Multistate models and lifetime risk estimation: Application to Alzheimer's disease

The lifetime risk of a clinical condition is the probability of onset of the condition during one's lifespan. Recent advances in Alzheimer's disease (AD) research have identified screening tests for biomarkers that can identify persons who are in the earliest stages of the AD process but who do not yet have any clinical signs or symptoms. A critical question asked by patients and clinicians is, what is the probability an individual will develop AD dementia during his or her lifetime? Here, we discuss estimation of lifetime risks using biomarkers for preclinical disease based on a discrete nonhomogeneous Markov multistate model for the disease process. We allow the transition probabilities to depend on chronological age. In addition, we allow the probabilities to depend on calendar time to account for possible calendar trends in death rates and to evaluate the impact on lifetime risks of future interventions designed to slow disease progression. We develop estimating equations for calculating the lifetime risks from the nonhomogeneous multistate Markov model. Estimates of lifetime risks for AD dementia based on biomarkers for preclinical disease are presented.