System Dynamics and Agent-Based Models Applied to Public Health Problems
MetadataShow full item record
During the past decades, there has been a growing body of research on the development of new methodologies in system sciences in public health. While systems thinking is prevalent in the practice of public health, there is a need for tools to quantify the multidimensional and multidisciplinary aspects of such thinking. In this thesis, we focus on two system science methods: Agent-Based Modeling (ABM) and System Dynamics (SD). We begin with an ABM to simulate the effects of an urban food desert environment on school-aged children. The data that was used to inform this model is based on children in low-income neighborhoods of Baltimore City. The baseline model was used to predict changes in body mass index due to eating behaviors of simulated children interacting with their food environment. The model was then used as a virtual social laboratory by introducing interventions into the environment and assessing their effects on child behaviors and weight gains. For our second application of systems science, we developed an SD model to study the stability of community functioning (CF) after a natural disaster. We define CF as a measure of a broad range of community activities in providing services to its residents. We studied the dynamic response of CF post-disaster from two different aspects: resilience, which indicates the speed of recovery after event, and resistance, which measures the degree to which a community can continue to function during the event. Key components that support or reduce CF were identified and were quantified as variables in a system of ordinary differential equations. The data for the model was obtained at the county level for 3143 United States counties, and the model results for resilience and resistance ratings were presented in a series of maps so that the regional patterns of our findings could be visualized. Finally, our last application was an SD model applied in a different public health context: an analysis of the mechanisms underlying the health system in Afghanistan between 2010 and 2012. We were interested in the Pay-for-Performance (P4P) intervention, in which relatively small health facilities were given bonus payments as a reward for year-to-year improvements in quality and quantity of health services. A recently published data analysis of the P4P intervention showed no improvement in health services. By working with some of the researchers who participated in this intervention, we were able to develop causal loops in the system associated with some of the key interactions that were generated within the health facilities. We then synthesized these loops into a model of differential equations with delays. We were able to generate several scenarios that indicate that the failure of P4P may be caused by poor implementation processes and gaming within the system. In summary, we demonstrate how ABM and SD can naturally embody systems thinking into a quantitative form, and can produce a wide array of numerical and visual results that capture the complex processes that characterize public health.