Cardiometabolic multimorbidity and atrial fibrillation: insights from traditional statistical and artificial...

Cardiometabolic multimorbidity (CMM) is increasingly prevalent among patients with atrial fibrillation (AF), yet its independent impact on post-ablation outcomes, the underlying mechanistic pathways, and the optimal approach to risk prediction in this population remain incompletely defined.

We analyzed three independent cohorts of patients undergoing AF catheter ablation: a derivation cohort (n = 3,308), an external validation cohort, and a prospective testing cohort. CMM was defined as the coexistence of two or more cardiometabolic conditions. Cox proportional hazards models and propensity score-matched analyses assessed the association between CMM and AF recurrence, all-cause death, and cardiovascular death. Mediation analysis quantified the contributions of structural, metabolic, and inflammatory pathways. Ten machine learning algorithms were developed and validated for predicting AF recurrence, and model performance was compared against eight established clinical risk scores. Time-dependent ROC analysis was used to evaluate discrimination across multiple follow-up horizons.

Among 3,308 patients in the derivation cohort, 686 (20.7%) had CMM. CMM was independently associated with an increased risk of AF recurrence (adjusted HR 1.17, 95% CI 1.03-1.33), all-cause death (HR 1.83, 95% CI 1.11-3.00), and cardiovascular death (HR 2.22, 95% CI 1.24-3.98), with a graded dose-response relationship across the number of CMM components (p for trend < 0.001). These associations persisted in propensity score-matched analyses (HR 1.32, 95% CI 1.16-1.50) and were replicated in external and prospective cohorts. Mediation analysis identified left atrial diameter (46.5% of total effect), insulin resistance (METS-IR, 25.6%), left atrial appendage emptying velocity (24.9%), and the uric acid-to-HDL ratio (UHR, 24.1%) as key mediators of recurrence risk. For pre-ablation prediction, the LightGBM model achieved the best discriminative performance (ROC-AUC 0.766, 95% CI 0.687-0.844; PR-AUC 0.755), outperforming all conventional risk scores. Time-dependent AUC values at 1, 2, and 3 years were 0.776, 0.760, and 0.730, respectively. A post-blanking model incorporating early recurrence (TabPFN) achieved an ROC-AUC of 0.873. Both models were validated in external and prospective cohorts.

CMM is an independent predictor of poor outcomes after AF ablation, driven by structural, metabolic, and inflammatory remodeling that is partially mediated through left atrial enlargement. A machine learning-based prediction provides more accurate risk stratification than traditional clinical scores, supporting personalized management in this high-risk population.