Challenges in the Analysis of Observational Longitudinal Aging Studies

• Activity Number: 187
• Type: Invited
• Date/Time: Tuesday, August 5, 2003 : 8:30 AM to 10:20 AM
• Sponsor: WNAR
• Author(s): Geert Verbeke*+
• Companies: Biostatistical Centre
• Keywords: linear mixed model ; longitudinal data ; observational data ; baseline characteristic ; parameterization ; unbalanced data

Abstract

Observational longitudinal data are often highly unbalanced in the sense that not a fixed number of measurements is available for all subjects, and measurements are taken at arbitrary time points. The analysis requires flexible models, such as linear mixed models in case of continuous data. We will show that, for observational data, many aspects of the model formulation deserve special attention. First, the actual parameterization of the model requires special attention. Indeed linear mixed models imply specific mean and covariance structures, which, depending on the parameterization chosen, may or may not be realistic and/or meaningful for the dataset at hand. Second, it is not always clear to what extent corrections are needed for baseline differences between study participant, nor is it often clear how such corrections will affect inferences of interest. Finally, the concepts of missing data and dropout are much less clearly defined than for balanced designed experiments. These issues will be discussed in the context of the Baltimore Longitudinal Study of Aging, an ongoing multidisciplinary observational study, with the study of normal human aging as primary objective.