A Shared Parameter Model for Continuous Longitudinal Data with Dropouts under Latent Class Structure
David Oslin, University of Pennsylvania
Mary Sammel, University of Pennsylvania
Thomas Ten Have, University of Pennsylvania
*Lingfeng Yang, Wyeth
Keywords: identifiability, informative/non-ignorable drop-out, latent class model, shared parameter model
We propose a shared parameter model for continuous longitudinal responses under the informative drop-out assumption. This is a finite mixture model with a latent class structure for the longitudinal outcome and a discrete-time survival model to describe the drop-out process; the outcome and the drop-out share the same random effect. A simulation study shows the validity of our estimation algorithm and several impacts of the drop-out process. We apply this model to a psychiatric dataset to classify patients according to their longitudinal profiles and assess the association between the longitudinal profiles and baseline characteristics as our primary focus. The results are also compared with those from the alternative naïve model without adjusting for drop-out to test the sensitivity to the informative drop-out assumption.