Abstract TP181: A Risk Score for All-Cause Readmissions Using Get With the Guidelines-Stroke Data Elements

Abstract

Introduction: Readmission rates of stroke patients are commonly used as a measure of quality of care by CMS and other payers. In this study, our goal was to identify risk factors for 30, 60, and 90 day all-cause readmissions and to develop a risk score that indicates likelihood of readmission. Methods: We used GWTG-Stroke data elements describing 1086 stroke admissions at St. David’s Medical Center between 2013 - 2014. The principal ICD-9 diagnosis code was used to identify the reason for readmit. We used logistic regression models to predict readmission events using clinical (such as stroke type and medical history) and demographic (such as age and gender) variables. We used the models to develop a readmission risk score that is predictive of future admissions following a stroke. Results: Of the 1086 patients, 79 (7.3%), 116 (10.7%), and 142 (13.1%) have at least one unplanned in-hospital readmission within 30, 60, and 90 days respectively. The main readmission reasons were related to stroke, followed by syncope, fatigue/malaise symptoms, and atrial fibrillation. Our models were robust and predicted 30, 60, and 90 day readmissions (averaged over 10 random train/test splits) with 0.63, 0.64, and 0.66 accuracy, respectively (c-statistic). Significant predictors contributing to patient readmission risk include history of hypertension (OR 5.93), diagnosis of transient ischemic attack (OR 2.0), initiation of tPA (OR 2.08), discharge to rehabilitation facility (OR 3.27). Conclusions: The leading causes of readmission following a stroke are related to the stroke event or episodes of comorbid conditions. Our study shows promise in using GWTG data to predict readmissions, however larger cohorts of patients are needed to develop highly accurate predictive models.