PVCuRe: A machine-learning-based tool to predict left ventricular systolic function recovery in patients undergoing ablation for premature ventricular contraction.
BACKGROUND: Ablation of frequent premature ventricular complexes (PVCs) can improve left ventricular ejection fraction (LVEF) in patients with systolic dysfunction, especially in those with suspected PVC-induced cardiomyopathy. However, many patients fail to achieve normalization of LVEF despite successful ablation, and current tools do not reliably distinguish true PVC-induced cardiomyopathy from underlying cardiomyopathy exacerbated by PVCs. OBJECTIVE: The aim of this study was to develop and externally validate a machine-learning (ML) model using routinely available clinical, echocardiographic, and electrocardiographic variables to predict LVEF recovery after PVC ablation. METHODS: In this retrospective multicenter study, 256 patients with LVEF <50% who underwent successful PVC ablation at 3 international referral centers were included. Predictors were selected using the Boruta algorithm, and 5 ML models were trained. Performance was assessed with 10-fold cross-validation and receiver operating characteristic curve analysis. The best-performing model underwent calibration and threshold analysis and was externally validated in an independent cohort from 3 additional centers. RESULTS: The Random Forest model showed the best performance, with an area under the curve of 0.88 (95% confidence interval = 0.79-0.98) in the internal test set and good calibration (Hosmer-Lemeshow P = .562). External validation confirmed consistent discrimination (area under the curve = 0.83, 95% confidence interval = 0.72-0.95). Key predictors included baseline PVC burden, QRS duration in sinus rhythm, and preprocedural LVEF. CONCLUSION: This ML-based tool, built on widely available variables, accurately estimates the probability of LVEF recovery after PVC ablation and may support clinical decision making and patient counseling.