基于可解释机器学习的肝硬化患者早期并发食管静脉曲张风险预测研究  

Risk prediction of early esophageal varices in patients with liver cirrhosis based on interpretable machine learning

作　　者：尹钰恒 王雨雯 范捷 杨春[3] 王伟 Yin Yuheng;Wang Yuwen;Fan Jie;Yang Chun;Wang Wei(College of Public Health,Chongqing Medical University;School of Public Health Emergency Management,Chongqing Medical and Pharmaceutical College;Department of Infection Diseases,The First Affiliated Hospital of Chongqing Medical University)

机构地区：[1]重庆医科大学公共卫生学院,重庆400016 [2]重庆医药高等专科学校公共卫生与应急管理学院,重庆401331 [3]重庆医科大学附属第一医院感染科,重庆400016

出　　处：《重庆医科大学学报》2025年第3期389-396,共8页Journal of Chongqing Medical University

基　　金：重庆医科大学未来医学青年创新资助项目(编号:W0150)。

摘　　要：目的:探讨肝硬化患者并发食管静脉曲张的风险因素,并建立预测模型,以便为预防肝硬化患者早期并发食管静脉曲张提供合理的指导。方法:回顾性纳入2006年12月至2021年5月在重庆医渡云大数据平台有电子记录的1113例肝硬化患者。通过结合递归特征消除算法(recursive feature elimination,RFE)和4种机器学习方法筛选特征,建立5种机器学习预测模型,包括逻辑回归、随机森林、支持向量机(support vector machine,SVM)、决策树和极端梯度提升(eXtreme Gradient Boosting,XGBoost)模型。通过绘制受试者工作特征(receiver operating characteristic,ROC)曲线评估不同模型的性能,并根据表现最优的模型分析肝硬化患者并发食管静脉曲张的风险因素。SHAP图用于解释各风险因素对患者的影响。结果:XGBoost模型在预测肝硬化患者并发食管静脉曲张的风险方面表现最佳,ROC曲线下面积为0.872(95%CI=0.813~0.918)。SHAP图表明,影响肝硬化患者并发食管静脉曲张的风险因素包括血小板计数、门静脉内径、胆碱酯酶、白蛋白、谷丙转氨酶、血红蛋白、凝血酶原比值、凝血酶原时间和血清总蛋白等9项临床指标。结论:本研究构建的XGBoost预测模型具有较高的预测价值,其筛选出的风险因素对临床防治肝硬化患者早期并发食管静脉曲张具有一定的指导意义。Objective:To investigate the risk factors for esophageal varices in patients with liver cirrhosis,to establish a predictive model,and to provide reasonable guidance for the prevention of early esophageal varices in patients with liver cirrhosis.Methods:A retrospective analysis was performed for 1113 patients with liver cirrhosis who attended the hospitals in Chongqing,China from December 2006 to May 2021.Recursive feature elimination(RFE)and four machine learning methods were used for the screening of features,and five machine learning predictive models were established by logistic regression,random forest,support vector machine(SVM),decision tree,and eXtreme Gradient Boosting(XGBoost).The receiver operating characteristic(ROC)curve was used to evaluate the performance of each model,and the model with the best performance was used to investigate the risk factors for esophageal varices in patients with liver cirrhosis.SHAP plots were used to explain the impact of each risk factor on patients.Results:The XGBoost model showed the best performance in predicting the risk of esophageal varices in patients with liver cirrhosis,with an area under the ROC curve of 0.872(95%CI=0.813-0.918).SHAP plots showed that platelet count,diameter of the portal vein,cholinesterase,albumin,alanine aminotransferase,hemoglobin,prothrombin ratio,prothrombin time,and serum total protein were risk factors for esophageal varices in patients with liver cirrhosis.Conclusion:This study shows that the XGBoost predictive model has a relatively high predictive value,and the risk factors obtained by this model have a certain guiding significance for the clinical prevention and treatment of early esophageal varices in patients with liver cirrhosis.

关 键 词：机器学习 肝硬化 食管静脉曲张 风险因素 

分 类 号：R543.6[医药卫生—心血管疾病]