Abstract:
|
Hematopoietic stem cell transplantation is currently the standard of care for patients with hematopoietic malignancies. However, a large number of patients who undergo transplantation may develop a form of transplant rejection called acute graft-versus-host disease (GVHD) which is tied to increased mortality. A main challenge of statistical modeling of acute GVHD is that the definition of acute GVHD is tied to a time limit: acute GVHD is only diagnosed within 100 days post-transplant, after which the diagnosis changes to late-onset acute GVHD. Another challenge is that acute GVHD is typically modeled as a univariate outcome, which ignores the strong force of mortality. To address these two issues, we propose a multi-state model to jointly model acute GVHD and death, where time from transplant to acute GVHD corresponds to a survival model defined on a finite interval. We illustrate our methods using data from a large stem cell transplant registry. Through simulations, we show that failing to account for the finite nature of the nonterminal event in model fitting can result in biased estimation and incorrect inference.
|