基于机器学习构建心脏外科术后低心排血量综合征风险预测的决策树模型  

Structuring a decision tree model based on machine learning for risk prediction of low output syndrome after cardiac surgery

作　　者：许欢[1] 洪亮[1] 沈骁 宋晓春[1] 卓荦 章淬[1] Xu Huan;Hong Liang;Shen Xiao;Song Xiaochun;Zhuo Luo;Zhang Cui(Department of ICU,Nanjing First Hospital,Nanjing Medical University,Nanjing 210006,China)

机构地区：[1]南京医科大学附属南京医院(南京市第一医院)重症医学科,南京210006

出　　处：《数字医学与健康》2025年第1期27-33,共7页DIGITAL MEDICINE AND HEALTH

基　　金：南京市卫生科技发展专项资金(ZKX19021)。

摘　　要：目的:应用机器学习(ML)算法构建预测心脏外科术后低心排血量综合征发病风险的决策树模型,并评价其预测效果。方法:从南京市第一医院重症医学科心脏重症临床数据库中提取1681例患者的临床资料,纳入基础信息、检验检查信息、手术信息、血流动力学信息等49项临床参数,通过ML建立预测模型,进而应用SHAP方法和部分依赖图筛选主要预测特征及其阈值,最终生成预测心脏术后低心排血量综合征的决策树模型,并通过受试者操作特征(ROC)曲线对模型预测效果进行评价。结果:SHAP算法分析显示术前左室射血分数[SHAP值=0.008449(0.000852)]、血乳酸水平[SHAP值=0.007434(0.000718)]、术中平均动脉压<65 mmHg(1 mmHg=0.133 kPa)持续时间[SHAP值=0.004983(0.000761)]3项参数对决策树模型具有较高的全局重要性,其构建的模型对心脏外科术后低心排血量综合征发病风险具有较高预测价值,ROC曲线下面积为0.791[95%可信区间(CI):0.718~0.864],预测准确度为0.763(95%CI:0.722~0.800),敏感度为0.736(95%CI:0.597~0.847),特异度为0.766(95%CI:0.723~0.806)。结论:应用ML算法筛选了预测心脏外科术后低心排血量综合征发病风险的特征参数,建立了预测效果良好、易于临床应用的决策树模型。ObjectiveTo evaluating the predictive effectiveness of a decision tree model constructed by machine learning(ML)algorithms for predicting the risk of low cardiac output syndrome(LCOS)after cardiac surgery.MethodsThe clinical data of 49 clinical parameters of 1681 patients,including basic information,examination and investigation information,surgical information,hemodynamic information were extracted from the clinical database of the cardiac intensive care in Nanjing First Hospital.Then,a prediction model was established through ML.The shapley additive explanations(SHAP)method and partial dependency graph were applied to screen the main prediction features and their thresholds.Finally,a decision tree model for predicting LCOS was generated,and the prediction effect of the model was evaluated through the receiver operating characteristic(ROC)curve.ResultsAnalysis using the SHAP algorithm showed that three parameters-preoperative left ventricular ejection fraction[SHAP value=0.008449(0.000852)],blood lactate level[SHAP value=0.007434(0.000718)],and the duration of mean arterial pressure<65 mmHg(1 mmHg=0.133 kPa)during surgery[SHAP value=0.004983(0.000761)]——had high global importance in the decision tree model.The model constructed with these parameters had high predictive value for the risk of postoperative LCOS in cardiac surgery patients.The area under the ROC curve was 0.791[95%Confidence Interval(CI):0.718-0.864],with a prediction accuracy of 0.763(95%CI:0.722-0.800),sensitivity of 0.736(95%CI:0.597-0.847),and specificity of 0.766(95%CI:0.723-0.806).ConclusionsThe study applied ML algorithms to select characteristic parameters for predicting the risk of postoperative LCOS in cardiac surgery patients,and established a decision tree model with good predictive performance and ease of clinical application.

关 键 词：低心排血量综合征 预测模型 机器学习 决策树 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]