CHEN Wenxiu , TAO Hong , XIE Rongrong , XUE Yinying , SHEN Xiao , XU Huan , HONG Liang , ZHANG Cui , SUN Fang
  • Department of Intensive Care Unit, Nanjing Medical University Affiliated Nanjing Hospital (Nanjing First Hospital), Nanjing, 210006, P. R. China;
SUN Fang, Email: xevia1993@126.com
Cardiac surgery; acute kidney injury; machine learning; prediction model; cardiopulmonary bypass; risk factors
10.7507/1007-4848.202604102
Export PDF Favorites Scan Get Citation

Abstract Full text Figures/Tables References Cited by

Objective To construct a machine learning (ML) model incorporating preoperative, intraoperative, and early postoperative variables to predict cardiac surgery-associated acute kidney injury (CSA-AKI), thereby providing decision support for early clinical warning and intervention. Methods Patients who underwent cardiac surgery at Nanjing Medical University Affiliated Nanjing Hospital from 2020 to 2022 were included in this study. The cohort was randomly divided into a training set and a validation set at a ratio of 7:3. Five ML models were constructed, including random forest (RF), logistic regression (LR), gradient boosted decision tree (GBDT), extreme gradient boosting (XGBoost), and support vector machines (SVM). Model performance was evaluated using metrics such as the area under the receiver operating characteristic curve (AUC), and the SHapley Additive exPlanations (SHAP) method was applied for model interpretation. Results A total of 3 679 patients were included, comprising 2 217 males and 1 462 females, with a median age of 64 (56, 70) years. Fifteen feature variables were incorporated: history of heart failure, age, body mass index, smoking history, creatinine, cardiopulmonary bypass time, total blood transfusion volume, total urine output, uric acid, hemoglobin, Acute Physiology and Chronic Health Evaluation Ⅱ (APACHE Ⅱ) score, European System for Cardiac Operative Risk Evaluation (EuroSCORE), postoperative white blood cell count, postoperative lactate, and postoperative lymphocyte count. Among the five ML models, LR, RF, GBDT, and XGBoost showed no statistical differences in performance within the validation set; however, considering the dimensions of performance, model robustness, and clinical interpretability, the LR model was deemed superior. SHAP analysis quantified the predictive weights of each variable (indicating that a history of heart failure and advanced age were high-risk features) and visually demonstrated the underlying causes of personalized risk for individual patients. Based on the optimized LR model, an online web-based risk calculator was successfully developed to output personalized probability of CSA-AKI occurrence. Conclusion A clinical prediction model for CSA-AKI risk is constructed using ML techniques, demonstrating favorable predictive performance and interpretability. This risk assessment tool can provide early warnings for clinicians and assist in optimizing postoperative treatment strategies, thereby reducing the risk of CSA-AKI occurrence.

Copyright ? the editorial department of Chinese Journal of Clinical Thoracic and Cardiovascular Surgery of West China Medical Publisher. All rights reserved