviors by providing insights into the causal relationships [53] because of big data’s advantage in detecting patterns and relationships in large and complex datasets [75]. Accordingly, we need more advanced analysis methods and behavior theories to capture the richness of big data [76]. Overall, ABMs have been used to model a diversity of behaviors, populations, and interventions. The insights obtained from the reviewed ABMs have illustrated the promise of ABMs. However, the use of ABMs in health behavior is at an early stage, and a major challenge is our limited knowledge of behaviors at the individual level. The qualitative research and big data may help to address this challenge. Acknowledgements: The author thanks the anonymous reviewers for their valuable comments and help to improve the quality of the article. Compliance with Ethical Standards Funding: None. Conflict of Interest: The author has no conflict interest. Ethical approval: Not applicable because no humans or animal were involved. References 1. Committee on Health and Behavior: Research, Practice, and Policy, Board on Neuroscience and Behavioral Health, Institute of Medicine . Health and Behavior: The Interplay of Biological, Behavioral, and Societal Influences . Washington, DC : National Academy Press ; 2001 . WorldCat COPAC 2. Schroeder SA . Shattuck lecture. We can do better – improving the health of the American people . N Engl J Med . 2007 ; 357 ( 12 ): 1221 – 1228 . Google Scholar Crossref Search ADS PubMed WorldCat 3. World Health Organization . Report of the Commission on Ending Childhood Obesity . Geneva, Switzerland : World Health Organization ; 2017 . WorldCat COPAC 4. Campbell MK . Biological, environmental, and social influences on childhood obesity . Pediatr Res . 2016 ; 79 ( 1–2 ): 205 – 211 . Google Scholar Crossref Search ADS PubMed WorldCat 5. Sweet SN , Fortier MS . Improving physical activity and dietary behaviours with single or multiple health behaviour interventions? A synthesis of meta-analyses and reviews . Int J Environ Res Public Health . 2010 ; 7 ( 4 ): 1720 – 1743 . Google Scholar Crossref Search ADS PubMed WorldCat 6. Thomas S, Fayter D, Misso K, et al . Population tobacco control interventions and their effects on social inequalities in smoking: Systematic review . Tob Control . 2008 ; 17(4):230–237 . WorldCat 7. Spruijt-Metz D, Hekler E, Saranummi N, et al . Building new computational models to support health behavior change and maintenance: New opportunities in behavioral research . Transl Behav Med . 2015 ;5(3):335–346. WorldCat 8. Bonabeau E . Agent-based modeling: Methods and techniques for simulating human systems . Proc Natl Acad Sci USA . 2002 ; 99 (Suppl 3): 7280 – 7287 . Google Scholar Crossref Search ADS WorldCat 9. Nianogo RA , Arah OA . Agent-based modeling of noncommunicable diseases: A systematic review . Am J Public Health . 2015 ; 105 ( 3 ): e20 – e31 . Google Scholar Crossref Search ADS PubMed WorldCat 10. Li Y , Lawley MA , Siscovick DS , Zhang D , Pagán JA . Agent-based modeling of chronic diseases: A narrative review and future research directions . Prev Chronic Dis . 2016 ; 13 : E69 . Google Scholar PubMed WorldCat 11. Carey G , Malbon E , Carey N , Joyce A , Crammond B , Carey A . Systems science and systems thinking for public health: A systematic review of the field . BMJ Open . 2015 ; 5 ( 12 ): e009002 . Google Scholar Crossref Search ADS PubMed WorldCat 12. Skinner AC , Foster EM . Systems science and childhood obesity: A systematic review and new directions . J Obesity . 2013; doi:10.1155/2013/129193 . WorldCat 13. Yang Y , Diez Roux AV , Auchincloss AH , Rodriguez DA , Brown DG . Exploring walking differences by socioeconomic status using a spatial agent-based model . Health Place . 