Machine learning–based prediction of in-hospital postoperative ischemic stroke following off-pump coronary artery bypass grafting
Year: 2026
Type: preprint
Study design: Observational Study
Abstract: Background This study aimed to develop and validate a machine learning (ML)-based model to assess the risk of ischemic stroke (IS) in hospitalised patients with coronary artery disease (CAD) after off-pump coronary artery bypass grafting (OPCABG). Methods Data were collected from the electronic health records of 25,488 hospitalised patients with CAD receiving coronary artery bypass grafting at the Beijing Anzhen Hospital of Capital Medical University (BAH-CMU cohort, training and internal validation sets) between January 1, 2019, and December 31, 2021, and at the Peking University People’s Hospital (PKUPH cohort, external validation set) between May 1, 2016, and June 30, 2024. Patient and clinical characteristics available pre-, intra-, and postoperatively were used as inputs for five candidate ML models: support vector machine, random forest, logistic regression, k-nearest neighbour (KNN), and XGBoost. Model performance was compared using the area under the receiver operating characteristic curve (AUC). Shapley Additive exPlanation (SHAP) method was utilised to visualise and interpret the complex relationships between the variables and outcomes. Post-OPCABG IS Model (POISM), a web-based tool, was further designed utilising ML algorithms. Results A total of 11,855 patients (2452 [20.9%] female) were included in this study. The best-performing model, using perioperative data, achieved the following AUCs: internal validation (XGBoost), 0.863 (95% CI, 0.793–0.933); external validation (XGBoost), 0.786 (95% CI, 0.68–0.891); and combined validation (XGBoost), 0.827 (95% CI, 0.765–0.890). The SHAP analysis illustrated significant clinical contributors associated with risk of postoperative IS. Based on this model, we created an easy-to-use website tool, POISM, which clinicians can easily utilise for clinical decision-making in patients at risk of IS. Conclusions This XGBoost model showed strong predictive performance on both the training and test datasets. Moreover, the user-friendly web tool, POISM, can quickly predict the risk of postoperative IS in hospitalised patients with CAD undergoing OPCABG, thereby supporting clinical decision-making and reducing adverse outcomes.
Source: Research Square
Institutions: Chinese Academy of Medical Sciences & Peking Union Medical College, Plastic Surgery Hospital, Xuan Wu Hospital of the Capital Medical University, Peking University People's Hospital, Capital Medical University, Beijing Anzhen Hospital, Beijing Hua Xin Hospital, Beijing Tian Tan Hospital, Qingdao University, Peking University
Language: English
Subfield: Surgery
Field: Medicine
Domain: Health Sciences
Keywords: ischemic stroke, off-pump coronary artery bypass, coronary artery disease, XGBoost, SHAP, clinical decision support systems, risk stratification, perioperative risk factors, external validation, machine learning prediction models, random forest, support vector machine, k-nearest neighbors, logistic regression
Open Access status: green
OpenAlex ID: W7221749136
