Improving risk adjustment for Medicare capitated reimbursement using nonlinear models

Area of Science:

• Health Economics
• Biostatistics
• Health Services Research

Background:

• Medicare expenditure data often exhibits a skewed distribution.
• Accurate risk adjustment is crucial for Medicare program integrity and resource allocation.
• Traditional linear models may not fully capture the complexities of expenditure distributions.

Purpose of the Study:

• To compare a linear risk-adjusted model with a model accounting for expenditure skewness in Medicare patients.
• To evaluate the impact of a square root transformation on expenditure modeling.
• To assess forecast performance and overfitting using different risk adjustment strategies.

Main Methods:

• Utilized Medicare Current Beneficiary Survey data (1992-1994).
• Estimated both a linear expenditure model and a square root transformed expenditure model.
• Assessed models based on linearity, heteroscedasticity, in-sample fit (R2), forecast bias, forecast mean squared error, and overfitting.

Main Results:

• The square root model with parsimonious risk adjusters showed superior forecast squared error and reduced overfitting.
• The untransformed model demonstrated better forecast bias for most disability groups, except the severely disabled.
• In a second analysis, the square root model improved forecast squared error, though bias was not significantly different from zero for either model.

Conclusions:

• Accounting for expenditure skewness generally enhances model precision but may not consistently reduce bias.
• The square root transformation proved beneficial for risk adjustment, especially for the severely disabled Medicare population.
• Incorporating health status as a risk adjuster demonstrably improves overall risk adjustment accuracy.