Analysis of Covariate Subsampling Designs Based on Continuous Longitudinal Data

• Activity Number: 394
• Type: Invited
• Date/Time: Tuesday, August 6, 2013 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Patrick Heagerty*+ and Jonathan S. Schildcrout and Paul J. Rathouz
• Companies: University of Washington and Vanderbilt University and University of Wisconsin School of Medicine and Public Health

Abstract

Recent work has proposed sampling designs where detailed covariate information is only obtained on a subsample of subjects, and where the subsample may be chosen on the basis of already observed longitudinal outcomes (Schildcrout, Garbett, and Heagerty 2012). One valid approach to analysis only uses data on the subsample subjects and ignores the incomplete information available on the remainder of the subjects. However, we show that efficiency gains can be obtained by also using outcome and partial covariate data on non-sampled subjects using either direct likelihood-based methods, or using multiple imputation as a computationally simple procedure. We study the robustness of methods to model misspecification and illustrate analysis using data from a longitudinal cohort study.