B1155
Title: Estimating dementia incidence in an illness-death model using health-claims data
Authors: Achim Doerre - University of Rostock (Germany) [presenting]
Abstract: The motivation comes from the study of dementia in humans, which usually does not occur before reaching age 60. We use an illness-death model to describe the health process over time, where individuals migrate from the initial healthy (non-diseased) state to either the death or disease state. While the death state is absorbing, migration to the disease state is eventually followed by migration to the death state. This model contains three migration rates that are unknown in general, among which the incidence rate is of primary interest. When incidence rates of chronic diseases in humans are studied, cohort and period effects are often of interest. Unfortunately, simple random samples are usually not obtainable for age-related diseases. As an alternative, health-claims data offer large-scale contemporary information on the health status of individuals, and may be regarded as observational data. Under this sampling scheme, incidence times are both left and right-censored. Furthermore, truncation occurs because only those individuals are sampled which are alive when the data collection begins, leading to individual-specific truncation times. We derive consistent and asymptotically normal Maximum Likelihood estimators using an inverse probability weighted likelihood function. In order to account for different cohorts possibly sharing certain migration rates, we describe a simple model selection procedure. A large health-claims dataset is used to illustrate the method.
