Dynamic Factor Analysis for Longitudinal Studies: An Application in Alzheimer's Disease

• Activity Number: 469
• Type: Contributed
• Date/Time: Wednesday, August 1, 2012 : 8:30 AM to 10:20 AM
• Sponsor: Biometrics Section
• Author(s): Georgios Tripodis*+ and Nikolaos Zirogiannis
• Companies: Boston University and University of Massachusetts
• Keywords: EM algorithm ; Factor Analysis ; Kalman Filter ; Alzheimer's disease

Abstract

Item response theory (IRT) has been applied extensively in psychometric settings. A particularly interesting application is in neurological studies, such as Alzheimer's disease where each patient's cognitive abilities is measured repeateadly by a battery of neuropsychiatric tests. A serious limitation of IRT is that it does not account for correlation in time. We develop a dynamic factor analysis (DFA) model that can be applied on longitudinal studies. While DFA is a widely used multivariate statistical technique, there is limited research of its application on longitudinal studies. In simple DFA we have multiple variables of the same subject. In a more general panel data setting, we have several variables of multiple subjects measured at each time period. This creates an extra element of complexity, since we need to account for between neuropsychiatric tests variability, as well as within tests variability. In this paper, we propose a solution to this problem by using the EM algorithm and the Kalman filter to estimate the state of cognitive ability of each patient at each time point. The model is then applied on a dataset from the National Alzheimer's Coordinating Center.