Equitable algorithmic resource allocation : examining and mitigating racial disparities in crisis standards of care during COVID-19

Algorithms have increasingly been adopted in many industries such as finance and healthcare to make organizational processes more efficient. However, recent research has highlighted severe and harmful biases within these systems, which shows critical importance of bias testing. In this thesis, we evaluate harmful biases in algorithmic allocation and its disparate impacts in the context of scarce healthcare resources, and propose a group deliberation framework to design equitable algorithmic allocation. In Study 1, we empirically examine racial disparities in the Crisis Standards of Care (CSC) ventilator allocation schemes. Many scholars have raised concerns on racial disparities in CSC: they maximize survival likelihood, which may negatively impact nonwhite patients as they have a higher burden of chronic disease and disabilities. To our knowledge, there is no existing empirical evidence on racial disparities in CSC. Using Austin Round-Rock Metro COVID-19 patient data, we created a ventilator allocation simulator comparing the impacts of CSC and non-CSC allocation schemes on patient mortality outcomes. The simulation results showed higher mortality for nonwhite patients as they were deprioritized in allocation. The disparities in allocation necessitates the inclusion of marginalized communities’ input on the construction of allocation rules to address inequities. In Study 2, we propose a group deliberation framework where community members assess their individual values, deliberate on ethical considerations of racial disparate-aware allocation, and test the behavior and outcomes of the rules established by the group. This thesis makes several contributions to the literature on fair algorithmic systems, public health, and community engagement. It offers i) one of the first empirical evidence that reveals potential disparate impacts of the CSC, and ii) a deliberation framework on the inclusion and participation of marginalized community members in designing algorithmic allocation.Computer Science