Abstract
BACKGROUND: The PRAGUE-25 trial demonstrated catheter ablation (CA) superiority over lifestyle modification plus antiarrhythmic drugs (LFM + AAD) in obese patients with atrial fibrillation (AF). However, obese AF patients represent a heterogeneous population with varying pathophysiological substrates. We hypothesized that distinct patient phenotypes may exhibit differential treatment responses. METHODS: This post-hoc analysis applied hierarchical cluster analysis (Ward's D2 method, Euclidean distance) to 122 PRAGUE-25 patients with complete data using six delta (12-month minus baseline) echocardiographic and metabolic variables (Δ-LAVI, Δ-LVEDD, Δ-NT-proBNP, Δ-triglycerides, Δ-leukocytes, Δ-platelets). Treatment effects within each phenotype were compared using Fisher's exact test and odds ratios with 95% confidence intervals. RESULTS: Three distinct phenotypes were identified: Metabolic (n = 20, 16%), characterized by highest triglycerides with minimal structural remodeling; Intermediate Remodeling (n = 67, 55%), distinguished by largest LV chamber with lowest inflammatory burden; and Advanced Neurohormonal/Inflammatory (n = 35, 29%), exhibiting highest NT-proBNP with elevated inflammatory markers. While cluster membership did not predict overall AF freedom (χ(2) = 3.45, p = 0.178), CA superiority was statistically significant only in the Intermediate phenotype (OR 3.98, 95% CI: 1.01-15.57, p = 0.047; AF freedom 88.5% CA vs 65.9% LFM + AAD). In the Metabolic and Advanced Neurohormonal/Inflammatory phenotypes (45% of patients), no statistically significant treatment difference was observed; however, wide confidence intervals preclude conclusions of treatment equivalence in these underpowered subgroups. These findings appear to be primarily driven by the Intermediate Remodeling phenotype. CONCLUSIONS: These hypothesis-generating findings suggest phenotype-dependent treatment response heterogeneity in obese atrial fibrillation. As cluster membership can only be determined retrospectively, prospective validation using baseline predictor models is required before clinical application.