基于隐结构模型的原发性肝癌不同症状群预后差异性分析  

Analysis of Prognostic Differences among Different Symptom Groups of Primary Liver Cancer Based on Implicit Structure Model

作　　者：叶俊秋 郑琪光 钟昆禹 舒梓心 周雪忠[4] 周亚娜 吴辉坤 李晓东[1,2,3] 卢晨霞 Ye Junqiu;Zheng Qiguang;Zhong Kunyu;Shu Zixin;Zhou Xuezhong;Zhou Yana;Wu Huikun;Li Xiaodong;Lu Chenxia(Afiliated Hospital of Hubei University of Chinese Medicine Wuhan 430061,China;Hubei Key Laboratory of the Theory and Application Research of Liver and Kidney in Traditional Chinese Medicine,Hubei Provincial Hospital of Traditional Chinese Medicine,Wuhan 430061,China;Hubei Province Academy of Traditional Chinese Medicine,Wuhan 430061,China;School of Computer and Information Technology,Beijing Jiaotong University,Beijing 100044,China)

机构地区：[1]湖北中医药大学附属医院,武汉430061 [2]湖北省中医院/中医肝肾研究及应用湖北省重点实验室,武汉430061 [3]湖北省中医药研究院,武汉430061 [4]北京交通大学计算机与信息技术学院,北京100044

出　　处：《世界科学技术-中医药现代化》2023年第1期422-429,共8页Modernization of Traditional Chinese Medicine and Materia Medica-World Science and Technology

基　　金：国家中医药管理局湖北中医药大学“十四五”优秀学科团队建设项目(5431-1007020200805):湖北省公共卫生领军人才培养计划(鄂卫通【2021】73号),负责人:李晓东。

摘　　要：目的基于隐结构模型建立原发性肝癌(Primary liver cancer,PLC)症状人群的分类方法,探究PLC不同症状群的生存预后差异性。方法利用人机协同表型谱标注系统(www.tcmai.org),对565例西医诊断为PLC的结构化中医电子病历文本进行实体标注,建立症状数据库并进行规范化处理,运用Lantern 5.0软件构建PLC的中医症状隐结构模型,结合中医专业理论知识基础,对分类结果进行模型学习诠释,归纳出有临床意义的症状人群分类,分析PLC特征性症状占比进行症状群-人群匹配,结合患者住院时长及生存信息,分析PLC不同症状群人群的预后差异。结果共纳入131个中医症状(即显变量)构建PLC症状隐结构模型,获得14个隐变量,通过综合聚类归纳出10个症状群,具体表现为精神状态相关症状群、消化系统相关症状群、癌性疲乏相关症状群、呼吸循环相关症状群和其他一般症状群5大类。其中癌性疲乏症状群和黄疸类症状群的预后具有显著差异性。结论通过建立隐结构模型挖掘PLC症状群,划分人群的研究方法,证实了PLC不同症状群存在预后差异性,为症状群分类及人群划分研究提供了一个新的思路。Objective To establish a classification method for patients with primary liver cancer(PLC) symptoms based on Implicit Structure Model,and explore the difference of survival prognosis among different PLC symptom cluster.Methods Using the human-computer collaborative phenotypic spectrum labeling system(www.tcmai.org),565 cases of structured TCM electronic medical records diagnosed as PLC by western medicine were physically labeled,the symptom database was established and standardized,the Implicit Structure Model of TCM symptoms of PLC was constructed by using Lantern 5.0 software,and the classification results were interpreted by model learning based on the theoretical knowledge of TCM specialty.Summarize the classification of clinically significant symptom groups,analyze the proportion of characteristic symptoms of PLC for symptom group-crowd matching,and analyze the difference of prognosis among different symptom groups of PLC combined with the length of hospital stay and survival information of patients.Results A total of 131 TCM symptoms(manifest variable) were included to construct the PLC symptom Implicit Structure Model,and 14 implicit variables were obtained.Through comprehensive clustering,10 symptom clusters were summed up,which were mainly manifested as five categories:mental state related symptom cluster,digestive system related symptom cluster,cancer fatigue related symptom cluster,respiratory cycle related symptom cluster and other general symptom clusters.The prognosis of cancer fatigue symptom group and jaundice symptom group has significant difference.Conclusion The research method of setting up Implicit Structure Model to mine PLC symptom groups and divide the population has confirmed that there are differences in prognosis of different symptom groups of PLC,and provided a new idea for the research of symptom group classification and population division.

关 键 词：原发性肝癌 隐结构模型 症状群 人群划分 生存分析 中医电子病历 

分 类 号：R-058[医药卫生]