Evaluating COVID-19 vaccine allocation policies using Bayesian m-top exploration.

Individual-based epidemiological models support the study of fine-grained preventive measures, such as tailored vaccine allocation policies, in silico. As individual-based models are computationally intensive, it is pivotal to identify optimal strategies within a reasonable computational budget. Moreover, due to the high societal impact associated with the implementation of preventive strategies, uncertainty regarding decisions should be communicated to policy makers, which is naturally embedded in a Bayesian approach. We present a novel technique for evaluating 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 experts to further inspect this small set of alternative strategies, along with their quantified uncertainty. The anytime component provides policy advisors with flexibility regarding the computation time and desired confidence, which is important as it is difficult to make this trade-off beforehand. We consider the Belgian COVID-19 epidemic using the individual-based model STRIDE, where we learn a set of vaccination policies that minimise infections and hospitalisations. In this setting, each policy specifies how the limited weekly supply of different COVID-19 vaccine types is allocated across age groups over the course of the vaccination campaign, under given social contact reduction policies. Formally, we define each such unique allocation policy as an arm within our multi-armed bandit framework. Through experiments we show that our method efficiently identifies the m-top policies. Finally, we explore how vaccination policies can best be organised under different contact reduction schemes and vaccine uptake proportions. We show that the top policies follow a clear trend regarding prioritised age groups and assigned vaccine types, which provides insights for future vaccination campaigns. Furthermore, our experiments suggest that the uptake proportion has only a limited influence on overall policy optimality.