Abstract:
|
In many observational studies, it is of interest to evaluate a state change (e.g., treatment initiation) which is time-dependent (i.e., assigned after time 0). In the setting of our interest, patients are at risk for a recurrent event (e.g., hospitalization) and a terminating event (e.g., death), with the events being correlated. We develop methods for estimating the effect of the state change on the recurrent and terminal events. We propose a two-stage method wherein the first stage involves estimating prognostic scores through a shared frailty model. At the second stage subjects are matched and stratified models are fitted to the observed recurrent and terminal events. The method is tested in moderate sized samples through simulation, then applied to liver transplant data.
|