Predictive potential assessment of preoperative risk factors for atrial fibrillation in patients with coronary artery disease after coronary artery bypass grafting

Abstract

Introduction. Postoperative atrial fibrillation (POAF) is one of the most common complications of coronary artery bypass grafting (CABG) and occurs in 25–65% of patients.Aim. The study aimed to assess the predictive potential of preoperative risk factors for POAF in patients with coronary artery disease (CAD) after CABG based on machine learning (ML) methods.Material and Methods. An observational retrospective study was carried out based on data from 866 electronic case histories of CAD patients with a median age of 63 years and a 95% confidence interval [63; 64], who underwent isolated CABG on cardiopulmonary bypass. Patients were assigned to two groups: group 1 comprised 147 (18%) patients with newly registered atrial fibrillation (AF) paroxysms; group 2 included 648 (81.3%) patients without cardiac arrhythmia. The preoperative clinical and functional status was assessed using 100 factors. We used statistical analysis methods (Chi-square, Fisher, Mann – Whitney, and univariate logistic regression (LR) tests) and ML tests (multivariate LR and stochastic gradient boosting (SGB)) for data processing and analysis. The models’ accuracy was assessed by three quality metrics: area under the ROC-curve (AUC), sensitivity, and specificity. The cross-validation procedure was performed at least 1000 times on randomly selected data.Results. The processing and analysis of preoperative patient status indicators using ML methods allowed to identify 10 predictors that were linearly and nonlinearly related to the development of POAF. The most significant predictors were the anteroposterior dimension of the left atrium, tricuspid valve insufficiency, ejection fraction <40%, duration of the P–R interval, and chronic heart failure of functional class III–IV. The accuracy of the best predictive multifactorial model of LR was 0.61 in AUC, 0.49 in specificity, and 0.72 in sensitivity. The values of similar quality metrics for the best model based on SGB were 0.64, 0.6, and 0.68, respectively.Conclusion. The use of SGB made it possible to verify the nonlinearly related predictors of POAF. The prospects for further research on this problem require the use of modern medical care methods that allow taking into account the individual characteristics of patients when developing predictive models.