基于随机森林和偏相关分析的小儿肺炎痰热闭肺证中医证候诊断模型研究  

Research on traditional Chinese medicine syndrome diagnosis model of pediatric pneumonia with accumulation of phlegm-heat syndrome based on random forest and partial correlation analysis

作　　者：宫文浩 兰天莹 杨燕[3] 戴启刚 冯容 何贤芬[5] 艾军 GONG Wenhao;LAN Tianying;YANG Yan;DAI Qigang;FENG Rong;HE Xianfen;AI Jun(School of Basic Medicine,Guangxi University of Chinese Medicine,Nanning 530200,China;Zhuang Medical College,Guangxi University of Chinese Medicine,Nanning 530200,China;Department of Traditional Chinese Medicine,Beijing Children’s Hospital,Capital Medical University,Beijing 100045,China;Department of Pediatrics,Affiliated Hospital of Nanjing University of Chinese Medicine,Nanjing 210023,China;The First Affiliated Hospital of Guangxi University of Chinese Medicine,Nanning 530023,China)

机构地区：[1]广西中医药大学基础医学院,南宁530200 [2]广西中医药大学壮医药学院,南宁530200 [3]首都医科大学附属北京儿童医院中医科,北京100045 [4]南京中医药大学附属医院儿科,南京210023 [5]广西中医药大学第一附属医院,南宁530023

出　　处：《中华中医药杂志》2023年第9期4497-4501,共5页China Journal of Traditional Chinese Medicine and Pharmacy

基　　金：国家自然科学基金项目(No.81660761);广西自然科学基金重点项目(No.2018GXNSFDA281008);广西研究生教育创新计划资助项目(No.YCBXJ2021001)。

摘　　要：目的:基于随机森林(RF)和偏相关分析(PCA)的方法,构建小儿肺炎痰热闭肺证中医证候诊断模型。方法:以大样本、多中心小儿肺炎痰热闭肺证病例作为数据源,运用RF方法建立小儿肺炎痰热闭肺证诊断模型,以准确率、召回率、灵敏度、特异度、精确度和受试者工作特征(ROC)曲线下面积(AUC)进行模型评价,并对提取的模型重要特征症状进行PCA。结果:超参数优化后的RF模型准确率为90.58%,灵敏度为94.52%,特异度为86.15%。其中关键特征变量为痰多黏稠、痰色黄、苔薄黄、舌红、流涕、气喘、气促、咽红、苔黄腻、发热等。其主要病机为“痰”和“热”。结论:基于RF和PCA可以有效建立小儿肺炎痰热闭肺证辨证诊断模型,为小儿肺炎临床辨证论治提供客观依据,为中医药标准化进程提供参考。Objective:To construct a diagnosis model of pediatric pneumonia with accumulation of phlegm-heat syndrome based on random forest(RF)and partial correlation analysis(PCA).Methods:Using a large sample and multi-center cases of children with pneumonia with accumulation of phlegm-heat syndrome as the data source,the RF method was used to establish a diagnosis model of pediatric pneumonia with accumulation of phlegm-heat syndrome.The accuracy rate,recall rate,sensitivity,specificity,precision and receiver operating characteristic(ROC)curve were calculated.The area under curve(AUC)was used to evaluate the model,and PCA was performed on the extracted important characteristic symptoms of the model.Results:The accuracy of the RF model after hyperparameter optimization was 90.58%,the sensitivity was 94.52%,and the specificity was 86.15%.The key characteristic variables were thick phlegm,yellow phlegm,a tongue coating with thin and yellow,red tongue,runny nose,asthma,shortness of breath,red throat,yellow and greasy coating,and fever,etc..Its main pathogenesis was‘phlegm’and‘heat’.Conclusion:Based on RF and PCA,a syndrome differentiation model of pediatric pneumonia with accumulation of pediatric phlegm-heat syndrome can be effectively established,which to provide an objective basis for the clinical diagnosis and treatment of pediatric pneumonia,and provide reference for the standardization process of traditional Chinese medicine.

关 键 词：随机森林 偏相关分析 小儿肺炎 痰热闭肺证 中医证候诊断模型 

分 类 号：R272[医药卫生—中医儿科学]