AFA-Recur

AFA-Recur predicts the probability of recurrent atrial arrhythmia within one year after catheter ablation for atrial fibrillation using pre-procedural clinical variables.

Key Features:

• Machine Learning Models: Utilizes four supervised machine learning models: decision tree, random forest, AdaBoost, and k-nearest neighbor.
• Best-performing Model: The random forest model achieved the highest discriminative performance with an area under the curve (AUC) of 0.721.
• Dataset and Cohorts: Developed using 3,128 patients from the ESC-EHRA Atrial Fibrillation Ablation Long-Term Registry (AFA-LT) with an 80% training and 20% testing split.
• Hyperparameter Tuning and Calibration: Hyperparameters were optimized via 10-fold cross-validation and model outputs were calibrated using Platt scaling.
• Comparative Performance: Demonstrated superior predictive performance compared with the APPLE score (APPLE AUC 0.557) based on AUC comparisons.

Scientific Applications:

• Patient Selection: Supports selection of patients for catheter ablation by providing individualized recurrence risk estimates.
• Clinical Decision Support: Provides data-driven risk estimates to inform post-ablation management and follow-up strategies.
• Research and Development: Serves as a foundation for further research into predictive modeling in cardiology.

Methodology:

Used pre-procedural clinical variables from the AFA-LT registry; trained supervised models (decision tree, random forest, AdaBoost, k-nearest neighbor) with an 80%/20% training/testing split; optimized hyperparameters via 10-fold cross-validation; calibrated outputs with Platt scaling; and evaluated performance using AUC with comparison to the APPLE score.