Novel predictors of late recurrence after catheter ablation for atrial fibrillation: from biomarkers to artificial intelligence models.

Atrial fibrillation (AF) is the most common sustained arrhythmia, and catheter ablation is an established rhythm-control strategy. However, late atrial arrhythmia recurrence remains common, limiting long-term procedural benefits. This review provides a structured and critical overview of current evidence on the predictors of late recurrence after AF catheter ablation across four domains: circulating biomarkers, imaging and electrophysiological substrate markers, genetic susceptibility, and artificial intelligence and machine-learning models. Novel predictors, including fibrosis-associated biomarkers, N-terminal pro-B-type natriuretic peptide, left atrial fibrosis quantified by late gadolinium enhancement cardiac magnetic resonance, low-voltage regions, epicardial adipose tissue features, and PITX2-associated genetic variants, may yield biological insights beyond conventional clinical scoring systems. Artificial intelligence-based approaches may additionally improve risk stratification through multimodal data integration; nevertheless, current evidence predominantly originates from retrospective single-center cohorts and is constrained by overfitting, variable rhythm-monitoring protocols, dataset shift, and inadequate external validation. Predictive performance also depends on the clinical and procedural context and may vary between paroxysmal and persistent AF, initial and repeat ablation, and thermal and pulsed-field energy modalities. Before routine implementation, future investigations should prioritize standardized phenotyping, rigorous prospective multicenter validation, transparent reporting, and clinically interpretable decision-support systems.