Development and Validation of a Prognostic Nomogram for Post-Transcatheter Aortic Valve Replacement Heart Failure Hospitalization in Patients With Concurrent Symptomatic Aortic Stenosis and Heart Failure With Preserved Ejection Fraction: A Multicenter Study.

backgroundExisting risk stratification models are insufficient in identifying patients at high risk for heart failure (HF) hospitalization, particularly among those presenting with symptomatic aortic stenosis and the HF with preserved ejection fraction phenotype.

methodsThis multicenter cohort study enrolled 321 patients diagnosed with severe aortic stenosis and HF with preserved ejection fraction who underwent transcatheter aortic valve replacement between January 2017 and February 2024. Various predictive modeling techniques were used, including random forest, XGBoost, SuperPC, plsRcox, least absolute shrinkage and selection operator-Cox, Gradient Boosting Machine, Coxboost, and Cox regression analysis, at multiple time points.

resultsPatients were divided into a derivation cohort (n=191) and an external validation cohort (n=130) based on institutional affiliation, with a median follow-up of 20 months. Feature selection using the Boruta algorithm and least absolute shrinkage and selection operator regression, combined with variance inflation factor analysis to assess multicollinearity, identified 6 independent predictors. Among 8 prediction models evaluated, the Cox regression-based nomogram demonstrated superior performance in external validation, achieving time-dependent area under the curve values of 0.824 (95% CI, 0.693-0.956) at 12 months and 0.818 (95% CI, 0.715-0.920) at 20 months. The nomogram exhibited excellent calibration and substantial clinical utility across both time points, consistently outperforming the European System for Cardiac Operative Risk Evaluation in discrimination and reclassification analyses. An interactive web-based clinical decision support tool was developed to facilitate point-of-care implementation.

conclusionsThis nomogram, based on machine learning and incorporating metabolic biomarkers, exhibits high predictive accuracy for HF hospitalization in patients with symptomatic aortic stenosis and high-risk HF with preserved ejection fraction phenotype following transcatheter aortic valve replacement. REGISTRATION: URL: https://www.clinicaltrials.gov; Unique Identifier: ChiCTR2400092655.