Generalized Estimating Equation for Longitudinal Cognitive Data

• Activity Number: 157
• Type: Contributed
• Date/Time: Monday, August 9, 2004 : 2:00 PM to 3:50 PM
• Sponsor: ENAR
• Author(s): Juan Li*+ and Wei Zhu and Susan DeSanti and Mony J. DeLeon
• Companies: Stony Brook University and SUNY, Stony Brook and New York University and New York University
• Keywords: generalized estimating equation ; longitudinal ; alternating logistic regression ; marginal model

Abstract

In longitudinal studies of cognitive data, traditional likelihood-based methods are less tractable since these approaches either assume multivariate normal distributions or have difficulties with time-varying covariates and missing data. An alternative strategy is the generalized estimating equation, which is most often applicable for dichotomous outcomes. GEE is an extension to generalized linear model and it accounts for the structure of the covariance of the response variables. In a longitudinal cognitive study conducted at the NYU Center for Brain Health, 210 subjects including normal and patients who converted to mild cognitive impairment were chosen. We used the first-order GEE model to characterize the marginal expectation of the set of diagnosis outcomes as a function of longitudinal cognitive test scores and other prognostic factors. Second-order GEE (alternating logistic regression) with log-odds ratio structures was applied to the data also to investigate the longitudinal associations within the diagnostic outcomes themselves.