Three-Part Mixture Models for Longitudinal Data with Non-Random Drop out and Semi-continuous OutcomesView Presentation *Depeng Jiang, University of Manitoba
Keywords: missing data, longitudinal study, growth mixture model, Semi-continuous data
This study proposes a three-part growth mixture model for handling non-random missing (m-part) with semi-continuous outcomes in a longitudinal study. Semi-continuous data in the form of mixture of zeros (u-part) and continuously distributed positive values (y-part) frequently arise in biomedical research. Two-part mixed models (one for u-part and one for y-part) with correlated random effects are attractive approach to the longitudinal semi-continuous data (Olsen and Schafer, 2001; Tooze et. al., 2002). However, the estimations from the two-part mixed model are only valid and unbiased when the drop outs are random. When the drop outs are non-random, we proposed a three-part growth mixture model to jointly modeling the longitudinal processes of three parts (m-part, u-part, and y-part). We used latent class analysis to model m-part and u-part and used the growth mixture model to determine the actual trajectories of the outcome (y-part) if it is non-missing. These methods were applied to a longitudinal study of workers with work-relevant musculoskeletal disorders, to show how the new approaches can overcome the problems of current available statistical methods and help to identify the distinct trajectories of worker productivity loss and the associated prognostic factors.