Effect of Correlation Structure in Generalized Estimating Equation and Quasi Least Square: An Application in Type 2 Diabetes Patient

Dilip C Nath,Atanu Bhattacharjee

doi:10.18187/pjsor.v7i2-sp.305

Abstract

The Quasi-Least Squares (QLS) is useful for different correlation structure with attachment of Generalized Estimating Equation (GEE). The purpose of this work is to compare the regression parameter in the presence of different correlation structure with respect to GEE and QLS method. The comparison of estimated regression parameter has been performed in clinical trial data set; studying the effect of drug treatment (metformin with pioglitazone) Vs (gliclazide with pioglitazone) in type 2 diabetes patients. In case of QLS, the correlation coefficient of post-parandinal blood sugar (PPBS) under tridiagonal correlation is 0.008 while it failed to produce by GEE. It has been found that the combination of metformin with pioglitazone is more effective as compared to the combination of gliclazide with pioglitazone.

Highlights

The appropriate and early diagnosis of type 2 diabetes can be helpful to start early treatment and improve the likelihood of achieving adequate glucose control
The challenge has been faced by recognizing that some correlation structures do not work with Generalized Estimating Equations (GEE) and some of them do not work with Quasi-Least Squares (QLS)
In the case of GEE, the unstructured correlation has produced the post-parandinal blood glucose (PPBG) correlation coefficient that could not be produced by QLS

Summary

Introduction

The appropriate and early diagnosis of type 2 diabetes can be helpful to start early treatment and improve the likelihood of achieving adequate glucose control. There is a chance that 30% of patients being tested for diabetes may be miss-diagnosed if screened by FBG alone (Ronald, 2005) These two biochemical parameters (PPBG and FBG) do not reflect the previous status of blood glucose levels, whereas glycosylated haemoglobin (HBA1C) does. The Collaborative Analysis of Diagnostic Criteria in Europe (DECODE, 2001) trial shows that as many as one-third of diabetic patients can be missed by measuring the FBG level. In this analysis the level of PPBG, HBA1C and the drug effect as the covariates over the response FBG have been considered

Methods

Results

Conclusion