A system for systems epidemiology: The example of inference from agent-based models

The past 10 years have seen a growth in interest in systems approaches in non-infectious disease epidemiology. A fair body of literature in the field has discussed how these methods, in theory, can capture non-linearity, dynamic feedback loops, and emergence in policy-relevant epidemiologic research. However, while calls for these approaches have grown, empiric research that has effectively used these approaches to yield new insight regarding the etiology of high-burden diseases lags behind. Systems epidemiology is hampered by a paucity of guidelines, best practices, and illustrative applications of systems tools to epidemiologic questions. Agent-based modeling is one such systems tool with promise for meaningful applications, yet lacking with respect to a methodologic framework for valid implementation. Here, I establish a conceptual paradigm to designing, parameterizing, and implementing agent-based models in epidemiology. Using this conceptual paradigm, I compare and contrast differing approaches to the parameterization and validation of agent-based models via the design and implementation of two agent-based epidemiologic models, one exploring the role of social networks in obesity, and the other exploring the relationship between collective efficacy and community violence. With robust methodologic frameworks, systems epidemiology has the potential to expand our toolkit with respect to understanding the causes and consequences of chronic disease.