2012 ; 18 ( 1 ): 96 – 99 . Google Scholar Crossref Search ADS PubMed WorldCat 14. Yang Y , Diez Roux AV , Auchincloss AH , Rodriguez DA , Brown DG . A spatial agent-based model for the simulation of adults’ daily walking within a city . Am J Prev Med . 2011 ; 40 ( 3 ): 353 – 361 . Google Scholar Crossref Search ADS PubMed WorldCat 15. Yang Y , Auchincloss AH , Rodriguez DA , Brown DG , Riolo R , Diez-Roux AV . Modeling spatial segregation and travel cost influences on utilitarian walking: Towards policy intervention . Comput Environ Urban Syst . 2015 ; 51 : 59 – 69 . Google Scholar Crossref Search ADS PubMed WorldCat 16. Yang Y , Diez-Roux A , Evenson KR , Colabianchi N . Examining the impact of the walking school bus with an agent-based model . Am J Public Health . 2014 ; 104 ( 7 ): 1196 – 1203 . Google Scholar Crossref Search ADS PubMed WorldCat 17. Yang Y , Diez-Roux AV . Using an agent-based model to simulate children’s active travel to school . Int J Behav Nutr Phys Act . 2013 ; (10):67 . WorldCat 18. Zhang J , Shoham DA , Tesdahl E , Gesell SB . Network interventions on physical activity in an afterschool program: An agent-based social network study . Am J Public Health . 2015 ; 105 ( Suppl 2 ): S236 – S243 . Google Scholar Crossref Search ADS PubMed WorldCat 19. Lemoine PD , Cordovez JM , Zambrano JM , et al. Using agent based modeling to assess the effect of increased bus rapid transit system infrastructure on walking for transportation . Prev Med . 2016 ; 88 : 39 – 45 . Google Scholar Crossref Search ADS PubMed WorldCat 20. Auchincloss AH , Riolo RL , Brown DG , Cook J , Diez Roux AV . An agent-based model of income inequalities in diet in the context of residential segregation . Am J Prev Med . 2011 ; 40 ( 3 ): 303 – 311 . Google Scholar Crossref Search ADS PubMed WorldCat 21. Widener MJ , Metcalf SS , Bar-Yam Y . Agent-based modeling of policies to improve urban food access for low-income populations . Appl Geogr . 2013 ; 40 : 1 – 10 . Available at https://www.sciencedirect.com/science/article/pii/S0143622813000283. Google Scholar Crossref Search ADS WorldCat 22. Orr MG , Galea S , Riddle M , Kaplan GA . Reducing racial disparities in obesity: Simulating the effects of improved education and social network influence on diet behavior . Ann Epidemiol . 2014 ; 24 ( 8 ): 563 – 569 . Google Scholar Crossref Search ADS PubMed WorldCat 23. Zhang D , Giabbanelli PJ , Arah OA , Zimmerman FJ . Impact of different policies on unhealthy dietary behaviors in an urban adult population: An agent-based simulation model . Am J Public Health . 2014 ; 104 ( 7 ): 1217 – 1222 . Google Scholar Crossref Search ADS PubMed WorldCat 24. Li Y , Zhang D , Pagán JA . Social norms and the consumption of fruits and vegetables across New York City neighborhoods . J Urban Health . 2016 ; 93 ( 2 ): 244 – 255 . Google Scholar Crossref Search ADS PubMed WorldCat 25. Blok DJ , de Vlas SJ , Bakker R , van Lenthe FJ . Reducing income inequalities in food consumption: Explorations with an agent-based model . Am J Prev Med . 2015 ; 49 ( 4 ): 605 – 613 . Google Scholar Crossref Search ADS PubMed WorldCat 26. Scott N , Livingston M , Reporter I , Dietze P . Using simulation modelling to examine the impact of venue lockout and last-drink policies on drinking-related harms and costs to licensees . Aust N Z J Public Health . 2017 ; 41 ( 3 ): 243 – 247 . Google Scholar Crossref Search ADS PubMed WorldCat 27. Scott N, Hart A, Wilson J, Livingston M, Moore D, Dietze P . The effects of extended public transport operating hours and venue lockout policies on drinking-related harms in Melbourne, Australia: Results from SimDrink, an agent-based simulation model . Int J Drug Policy . 2016 ;32:44–49. Available at https://www.sciencedirect.com/science/article/abs/pii/S0955395916300251?via%3Dihub WorldCat 28. Scott N, Livingston M, Hart A, Wilson J, Moore D, Dietze P . SimDrink: An agent-based NetLogo model of young, heavy drinkers for conducting alcohol policy experiments . J Artif Soc Soc Simul . 