Algorithms Enhance Fairness in Medicare Advantage Risk Adjustment Payments

Medicare Advantage (MA) plans represent a significant portion of the Medicare program, with annual spending exceeding $450 billion and enrollment of a large share of beneficiaries from minoritized racial and ethnic groups.

The current Medicare Advantage risk adjustment algorithm, which sets payments to these plans based on beneficiary demographics and diagnoses, relies on historical fee-for-service Medicare data. This reliance can perpetuate existing disparities in healthcare spending, access, and outcomes among minoritized populations.

This article presents two novel algorithmic approaches — constrained regression and post-processing — designed to promote fairer Medicare Advantage spending for multiple minoritized racial and ethnic groups. These algorithms adjust spending predictions to better align payments with actual healthcare costs for these groups, achieving equitable spending targets without compromising predictive accuracy. Constrained regression modifies payments related to specific health conditions and may offer broader health equity benefits compared to post-processing techniques. The study analyzed data from over 4.3 million beneficiaries to evaluate these algorithms' impact on spending fairness. The research highlights that algorithmic reform in risk adjustment payments for Medicare Advantage plans holds the potential to reduce systemic spending disparities. However, it emphasizes that such technological reforms must be supplemented by broader policy measures to comprehensively address health inequities.

This work addresses a gap in prior studies that have often overlooked algorithms aimed at multi-group fairness within Medicare Advantage's payment system. The changing demographics of the Medicare population and the disproportionate enrollment of minority groups in Medicare Advantage underscore the importance of equitable payment models. The paper notes that adding marginalized group indicators directly into risk adjustment models can exacerbate disparities, highlighting the need for carefully designed fairness algorithms. By improving the risk adjustment mechanism, policymakers can influence Medicare spending distribution, impacting provider incentives, healthcare access, and outcomes for underserved populations. This research is timely given Medicare's significant share of federal spending and the growth of Medicare Advantage enrollment.

The findings support transparency and reproducibility by making the algorithmic code publicly accessible to encourage further research and refinement. Overall, the development and implementation of fairness-focused algorithms in Medicare Advantage risk adjustment could serve as an important step toward mitigating longstanding disparities encoded in historical healthcare data, shaping more equitable enrollee payment and care delivery systems.