|
Abstract:
|
Transition models are useful in estimating the probability of recurrent events in longitudinal studies. However, direct application of a transition model may suffer from two complications, informative cluster size and informative gap time between observations. For example, Consecutive Pregnancy Study (CPS) is a retrospective cohort study aiming at understanding the recurrence patterns and predictors of adverse pregnancy outcomes, such as preterm birth. The number of pregnancies observed and the gap time may be both indicative of a women's underlying fertility, and hence correlated with the pregnancy outcomes. We propose a shared random effect structure for jointly modelling the transition model with the informative observation process. The gap time is modelled by a parametric distribution with right censoring; the cluster size is characterized by a continuation ratio model. Through extensive simulation studies and analyses of the CPS data, we show that naive approaches ignoring the cluster size model, the gap time model, or both, could lead to seriously biased inference.
|
