Keywords: Missing at random (MAR), missing not at random (MNAR), mixed-effects model for repeated measures (MMRM), latent mixture model
In clinical trials with extensive dropout, patients may respond to the same treatment differently due to different underlying mechanisms therefore generate different outcome and dropout patterns. This may result in missing not at random (MNAR). Inappropriate methods to handle missing data may lead to misleading results and affect regulatory decisions. In this paper, latent mixture model will be introduced. This approach of MNAR allows us to handle the inhomogeneity among subpopulations and make more appropriate inferences on the treatment effect at various time points. Comparisons will be made with conventional statistical methods such as missing at random (MAR), mixed-effects model for repeated measures (MMRM) and pattern mixture model (PMM) through simulation studies. Situations for the appropriate use of these models will be discussed.