A retrospective cohort and machine learning study on the efficacy, safety, and recurrence prediction of combined EI-VOM and linear ablation in persistent Non-valvular atrial fibrillation

TL;DR

This study shows that combining CPVI with EI-VOM and linear ablation improves outcomes in patients with persistent atrial fibrillation, with machine learning enhancing recurrence prediction.

Contribution

The novel contribution is the evaluation of CPVI + PLUS (EI-VOM and linear ablation) alongside a stacking ensemble machine learning model for predicting AF recurrence.

Findings

CPVI + PLUS improved arrhythmia-free survival, cardiac function, and quality of life without increasing procedural risk.

Machine learning stacking model accurately predicted AF recurrence with high performance metrics.

Heart failure, valvular regurgitation, and longer AF duration were identified as key risk factors for recurrence.

Abstract

Although circumferential pulmonary vein isolation (CPVI) is the standard treatment for persistent nonvalvular atrial fibrillation (PeNVAF), its efficacy remains limited. Combining CPVI with linear ablation and ethanol infusion into the vein of Marshall (EI-VOM) (CPVI + PLUS) may improve patient outcomes.

This single-center retrospective study included hospitalized patients with PeNVAF who underwent their first radiofrequency ablation at Zigong First People's Hospital from August 2022 to August 2024. After propensity score matching (PSM), 132 patients were assigned to each group: the CPVI + PLUS group (combined EI-VOM and linear ablation) and the CPVI-only group. Outcomes were assessed at 3 and 6 months, including AF recurrence, complications, and cardiac function. Quality of life was evaluated at 6 months using the Atrial Fibrillation Effect on Quality-of-Life Questionnaire (AFEQT). The primary endpoints were AF recurrence rate and arrhythmia-free survival at 6 months after ablation; secondary endpoints included changes in cardiac function parameters, quality-of-life scores, and complication rates. Kaplan–Meier analysis was used to compare arrhythmia-free survival between the two groups, and univariate and multivariate logistic regression analyses were performed to identify risk factors for AF recurrence. In addition, a stacking ensemble model was developed to enhance prediction of AF recurrence after ablation.

There were no significant differences in baseline demographics, comorbidities, laboratory, or echocardiographic parameters between the two groups. The CPVI + PLUS group had a longer procedure time than the CPVI group, but intraoperative complication rates were similar. At 6-month follow-up, the CPVI + PLUS group showed significantly better AFEQT, left ventricular ejection fraction (LVEF), left atrial diameter(LAD), and left ventricular end-diastolic diameter (LVEDD) compared to the CPVI group. Recurrence rates of AF were lower in the CPVI + PLUS group, and arrhythmia-free survival time was longer. Multivariate analysis identified coronary artery disease, heart failure, severe preoperative valvular regurgitation, intraoperative myocardial injury, longer AF course, and larger LAD as independent risk factors for recurrence. Additional ablation strategies and left atrial appendage occlusion were protective factors. The stacking ensemble model achieved high accuracy, precision, recall, F1 score, and ROC-AUC in both training and validation cohorts. Feature importance analysis consistently identified AF course, age, body mass index, B-type natriuretic peptide, LVEF, platelet count, and uric acid as core predictors of recurrence risk.

CPVI + PLUS significantly improved arrhythmia-free survival, cardiac function, and quality of life in patients with PeNVAF without increasing procedural risk. Multivariate and machine learning analyses confirmed that additional ablation and left atrial appendage occlusion were protective against recurrence, while heart failure, valvular regurgitation, and longer AF course increased risk. The stacking model showed excellent predictive performance, and key features identified may support precise risk stratification and individualized therapy.

Figures