Abstract
BACKGROUND: Same-day discharge after transcatheter aortic valve replacement (TAVR) may be feasible for selected patients if a low risk for adverse clinical events can be defined. We aimed to develop a clinical risk prediction model to facilitate same-day discharge planning. METHODS: A random forest machine learning algorithm was used to build a prediction model of adverse events occurring in-hospital after TAVR. Patients were categorized into low, moderate, or high-risk groups based on their estimated scores. RESULTS: Overall, 730 patients (median age, 81 years; 58.9% men) who had transfemoral TAVR performed with conscious sedation were examined. The risk score was built utilizing 9 clinical parameters. The prediction model had a median area under the receiver operating characteristic curve of 0.76. For determining the probability of events that would disallow same-day discharge, the model successfully identified 172 patients (23.6% of the population) as low-risk for same-day discharge, or for having an event rate of <3%, with all events occurring within 6 hours after TAVR. The low-risk group had no in-hospital events after a 6-hour observation, and no mortality at the 30-day follow-up. External testing in 158 patients showed 94% sensitivity in predicting overall adverse events and identified a low-risk group using the clinical risk score. CONCLUSIONS: In this analysis, ∼1 in 4 patients may be candidates for same-day discharge after TAVR. This prediction model can identify such patients, with findings that may have implications for hospital resource allocation in those undergoing TAVR.