Predicting Early recurrence of atrial fibrilation post-catheter ablation using machine learning techniques.

BACKGROUND

Catheter ablation is a common treatment for atrial fibrillation (AF), but recurrence rates remain variable. Predicting the success of catheter ablation is crucial for patient selection and management. This research seeks to create a machine learning model to forecast the early recurrence of atrial fibrillation following catheter ablation.

METHODS

A prospective longitudinal study was conducted using data from the Iranian AF registry. The dataset included 402 consecutive AF patients who underwent radiofrequency catheter ablation. The primary outcome was early recurrence of AF within 3 months' post-ablation. Data preprocessing and feature selection were performed, followed by the development and evaluation of various machine learning models. The CatBoost model was selected as the best-performing model.

RESULTS

The study population had a mean age of 57.30 ± 14.05 years, and 61.4% were male. AF recurrence occurred in 26.1% of patients. The CatBoost model, utilizing 35 features, achieved an accuracy of 92.5% in predicting AF recurrence. The model demonstrated high sensitivity (88.6%) and specificity (94.0%), with an area under the ROC curve of 0.96. Paroxysmal AF, BUN, Cr, age, mitral regurgitation, LA velocity, and mild valvular heart disease were among the most important predictive features.

CONCLUSION

Machine learning methods, particularly the CatBoost model, demonstrate high accuracy in predicting early catheter ablation outcomes in AF patients. The developed model has the potential to improve patient care and decision-making by identifying patients most likely to benefit from the procedure. Further studies with larger sample sizes and external validation are warranted to assess the generalizability of this method for catheter ablation outcome prediction in AF patients.