Multivariate Mixture Models to Describe Longitudinal Patterns of Frailty in American Seniors

Large, longitudinal, multivariate population surveys are increasingly common. Many analytic methods inspect changing rates of individual outcomes but ignore heterogeneous subpopulations that may exist. In this document I propose two analytical methods which extend group-based trajectory models to multivariate outcomes. I use group-based longitudinal finite mixture models (i.e., developmental trajectory models) to identify and describe latent subpopulations for the multivariate outcomes of interest. I use the National Long Term Care Survey, which measures various disabilities in American elderly from 1982 to 2004, as an example. Both models of longitudinal pathways of disability clearly illustrate the various frailty patterns in latent subpopulations of American seniors. These models recognize that within group some disabilities may coincide while others may follow at a later time. Finally, I demonstrate latent frailty patterns with illustrative plots that show multivariate temporal patterns within latent class.