Development and Application of a Risk Prediction Model for In-Hospital Stroke After Transcatheter Aortic Valve Replacement: A Report From The Society of Thoracic Surgeons/American College of Cardiology Transcatheter Valve Therapy Registry.

BACKGROUND: Stroke is a serious complication after transcatheter aortic valve replacement (TAVR), yet predictive models are not available. A new risk model for in-hospital stroke after TAVR was developed and used to estimate site-specific performance. METHODS: We included 97,600 TAVR procedures from 521 sites in The Society of Thoracic Surgeons/American College of Cardiology Transcatheter Valve Therapy Registry from July 2014 to June 2017. Association between baseline covariates and in-hospital stroke was estimated by logistic regression. Discrimination was evaluated by C-statistic. Calibration was tested internally via cross-validation. Hierarchical modeling was used to estimate risk-adjusted site-specific performance. RESULTS: Median age was 82 years, 44,926 (46.0%) were women, and 1,839 (1.9%) had in-hospital stroke. Covariates associated with stroke (odds ratio) included transapical access (1.44), access excluding transapical and transfemoral (1.77), prior stroke (1.57), prior transient ischemic attack (1.50), preprocedural shock, inotropes or mechanical assist device (1.48), smoking (1.28), porcelain aorta (1.23), peripheral arterial disease (1.21), age per 5 years (1.11), glomerular filtration rate per 5 mL/min (0.97), body surface area per m2 (0.55 male; 0.43 female), and prior aortic valve (0.78) and nonaortic valvular (0.42) procedures. The C-statistic was 0.622. Calibration curves demonstrated agreement between observed and expected stroke rates. Hierarchical modeling showed 10 (1.9%) centers with significantly higher odds ratios for in-hospital stroke than their peers. CONCLUSIONS: A risk model for in-hospital stroke after TAVR was developed from The Society of Thoracic Surgeons/American College of Cardiology Transcatheter Valve Therapy Registry and used to estimate site-specific stroke performance. This model can serve as a valuable resource for quality improvement, clinical decision making, and patient counseling.

Duke Scholars

Author Pamela Susan Douglas Medicine, Cardiology

Author Sean Michael O'Brien Biostatistics & Bioinformatics, Division of Biostatistics

Author Sreekanth Vemulapalli Medicine, Cardiology