Predictors of long-term atrial fibrillation recurrence after catheter ablation: non-linear analytical approach for individualized prognostic stratification

Overview

In this retrospective study, investigators from the National Institute of Cardiovascular Diseases in Bratislava, Slovakia, and Powerful Medical used non-linear machine learning to identify predictors of long-term atrial fibrillation (AF) recurrence in 92 patients with paroxysmal AF undergoing catheter ablation for pulmonary vein isolation. All patients had cardiac CT and an implantable loop recorder before ablation, and 23 baseline clinical and CT-derived epicardial fat parameters were analyzed with random forest models and Shapley values. Over 3 years, AF recurred in 58 patients (63%). Upper epicardial fat volume, BMI, baseline AF burden, age and lower-segment pericardial volume were the five most important predictors, and upper epicardial fat volume was twice as important in men as in women.

Published in: ESC Congress 2022, Barcelona; abstract published in European Heart Journal, Volume 43, Supplement 2
Published on: 26 August 2022

Background

Recurrences are common in patients with atrial fibrillation (AF) during long-term follow-up after catheter ablation for pulmonary vein isolation (PVI). Machine learning models that identify non-linear relationships among patient parameters have recently been applied for prognostic stratification in different cardiac diseases. This retrospective study aimed to determine whether machine learning models can identify non-linear relationships in individual baseline clinical characteristics and CT-quantified volumetric parameters of epicardial fat tissue to help predict the outcome of catheter ablation for PVI in patients with paroxysmal AF.

Methods

The study analyzed 92 patients (median age 60.2 years [51.9–64.0]; 74% male) with paroxysmal AF (a single persistent AF episode was accepted) undergoing catheter ablation targeting PVI. All patients underwent cardiac CT imaging and were fitted with an implantable loop recorder before ablation. PVI was performed with radiofrequency ablation and electroanatomical mapping in 79 patients and with cryoballoon ablation in 13 patients. AF recurrence, defined as an AF burden >0.1% after a 90-day blanking period, was continuously assessed by the loop recorder. Feature selection on 23 baseline parameters was performed with random forest models (XGBoostRegressor), and mean absolute Shapley values (|mSHAP|, SHapley Additive exPlanations) were used to quantify the relative discriminative power of the analyzed parameters.

Results

During 3 years of follow-up, AF recurrence was detected in 58 patients (63%), and 29 of them (50%) underwent a repeat ablation. The five most important predictors of AF recurrence were upper epicardial fat volume, BMI, baseline AF burden, age and pericardial volume (lower segment). Upper epicardial fat volume was twice as important for men as for women (|mSHAP| 0.44 and 0.21, respectively). For a patient without AF recurrence, low age (41 years) and low upper epicardial fat volume (13.2 ml) were the most important drivers of a predicted positive ablation outcome. In contrast, for a patient with AF recurrence after ablation, an above-average epicardial fat volume of 55.5 ml and a high BMI made the largest net contribution to predicting the failed outcome.

Conclusion

In this limited cohort of patients with paroxysmal AF undergoing catheter ablation, non-linear machine learning analysis suggests a significant association between high epicardial fat volume and loop-recorder-detected AF recurrence during 3-year follow-up, and a potential role for such analyses in more granular, highly individualized prediction of the outcome of planned ablation. These results need further testing and validation in prospective trials.