The goal of this research project is to develop machine learning techniques for the fair allocation of healthcare services such as those provided by Medicaid. Although such programs provide crucial services to vulnerable populations, many of the individuals who most need these services languish on waiting lists due to limited resources. Machine learning models can potentially improve this situation by predicting individuals' levels of need, which can then be used to prioritize the waiting lists. Providing care to those in need can prevent institutionalization for those individuals, which both improves quality of life and reduces overall costs. While the benefits of such an approach are clear, care must be taken to ensure that the prioritization process is fair. The researchers also plan to address this issue directly by developing fairness definitions and corresponding fair learning algorithms for the task of learning to rank. The proposed techniques for fair prioritization of healthcare have the potential to save lives, as well as taxpayer dollars. This project aims to lead to a deployed solution for Medicaid prioritization in the state of Maryland, where over 8,000 individuals have died on the Medicaid waitlist since the state's Medicaid expansion under the Affordable Care Act began, according to a 2018 report from the Foundation for Government Accountability.
This project will develop a machine learning intervention to the processes of ranking individuals in order of priority for receiving healthcare services. The researchers will apply their methods to Medicaid data, which they will access via their ongoing collaboration with colleagues from the Hilltop Institute, a nonpartisan research organization which is dedicated to community-oriented healthcare analytics; they will also evaluate their methods on a public dataset to facilitate research reproducibility. A key goal of the project is to promote fairness in the ranking. In meeting this goal, the project will extend the capabilities of fair machine learning definitions and algorithms to tasks that have not previously been addressed including survival and temporal modeling. To predict individuals' health status, the research team will use survival models to estimate the risk of future institutionalization, such as relocating to a nursing home. The team will use also Cox proportional hazard models; the multiplicative relationship between covariates and risk will serve to aid explainability. The fairness definitions and the corresponding fair learning algorithms for these models will yield risk scores that can then be used to prioritize waiting lists. For waitlists deployed in practice, it will be necessary to continually re-rank the list since individuals enter and leave the list (due to death or institutionalization, for example), and since covariates change for those who remain on the list; reranking should ensure that individuals who need care will eventually reach the front of the list. The proposed work crosses the boundaries of multiple disciplines (machine learning, fairness, health IT, feminism and civil rights) to solve an urgent real-world problem.
This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.
