Patient factors may only marginally improve readmission prevention programs

UNIVERSITY PARK, Pa. — A recent study led by researchers at Penn State, the University of Michigan, and UCLA, found that Medicare’s Hospital Readmissions Reduction Program (HRRP) disproportionately penalizes safety net hospitals, leading to concerns that the program’s omission of patient social complexity may aggravate healthcare disparities by penalizing financially-challenged hospitals and reducing their resources to improve care.

The researchers found that although adding patient functional and social factors improves the program’s ability to distinguish patients who are or are not readmitted, it does so only slightly, suggesting that financial support for readmission prevention programs in hospitals serving vulnerable patients, known as safety net hospitals, may be of greater benefit for health equity than modifying the HRRP program. 

The study found that inclusion of a broad set of risk adjustors measuring functional and social patient factors improved model discrimination, but only marginally. This suggests, said the researchers, that existing risk adjustment largely captures patient complexity and that unexplained portions of readmission risk may instead involve provider performance or access to quality care in the community.

Therefore, financial support for readmission-prevention programs in hospitals serving vulnerable patients may be of greater benefit for health equity than modification of HRRP, particularly because the collection and use of patient factors or their proxies (such as race and income) to modify HRRP risk-adjustment may be controversial given small gains in observed model discrimination, the costs of collecting patient complexity data, and concerns about creating different standards of care for vulnerable versus other patients.

The authors examined 1742 observations for fee-for-service Medicare beneficiaries ages ≥ 65 years from 2006 to 2012 Medicare hospital and linked Health and Retirement Study (HRS) data. They compared the discrimination of three models — the CMS base model; the base model plus patient income and education, health status, social factors, and healthy behavior; and the CMS base model plus income and race. Predictiveness was measured using model concordance, or the c-statistic.

The findings appeared in the Journal of General Internal Medicine on Aug. 24.

Authors of the paper include Geoffrey J. Hoffman, assistant professor, Department of Systems, Populations and Leadership, University of Michigan School of Nursing; Charleen Hsuan, assistant professor, Department of Health Policy and Administration, Penn State; Thomas Braun, professor of Biostatistics, University of Michigan School of Public Health; and Ninez Ponce, professor of health policy and management, UCLA Fielding School of Public Health.

Last Updated December 13, 2018