A network-based model to assess vaccination strategies for the COVID-19 pandemic by using Bayesian optimization

In this paper we build a network-based model to evaluate and compare vaccination plans in order to find the optimal strategies. The age-structured model is designed to take into account the comorbidity status and vaccination hesitancy of the population. The network-based model is calibrated to reported infected cases and deaths in the USA in order to obtain an approximated realistic scenario to test the vaccination strategies. We adapt an algorithm that is based on Bayesian optimization over permutation spaces with heuristics in order to deal with the discrete space of the vaccination strategies. We also developed an ad-hoc randomized algorithm which has a higher computational cost. Both algorithms provide similar patterns of the best found vaccination strategies. We find that these best vaccination plans prioritize the age-groups 40–59 and 60–69 years old, both with comorbidities. This result shows the highly nonlinear complexity related to the problem and its dependence on social contacts and case fatality rates. The developed network-based model adapts well to the uncertainty and heterogeneity of the real world situation and allows us to assess the efficacy of many vaccination strategies. The stochastic nature of the simulations enables us to explore additional potential scenarios and the findings offer useful information for developing vaccination plans for other future potential pandemics.