A Multi-State Parametric Approach for Modeling Cognitive Dynamics

• 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): Arnold Mitnitski*+ and Nader Fallah and Charmaine B. Dean and Kenneth Rockwood
• Companies: Dalhousie University and Dalhousie University and Simon Fraser University and Dalhousie University
• Keywords: multi-state model ; stochastic process ; truncated Poisson distribution ; mortality ; cognitive decline ; cognitive improvement

Abstract

We present a novel parametric multi-state stochastic model to describe longitudinal changes in cognitive tests. Scores are modeled as a truncated Poisson distribution, conditional on survival to a fixed endpoint; the Poisson mean depends on the baseline score and covariates. The basic model has four parameters. Two represent the cognitive transitions of individuals with no cognitive errors at baseline ("cognitively fittest") and their survival. The two other parameters represent corresponding increments in the probability of transitions and of mortality as a function of baseline cognition. Note that this approach permits estimation of the probabilities of transitions in different directions: improvement, decline and death. It allows gradual changes to be discerned, before progression to diagnostic and staging thresholds that are often irreversible. The model does not suffer from the common limitation of being applicable to only a small number of states. This is illustrated by comparing the model's performance with polytomous logistic regression. The model's performance is illustrated by applying it to different cognitive measures (MMSE, CASI, ADAS-Cog) in several databases.