基于机器学习构建老年糖尿病患者轻度认知障碍风险评估模型  

Risk assessment of mild cognitive impairment in elderly patients with diabetes mellitus based on machine learning

作　　者：张海鑫 张一方 谢芷兰 张纹菱 王宇萍[1] 李晋磊 ZHANG Haixin;ZHANG Yifang;XIE Zhilan;ZHANG Wenling;WANG Yuping;LI Jinlei(Department of Epidemiology and Biostatistics,School of Population Medicine and Public Health,Chinese Academy of Medical Sciences,Peking Union Medical College,Beijing 100005,China)

机构地区：[1]中国医学科学院北京协和医学院群医学及公共卫生学院流行病与生物统计学系,北京100005

出　　处：《中华疾病控制杂志》2024年第3期284-289,309,共7页Chinese Journal of Disease Control & Prevention

基　　金：美国中华医学基金会(CMB 22-467)。

摘　　要：目的基于机器学习构建老年糖尿病患者轻度认知障碍(mild cognitive impairment,MCI)风险评估模型,为老年糖尿病患者认知障碍的早期识别和预防提供参考。方法纳入2021年10月―2022年5月就诊于山东省烟台市蓬莱人民医院内分泌科的≥60岁2型糖尿病患者1319例作为研究对象,调查患者一般人口学特征、身体疾病、生活方式、心理健康状况、生理指标资料,采用蒙特利尔认知评估量表(Montreal cognitive assessment,MoCA)评估患者认知功能。利用R 4.1.3软件构建反向传播(back propagation,BP)神经网络模型、随机森林模型、XGBoost模型,并计算模型准确率、灵敏度、特异度、阳性预测值、阴性预测值、F1评分、AUC值及其95%CI。结果BP神经网络模型、随机森林模型、XGBoost模型的灵敏度分别为57.79%、77.89%、80.40%,特异度分别为78.17%、60.41%、61.42%,AUC分别为0.746(95%CI:0.698~0.794)、0.755(95%CI:0.708~0.802)、0.756(95%CI:0.709~0.803)。结论XGBoost模型和随机森林模型具有较好的性能,在老年糖尿病患者MCI风险评估领域具有一定的应用前景。Objective This study aims to develop a high-accuracy risk assessment model for identifying the risk of mild cognitive impairment(MCI)in elderly patients with diabetes mellitus using machine learning algorithms,providing insights for early identification and prevention of cognitive impairment in this population.Methods A total of 1319 patients aged 60 and above with type 2 diabetes mellitus,who visited the Endocrinology Department of People′s Hospital of Penglai in Yantai City,Shandong Province,between October 2021 and May 2022,were enrolled as the study population.The demographic information,medical history,lifestyle factors,psychological health status,and physiological indicators were collected.The Montreal Cognitive Assessment(MoCA)scale was used to evaluate the cognitive function of patients.BP neural network model,random forest model,and XGBoost model were constructed using R version 4.1.3 software.The accuracy,sensitivity,specificity,positive predictive value,negative predictive value,F1 score,and the area under the curve(AUC)with 95%CI of models were calculated.Results The sensitivity values of the BP neural network model,random forest model,and XGBoost model were 57.79%,77.89%,and 80.40%,respectively.The specificity values were 78.17%,60.41%,and 61.42%for the respective models.The AUC values for the ROC curves were 0.746(95%CI:0.698-0.794),0.755(95%CI:0.708-0.802),and 0.756(95%CI:0.709-0.803),respectively.Conclusions The XGBoost model and random forest model demonstrated good performance and showed potential for application in the field of MCI risk assessment among elderly patients with diabetes mellitus.

关 键 词：2型糖尿病 轻度认知障碍 反向传播神经网络 随机森林 XGBoost 

分 类 号：R161.7[医药卫生—公共卫生与预防医学]