A Microsimulation Model to Assess the Cost-Effectiveness of Physical Activity Policies Among US Adults: The Physical Activity, Diabetes, and Cardiovascular Disease Model

INTRODUCTION: We developed a new, individual-level microsimulation model, the Physical Activity, Diabetes, and Cardiovascular disease model, to project the long-term health and economic consequences of population-level interventions for increasing physical activity (PA) among US adults. METHODS: The Physical Activity, Diabetes, and Cardiovascular disease model predicts the effect of changes in PA on the risk of developing type 2 diabetes (T2D) and cardiovascular disease (CVD), as well as subsequent effects on health care costs, health-related quality of life, and mortality. Risk equations for incident T2D and CVD were obtained from the literature, and the simulation population was based on nationally representative survey data. Changes in T2D and CVD risk by increased PA level were estimated using dose-response curves derived from published meta-analyses. We conducted 4 types of model validation to confirm the reliability and accuracy of results. RESULTS: We validated the overall modeling approach and parameters with an external panel of experts at 3 separate points throughout model development (face validity). We observed the expected reductions in risk (internal validity), compared reductions with at least one other comparable model (cross-validity), and compared incidence results to national rates of CVD and diabetes using risk equations and person-level data from the National Health and Nutrition Examination Survey (external validity). CONCLUSION: Researchers and policymakers can employ the model to estimate the long-term benefits and cost-effectiveness of PA interventions for preventing T2D and CVD. The strengths of the Physical Activity, Diabetes and Cardiovascular disease model include (1) up-to-date, high-quality data; (2) using microsimulation to capture long-term intervention effects accurately; (3) using probabilistic sensitivity analysis to define the uncertainty of underlying parameters and reflect this uncertainty in model results.