Abstract:
|
In clinical trials with extensive dropout, patients may respond to the same treatment differently and generate different outcome trajectories and dropout patterns. As a result, the patient population may consist of heterogeneous clusters each with its own outcome and dropout pattern. This is a scenario of missing not at random (MNAR) and the observed treatment effects may be seriously biased when the statistical methods based on the missing at random (MAR) assumption are applied. In this paper, we introduce a new statistical method based on the latent growth mixture model, which is a special case of the latent mixture model, to analyze the data in such trials. This model allows us to handle the inhomogeneity among subpopulations and make more accurate inferences on the treatment effect at any particular visit time. Comparing to the conventional statistical methods such as mixed-effects model for repeated measures (MMRM) and multiple imputation method (MI), we demonstrate through simulations that our proposed method gives better control on the Type I error rate in testing treatment effect.
|