Models and Algorithms for Balancing Efficiency and Equity in Vaccine Allocation

The COVID-19 pandemic was a powerful reminder that existing societal inequalities get amplified during public health emergencies. In response to the pandemic, organizations such as the CDC, WHO, and public health departments developed frameworks for equitable allocation of vaccines, using well-established ethical principles as a foundation. The overall goal of this paper is to translate these policy frameworks into a computational framework that can be used by public health departments to equitably allocate vaccines in a transparent manner during the initial stages of a pandemic, when vaccine demand far exceeds supply. We start by developing a mathematical model for disease-spread that accounts for social vulnerability, geographic barriers to healthcare access, and differences in work constraints. On the basis of this model, we present multiple optimization formulations of a vaccine allocation problem that aims to reduce overall disease prevalence while also reducing disparity in outcomes for a given ``protected class'' relative to the general population. We present simple, scalable, and transparent algorithms for our optimization formulations. Our experiments focus on allocating vaccines at the census block group granularity in Johnson County, Iowa. Our experimental test bed incorporates social vulnerability index, a hospital accessibility index, and essential worker status into CovaSim, a state-of-the-art agent-based COVID-19 model. Our experiments lead to two main takeaways. First, it is possible to substantially reduce disparity in the outcomes of the protected class (for various choices of this class) with negligible worsening in overall disease-prevalence. Second, it is critical for disparity to be considered at all stages of the computational framework, e.g., incorporating it into the optimization formulation without considering it in the modeling stage has very limited value.