Development and validation of a prediction model for atrial fibrillation recurrence after radiofrequency ablation: integrating echocardiographic and clinical indicators.

backgroundRadiofrequency ablation (RFA) is an established treatment for symptomatic atrial fibrillation (AF), yet post-ablation recurrence remains a significant clinical challenge. This study aimed to develop and internally validate a prediction model for post-RFA AF recurrence by integrating echocardiographic parameters and clinical indicators, and to explore potential nonlinear associations of key indices for risk stratification.

methodsA total of 532 consecutive patients who underwent first-time RFA for AF between January 2022 and October 2024 were retrospectively enrolled. The database was locked on October 31, 2025, ensuring a complete 12-month follow-up for all patients. Post-ablation recurrence was defined as any documented AF episode lasting ≥ 30 s on electrocardiogram (ECG) or 24-hour Holter monitoring, occurring after a 3-month blanking period. Independent predictors were identified by forward stepwise multivariable logistic regression (α-in = 0.05, α-out = 0.10), and a nomogram was constructed for model visualization. Internal validation was performed using 1000 bootstrap resamples. Model performance was assessed by the area under the receiver operating characteristic curve (AUC), calibration plots, decision curve analysis (DCA), and clinical impact curves (CIC). Restricted cubic spline (RCS) analysis with 3 quantile knots was used to explore the non-linear associations between left atrial diameter (LAD), left atrial sphericity index (LASI) and post-ablation recurrence risk. Inter- and intra-observer consistency of echocardiographic measurements was verified by intraclass correlation coefficient (ICC).

resultsLAD, LASI (per 0.1 unit increase), AF duration category, and CHA₂DS₂-VASc score were identified as independent predictors of post-RFA AF recurrence (CHA₂DS₂-VASc, LAD, AF duration: all P < 0.001; LASI: P = 0.001). The model equation was: logit(p) = -9.093 + 0.281 × (CHA₂DS₂-VASc score) + 0.081 × (LAD) + 0.569 × (LASI, per 0.1 unit) + Σβk × (AF duration group). The model exhibited moderate discriminatory ability (AUC = 0.752; 95% CI: 0.708-0.796). DCA showed higher net benefit than the treat-all and treat-none strategies across threshold probabilities of 0.10-0.60, and CIC indicated that at thresholds between 0.10 and 0.30 the model identified approximately 250-330 true positive cases per 1000 patients. At the Youden‑based optimal cut-off (0.35), PPV and NPV were 48.2% and 82.5%. RCS analysis revealed a nonlinear threshold effect for LAD, with recurrence risk increasing sharply above 50 mm (inter-observer ICC = 0.92; intra-observer ICC = 0.95). For LASI, the association was predominantly linear, with a visual bend near 0.78 that lacked statistical support (nonlinearity P = 0.9516; inter-observer ICC = 0.90; intra-observer ICC = 0.93).

conclusionsThis internally validated prediction model, integrating LAD, LASI, AF duration category, and CHA₂DS₂-VASc score, demonstrated moderate discrimination and adequate calibration, with higher net benefit than default strategies on decision curve analysis. LAD showed an exploratory nonlinear inflection near 50 mm, whereas LASI showed a predominantly linear association; the bend near 0.78 is only a hypothesis-generating observation. These findings require rigorous external validation before any clinical application. The nomogram offers a tool for research-oriented risk estimation, pending multicenter confirmation.