Modifications to the Hospital Readmissions Reduction Program: Improvements Still Needed?

Research Objective To address concerns that Medicare’s Hospital Readmissions Reduction Program (HRRP) unfairly penalized safety‐net hospitals treating patients with high social and functional risk, in 2019 HRRP introduced a “peer‐grouping” methodology comparing hospital readmission performance only to other hospitals with similar proportions of patients dually eligible for Medicaid and Medicare. Whether the new methodology fairly accounts for differences in social and functional risk factors is unknown. Using a unique dataset that adjusts for patient‐level social and functional factors and community‐level proxies, we examined changes in risk‐standardized readmission rates (RSRRs) for each peer group when adding these factors. Study Design Using 2000‐14 Medicare hospital discharge data, Health and Retirement Study data, Area Health Resources File data, and CMS’ New Stratified Methodology Impact File (2017), we calculated RSRRs for hospitals in the five peer groups, where peer group 5 hospitals served the highest proportion of dual‐eligibles. We first created latent factors associated with patients and with communities using factor analysis. We then modeled RSRRs using CMS criteria in four hierarchical logistic regression models: (a) “base model,” adjusting for patient age, sex, and Charlson‐weighted comorbidity score; (b) “patient model,” additionally adjusting for latent factors reflecting patient health status, functional status, frailty, and socioeconomics; (c) “community model,” adjusting for the base model and latent county‐level factors reflecting postacute care access, income, and education; and (d) “full model,” adjusting for the base model, patient factors, and community factors. Population Studied Older (≥65) fee‐for‐service Medicare beneficiaries with an eligible index hospitalization at 1724 HRRP‐participating hospitals, using CMS criteria. Principal Findings Hospitals in peer group 5, which accounts for 53% of the safety‐net hospitals, treated patients in poorer health, with more functional challenges, and from socially disadvantaged groups, compared to hospitals in the other peer groups when using patient‐level characteristics. However, differences between peer group 5 and other peer groups were less pronounced when using community‐level proxies. Unadjusted readmission rates were higher for peer group 5 than peer group 1 hospitals (16.8% vs 12.7%). Compared to the base model, the patient and full models decreased RSRRs by 0.7% for peer groups 2 and 4 and by 1.3% for peer group 5. Compared to the base model, the community model decreased RSRRs by 0.7% for peer group 5. Under the patient (versus base) model, fewer safety‐net hospitals (57.1% vs 54.6%) but more nonsafety‐net hospitals (45.3% vs 47.5%) were penalized. Adjusting for patient‐level factors increased the proportion of peer group 1 hospitals penalized from 37.8% to 41.1%, but reduced the proportion of peer group 5 hospitals from 60.8% to 58.1%. Conclusions Peer group 5 hospitals treat a substantially different patient population than other hospitals. Patient‐level risk‐adjustment decreased RSRRs and the proportion of safety‐net and peer group 5 hospitals penalized, while having less effect on other hospitals. Implications for Policy or Practice CMS said they would seek further information about adding social risk factors in HRRP. Our results suggest the importance of collecting and including in HRRP risk adjustment patient‐level social and functional factors (but not community‐level proxies), even after peer group modification intended to help safety‐net hospitals. Primary Funding Source NIH and AHRQ.