Fixing Medicare Advantage: Better Risk Adjustment, Machine Learning, and AI for Payment and Accountability (Intermediate; Added Fee)

Medicare Advantage (MA) health plans that contract with the Centers for Medicare & Medicaid Services (CMS) now cover over half of Medicare beneficiaries. However, MA spending is estimated to substantially exceed what CMS would have paid for similar beneficiaries in Traditional Medicare, raising questions about program sustainability.     This workshop provides a structured introduction to MA for a health policy research audience, followed by a focused examination of risk-adjusted payment systems. We begin with the historical rationale for MA and the role of diagnosis-based risk adjustment in aligning payments with expected costs. We then examine how current models (e.g., CMS-HCCs) operate in practice, including their strengths and limitations.     Building on this foundation, we introduce a new generation of risk adjustment models based on clinically grounded classification systems and machine learning methods. We discuss how to evaluate these models using criteria that extend beyond overall predictive accuracy, including transparency and resistance to gaming. Accurate risk adjustment is a central determinant of whether Medicare Advantage can function as intended. The underlying classification system (DXIs) and model-building tools are publicly available, supporting transparency, reproducibility, and broad adoption.      Finally, we explore emerging opportunities to use these methods—augmented with AI—to detect and quantify plan behaviors, including upcoding, selection, and care management patterns. The session emphasizes actionable insights for policymakers, researchers, and health system leaders, including those outside the U.S. seeking to understand or adopt similar payment approaches.