2016 ; 19 (10). WorldCat 29. Fitzpatrick BG , Martinez J , Polidan E , Angelis E . On the effectiveness of social norms intervention in college drinking: The roles of identity verification and peer influence . Alcohol Clin Exp Res . 2016 ; 40 ( 1 ): 141 – 151 . Google Scholar Crossref Search ADS PubMed WorldCat 30. Fitzpatrick B, Martinez J, Polidan E, Angelis E . The big impact of small groups on college drinking . J Artif Soc Soc Simul . 2015 ; 18 ( 3 ):4. WorldCat 31. Haas SA , Schaefer DR . With a little help from my friends? Asymmetrical social influence on adolescent smoking initiation and cessation . J Health Soc Behav . 2014 ; 55 ( 2 ): 126 – 143 . Google Scholar Crossref Search ADS PubMed WorldCat 32. Schaefer DR , Adams J , Haas SA . Social networks and smoking: Exploring the effects of peer influence and smoker popularity through simulations . Health Educ Behav . 2013 ; 40 ( Suppl 1 ): 24S – 32S . Google Scholar Crossref Search ADS PubMed WorldCat 33. Luke DA , Hammond RA , Combs T , et al. Tobacco town: Computational modeling of policy options to reduce tobacco retailer density . Am J Public Health . 2017 ; 107 ( 5 ): 740 – 746 . Google Scholar Crossref Search ADS PubMed WorldCat 34. Moore D , Dray A , Green R , et al. Extending drug ethno-epidemiology using agent-based modelling . Addiction . 2009 ; 104 ( 12 ): 1991 – 1997 . Google Scholar Crossref Search ADS PubMed WorldCat 35. Perez P , Dray A , Moore D , et al. SimAmph: An agent-based simulation model for exploring the use of psychostimulants and related harm amongst young Australians . Int J Drug Policy . 2012 ; 23 ( 1 ): 62 – 71 . Google Scholar Crossref Search ADS PubMed WorldCat 36. Dray A , Perez P , Moore D , et al. Are drug detection dogs and mass-media campaigns likely to be effective policy responses to psychostimulant use and related harm? Results from an agent-based simulation model . Int J Drug Policy . 2012 ; 23 ( 2 ): 148 – 153 . Google Scholar Crossref Search ADS PubMed WorldCat 37. Bobashev G, Goree S, Frank J, Zule W . Pain Town: An Agent-Based Model of Opioid Use Trajectories in a Small Community . Cham, Switzerland : Springer International Publishing ; 2018 . Google Preview WorldCat COPAC 38. Resnicow K , Page SE . Embracing chaos and complexity: A quantum change for public health . Am J Public Health . 2008 ; 98 ( 8 ): 1382 – 1389 . Google Scholar Crossref Search ADS PubMed WorldCat 39. Michie S , van Stralen MM , West R . The behaviour change wheel: A new method for characterising and designing behaviour change interventions . Implement Sci . 2011 ; 6 : 42 . Google Scholar Crossref Search ADS PubMed WorldCat 40. Glanz K , Bishop DB . The role of behavioral science theory in development and implementation of public health interventions . Annu Rev Public Health . 2010 ; 31 : 399 – 418 . Available at https://www.annualreviews.org/doi/full/10.1146/annurev.publhealth.012809.103604?url_ver=Z39.88-2003&rfr_id=ori%3Arid%3Acrossref.org&rfr_dat=cr_pub%3Dpubmed. Google Scholar Crossref Search ADS PubMed WorldCat 41. McLeroy KR , Bibeau D , Steckler A , Glanz K . An ecological perspective on health promotion programs . Health Educ J . 1988 ; 15 ( 4 ): 351 – 377 . WorldCat 42. Huang TT , Drewnosksi A , Kumanyika S , Glass TA . A systems-oriented multilevel framework for addressing obesity in the 21st century . Prev Chronic Dis . 2009 ; 6 ( 3 ): A82 . Google Scholar PubMed WorldCat 43. Yang Y . A dynamic framework on travel mode choice focusing on utilitarian walking based on the integration of current knowledge . J Transp Health . 2016;3(3):336–345 . WorldCat 44. Sen A . Rational behavior . In: Durlauf SN , Blume LE , eds. The New Palgrave Dictionary of Economics . London, UK: Palgrave Macmillan; 2008:117 . Google Preview WorldCat COPAC 45. Maslow AH . A theory of human motivation . Psychol Rev . 