Abstract
BACKGROUND: Recurrence after the first ablation for persistent atrial fibrillation (AF) is common. Existing clinical and structural prediction models have limited performance. Peak atrial longitudinal strain (PALS) quantifies left atrial function and may improve risk stratification. OBJECTIVE: To develop and validate a 12-month recurrence prediction model for persistent AF by integrating PALS with clinical features, and to evaluate its performance and clinical utility. METHODS: Single-center prospective cohort; of 629 patients screened, 429 were enrolled, and 410 were included in the final analysis after exclusions for image quality and loss during the blanking period. Pre-procedural PALS was measured with standardized quality control. The outcome was the first recurrence (>30 s) between the end of the 90-day blanking period and 12 months. Cox models with restricted cubic splines assessed the dose-response. Model performance (C-index, time-dependent AUC, Brier score, calibration) was compared among "Clinical-only", "Clinical + PALS", and "Clinical + LAVI" models. Reclassification (NRI/IDI) and clinical net benefit (decision curve analysis) were evaluated, along with sensitivity and interaction analyses. RESULTS: PALS independently predicted recurrence (per 1% decrease: HR 1.06, 95% CI 1.03-1.10; p < 0.001), with excellent reproducibility (ICC intraobserver 0.91-0.93, interobserver 0.87-0.91). Adding PALS to the Clinical-only model improved discrimination (C-index 0.74 vs. 0.66; ΔAUC 0.08, p < 0.001) and calibration (α = -0.02, β = 0.94). Reclassification improved (NRI 0.43; IDI 0.07; p < 0.01). The Clinical + PALS model provided greater net benefit across thresholds of 10%-30% (max ΔNB 0.06 at 24%). The PALS effect was consistent across acquisition rhythms (p-interaction = 0.417). Adding LAVI yielded more modest improvement (C-index 0.69; ΔAUC 0.03, p = 0.084). CONCLUSIONS: Pre-procedural PALS significantly improves individualized prediction of 12-month recurrence after first ablation for persistent AF when combined with clinical features. The model shows robust performance and clinical net benefit, supporting PALS as a core functional metric in pre-procedural assessment.