失能老年人抑郁风险预测模型的构建与评价  

Construction and evaluation of depression risk prediction model for disabled elderly people

作　　者：洪珊珊 姜燕[1] HONG Shan-shan;JIANG Yan(School of Public Health and Health Sciences,Hubei University of Medicine,Shiyan,Hubei 442000,China)

机构地区：[1]湖北医药学院公共卫生与健康学院,湖北十堰442000

出　　处：《现代预防医学》2024年第15期2818-2823,共6页Modern Preventive Medicine

基　　金：湖北省普通高等学校人文社会科学重点研究基地基金重点项目(2015ZD001)。

摘　　要：目的探究机器学习与logistic回归对失能老年人抑郁的预测价值,并分析失能老年人抑郁的独立危险因素。方法基于2018年中国老年健康影响因素跟踪调查(CLHLS)数据与2018年中国健康与退休纵向研究(CHARLS)数据,排除缺失值及异常值,分别将CLHLS中2277位与CHARLS中476位≥65岁的失能老年人纳入本研究。将CLHLS筛选出的失能老年人按照6∶4的比例随机划分为训练集和内部验证集,并将CHARLS作为外部验证集。采用单因素分析、lasso回归以及多因素logistic回归筛选出失能老年人抑郁的独立危险因素,用于构建机器学习及logistic回归风险预测模型。结果本研究中健康自评、睡眠时长、社交、认知及居住地为失能老年人抑郁的独立危险因素。XGBoost模型的综合性能最好,训练集、内部验证集和外部验证集的AUC、准确率、敏感性、特异性、阳性预测值、阴性预测值、召回率、精确率、Brier-score分别为0.74、0.68、0.72、0.62、0.69、0.65、0.69、0.73、0.20,0.76、0.70、0.75、0.63、0.72、0.66、0.72、0.75、0.19,0.76、0.74、0.86、0.50、0.77、0.65、0.77、0.87、0.17。结论基于XGBoost机器学习方法构建的失能老年人抑郁风险预测模型具有较好的预测效果,可为失能老年人抑郁的干预工作提供参考。Objective To explore the predictive value of machine learning and logistic regression for depression in disabled elderly people,and to analyze the independent risk factors for depression in disabled elderly people.Methods Based on the 2018 Chinese Longitudinal Healthy Longevity Survey(CLHLS)data and the 2018 China Health and Retirement Longitudinal Study(CHARLS)data,2277 disabled elderly individuals aged 65 and above from CLHLS and 476 from CHARLS were included in this study after excluding missing and outlier values.The disabled elderly individuals selected from CLHLS were randomly divided into training and internal validation sets in a 6:4 ratio,with CHARLS serving as the external validation set.Single-factor analysis,LASSO regression,and multiple-factor logistic regression were used to identify the independent risk factors for depression in disabled elderly people,which were then used to construct machine learning and logistic regression risk prediction models.Results Self-rated health,sleep duration,social interaction,cognition,and place of residence were identified as independent risk factors for depression in disabled elderly people.The XGBoost model demonstrated the best comprehensive performance,with an area under the curve(AUC),accuracy,sensitivity,specificity,positive predictive value,negative predictive value,recall,precision,and Brier-score of 0.74,0.68,0.72,0.62,0.69,0.65,0.69,0.73,0.20 in the training set,0.76,0.70,0.75,0.63,0.72,0.66,0.72,0.75,0.19 in the internal validation set,and 0.76,0.74,0.86,0.50,0.77,0.65,0.77,0.87,0.17 in the external validation set.Conclusion The depression risk prediction model for disabled elderly people constructed based on the XGBoost machine learning method exhibits good predictive performance and can provide reference for interventions targeting depression in disabled elderly people.

关 键 词：失能老年人 抑郁 机器学习 LOGISTIC回归 预测模型 

分 类 号：R749.4[医药卫生—神经病学与精神病学] B84.4[医药卫生—临床医学]