1943 ; 50 ( 4 ): 370 – 396 . Google Scholar Crossref Search ADS WorldCat 46. Berkowitz A . An overview of the social norms approach . In: Lederman L, Stewart L, Goodhart F, Laitman L, eds. Changing the Culture of College Drinking. Cresskill, NJ: Hampton Press; 2005:187–208. COPAC 47. Luke DA , Harris JK . Network analysis in public health: History, methods, and applications . Annu Rev Public Health . 2007 ; 28 : 69 – 93 . Google Scholar Crossref Search ADS PubMed WorldCat 48. Prochaska JO , Velicer WF . The transtheoretical model of health behavior change . Am J Health Promot . 1997 ; 12 ( 1 ): 38 – 48 . Google Scholar Crossref Search ADS PubMed WorldCat 49. Wallace R , Geller A , Ogawa VA. Assessing the Use of Agent-Based Models for Tobacco Regulation . Committee on the Assessment of Agent-Based Models to Inform Tobacco Product Regulation; Board on Population Health and Public Health Practice. Washington, DC: Institute of Medicine National Academies Press ; 2015 . Google Scholar Crossref Search ADS Google Preview WorldCat COPAC 50. Ben-Akiv ME , Lerman SR. Discrete Choice Analysis: Theory and Application to Travel Demand . Cambridge, MA: MIT Press ; 1985 . Google Preview WorldCat COPAC 51. Lu YD , Kawamura K . Data-mining approach to work trip mode choice analysis in Chicago, Illinois, area . Transp Res Rec . 2010 ; 2156 : 73 – 80 . Available at https://trrjournalonline.trb.org/doi/abs/10.3141/2156-09. Google Scholar Crossref Search ADS WorldCat 52. Idris A , Shalaby A , Habib K . Towards a learning-based mode shift model: A conceptual framework . Transp Lett . 2012 ; 4 ( 1 ): 15 – 27 . Google Scholar Crossref Search ADS WorldCat 53. Khoury MJ , Ioannidis JP . Medicine. Big data meets public health . Science . 2014 ; 346 ( 6213 ): 1054 – 1055 . Google Scholar Crossref Search ADS PubMed WorldCat 54. Baqueiro O, Wang YJ, McBurney P, Coenen F . Integrating data mining and agent based modeling and simulation . In: Perner P , ed. Advances in Data Mining, Applications and Theoretical Aspects . Heidelberg, Germany : Springer ; 2009 : 220 – 231 . Google Preview WorldCat COPAC 55. Arentze T , Timmermans H . Parametric action decision trees: Incorporating continuous attribute variables into rule-based models of discrete choice . Transp Res B: Methodol . 2007 ; 41 ( 7 ): 772 – 783 . Google Scholar Crossref Search ADS WorldCat 56. Xie C , Lu J , Parkany E . Work travel mode choice modeling with data mining: Decision trees and neural networks . Transp Res Rec . 2003 ; 1854 : 50 – 61 . Google Scholar Crossref Search ADS WorldCat 57. Hammond RA . Considerations and best practices in agent-based modeling to inform policy . In: Institute of Medicine , ed. Assessing the Use of Agent-Based Models for Tobacco Regulation . Washington, DC : National Academies Press ; 2015 : 161 – 193 . Google Preview WorldCat COPAC 58. Weiner BJ , Lewis MA , Clauser SB , Stitzenberg KB . In search of synergy: Strategies for combining interventions at multiple levels . J Natl Cancer Inst Monogr . 2012 ; 2012 ( 44 ): 34 – 41 . Google Scholar Crossref Search ADS PubMed WorldCat 59. Cleary PD , Gross CP , Zaslavsky AM , Taplin SH . Multilevel interventions: Study design and analysis issues . J Natl Cancer Inst Monogr . 2012 ; 2012 ( 44 ): 49 – 55 . Google Scholar Crossref Search ADS PubMed WorldCat 60. King AC , Satariano WA , Marti J , Zhu W . Multilevel modeling of walking behavior: Advances in understanding the interactions of people, place, and time . Med Sci Sports Exerc . 2008 ; 40 ( Suppl 7 ): S584 – S593 . Google Scholar Crossref Search ADS PubMed WorldCat 61. Giles EL , Robalino S , McColl E , Sniehotta FF , Adams J . The effectiveness of financial incentives for health behaviour change: Systematic review and meta-analysis . PLoS One . 