Risk-adjustment in Medicare’s Value-Based Purchasing Programs: Implications for Providers with Clinically and Socially Vulnerable Dual-Eligible Patients

Background: Medicare’s value-based purchasing programs, including the Merit-Based Payment System (MIPS) and Hospital Readmission Reduction Program (HRRP), aim to improve care quality and efficiency by holding physicians and hospitals financially accountable for their patients’ spending and outcomes of care. A concern in both programs is that providers serving higher-risk populations, particularly patients with dual Medicare and Medicaid eligibility, may be more likely to receive penalties because of variation in their patient mix, rather than only because of variation in quality.

Objective: To assess the adequacy of risk-adjustment in the MIPS and HRRP and to estimate expected changes in risk-adjusted performance after adjusting for clinical and social characteristics of patients not currently incorporated in CMS’ risk-adjustment algorithms for these programs.

Design: Using 2013 and 2014 Medicare claims for a random 20% sample of fee-for-service Medicare beneficiaries, we conducted two sets of analyses. First, separately for hospitals and physician practices, we categorized providers into quintiles based on the proportion of their Medicare patients with dual Medicaid eligibility. We estimated performance differences between quintiles after adjusting for a base set of variables used by CMS for risk adjustment, and then after adjusting for additional patient characteristics observable in Medicare claims and linked Census data. Adjustments reflected the expected within-practice association of patient characteristics with outcomes (using provider fixed effects), preserving any between-provider differences that may reflect physician- or hospital-level variation in quality. Second, separately for hospitals and physicians, we used hierarchical regression models to estimate the variances and covariances of providers’ performance scores based on the different risk-adjustment methods. Using estimates from these models, we performed simulation analyses to determine how more complete risk adjustment would affect hospital or practice performance rankings, net of within-provider sampling error.

Results: For measures used to evaluate physicians in the MIPS, further adjustments narrowed performance differences between practices in the highest vs. lowest quintile of dual-eligible Medicaid patients by 11.9% to 55.9%. For hospital readmissions, further adjustments narrowed performance differences between hospitals in the highest vs. lowest quintile of dual-eligible patients by 91.5%. Simulation analyses indicated that adjustments would result in substantial re-ordering in provider rankings, with a net improvement among initially low-ranked practices and hospitals. For measures in the MIPS, these changes moved between 2.9% and 16.7% of practices from below to above the MIPS exceptional performance threshold for bonuses. For the HRRP, these changes are expected to move 47.9% of hospitals out of eligibility for penalties.

Conclusions: Current risk-adjustment methods in Medicare’s value-based purchasing programs inadequately distinguish variation in patient mix from differences in provider performance, thus exposing providers who disproportionately serve dual-eligible patients to greater penalties, creating unintended incentives for favorable risk selection while depleting providers’ resources to improve care for dual-eligible patients. Our results highlight the need for improved risk-adjustment and payment methods to guard against such unintended outcomes.