中国临床药理学与治疗学2025，Vol.30Issue(9)：1182-1192,11.DOI:10.12092/j.issn.1009-2501.2025.09.004
构建和验证乳腺癌患者麻醉苏醒延迟风险的机器学习网络计算器
Construction and validation of a machine learning network calculator for the risk of delayed awakening from anaesthesia in breast cancer patients
摘要
Abstract
AIM:To construct a network calcula-tor based on machine learning(ML)models to pre-dict the risk of delayed awakening from anaesthesia in breast cancer(BC)patients.METHODS:A total of 435 BC patients surgically treated at our hospital from January 2023 to June 2024 were selected.The Boruta algorithm was used to screen for important characteristic variables for the risk of delayed awak-ening from anaesthesia.All patients were randomly assigned to a training set(n=261)and a test set(n=174)based on a 3:2 ratio and nine ML models were constructed and trained.Nine ML models were evaluated on the basis of receiver operating charac-teristic(ROC)curves for a random sample of 10 sub-jects and the clinical utility of the models was as-sessed using decision curve analysis.Combined with SHapley Additive exPlanations(SHAP)bar graphs,summary graphs and force diagrams additional in-terpretation and visualization of the ML model.Con-struction of a network calculator for predicting the risk of delayed awakening from anesthesia in BC pa-tients using the R package.RESULTS:Of the 435 BC patients,25.1%experienced delayed awakening from anesthesia.Boruta algorithm screened seven feature variables.The ROC curve shows that the XG-Boost model has the highest area under the curve(AUC)for 10 random samples among the 9 ML mod-els,and the decision curve shows that the XGBoost model has a significant clinical net benefit.The SHAP bar graph shows the importance of ASA classi-fication,surgery time,anesthesia time,intraopera-tive blood loss,propofol,preoperative anemia,and intraoperative hypothermia,and the SHAP summa-ry graph reflects the distribution of the ranges of in-fluence of the seven important characteristic vari-ables,which are"separated at the ends."The SHAP force diagram visualization XGBoost model predict-ed the risk of delayed awakening from anesthesia for individual patients with a predictive value of 0.998 for patients with delayed awakening from an-esthesia and 0.008 91 for patients without delayed awakening from anesthesia.A web-based calculator(https://xz-nomogram.shinyapps.io/DE_web/)based on an interpretable XGBoost model effective-ly predicts the risk of delayed awakening from anes-thesia in BC patients.CONCLUSION:ASA classifica-tion,surgery time,propofol,intraoperative blood loss,anaesthesia time,preoperative anaemia and intraoperative hypothermia are important charac-teristic variables for the risk of delayed awakening from anaesthesia in BC patients.The network calcu-lator based on the interpretable XGBoost model can accurately and quickly quantify the risk of de-layed awakening from anaesthesia,which can help clinicians to effectively adjust the treatment strate-gy and better improve the prognosis of patients.关键词
乳腺癌/麻醉苏醒延迟/机器学习/Boru-ta算法/SHAP/XGBoost模型/网络计算器Key words
breast cancer/delayed awakening from anaesthesia/machine learning/Boruta algo-rithm/SHAP/XGBoost model/network calculator分类
医药卫生引用本文复制引用
葛亮,冷玉芳,张鹏,孔令国,韩旭东..构建和验证乳腺癌患者麻醉苏醒延迟风险的机器学习网络计算器[J].中国临床药理学与治疗学,2025,30(9):1182-1192,11.基金项目
甘肃省自然科学基金项目(20JR10RA423) （20JR10RA423）
吴阶平医学基金会临床科研专项资助基金(320.6750.2023-18-103) （320.6750.2023-18-103）
兰州市科技(人才)项目(2020-ZD-1) （人才）
