What type of study is this?

September 10, 2025Open Access

Machine learning-based prediction model for recurrence after radiofrequency catheter ablation in patients with atrial fibrillation

Key Points

The LightGBM model achieved an accuracy of 0.721 and an AUC of 0.848 for predicting AF recurrence.
Key predictors include B-type natriuretic peptide and neutrophil-to-lymphocyte ratio, impacting recurrence risks.
Four machine learning algorithms were assessed on data from 430 patients who underwent RFCA for AF.
These predictive models may inform personalized follow-up strategies, pending further validation.

Abstract

Background This study seeks to develop and validate a machine learning (ML) model for predicting atrial fibrillation (AF) recurrence at 12 months following radiofrequency catheter ablation (RFCA). Methods A total of 430 consecutive patients with atrial fibrillation undergoing first-time radiofrequency catheter ablation were retrospectively enrolled between June 2022 and December 2023. Patients were randomly assigned to either a training cohort (70%) or a testing cohort (30%). Four ML algorithms were employed to develop prediction models. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC) and accuracy. The SHapley Additive exPlanations (SHAP) methodology was employed to interpret the best-performing model and quantify each feature's contribution to its predictions. Results Among the four machine learning algorithms evaluated, the Light Gradient Boosting Machine (LightGBM) model showed promising predictive performance on the testing set, achieving an accuracy of 0.721 and an AUC of 0.848 (95% CI: 0.778–0.919). Interpretation of the LightGBM model using SHAP analysis identified B-type natriuretic peptide (BNP) and the neutrophil-to-lymphocyte ratio (NLR) as the most impactful predictors for AF recurrence. The analysis revealed that higher levels of BNP and NLR were strongly associated with an increased risk of recurrence, whereas higher levels of albumin and lymphocyte count were protective. Other significant predictors included left atrial diameter (LAD) and nonparoxysmal atrial fibrillation (NPAF). Conclusion Machine learning-based models show modest but promising performance for assessing AF recurrence risk after RFCA using routine clinical data. While requiring extensive external validation before clinical application, these models highlight the potential of ML to inform future risk stratification and guide personalized follow-up strategies.

Read Full Paperexternally

Ask AI

Helpful

Bookmark

View Full Paper