Time-Varying Coefficient Models for Missing-by-Design Intensive Longitudinal Data

• Activity Number: 183
• Type: Contributed
• Date/Time: Monday, August 10, 2015 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Xiaoxue Li* and Stewart Anderson and Abdus Wahed and Saul Shiffman
• Companies: and University of Pittsburgh and University of Pittsburgh and University of Pittsburgh
• Keywords: Time varying coefficient model ; missing by design ; intensive longitudinal study

Abstract

Intensive longitudinal studies (ILS), an extension of classical longitudinal studies, usually have a large number of intensive observations per subject. As a result, ILS can be used to identify time patterns (periodic or otherwise) in health event outcomes that can occur multiple times. Practically, when participants are given too many assessments, they may not adhere to study protocols. As a result, some investigators applied sampling algorithm to balance between the amount of information assessed and the compliance load of each individual. This complicated the analyses as some observations are missing by design. In our study, we are particularly interested in the time patterns of one single predictor. We hypothesize that effects of a single predictor may change over time, more specifically, periodical. Subsequently, time varying coefficient models can be applied to accomplish this purpose (ref). However, the little has been done to accommodate missing by design issues. In our work, we combined the inverse probability weighting method and the time varying coefficient modeling techniques to handle this problem. Simulation studies and a data example will be presented.