First and second order generalized estimating equations and its application in the analysis of longitudinal microleakage data

Longitudinal data are frequently obtained in medical studies. When the main aim of a study is marginal modeling of the mean and the correlation structure is considered as a nuisance parameter, the first- order generalized estimating equations (GEE1) is usually an appropriate option. However, when the modeling of correlation structure is considered the aim of a study, the second- order generalized estimating equations (GEE2) may be the first choice for analyzing the available data. The aim of the study was to evaluate application of first- and second-order generalized estimating equations to analyze longitudinal microleakage data. In this study, GEE1 and GEE2 methods were used to analyze data obtained from a study of microleakage in two root- end filling materials (CEM and MTA) in two different thicknesses and two diameters at three different times of measurement (one day, one week and one month after treatment). The obtained results from these statistical approaches were compared in continuous and binary (presence of absence) microleakage data. The results from the GEE1 and GEE2 methods showed that time of measurement, material type, diameter and thickness of filling material had significant effects on (continuous) microleakage rate. In addition, in binary microleakage data, these methods revealed that only time and material type were the significant factors. The correlations between measurements were not significant in continuous data, while they were significant in binary response microleakage data. Since the correlations between pairs of measurements were not significant in continuous microleakage data and the obtained estimates were similar in both GEE1 and GEE2 methods, so the simpler GEE1 method seems to be adequate for these data. In contrast, in binary microleakage data, significant correlations were found between measurements. Therefore, in this case the GEE2 methodology may be used to estimate the correlation structure more efficiently.