Examining Racial Bias in Predictive Analytic Tools to More Equitably Distribute Care to At-Risk Veterans
In order to target care resources equitably and efficiently, the Veterans Health Administration (VA) has implemented novel predictive analytic tools in clinical care settings, including the Care Assessment Needs (CAN) score. Commonly used by VA clinicians nationwide, the CAN score is used to direct clinical programs and resources, including telehealth, palliative care, and home-based primary care, to high-risk veterans. However, recent studies have shown that other similar algorithms may mischaracterize and underestimate risk for vulnerable patient subgroups and do not routinely factor in race nor social determinants of health. The growing concern is that algorithms like the CAN score could generate “algorithmically unfair” predictions that systematically mischaracterize risk for subgroups, particularly African Americans. Thus, this project will examine algorithmic unfairness in the VA CAN score and develop approaches to mitigate this unfairness within the existing CAN score its current metrics. This project will also develop a new CAN score that incorporates race and select social determinants of health metrics and test how this score impacts the composition of veterans determined to be at-risk.
Examining and reducing unfairness in models such as the VA CAN score has the potential to lead to more equitable distribution of VA resources and enrollment in VA programs. By factoring in race and social determinants of health, the CAN score can more accurately portray to clinicians whom may be at-risk and thus ensure that all veterans have fair access to needed programs.
VA Merit (R01-funded) Grant
Augmented & Artificial Intelligence