Integrative Evaluation of Atrial Function and Electromechanical Coupling as Predictors of Postoperative Atrial...

Background and Objectives: Postoperative atrial fibrillation (POAF) remains one of the most frequent complications after cardiac surgery, increasing the risk of morbidity, prolonged hospitalization, and adverse long-term outcomes. Although several clinical and echocardiographic factors have been associated with POAF, the integrated contribution of atrial conduction delay, biatrial mechanics, and atrioventricular coupling to arrhythmogenesis remains unclear. Materials and Methods: This retrospective study included 131 adult patients undergoing coronary artery bypass grafting and/or aortic valve replacement. Preoperative echocardiography within one week before surgery provided detailed assessment of atrial phasic function, valvular motion, and total atrial conduction time (TACT). Univariate analysis was followed by multivariable modeling using penalized logistic regression (Elastic Net) to identify the most robust predictors of POAF. Discriminative performance and calibration were evaluated via receiver operating characteristic (ROC) and calibration analysis. An exploratory Extreme Gradient Boosting (XGBoost) model with SHapley Additive exPlanations (SHAP) analysis was used to confirm the stability and directionality of nonlinear feature interactions. Results: POAF occurred in 47 (36%) patients. The Elastic Net model identified prolonged TACT, reduced right atrial active emptying fraction (RAAEF), increased indexed minimal left atrial volume (MIN LA/BSA), and lower tricuspid annular plane systolic excursion (TAPSE) as the most informative predictors. The model demonstrated excellent internal discrimination (AUC = 0.95; 95% CI 0.91-0.99) and satisfactory calibration (Hosmer-Lemeshow p = 0.41). Exploratory XGBoost analysis yielded concordant feature hierarchies, confirming the physiological consistency of the results. Conclusions: POAF arises from an identifiable electromechanical substrate characterized by atrial conduction delay, biatrial mechanical impairment, and reduced atrioventricular coupling. A parsimonious, regularized statistical model accurately delineated this profile, while complementary machine-learning analysis supported its internal validity. These findings underscore the potential of echocardiographic electromechanical parameters for refined preoperative risk stratification, pending prospective multicenter validation.