Application of Pattern-Mixture Latent Trajectory Modeling to Assess Longitudinal Trajectory with Non-Ignorable Missing Data

Hiroko H. Dodge, Oregon State University
Changyu Shen, Indiana University
Mary Ganguli, University of Pittsburgh

Longitudinal designs, requiring follow-up of the same individuals over time, are increasingly common in epidemiological and demographical studies. However, missing data bias is a major problem in longitudinal studies where attrition is inevitable over time. Restricting analyses to only the observed data could bias the results depending on the types of missingness; a missing-data process is called non-ignorable if a likelihood-based approach cannot provides valid inferences to the model parameters. One approach to address non-ignorable missing data bias is a pattern mixture model, but un-identifiability is a problem. We offer a practical solution to this problem by using latent trajectory analysis implemented in the SAS TRAJ procedure, which identifies latent groups with different trajectory patterns. The approach presented here is appealing since it can be easily implemented using common software and can be applied to wide variety of disciplines which analyze longitudinal data with potentially non-ignorable missing data.