Modeling plays a major role in policy making, especially for infectious disease interventions but such models can be complex and computationally intensive. In order to improve confidence in model-based conclusions, it is necessary to gain a thorough understanding of the system and assess how model assumptions and parameters alter the results. Surrogate models can be very useful for this purpose since they can be readily explored.
In Willem et al. [1] we used Pareto-aware symbolic regression for nonlinear response surface modeling with automatic feature selection. First, we analyzed the input-response data from an open source individual-based model for pandemic influenza, called FluTE [2]. Second, we used surrogate modeling techniques on input-response data from a deterministic model to explore cost-effectiveness of varicella-zoster virus vaccination [3]. The model included empirical observations on social mixing patterns with childhood and adult vaccination strategies. In total, the age-structured dynamic transmission model for the United Kingdom [27] contained 185 input parameters, including 100 correlated transmission rates. In the emulators below, the predicted response plots are shown for each parameter with all other parameters fixed.
Explore response surfaces based on FluTE [1] simulations to predict the cumulative clinical attack rate. |
Analyze the driving parameters of a transmission model for varicella-zoster virus vaccination [3] to estimate the incremental gain of Quality Adjusted Life Years (QALY). |
Test three instances of our FluTE emulator to enhance dissemination to policy makers. |