At a Glance
Multistate Models under Outcome-Dependent Selection
Jutta Gampe, Roland Rau, Marcus Ebeling; in Collaboration with Hein Putter (Leiden University, Medical Center, Netherlands), Karin Modig, Anna Meyer (both: Karolinska Institutet, Stockholm, Sweden)
Multistate models analyze the transitions between several states, such as different stages of disease progression or the movements through different living arrangements over the life course.
The key quantities are the transitions rates, possibly modified by covariates, which describe the dynamics over time of the multistate process. Once the transition rates are estimated, further key quantities, such as the transition probabilities, the state prevalence, and life expectancies in particular states can be calculated.
The estimation of transition rates requires longitudinal data, which often come with incomplete observation schemes, such as right-censoring or left-truncation. Methods for these complications exist, and also interval-censored observations (panel data) can be dealt with. If, however, the sample is collected conditional on some event, for example death having happened, and the process up to this event is to be studied, then this outcome-dependent selection needs to be taken into account. A typical example is the health trajectory of individuals who have been selected based on their death having occurred by a certain point in time. This conditional selection of observations violates some standard assumptions in the estimation of transitions rates and is so far understudied. This project studies the statistical implications of this sampling procedure.
Aging, Mortality and Longevity, Health Care, Public Health, Medicine, and Epidemiology, Life Course, Statistics and Mathematics