Machine learning algorithms to predict major bleeding after isolated coronary artery bypass grafting

• DOI: 10.3389/fcvm.2022.881881
• Publication Date: 2022-07-28T06:45:36Z
• Authors: Yuchen Gao; Xiaojie Liu; Lijuan Wang; Sudena Wang; Yang Yu; Yao Ding; Jingcan Wang; Hushan Ao
• Keywords: Gradient boosting; Derivation; Bleed

Objectives Postoperative major bleeding is a common problem in patients undergoing cardiac surgery and associated with poor outcomes. We evaluated the performance of machine learning (ML) methods to predict postoperative bleeding. Methods A total 1,045 who underwent isolated coronary artery bypass graft (CABG) were enrolled. Their datasets assigned randomly training (70%) or testing set (30%). The primary outcome was defined as universal definition perioperative (UDPB) classes 3–4. constructed reference logistic regression (LR) model using known predictors. also developed several modern ML algorithms. In test set, we compared area under receiver operating characteristic curves (AUCs) these algorithms LR results, TRUST WILL-BLEED risk score. Calibration analysis undertaken calibration belt method. Results prevalence 7.1% (74/1,045). For bleeds, conditional inference random forest (CIRF) showed highest AUC [0.831 (0.732–0.930)], stochastic gradient boosting (SGBT) models demonstrated next best results [0.820 (0.742–0.899) 0.810 (0.719–0.902)]. AUCs all higher than [0.629 (0.517–0.641) 0.557 (0.449–0.665)], achieved by WILL-BLEED, respectively. Conclusion successfully predicted after surgery, greater previous scoring models. Modern may enhance identification high-risk subpopulations.