Effect of Clinical and Social Risk Factors on Hospital Profiling for Stroke Readmission A Cohort Study

Background: The Centers for Medicare & Medicaid Services (CMS) and Veterans Health Administration (VA) will report 30-day stroke readmission rates as a measure of hospital quality. A national debate on whether social risk factors should be included in models developed for hospital profiling is ongoing. Objective: To compare a CMS-based model of 30-day readmission with a more comprehensive model that includes measures of social risk (such as homelessness) or clinical factors (such as stroke severity and functional status). Design: Data from a retrospective cohort study were used to develop a CMS-based 30-day readmission model that included age and comorbid conditions based on codes from the International Classification of Diseases, Ninth Revision, Clinical Modification (model 1). This model was then compared with one that included administrative social risk factors (model 2). Finally, the CMS model (model 1) was compared with a model that included social risk and clinical factors from chart review (model 3). These 3 models were used to rank hospitals by 30-day risk-standardized readmission rates and examine facility rankings among the models. Setting: Hospitals in the VA. Participants: Patients hospitalized with stroke in 2007. Measurements: 30-day readmission rates. Results: The 30-day readmission rate was 12.8%. The c-statistics for the 3 models were 0.636, 0.646, and 0.661, respectively. All hospitals were classified as performing as expected using all 3 models (that is, performance did not differ from the VA national average); therefore, the addition of detailed clinical information or social risk factors did not alter assessment of facility performance. Limitation: A predominantly male veteran cohort limits the generalizability of these findings. Conclusion: In the VA, more comprehensive models that included social risk and clinical factors did not affect hospital comparisons based on 30-day readmission rates.