Non-invasive Predictors of recurrence after catheter ablation for idiopathic PVCs

Background Idiopathic Premature Ventricular Contractions (PVCs) are generally benign, but catheter ablation is the preferred treatment modality for drug-refractory PVCs and PVC-induced cardiomyopathy. While success rates highly depend on the PVC location, definitive predictors for recurrence after ablation have not been fully established. Purpose This study aims to investigate the predictors of recurrence using both established clinical risk factors and a novel Deep Learning (DL) approach based on 12-lead surface electrocardiograms (ECGs) obtained from patients undergoing catheter ablation. Methods A total of 159 patients who underwent catheter ablation for PVC or idiopathic VT between 2020 and 2025 were retrospectively analyzed. Patients were followed up with ECG and Holter monitoring at the 4th week and 3 or 6 months post-procedure. A 1-D Convolutional Neural Network (1D-CNN) architecture, optimized for multi-lead classification with a 13-channel input, was developed using twelve-lead ECGs. The Recurrence DL model was trained to distinguish between cured and recurrence cases. Results The majority of patients had PVCs originating from the RVOT (36.5%), and the overall success rate of the procedure was 79.2%. Baseline characteristics (Age, LVEF, PVC burden, PVC QRS duration) did not show a statistically significant difference between the recurrence and no-recurrence groups(Table 1a,1c). The ABC-VT score assigns 1 point for superior axis, 2 points for PVC burden 10%–20%, or 3 points for burden ≥20%, 4 points for coupling interval ≤500 ms, and 4 points for the presence of NSVT. Univariable and multivariable logistic regression analysis identified the ABC-VT Score (OR: 1.25, p = 0.014) and a history of Atrial Fibrillation (OR: 5.91, p = 0.030) as independent predictors of recurrence(Table 1b). The ABC-VT Score showed moderate discriminatory power in predicting recurrence (AUC = 0.649). The optimal cut-off value of the ABC-VT score for recurrence prediction was 8.0 (high risk), yielding a sensitivity of 28.57% and a specificity of 96.30% (OR: 10.40, 95% CI: 2.63–41.18, p < 0.001). The developed Recurrence DL Model demonstrated an accuracy of 69.83% (p < 0.01) in distinguishing recurrence cases. Conclusion(s) The presence of Atrial Fibrillation and a high ABC-VT Score, alongside our EKG-derived Recurrence DL Model, may serve as useful tools for predicting recurrence following catheter ablation in patients with idiopathic PVCs.TableFigure