基于CHARLS数据的中老年人抑郁症状影响因素分析及预测模型构建  

Analysis on influencing factors and construction of prediction model for depressive symptoms in middle-aged and elderly people based on CHARLS data

作　　者：王正文 夏昉[1] 苏鑫[1] WANG Zheng-wen;XIA Fang;SU Xin(Changchun University of Chinese Medicine,Jilin130117,China)

机构地区：[1]长春中医药大学,吉林130117

出　　处：《预防医学论坛》2024年第11期840-846,共7页Preventive Medicine Tribune

摘　　要：目的 探讨中老年人抑郁症状的影响因素,为抑郁预防提供科学依据。方法 数据来源于2020年中国健康与养老追踪调查(CHARLS),筛选出1 626位研究对象,按照7∶3的比例随机将其分为训练集(n=1 138)和测试集(n=488)。依据流行病学研究中心抑郁量表(CES-D)评分,将研究对象分为抑郁症状组(CES-D≥10分)、无抑郁症状组(CES-D<10分)。LASSO回归模型筛选变量,基于逻辑回归、支持向量机和决策树3种方法构建预测模型。绘制受试者工作特征(ROC)曲线,用曲线下面积评估各模型的预测价值。结果 经过模型评估,逻辑回归AUC值高于支持向量机和决策树,选择逻辑回归构建预测模型。回归结果显示,女性(OR=1.797,95%CI:1.430~2.257,P<0.05),因疼痛而难受程度为有一点(OR=1.530,95%CI:1.178~1.988,P<0.05)、有一些(OR=1.991,95%CI:1.301~3.048,P<0.05)、比较多(OR=2.064,95%CI:1.013~4.204,P<0.05)、非常多(OR=3.802,95%CI:1.790~8.079,P<0.05),子女个数为2个(OR=1.389,95%CI:1.018~1.895,P<0.05)和3个及以上(OR=1.650,95%CI:1.183~2.300,P<0.05)均为中老年人出现抑郁症状的危险因素。自评健康状况好(OR=0.304,95%CI:0.118~0.780,P<0.05)和很好(OR=0.308,95%CI:0.120~0.791,P<0.05),睡眠时长为>5~<9 h(OR=0.503,95%CI:0.390~0.648,P<0.05),需要时有人照顾(OR=0.544,95%CI:0.427~0.691,P<0.05),自评记忆力为一般(OR=0.569,95%CI:0.431~0.751,P<0.05)、好(OR=0.442,95%CI:0.272~0.719,P<0.05)、很好(OR=0.336,95%CI:0.185~0.608,P<0.05),生活满意度为极其满意(OR=0.272,95%CI:0.086~0.866,P<0.05),以上因素均为中老年人抑郁症状的保护因素。结论 性别、因疼痛而难受程度、子女个数、自评健康状况、睡眠时长、需要时是否有人照顾、自评记忆力、生活满意度为中老年人抑郁症状的影响因素,基于以上因素构建的逻辑回归模型有较高的预测价值。Objective To explore the influencing factors of depressive symptoms in middle-aged and elderly individuals and provide scientific evidence for depression prevention.Methods The data originated from the 2020 China Health and Retirement Longitudinal Study(CHARLS),from which 1626 participants were screened and randomly divided into a training set(n=1138)and a test set(n=488)in a 7:3 ratio.Based on the scores from the Center for Epidemiologic Studies Depression Scale(CES-D),the participants were categorized into a depressive symptom group(CES-D≥10)and a non-depressive symptom group(CES-D<10).Variables were selected using the LASSO regression model,and predictive models were constructed based on 3 methods:logistic regression,support vector machines,and decision trees.Receiver Operating Characteristic(ROC)curves were plotted,and the predictive value of each model was assessed using the area underthe curve.Results After model evaluation,the logistic regression model had a higher AUC value than support vector machines and decision trees,and was selected to construct the prediction model.Regression results showed that being female(OR=1.797,95%CI:1.430-2.257,P<0.05),experiencing pain to some extent(OR=1.530,95%CI:1.178-1.988,P<0.05),some pain(OR=1.991,95%CI:1.301-3.048,P<0.05),quite a bit of pain(OR=2.064,95%CI:1.013-4.204,P<0.05),and a lot of pain(OR=3.802,95%CI:1.790-8.079,P<0.05),having 2 children(OR=1.389,95%CI:1.018-1.895,P<0.05)or 3 or more children(OR=1.650,95%CI:1.183-2.300,P<0.05)were risk factors for depressive symptoms among middle-aged and elderly individuals.Good(OR=0.304,95%CI:0.118-0.780,P<0.05)and very good(OR=0.308,95%CI:0.120-0.791,P<0.05)self-rated health status,sleep duration of>5 to<9 hours(OR=0.503,95%CI:0.390-0.648,P<0.05),having someone to care for when nee-ded(OR=0.544,95%CI:0.427-0.691,P<0.05),self-rated memory as average(OR=0.569,95%CI:0.431-0.751,P<0.05),good(OR=0.442,95%CI:0.272-0.719,P<0.05),and very good(OR=0.336,95%CI:0.185-0.608,P<0.05),and extreme satisfaction with life(OR=0.272,95%CI:0.

关 键 词：抑郁症状 中老年人 影响因素 预测模型 CHARLS数据库 

分 类 号：R395.4[哲学宗教—心理学]