广义线性混合模型与logistic回归应用于层次结构数据的对比分析  

Contrast research of hierarchical structure data analyzed by generalized linear mixed models with logistic regression

作　　者：万国峰 杨兴华[1,2] WAN Guo-fen YA NG Xing-hua(Department of Epidemiology and Health Statistics ,School of public health, Capital Medical University,Beijing, 100069, China Beijing Key Laboratory of clinical epidemiology, ,Beijing, 100069,China Institute of Health Education,Beijing Center for Disease Control and Prevention,Beijing, 100013, China)

机构地区：[1]首都医科大学公共卫生学院流行病与卫生统计系,北京100069 [2]临床流行病学北京市重点实验室,北京100069 [3]北京市疾病预防控制中心健康教育所,北京100013

出　　处：《职业与健康》2017年第19期2686-2689,共4页Occupation and Health

摘　　要：目的比较广义线性混合模型与logistic回归应用于层次结构数据的差异性。方法选择2013年10—12月期间北京市14家单位职工食堂满意度调查的资料数据,对可能影响食堂满意度的各变量分别进行非条件logistic回归和广义线性混合模型分析,并对结果进行对比分析。结果 Logistic回归显示,食堂经营性质(OR=2.757,P<0.05)、性别(OR=0.723,P<0.05)、年龄(OR=0.991,P<0.05)、学历(OR=0.694,P<0.05)、岗位(OR=0.745,P<0.05)、在食堂用餐天数(OR=1.685~2.590,P<0.05)、职工健康知识水平(OR=1.099,P<0.05)等7个变量为食堂满意度影响因素。广义线性混合模型显示,高水平(单位)常数项估计值为1.138,标准误为0.514,95%CI=0.469~2.760,差异有统计学意义(P<0.05)。说明食堂满意度在单位水平上存在聚集性。且仅有女性(OR=0.780,P<0.05)、在食堂用餐时间(OR=2.336~2.655,P<0.05)和膳食知识得分(OR=1.135,P<0.05)3个变量差异有统计学意义。两种分析方法都有统计学意义的变量为性别、在食堂就餐天数和职工膳食知识得分,且OR值水平接近。结论处理在高水平上存在聚集性的层次结构数据时,广义线性混合模型比logistic回归更为谨慎和科学。但对差异有统计学意义的变量解释上,两者具有一致性。[Objective] To compare the difference of generalized linear mixed models and logistic regression for hierarchical structure data.[Methods]The data about satisfaction survey of 14 employee canteens in Beijing from October to December 2013 were collected as the research object. The possible influencing factors of canteen satisfaction were analyzed by non-conditional logistic regression and generalized linear mixed models,and the results were compared.[Results]The results of logistic regression showed that the canteen management nature（OR=2.757,P〈0.05）,gender（OR=0.723,P〈0.05）,age（OR=0.991,P〈0.05）,education（OR=0.694,P〈0.05）,jobs（OR=0.745,P〈0.05）,meals days in the canteen（OR=1.685-2.590,P〈0.05）,and the health knowledge level（OR=1.099,P〈0.05） were the influencing factors of canteen satisfaction. Generalized linear mixed models showed that the estimate of high level（units） was 1.138,standard error was 0.514,95% CI =0.469-2.760,with statistical significance（P〈0.05）,which indicated that the canteen satisfaction was gathered at the unit level. Female（OR=0.780,P〈0.05）,meal time in the dining hall（OR=2.336-2.655,P〈0.05） and dietary knowledge score（OR=1.135,P〈0.05） has statistical significance. The same factors of two ways were gender,meal time in the dining hall and dietary knowledge score,furthermore,the OR value of the 3 factors was approximative.[Conclusion]For clustering hierarchy data,generalized linear mixed models are more cautious and scientific than logistic regression,but they had the consistency as explaining the influencing factor.

关 键 词：广义线性混合模型 LOGISTIC回归 层次结构数据 

分 类 号：R181.2[医药卫生—流行病学]