2014 ; 9 ( 3 ): e90347 . Google Scholar Crossref Search ADS PubMed WorldCat 62. Dickhaut J , McCabe K , Nagode JC , Rustichini A , Smith K , Pardo JV . The impact of the certainty context on the process of choice . Proc Natl Acad Sci USA . 2003 ; 100 ( 6 ): 3536 – 3541 . Google Scholar Crossref Search ADS PubMed WorldCat 63. Frederick S , Loewenstein G , O’Donoghue T . Time discounting and time preference: A critical review . J Econ Lit . 2002 ; 40 ( 2 ): 351 – 401 . Google Scholar Crossref Search ADS WorldCat 64. Loewenstein G , Brennan T , Volpp KG . Asymmetric paternalism to improve health behaviors . JAMA . 2007 ; 298 ( 20 ): 2415 – 2417 . Google Scholar Crossref Search ADS PubMed WorldCat 65. Ball K , Jeffery RW , Abbott G , McNaughton SA , Crawford D . Is healthy behavior contagious: Associations of social norms with physical activity and healthy eating . Int J Behav Nutr Phys Act . 2010 ; 7 : 86 . Google Scholar Crossref Search ADS PubMed WorldCat 66. Reid AE , Cialdini RB , Aiken LS . Social norms and health behavior . In: Steptoe A , ed. Handbook of Behavioral Medicine: Methods and Applications . New York, NY : Springer New York ; 2010 : 263 – 274 . Google Scholar Crossref Search ADS Google Preview WorldCat COPAC 67. Cutler DM . Behavioral health interventions: What works and why ? In: Anderson NB, Bulatao RA, Cohen B, eds. Ethnic Disparities in Aging Health. Washington, DC: National Academies Press; 2002:643–676. COPAC 68. Link BG , Phelan J . Social conditions as fundamental causes of disease . J Health Soc Behav . 1995 ; Spec No : 80 – 94 . Google Scholar Crossref Search ADS PubMed WorldCat 69. Gehlert S , Sohmer D , Sacks T , Mininger C , McClintock M , Olopade O . Targeting health disparities: A model linking upstream determinants to downstream interventions . Health Aff (Millwood) . 2008 ; 27 ( 2 ): 339 – 349 . Google Scholar Crossref Search ADS PubMed WorldCat 70. Yang Y . Using agent-based modeling to study multiple risk factors and multiple health outcomes at multiple levels . Ann N Y Acad Sci . 2017 ; 1408 ( 1 ): 7 – 14 . Google Scholar Crossref Search ADS PubMed WorldCat 71. Yang L , Gilbert N . Getting away from numbers: Using qualitative observation for agent-based modeling . Adv Complex Syst . 2008 ; 11 ( 2 ): 175 – 185 . Google Scholar Crossref Search ADS WorldCat 72. Ghorbani A , Dijkema G , Schrauwen N . Structuring qualitative data for agent-based modelling . J Artif Soc Soc Simul . 2015 ; 18 ( 1 ): 2 . Google Scholar Crossref Search ADS WorldCat 73. Bruch E , Atwell J . Agent-based models in empirical social research . Sociol Methods Res . 2015 ; 44 ( 2 ): 186 – 221 . Google Scholar Crossref Search ADS PubMed WorldCat 74. Tubaro P , Casilli AA . “An ethnographic seduction”: How qualitative research and agent-based models can benefit each other . Bull Sociol Methodol . 2010 ; 106 ( 1 ): 59 – 74 . Google Scholar Crossref Search ADS WorldCat 75. Gandomi A , Haider M . Beyond the hype: Big data concepts, methods, and analytics . Int J Inf Manage . 2015 ; 35 ( 2 ): 137 – 144 . Google Scholar Crossref Search ADS WorldCat 76. Atienza AA , Serrano KJ , Riley WT , Moser RP , Klein WM . Advancing cancer prevention and behavior theory in the era of big data . J Cancer Prev . 2016 ; 21 ( 3 ): 201 – 206 . Google Scholar Crossref Search ADS PubMed WorldCat © Society of Behavioral Medicine 2019. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com. This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/open_access/funder_policies/chorus/standard_publication_model)
TI - A narrative review of the use of agent-based modeling in health behavior and behavior intervention
JF - Translational Behavioral Medicine
DO - 10.1093/tbm/iby132
DA - 2019-11-25
UR - https://www.deepdyve.com/lp/oxford-university-press/a-narrative-review-of-the-use-of-agent-based-modeling-in-health-oi8ZpWC1HT
SP - 1065
VL - 9
IS - 6
DP - DeepDyve
ER -