基于机器学习算法构建多病共存老年人轻度认知功能障碍预测模型  

Constructing a predictive model for mild cognitive impairment in elderly individuals with coexisting multiple diseases based on machine learning algorithms

作　　者：张雷[1] 胡鑫涛 陈巍 相璇 邹婷[1] 苗海军[1] 吴瑞凯 周晓辉[1] ZHANG Lei;HU Xintao;CHEN Wei;XIANG Xuan;ZOU Ting;MIAO Haijun;WU Ruikai;ZHOU Xiaohui(Department of Geriatrics,The First Affiliated Hospital of Xinjiang Medical University,Urumqi 830011,China)

机构地区：[1]新疆医科大学第一附属医院老年医学科,乌鲁木齐830011

出　　处：《医学新知》2025年第4期409-418,共10页New Medicine

基　　金：新疆维吾尔自治区自然科学基金重点项目(2022D01D63)。

摘　　要：目的探讨多病共存住院老年人轻度认知功能障碍(mild cognitive impairment,MCI)的影响因素,并基于机器学习(machine learing,ML)方法构建多病共存住院老年人MCI风险预测模型。方法以新疆医科大学第一附属医院多病共存住院老年人为研究对象,使用单因素分析和最小绝对值收缩和选择算子回归算法筛选MCI风险因素,使用随机森林、轻量梯度提升机、极端梯度提升、逻辑回归、K最近邻分类算法、支持向量机、人工神经网络、决策树、弹性网络回归算法9种不同的ML方法构建MCI风险预测模型,并采用SHAP算法对最终模型进行解释。结果共纳入920例多病共存住院老年人,MCI组261例。随机森林模型的预测性能最优,其受试者工作特征曲线的曲线下面积均高于其他模型。SHAP算法对随机森林模型进行分析,显示年龄、共病数量、文化程度、脑血管病是预测多病共存住院老年人MCI发生的关键决策因素。校准曲线表明该模型预测效果和实际结果基本一致,决策曲线表明模型具有良好的临床适用性。结论高龄、共病数量增加、患有脑血管病是多病共存住院老年人发生MCI的危险因素,高文化水平是多病共存住院老年人MCI发生的保护因素。基于随机森林算法的MCI发生风险的预测模型表现最优,具有良好的预测性能及临床适用性,可辅助临床对多病共存老人年进行认知管理和更精准的医疗干预。Objective To explore the influencing factors of mild cognitive impairment(MCI)in hospitalized elderly patients with multiple comorbidities,and to construct a MCI risk prediction model based on machine learning(ML)methods.Methods The study included elderly patients with multiple comorbidities admitted to the First Affiliated Hospital of Xinjiang Medical University as research subjects.Single factor analysis and least absolute shrinkage and selection operator regression algorithms were used to screen for MCI risk factors.Nine different ML methods were used,including random forest,light gradient boosting machine,extreme gradient boosting,Logistic regression,K-nearest neighbor classification algorithm,support vector machine,artificial neural network,decision tree,and elastic network regression algorithm,to construct MCI risk prediction models.Shapley addition explanation(SHAP)algorithm was used to explain the final model.Results A total of 920 hospitalized elderly patients with multiple comorbidities were included,including 261 cases in the MCI group.The random forest model had the best predictive performance,with a higher area under the receiver operating characteristic curve than other models.The SHAP algorithm identified the age,comorbidities,education level,and cerebrovascular disease in the random forest model as key decision factors for predicting MCI in hospitalized elderly patients with multiple comorbidities.The calibration curve showed that the predictive performance of the model was basically consistent with the actual results,and the decision curve indicated that the model had good clinical applicability.Conclusion Advanced age,increased comorbidities,and cerebrovascular disease are risk factors for MCI in hospitalized elderly people with multiple comorbidities.High educational level is a protective factor for MCI in hospitalized elderly people with multiple comorbidities.Based on machine learning algorithms,the prediction model for MCI risk using random forest has the best predictive performance and good

关 键 词：轻度认知功能障碍 多病共存 机器学习 预测模型 Shapley加法解释算法 

分 类 号：R592[医药卫生—老年医学] R749.16[医药卫生—内科学]