We reviewed the use of agent-based modeling (ABM), a systems science method, in understanding noncommunicable diseases (NCDs) and their public health risk factors. We systematically reviewed studies in PubMed, ScienceDirect, and Web of Sciences published from January 2003 to July 2014. We retrieved 22 relevant articles; each had an observational or interventional design. Physical activity and diet were the most-studied outcomes. Often, single agent types were modeled, and the environment was usually irrelevant to the studied outcome. Predictive validation and sensitivity analyses were most used to validate models. Although increasingly used to study NCDs, ABM remains underutilized and, where used, is suboptimally reported in public health studies. Its use in studying NCDs will benefit from clarified best practices and improved rigor to establish its usefulness and facilitate replication, interpretation, and application. There has been an increasing interest in using systems science approaches such as agent-based modeling (ABM) to investigate and understand complex public health problems. 1–4 Complex systems are systems that are not fully explained by just understanding the individual elements of the system. 4 In other words, these systems cannot be reduced to their component parts because of the interactions among the parts. 5 Complex systems are made of heterogeneous elements or agents (e.g., individuals, organizations) whose interactions with one another yield an unpredictable yet organized emerging behavior that can persist over time. 5–7 When agents are capable of adapting to changing circumstances, the systems are said to be adaptive and thus called complex adaptive systems (CAS). 7,8 Examples of such complex systems include stock markets, insect colonies, immune systems, social systems, traffic jams, epidemics, and pandemics. All these phenomena have been studied in various fields such as economy, ecology, molecular biology, sociology, and epidemiology. 5,9 Noncommunicable diseases (NCDs) are by far the leading cause of mortality in the world, killing 36 million people in 2008 worldwide, which accounted for about 63% of all deaths. 10 Cardiovascular diseases, cancer, chronic respiratory diseases, and diabetes represent about 80% of all NCD deaths. 10 These diseases constitute a huge health and economic burden across the world. Four main behavioral risk factors—tobacco use, physical inactivity, unhealthy diet, and harmful use of alcohol—are responsible for most NCDs. 10 Noncommunicable diseases are diseases that are not passed from person to person 10 and can have a chronic or acute progression. 11 They differ from chronic diseases in that the latter can be communicable or not and they require a long-term management. 11 The study of NCDs can be recast as one of complex systems. Noncommunicable diseases are caused by factors that are influenced by one’s individual behaviors as well as interaction with the physical, social, or economic environment. 12–14 Researchers have described obesity as a health problem that exhibits attributes that are characteristic of a CAS and have argued that techniques used to model such systems can and should be used to model obesity. 15 Importantly, obesity involves substantial diversity and heterogeneity in relevant actors at many different levels of scales (e.g., individuals, communities, policy), with a multiplicity of mechanisms in which actors interact with one another with dynamic feedback loops and changes over time. 2,15,16 To study complex systems, traditional analysis (e.g., multivariate analyses) will often not suffice. The latter often assumes linearity (at least on some scale), normality, homogeneity, and independence between individuals and over time, and is concerned with variables often representing a single-level system. 4 This type of analysis is said to be reductionist or top-down. 4 In contrast, complex systems are often nonlinear, nonnormal, and involve heterogeneous actors or agents that interact at different levels with possibility of dynamic feedback loops. These systems approaches are said to be holistic and, in particular, bottom-up in the case of ABM. 17 Besides ABM, other key systems science approaches have been developed to study complex systems and include systems dynamics and network analysis, as well as discrete event simulation. 4,18 Briefly, system dynamics uses computer simulation models to uncover and understand endogenous sources of complex system behavior. 4 They are based on the premise that complex behaviors of a system result from the interplay of feedback loops, stocks, and flows that all occur within the bounded endogenous system. 4,19 Unlike ABM, which is an individual-based modeling technique, systems dynamics is an aggregate-level modeling type. Network analysis, on the other hand, focuses on the measurement and analysis of relationships and flows among a set of actors. 4 Discrete event simulation is a type of modeling simulatio
展开▼
