基于多水平模型的2型糖尿病影响因素分析

Analysis on influencing factors of type 2 diabetes based on multilevel model

目的以2型糖尿病为例,探讨具有层次聚集性变量的多水平模型的构建。方法用2015年蚌埠市龙子湖区慢性非传染性疾病调查中的2型糖尿病相关数据作为研究案例,采用R 3.3.1软件对2型糖尿病的影响因素进行单水平和多水平Logistic回归分析。结果单水平Logistic回归和多水平Logistic回归分析筛选的影响因素不完全一致,主要有年龄、糖尿病家族史、高血压病史、水果摄入频率、糖化血红蛋白和体质指数(body mass index,BMI)。非超重(BMI≤23.9 kg/m^2)与超重(BMI≥24.0 kg/m^2)在单水平Logistic回归中有差异(均有P<0.05),在多水平Logistic回归中显示无差异(均有P>0.05)。赤池信息准则(akaike information criterion,AIC)在单水平和多水平Logistic回归模型的值分别为1 550.2和1 535.3。结论单水平与多水平Logistic回归模型所得出的结论并不完全一致,多水平模型在具有层次聚集性的数据分析中更具有优势。基于2型糖尿病的层次聚集性,应该以社区服务站为基础开展2型糖尿病的防控工作。

Objective To discuss the construction of a multilevel model for binary outcome measure with hierarchical cluster data based on an example of type 2 diabetes. Methods Data was collected from a survey on chronic disease of Longzihu district of Bengbu city in 2015. One-level and multilevel Logistic regression models were conducted to investigate the influencing factors of type 2 diabetes using R software. Results The influencing factors of type 2 diabetes analyzed by one-level and multilevel Logistic regression were not exactly the same,including age,family history of diabetes,history of hypertension,frequency of fruit intake,glycosylated hemoglobin,body mass index（ BMI）. One-level Logistic regression showed that the prevalence of diabetes in non-overweight group was significantly different from that in overweight group（ all P 0. 05）,which was not significant between non-overweight and overweight group in multilevel Logistic regression（ all P 0. 05）. The Akaike information criterions of one-level and multilevel Logistic regression were 1 550. 2 and1 535. 3,respectively. Conclusions The conclusions from one-level and multilevel Logistic regression model are not exactly the same,and the multilevel model has advantages in data analysis for hierarchical cluster data. Based on hierarchical cluster of type 2 diabetes,the prevention and control of type 2 diabetes should be carried out on the basis of community service center.