Simulation framework tailors infectious disease policy to local data

Researchers have unveiled a modeling framework that turns sparse routine surveillance data into region-specific guidance for infectious disease control, addressing a persistent gap in public health planning. The study, published in BMC Public Health, applies the method to COVID-19 in nine demographically distinct Michigan counties and finds that the ranking and robustness of interventions differ substantially from one county to another, in ways that simple demographic predictors do not reliably anticipate.

The central challenge the team set out to solve is that agent-based models, which simulate individuals and the contact structures that drive local transmission, are well suited to analyzing local interventions but are computationally expensive and difficult to fit directly to real-world data. Rather than fitting the agent-based model itself, the framework uses an ordinary differential equation surrogate model as an intermediary. The surrogate is calibrated to observed case and hospitalization data using weighted least squares, and profile likelihood is then used to define a region of agent-based model parameters that remains consistent with the data.

Intervention effects are evaluated across this entire data-consistent parameter region rather than at a single best-fit point, while the stochastic variability inherent to the agent-based model is quantified separately through repeated simulations. This explicit propagation of uncertainty allows the researchers to report not just which intervention performs best on average, but how confident that conclusion is for each county. For every county in the case study, the team recovered parameter regions consistent with local case and hospitalization data and verified that the calibrated model reproduced the observed epidemic trajectories.

With the calibrated models in hand, the researchers evaluated four non-pharmaceutical interventions across each county’s parameter region. The results underscored the limits of one-size-fits-all policy: interventions that ranked highly in one county could prove fragile or less effective in another, and no simple demographic summary of a county reliably predicted which strategy would succeed there. The authors note that the intervention implementation assumptions were drawn from the COVID-19 literature, but that the framework is general and can accept user-specified assumptions to evaluate the conclusions those assumptions imply.

Because the approach couples mechanistically rich agent-based models with formal uncertainty quantification, the researchers describe it as a generic route to local, uncertainty-aware intervention analysis that could extend to influenza, measles, and other pathogens. For public health officials working with limited data, the framework offers a way to extract region-specific, decision-ready insights from the case counts and hospitalization figures already collected routinely, without requiring the intensive data streams that direct agent-based calibration would demand.

Subject of Research: Region-specific evaluation of infectious disease interventions using surrogate-calibrated agent-based simulation

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

Article References: Dudley, C., Bergman, D., Jain, H., Norton, K.-A., Rutter, E., Eisenberg, M., & Jackson, T. (2026). From sparse data to smart decisions: region-specific policy evaluation via simulation. BMC Public Health. https://doi.org/10.1186/s12889-026-29919-0

Image Credits: AI Generated

DOI: 10.1186/s12889-026-29919-0

Keywords: agent-based modeling, COVID-19, surrogate models, uncertainty quantification, non-pharmaceutical interventions, public health policy, Michigan counties, surveillance data, profile likelihood, mathematical epidemiology, sparse, data