Functional Status Predicts Acute Care Readmissions from Inpatient Rehabilitation in the Stroke Population

Objective Acute care readmission risk is an increasingly recognized problem that has garnered significant attention, yet the reasons for acute care readmission in the inpatient rehabilitation population are complex and likely multifactorial. Information on both medical comorbidities and functional status is routinely collected for stroke patients participating in inpatient rehabilitation. We sought to determine whether functional status is a more robust predictor of acute care readmissions in the inpatient rehabilitation stroke population compared with medical comorbidities using a large, administrative data set. Methods A retrospective analysis of data from the Uniform Data System for Medical Rehabilitation from the years 2002 to 2011 was performed examining stroke patients admitted to inpatient rehabilitation facilities. A Basic Model for predicting acute care readmission risk based on age and functional status was compared with models incorporating functional status and medical comorbidities (Basic-Plus) or models including age and medical comorbidities alone (Age-Comorbidity). C-statistics were compared to evaluate model performance. Findings There were a total of 803,124 patients: 88,187 (11%) patients were transferred back to an acute hospital: 22,247 (2.8%) within 3 days, 43,481 (5.4%) within 7 days, and 85,431 (10.6%) within 30 days. The C-statistics for the Basic Model were 0.701, 0.672, and 0.682 at days 3, 7, and 30 respectively. As compared to the Basic Model, the best-performing Basic-Plus model was the Basic+Elixhauser model with C-statistics differences of +0.011, +0.011, and +0.012, and the best-performing Age-Comorbidity model was the Age+Elixhauser model with C-statistic differences of -0.124, -0.098, and -0.098 at days 3, 7, and 30 respectively. Conclusions Readmission models for the inpatient rehabilitation stroke population based on functional status and age showed better predictive ability than models based on medical comorbidities.