Machine Learning Integration of Ablation Quality Metrics for Predicting Atrial Fibrillation Recurrence: A Multicenter...

Current prediction models for atrial fibrillation (AF) recurrence after ablation rely predominantly on nonmodifiable clinical characteristics, demonstrating limited predictive accuracy and offering little guidance for intraprocedural decision-making and outcome optimization.

This study aimed to develop and externally validate a machine learning model that integrated clinical characteristics with ablation quality metrics to individually predict AF recurrence.

We conducted a multicenter, retrospective cohort study of patients undergoing their first AF ablation in 14 hospitals from mainland China. Patients aged 18 to 80 years were enrolled, and data on clinical characteristics and quality metrics were collected. An XGBoost model was developed using 7 selected features to predict 1-year AF recurrence. The model was developed using 5-fold cross-validation for internal validation and was further externally validated on data from 2 independent hospitals.

Among 1,782 patients enrolled between December 2021 and December 2023, 294 (16.5%) experienced AF recurrence within 1 year. The final XGBoost model achieved strong discrimination in internal validation (area under the curve [AUC]: 0.918, 95% CI: 0.874-0.962). In feature-combination analyses, model performance improved from an AUC of 0.588 using clinical features alone to 0.918 after incorporating ablation quality metrics. In 2 independent external validation centers (n = 239), the model showed good discrimination, with AUCs of 0.992 and 0.951 and precision-recall AUCs of 0.960 and 0.815, respectively.

Integrating ablation quality metrics yielded an validated model that showed strong discrimination for 1-year AF recurrence, outperforming clinical features alone. This approach highlights the prognostic value of intraprocedural quality assessment and may support personalized patient management.