Evaluating two statistical methods for handling dropout impact in repeated measures data: a comparative study on missing at random assumption

Introduction Dropout is a major source of missing data in repeated measures studies and can bias statistical inference if not handled properly. This study compares the performance of two common methods for addressing dropout under the missing at random (MAR) assumption: multiple imputation (MI) and inverse probability weighting (IPW). Methods A simulation study was conducted using repeated measures data generated from a marginal regression model with continuous outcomes. Two sample sizes (100 and 250) and three dropout rates (5%, 15%, and 30%) were considered. The methods were evaluated based on bias, coverage, and mean square error (MSE). In addition, the approaches were applied to serum cholesterol data from the National Cooperative Gallstone Study to illustrate their practical performance. Results Simulation results showed that MI consistently produced lower bias and MSE and higher coverage compared to IPW, particularly at moderate to high dropout rates. The empirical analysis of serum cholesterol data indicated that while both methods yielded similar inferential conclusions regarding treatment effects, MI estimates were more stable and precise than those obtained from IPW. Discussion This analysis demonstrates that MI outperforms IPW in handling MAR dropout for continuous repeated measures data. The findings support the use of MI as a more reliable approach, especially in studies with moderate to high dropout rates, although IPW may remain acceptable under low dropout rate.