Development and Validation of an Explainable Machine Learning Model for Predicting Repeat Catheter Ablation for...

Area of Science:

• Cardiology
• Medical Informatics
• Machine Learning in Healthcare

Background:

• Atrial fibrillation (AF) is a prevalent cardiac arrhythmia, with catheter ablation as a primary treatment.
• High recurrence rates post-ablation necessitate repeat procedures, increasing costs and risks.
• Current risk scores for predicting repeat AF ablation have limited accuracy and clinical utility.

Purpose of the Study:

• To develop and validate an explainable machine learning (ML) model for predicting repeat AF ablation.
• To utilize routine clinical and echocardiographic features for risk prediction.
• To improve the discriminative ability and clinical utility of repeat ablation risk assessment.

Main Methods:

• Retrospective analysis of 1073 patients undergoing AF ablation (2012-2023).
• Feature selection using LASSO regression and Boruta algorithm; construction of eight ML models.
• Performance evaluation via AUC, sensitivity, specificity, Brier score, and decision curve analysis; interpretability using SHAP.

Main Results:

• 32.8% of patients required repeat AF ablation.
• Nine predictive features identified: NT-proBNP, age, GLO, DBIL, LVEF, Cys-C, smoking history, CK, and urea.
• XGBoost model achieved an AUC of 0.811, demonstrating superior performance and interpretability (NT-proBNP, age most influential).

Conclusions:

• An explainable XGBoost model effectively predicts the need for repeat AF ablation.
• The model offers valuable clinical insights for patient management and procedural optimization.
• This ML tool enhances the prediction of repeat AF ablation compared to existing risk scores.