Publication Details
Ioana Alexandra Cimpean, Lander Willem, Verstraeten, Timothy, Niel Hens, , Libin, Pieter

Contribution To Conference


Individual-based epidemiological models (IBMs) support the study of fine-grained preventive measures, such as tailored vaccine allocation policies, in silico. As IBMs are computationally intensive, it is pivotal to identify optimal strategies using a minimal amount of model evaluations. Moreover, due to the high societal impact of enforcing preventive strategies, uncertainty regarding decisions should be communicated to policy makers, which is naturally embedded in a Bayesian approach. We contribute a novel technique to evaluate vaccine allocation strategies using a multi-armed bandit framework in combination with a Bayesian anytime m-top exploration algorithm. m-top exploration allows the algorithm to learn m policies for which it expects the highest utility, enabling the experts to inspect this small set of alternative strategies, along with their quantified uncertainty for the decision making. The anytime component provides decision makers with flexibility regarding the computation time and the desired confidence, which is important as it is difficult to make this trade-off beforehand. We consider the Belgian COVID-19 epidemic using the STRIDE IBM, where we learn a set of optimal vaccination policies that minimize the number of hospitalisations. In this scenario, the shape of model outcome distribution cannot be assumed analytically a priori, to which end we use a Gaussian mixture to model the arms’ posteriors. Through our experiments we show that our method can efficiently identify the m best policies, which is validated in a scenario where the ground truth is available. Finally, we explore how vaccination policies can best be organized under different contact reduction schemes.