Analysis of Longitudinal Data Gathered Using Outcome-Related Sampling Designs

• Activity Number: 688
• Type: Contributed
• Date/Time: Thursday, August 13, 2015 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): John Neuhaus* and Charles McCulloch
• Companies: UC San Francisco and UC San Francisco
• Keywords: Generalized linear mixed models ; Joint models ; Multiple imputation ; Retrospective sampling

Abstract

Longitudinal studies with valuable measurements such as magnetic resonance images often require subsampling the data based on an auxiliary response related to the outcome of interest. Failure to accommodate the outcome-related sampling design in statistical analyses can lead to biased estimates and incorrect scientific conclusions. In this talk we present two analysis strategies that provide consistent estimation with longitudinal data gathered using outcome-related sampling designs: 1) likelihood-based fits of fully parametric, joint models for the auxiliary measurement and the outcome of interest; and 2) multiple imputation of the unselected outcomes. These methods may increase efficiency by including information from the auxiliary responses, and can be implemented using standard software. We assess the performance of the two analysis strategies using simulated longitudinal data gathered using outcome-related sampling designs. Data from the Osteoarthritis Initiative illustrate the sampling design and motivate the proposed analysis strategies.