Global Group Parameter Model for Multivariate Longitudinal Rates of Change

• Activity Number: 468
• Type: Contributed
• Date/Time: Wednesday, August 6, 2014 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Matthew Bryan*+ and Patrick Heagerty
• Companies: and University of Washington
• Keywords: Multivariate ; Longitudinal ; Rates of Change ; Global Parameter ; Non-Linear

Abstract

When comparing groups using multivariate longitudinal data, two common approaches are latent variable models and global group parameter (GGP) models. When inference on the measured outcomes is of primary interest, a GGP model will be most suitable for linking the multivariate outcomes. However, in order for it to be advantageous to link outcomes using a GGP model, the effect size of the group parameter must be the same or similar for each outcome. We evaluated the utility of a GGP approach for modeling group differences in the rate of change of a multivariate longitudinal outcome. The proposed model allows a unique reference time structure and mean level adjustment for baseline covariates for each outcome. The difference in the rate of change for each outcome associated with covariate-defined groups was estimated globally across all outcomes. The power to detect group differences in the rate of change based on the global rate model was compared with models that separately estimate differences in the rate of change for each outcome. We applied the GGP model for longitudinal rates of change to a study of growth among infants exposed to HIV infection.