From sparse data to smart decisions: region-specific policy evaluation via simulation

Effective infectious disease response requires region-specific guidance, because the contact structures that drive local transmission and shape intervention effects vary from community to community. Agent-based models (ABMs) represent these structures and the individual heterogeneity underlying them, which makes them well suited to local intervention analysis, but they are computationally expensive and difficult to fit directly to data. We present a framework that avoids direct ABM fitting by using a surrogate ordinary differential equation model as an intermediary between routine surveillance data and a mechanistically rich ABM. The surrogate is fit to observed case and hospitalization data using weighted least squares, with profile likelihood used to define a data-consistent ABM parameter region. Intervention effects are evaluated across this region, while stochastic ABM variability is quantified separately through repeated simulations. We apply the framework to COVID-19 in nine demographically distinct Michigan counties as an illustrative case study. For each county, we recover ABM parameter regions consistent with local case and hospitalization data, verify that the calibrated model reproduces observed trajectories, and evaluate four non-pharmaceutical interventions across the ABM parameter regions. The ranking and robustness of interventions differ substantially across counties, and simple demographic predictors do not reliably predict which will succeed. Although we use intervention implementation assumptions drawn from COVID-19 literature, the framework is general and can take user-specified assumptions to evaluate the conclusions they imply. By coupling mechanistic ABMs with explicit uncertainty propagation, this approach provides a generic route to local, uncertainty-aware intervention analysis for COVID-19, influenza, measles, and other pathogens.