Longitudinal structural models of growth and survival applied to the early detection of Alzheimer’s disease

Article Outline

• Copyright

Background: A number of studies in human aging research where longitudinal data have been collected on adults who are at risk for the development of Alzheimer’s Disease (e.g., Bäckman, et al, 2005; Neuropsychology). One of the key questions raised is the possibility of “early detection” of AD using scores on psychometric instruments (e.g., the MMSE, the CERAD battery, etc.). The collection and analyses of longitudinal data emphasizes the description of patterns of growth and change over time, with special interest on estimates of individual differences in rates of change (slopes), and their use in the prediction of AD. Objective(s): In classical analyses using mixed-effects models, there has been an emphasis on age as a focal basis of change, an effort to understand the lead-lag relationships across different phenotypes and genotypes. In more recent analyses the importance of accounting for survival (mortality) of individuals and families, an individual differences in survival (frailty). In this presentation we examine new statistical techniques which allow all of these components estimated in a joint analysis of longitudinal data focused on the earliest possible prediction of late onset AD. Methods: We examine the mixture of latent growth and survival models, termed “shared parameter models,” based on the overview of Lin, McCulloch, & Mayne (Statistics in Medicine, 2002) and the recent contributions of Guo & Carlin (American Statistician, 2004) and Vonesh, Greene & Schluchter (2005, submitted). We apply these models to the early prediction of AD in the Kungsholmen Project, a longitudinal study of aging and dementia (Fratiglioni et al, 1997). These analyses are based on N=1,600 People aged 75–95 who were measured up to 4 occasions on both psychometric tasks and clinical interviews (Incidence Rate∼10%). To combine the most useful features of these longitudinal models we examine several models of: (1) latent growth (decline) components over age, (2) age-based survival-frailty models for the onset of AD, and (3) examine several alternative models for their inter-relationships. Conclusions: These analyses illustrate the accuracy and reliability of this approach and highlight both the potential benefits and the potential limitations of integrated growth-survival analyses based of longitudinal data.