Improving target price calculations in Medicare bundled payment programs.

Area of Science:

• Health economics
• Healthcare policy
• Statistical modeling

Background:

• Bundled Payments for Care Improvement-Advanced (BPCI-A) utilizes target prices for episode-based payments.
• Accurate target price calculation is crucial for program success and financial predictability.
• Traditional Centers for Medicare and Medicaid Services (CMS) methodology may be susceptible to statistical biases like regression to the mean.

Purpose of the Study:

• To compare the predictive accuracy of the traditional CMS target price calculation methodology with an empirical Bayes approach.
• To evaluate which method provides more precise predictions for healthcare spending within BPCI-A episodes.
• To assess the impact of hospital size on the accuracy of these prediction methods.

Main Methods:

• Utilized Medicare fee-for-service claims data from 2010-2016.
• Trained prediction models on a baseline period (2010-2013) to forecast spending in a performance period (2015-2016).
• Compared average prediction errors for 23 clinical episode types across hospitals using both CMS and empirical Bayes approaches.

Main Results:

• The empirical Bayes approach yielded significantly more accurate spending predictions for 19 out of 23 BPCI-A clinical episode types.
• Average prediction error was lower with the empirical Bayes approach ($7521) compared to the CMS approach ($8456).
• Improved prediction accuracy was more pronounced in larger hospitals.

Conclusions:

• The empirical Bayes method demonstrates superior predictive accuracy for BPCI-A target prices.
• Healthcare organizations should consider adopting empirical Bayes methods for more reliable episode spending predictions.
• CMS should evaluate the integration of empirical Bayes methods into its BPCI-A target price calculations to enhance program efficiency.