基于人工智能算法与中医体质类型的老年糖尿病患者合并心血管疾病预测模型  

Prediction models for cardiovascular diseases in elderly diabetic patients based on artificial intelligence algorithms and traditional Chinese medicine constitution types

作　　者：朱晶晶 张扬 龚小会 郑婉怡 毛丽君 宋花玲 徐祥龙 ZHU Jingjing;ZHANG Yang;GONG Xiaohui;ZHENG Wanyi;MAO Lijun;SONG Hualing;XU Xianglong(School of Public Health,Shanghai University of Traditional Chinese Medicine,Shanghai 201203,China;Yuyuan Community Health Service Center,Huangpu District,Shanghai 200010,China;Obstetrics Department,Obstetrics and Gynecology Hospital of Fudan University,Shanghai 200011,China)

机构地区：[1]上海中医药大学公共健康学院,上海201203 [2]上海市黄浦区豫园社区卫生服务中心,上海200010 [3]复旦大学附属妇产科医院产科,上海200011

出　　处：《上海中医药杂志》2025年第3期1-6,共6页Shanghai Journal of Traditional Chinese Medicine

基　　金：上海市卫健委中医药科研项目(2024QN108);上海市教委大学生创新创业训练计划项目(202410268154);上海中医药大学预算内项目(2021LK008)。

摘　　要：目的结合人工智能算法、临床血液检查数据与中医体质理念,构建针对老年糖尿病患者的心血管疾病预测模型,为个性化健康管理提供新手段。方法纳入2018年至2022年上海市黄浦区豫园社区1068名65岁以上的糖尿病患者,通过逻辑回归模型(logistic regression)和极限梯度提升模型(XGBoost)两种机器学习算法,利用中医的9种体质变量和血液检查指标,构建老年糖尿病患者的心血管疾病预测模型,并使用曲线下面积指标评估性能。结果两种预测模型的性能均达到了可接受的水平(logistic regression,曲线下面积=0.77;XGBoost,曲线下面积=0.79),XGBoost是最佳模型;XGBoost重要性排序前15的变量中,中医体质的变量分别是平和质、气虚质、湿热质、血瘀质。结论基于中医体质的预测模型可以预测老年糖尿病患者的心血管疾病,为老年糖尿病患者的个性化健康管理和社区卫生服务提供了新策略,同时促进中医“治未病”理念的应用。Objective To develop prediction models for cardiovascular disease in elderly diabetic patients by integrating artificial intelligence algorithms,clinical blood test data,and the concept of traditional Chinese medicine(TCM)constitution types,in order to provide new approaches for personalized health management.Methods A total of 1,068 diabetic patients aged 65 and above,from the Yuyuan community of Huangpu District in Shanghai,were included in the study from 2018 to 2022.Two machine learning algorithms,logistic regression and extreme gradient boosting(XGBoost),were used to construct cardiovascular disease prediction models for elderly diabetic patients based on nine TCM constitution variables and blood test indicators.The models'performance was assessed by using the area under the curve(AUC)metric.Results Both prediction models showed acceptable performance(logistic regression,AUC=0.77;XGBoost,AUC=0.79),with XGBoost being the best model.Among the top 15 most important variables in the XGBoost model,the TCM constitution variables included were balanced constitution,qi-deficiency constitution,damp-heat constitution,and blood stasis constitution.Conclusions The prediction model based on TCM constitution can effectively predict cardiovascular diseases in elderly diabetic patients,and provide a new strategy for personalized health management and community healthcare services for these patients.This model also promotes the application of the TCM concept of"Zhiweibing",that is,the prevention of disease before it occurs.

关 键 词：糖尿病 心血管疾病 人工智能 中医体质 老年患者 慢性病 预测模型 

分 类 号：R259[医药卫生—中西医结合]