广义估计方程与准最小二乘在社区纵向数据分析中的应用

Application of Generalized Estimating Equations and Quasi-least Squares in Analyzing Community Based Longitudinal Data

目的应用广义估计方程和准最小二乘方法分析社区卫生服务中心纵向数据,探讨纵向数据分析的问题,为社区的随访的纵向数据的分析提供科学的方法。方法对收集的社区卫生服务中心的糖尿病病人血糖的纵向数据,分别使用广义估计方程和准最小二乘方法以及传统的线性回归模型进行分析并比较结果。同时比较三种方法的标准化残差图。结果广义估计方程不收敛时与传统线性模型的结果相同,显示糖尿病人血糖与教育水平相关,而广义估计方程收敛时与准最小二乘的结果相同,显示教育无统计学意义。从标准化残差图看广义估计方程和准最小二乘法对数据的拟合比传统回归好。结论广义估计方程和准最小二乘法都能有效的处理纵向数据。与广义估计方程相比,准最小二乘法有一些优势。

Objective To explore the application of generalized estimating equations and quasi-least squares in analyzing the longitudinal data of community health service centers,so as to provide scientific statistical methods for processing the follow-up data of communities.Methods Generalized estimating equations （GEE）,quasi-least squares （QLS） and general linear regression model were applied to analyze the longitudinal data of the blood glucose of diabetic patients of the community health service center.The results and the standardized residual plots of the three methods were compared.Results When the GEE did not converge,it had identical results with the general linear model,both indicating that blood glucose was correlated with education.While when the GEE converged,it had identical results with the QLS,suggesting no significant correlation between blood glucose and education.From the standardized residual plots,it was known that GEE and QLS were better.Conclusions Generalized estimating equations and quasi-least squares do better in dealing with longitudinal data.Quasi-least squares procedure has some advantages over generalized estimating equations.