Clinical trials in the diabetes therapeutic area often employ mixed-model repeated measure (MMRM) approach as the primary analysis model for repeated continuous efficacy measurements such as HbA1c. Typically, HbA1c measurements are taken at different time points during the course of the trial and the change from baseline at the last measurement is often considered the primary endpoint in most diabetes trials. Targeted minimum loss based estimation (TMLE), providing doubly robust estimation, serves as an alternative method to estimate the average treatment effect. We evaluate and compare the performance of TMLE versus MMRM through completed diabetes trial data sets and simulation studies across multiple aspects such as bias, variance, mean squared error and confidence interval coverage probability for estimation of average treatment effect. Also, we consider the incorporation of machine learning methods into TMLE approach and assess its performance to gauge how much value can be added to the estimation. The simulation studies explore a wide variety of situations including different treatment effect sizes, different patient dropout scenarios, etc.