中国中医急症2025，Vol.34Issue(9)：1479-1482,1491,5.DOI:10.3969/j.issn.1004-745X.2025.09.008
基于5种机器学习构建脓毒症中医证素与临床指标的死亡预测模型
Development of a Mortality Prediction Model for Sepsis Using Five Machine Learning Approaches Based on Traditional Chinese Medicine Syndrome Elements and Clinical Indicators
摘要
Abstract
Objective:This study aims to develop five distinct machine learning models to predict the risk of mortality in sepsis by integrating Traditional Chinese Medicine(TCM)syndrome elements and clinical indicators.Methods:Data were collected from 315 sepsis patients admitted to the intensive care unit(ICU)of Cangzhou Hos-pital of Integrated Traditional Chinese and Western Medicine,Hebei Province.Patients were divided into survival and death groups based on their 30-day outcomes.Relevant patient information was collected.Feature selection was performed using LASSO regression and multivariate analysis.Five machine learning models were constructed and evaluated using 10-fold cross-validation.Model performance was compared based on metrics such as AUC,sensitivity,and specificity.SHAP(SHapley Additive exPlanations)values were used to analyze the importance of features within the models.Results:8 key predictors,i.e.,age,albumin(Alb),magnesium(Mg2+),sodium(Na+),ac-tivated partial thromboplastin time(APTT),and TCM syndrome elements of phlegm,blood stasis,and brain involve-ment,were selected through LASSO regression and multivariate analysis.Among the five constructed machine learning models,the Random Forest(RF)model demonstrated the best predictive performance,with an AUC of 0.883(95％CI:0.848～0.919).Its high sensitivity,specificity,precision,and F1 score indicated excellent classifica-tion capability and stability.The calibration curve of the RF model also showed good agreement between predicted and observed outcomes.Conclusion:Among the five machine learning prediction models developed based on a combination of TCM syndrome elements and clinical indicators,the RF model exhibited the best performance,showing potential for clinical application in predicting 30 day mortality risk in sepsis patients.关键词
脓毒症/机器学习/中医证素/十折交叉验证/预测模型/随机森林模型Key words
Sepsis/Machine learning/Traditional Chinese medicine syndrome elements/10-fold cross-valida-tion/Prediction model/Random forest model分类
医药卫生引用本文复制引用
黄赵鑫,王健,周玉灿,申建国,陈羽,刘倩倩,刘春龙..基于5种机器学习构建脓毒症中医证素与临床指标的死亡预测模型[J].中国中医急症,2025,34(9):1479-1482,1491,5.基金项目
河北省中医药管理局科研计划项目(2022232) （2022232）
河北省名中医刘春龙工作室建设项目(冀中医药函[2024]37号) （冀中医药函[2024]37号）
