Analysis of longitudinal binary outcomes in clinical trials with low percentage of missing values

In interventional or observational longitudinal studies, the issue of missing values is one of the main concepts that should be investigated. The researcher's main concerns are the impact of missing data on the final results of the study and the appropriate methods that missing values should be handled. Regarding the role and the scale of the variable that missing values have been occurred and the structure of missing values, different methods for analysis have been presented. In this article, the impact of missing values on a binary response variable, in a longitudinal clinical trial with three follow up sessions has been investigated Propensity Score, Predictive Model Based and Mahalanobis imputation strategies with complete case and available data methods have been used for dealing with missing values in the mentioned study. Three models; Random intercept, Marginal GEE and Marginalized Random effects models were implemented to evaluate the effect of covariates. The percentage of missing responses in each of the treatment groups, throughout the course of the study, differs from 6.8 to 14.1. Although, the estimate of variance component in random intercept and marginalized random effect models were highly significant (p <0.001) the same results were obtained for the effect of independent variables on the response variable with different imputation strategies. In our study according to the low missing percentage, there were no considerable differences between different methods that were used for handling missing data.