Changes in cognition during the course of eight years in elderly Japanese Americans: a multistate transition model

PURPOSE: To analyze cognitive changes in relation to mortality with the use of a multistate transition model. METHODS: In a prospective cohort study of Japanese Americans living in King County, WA, study (n = 1985) cognitive states were defined as the errors in the Cognitive Abilities Screening Instrument score. Transition probabilities were modeled by the use of a modified Poisson distribution with the Poisson mean and mortality dependent on the cognitive state and covariates. RESULTS: During an 8-year follow-up, 21.5% died (95% confidence interval [95% CI], 19.3-23.7), 26.6% experienced cognitive decline (95% CI, 24.2-29.1), and 51.9% remained stable or improved cognitively (95% CI, 49.2-54.6). In multivariable analyses, improvements were notably more likely to occur among younger, more-educated people and in women. Older age, male sex, and less education were each significantly related to mortality. CONCLUSIONS: A multistate transition model can be used to estimate the impact of covariates on the individual probabilities of cognitive improvement, stability, decline, and death. This approach can have advantages for considering how interventions might work on patterns of cognitive change over time, especially if factors associated with the prevention of decline are also associated with risk of death.