|
Abstract:
|
Acute graft-versus-host disease (aGVHD) is a complication of allogeneic hematopoietic cell transplantation (aHCT) and is a leading cause of death in patients receiving aHCTs. Thus, investigators would like to have models that accurately predict those patients most likely to develop aGVHD in order to minimize over-treatment with steroids as well as reduce mortality. To this end, we propose using biomarkers that are collected weekly to predict the time-to-aGVHD through pattern mixture model. We consider settings in which the population is a mixture of various aGVHD risk classes so that the biomarker trajectories are irregular and possibly from different distributions. We compare different modeling approaches through simulations motivated from actual data collected at the University of Michigan Blood and Marrow Transplant Program.
|
