Spatio-temporal optimization of seasonal vaccination using a metapopulation model of influenza

Prophylactic interventions such as vaccine allocation are one of the most effective public health policy planning tools. The supply of vaccines is limited, and an important problem is when and how to allocate the available vaccination supply, referred to as the Vaccine Allocation Problem (VaccIntDesign) problem. The spread of epidemics is modeled by the SEIR process, which has a very complex dynamics, and depends on human contacts and mobility. This makes the design of efficient solutions to VaccIntDesign problem to minimize the number of infections a very challenging problem. In particular, this requires good models for human mobility, and optimization tools for vaccine allocation. In this paper, we study the VaccIntDesign problem in the context of seasonal Influenza spread in the United States. We develop a novel national scale flu model that integrate both short and long distance travel, which are known to be important determinants of the spread of Influenza. We also design GreedyAlloc, a greedy algorithm for allocating the vaccine supply at a county level. Our results show significant improvement over the current baseline, which involves allocating vaccines based on the state population.