Methods for Modeling the Association of Change Points with Risk Factors for Cognitive Decline in the Elderly

• Activity Number: 158
• Type: Topic Contributed
• Date/Time: Monday, August 1, 2011 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Charles B. Hall*+
• Companies: Albert Einstein College of Medicine
• Keywords: change point ; longitudinal data ; profile likelihood ; Bayesian ; Alzheimer's disease ; dementia

Abstract

Cognition in the elderly often is relatively stable up to some point at which individuals at risk for mild cognitive impairment or dementia begin to experience accelerated decline. Change point models are useful for modeling this transition, as they directly estimate the rates of change before and after the acceleration. The dependence of the change point, and the rates of decline, may depend on risk factors. This association may be modeled in three ways: (1) A profile likelihood method where the rates of decline conditional on the change point are modeled using standard mixed linear model software, (2) a maximum likelihood approach where the change point and the rates of decline are simultaneously estimated using a nonlinear mixed effects model, and (3) a Bayesian approach using Markov Chain Monte Carlo simulation. The Bayesian approach has the advantage of the ability to model heterogeneity in the change point across individuals beyond that captured in the known, measured risk factors. The methods will be compared using data from the Bronx Aging Study to estimate the effect of a purported marker for cognitive reserve.