A Machine Learning Model to Improve Risk Adjustment Accuracy in Medicare.

Area of Science:

• Health Informatics
• Machine Learning in Healthcare
• Predictive Modeling

Background:

• The Centers for Medicare and Medicaid Services (CMS) utilizes the Hierarchical Condition Category (HCC) score for risk-adjusting payments.
• Accurate risk adjustment is crucial for equitable healthcare payments and resource allocation among millions of Americans.

Purpose of the Study:

• To develop and evaluate a machine learning (ML) algorithm named "Franklin" for risk adjustment in Medicare.
• To compare the predictive accuracy of the ML algorithm against the existing HCC scoring system.

Main Methods:

• A prognostic study utilizing Medicare claims data from 2018-2019.
• Training the "Franklin" ML algorithm on identical data used for the HCC score.
• Evaluating predictive accuracy using R-squared log cost, Spearman rho, sensitivity, and specificity.

Main Results:

• The "Franklin" ML algorithm demonstrated superior accuracy compared to the HCC score (R-squared log cost 0.44 vs. 0.15; Spearman rho 0.61 vs. 0.41).
• Improved accuracy was observed for beneficiaries with zero or one HCC, and for identifying the lowest-cost beneficiaries.
• Franklin enhanced accuracy for racial/ethnic minorities and rural populations, though equity impacts require further clarification.

Conclusions:

• The "Franklin" ML model significantly improves risk-adjustment accuracy for Medicare beneficiaries compared to the HCC score.
• Franklin has the potential to improve payment accuracy, reduce selection incentives, and yield financial savings for Medicare.
• Further research is needed to clarify the equity implications of enhanced risk adjustment accuracy.