Longitudinal Models with Penalized Splines and Time-Dependent Covariates

• Activity Number: 170
• Type: Contributed
• Date/Time: Monday, August 2, 2010 : 10:30 AM to 12:20 PM
• Sponsor: Biometrics Section
• Author(s): Jesse D. Raffa*+ and Joel A. Dubin+
• Companies: University of Waterloo and University of Waterloo
• Keywords: longitudinal model ; penalized splines ; time-varying covariates ; mixed effects model

Abstract

Linear mixed effects models are frequently used for modeling continuous longitudinal data, including sometimes in the presence of time-varying covariates. For example, we have interest in looking at the CD4 cell count trajectories of a cohort of HIV-infected illicit drug users who repeatedly go on and off for HIV treatment. While modeling CD4 trajectories has been well-examined, this has typically been done in cohorts of patients who were exclusively on or off therapy (or were assumed to be). Penalized splines have become popular for modeling longitudinal data, particularly as they can be expressed under the mixed effects modeling framework. However, this approach is not typically used in the presence of time-varying covariates. We propose a penalized splines subject-specific curve approach to handle the case of time-varying covariates, and apply this to the illicit drug user data.