Generalized Estimating Equation For Longitudinal Data With Its Application To Children Nutrition Clinical Data Analysis

In children nutrition research,the effects of vitamin A deficiency on children's health is received with grave concern of clinicians.Cohort study is the best method to tackle such problem.In such study,it is an important problem to analyze the longitudinal data from the cohort by making appropriate statistical models to characterize the relationship between vitamin A deficiency and the risk of children disease.This paper investigates how to make appropriate statistical model to study whether children with vitamin A deficiency are at increased risk of developing respiratory infection based on the data from a cohort study in Indonesia,to determine the effects of vitamin A deficiency in preschool children.Within-subject correlation of longitudinal data must not be ignored.As a basis for comparison,we constructed the generalized linear model(GLM)ignoring such correlation firstly.And we constructed the generalized linear mixed effects model(GLMM)taking the within-subject correlation into account further.Since our data is the binary longitudinal data,we mainly devote ourselves to studying the generalized estimation equation(GEE)and the comparison with GLMM.GEE takes the within-subject correlation into account and actually uses the quasi-likelihood estimation which does not specify the exact distribution of the response.We investigate the parameter estimations of GEE under four commonly used working correlations,i.e.working independent,AR(1),exchangeable compound symmetry and unstructured correlation.According to the CIC criterion,the work correlation matrix is selected.Our results are as follows.For the binary longitudinal data,the parameter estimations of GEE feature robustness.The standard error of parameters estimation by GEE is smaller than that of GLMM.And the accuracy of fitting of GEE is higher than other models in our paper.Our data analysis shows that vitamin A deficiency significantly increases the risk of respiratory infection disease in girls,the risk decreases with age and is significantly different with gender in preschool children.All calculations in this paper are completed by R software.