Algorithms to Improve Fairness in Medicare Risk Adjustment.

Area of Science:

• Health economics
• Health policy
• Data science in healthcare

Background:

• Payment system design significantly influences healthcare spending, access, and outcomes.
• Medicare Advantage represents over half of Medicare spending, making its risk adjustment algorithm crucial.

Purpose of the Study:

• Develop risk adjustment algorithms for fair spending targets.
• Compare algorithm performance against the Centers for Medicare and Medicaid Services' baseline regression approach.

Main Methods:

• Retrospective analysis of Traditional Medicare data (2017-2020).
• Mapped diagnoses to Hierarchical Condition Categories (HCCs).
• Utilized demographic indicators and HCCs to predict subsequent year Medicare spending.

Main Results:

• Analysis included 4,398,035 beneficiaries; mean age 75.2 years; mean annual spending $8,345.
• Constrained regression and post-processing achieved fair spending targets (fit 12.6%-12.7%) vs. baseline (12.7%).
• Constrained regression benefited minoritized groups and others in socioeconomically disadvantaged areas; post-processing benefited minoritized groups.

Conclusions:

• Constrained regression and post-processing effectively integrate fairness objectives into Medicare risk adjustment.
• These methods achieve fairness with minimal reduction in overall payment system fit.