Risk Adjustment Tools for Learning Health Systems: A Comparison of DxCG and CMS-HCC V21

Area of Science:

• Health Services Research
• Health Economics
• Biostatistics

Background:

• Accurate patient risk adjustment is crucial for healthcare resource allocation and performance evaluation.
• Existing risk adjustment models, such as the Centers for Medicare and Medicaid Services (CMS) V21, may not fully capture the complexity of healthcare costs.
• The Department of Veterans Affairs (VA) manages a large and diverse patient population with unique healthcare needs.

Purpose of the Study:

• To compare the predictive accuracy of the DxCG (Verisk) risk score with the CMS V21 risk score using VA administrative data.
• To evaluate the impact of incorporating pharmacy data on risk score performance.
• To assess the feasibility of recalibrating the CMS V21 model for the VA population.

Main Methods:

• Analysis of administrative and pharmacy data from the VA for fiscal years 2010-2011.
• Regression analysis to model total annual VA costs based on predicted risk scores from DxCG and CMS V21.
• Model fit was assessed using R-squared, root mean squared error, mean absolute error, and Hosmer-Lemeshow tests.
• Creation of six distinct analytical files representing different patient subgroups.

Main Results:

• The DxCG risk score incorporating pharmacy data demonstrated significantly improved model fit compared to the CMS V21 model.
• Recalibrating the CMS V21 model using VA pharmacy data resulted in risk scores with fit statistics comparable to the DxCG model.
• Both models showed varying performance across different patient subgroups.

Conclusions:

• While DxCG with pharmacy data offers enhanced predictive fit, the CMS V21 model can be recalibrated using additional variables, such as pharmacy data, to achieve comparable performance within specific healthcare systems like the VA.
• Tailoring risk adjustment models to specific patient populations and data sources is essential for accurate healthcare cost prediction and resource management.