Development and validation of a machine learning–based predictive model for new-onset atrial fibrillation after CABG

Objective: New-onset atrial fibrillation (NOAF) after coronary artery bypass grafting (CABG) is one of the most common post-operative complications in patients undergoing cardiac surgery and is strongly associated with adverse outcomes. We aimed to develop a machine learning (ML) model for predicting NOAF after CABG. Method: We studied 925 patients who underwent coronary artery bypass grafting (CABG). We randomly split the data into a training set (70%) and a validation set (30%). Our primary outcome was new-onset atrial fibrillation (NOAF) after surgery (i.e., the first episode of atrial fibrillation within 7 days post-operatively in patients who were in sinus rhythm before CABG). We developed eight predictive models based on machine learning algorithms and evaluated their performance using area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, specificity, F1 score, and the Youden index. SHapley Additive exPlanations (SHAP) values were computed and plotted to interpret the contribution of individual features to model predictions. Results: The incidence of NOAF was 19%. For predicting new-onset atrial fibrillation after CABG, the Gradient Boosting model achieved the highest AUC [0.842 (0.785-0.899)], outperforming the logistic regression model [0.790 (0.722-0.858)]. In addition, AUCs of all machine learning models [0.761-0.842] exceeded those of conventional risk scores, such as CHA2DS2-VASc and HATCH [0.587 (0.55-0.724), 0.62 (0.543-0.697), respectively]. Conclusion: The key features selected by machine learning methods and the resulting predictive models can predict new-onset atrial fibrillation after CABG with reasonable accuracy, which may support exploratory risk stratification and hypothesis generation; external validation is required before clinical implementation.