A Robust Joint Modeling Approach For Longitudinal Data With Informative Dropouts

In this paper,we propose a robust method for longitudinal continuous data with informative dropout missing and potential outliers through multivariate t distribution.When analyzing longitudinal data,it is important to study its dependent structure.Dif-ferent from the existing method focuses on regression mean and informative dropout mechanism,we propose an approach by joint modeling the location function and de-pendent structure to reveal the dynamics in the location function and marginal scale function.Because of the complexity of observed likelihood function,EM algorithm is used to solve maximum likelihood estimation.Due to the complexity of the calculation in E-step,Monte Carlo algorithm is considered.However,the computational ineffi-ciency of MCEM and the convergence cannot be guaranteed for fixed Monte Carlo sample size.A parametric fractional imputation(PFI)algorithm is proposed to speed up the computation associated with EM algorithm for the maximum likelihood estimation with informative dropout mechanism.The resulting estimators are shown to be consis-tent and asymptotic normality distributed.The MDD data and simulations demonstrate the effectiveness of the proposed method.