Extension to Time to Event Data of the Outcome Adaptive Bayesian Dose Finding Procedure of Thall and Cook with Application to a Bone Marrow Transplant Trial
The Emmes Corporation
Johnson Center 3rd Floor- Meeting Room D
4400 University Drive, Fairfax, VA 22030
Time: 11 am – 12 pm
Date: Friday, April 13, 2012
My talk will discuss research that I am conducting with Eric Leifer, a statistical colleague at the NHLBI. We are studying two modifications to the outcome-adaptive, dose finding Bayesian procedure proposed by Thall and Cook (TC)(2004). In the Phase I-II setting, the TC procedure assigns doses of an experimental drug to successive cohorts of subjects, using short-term efficacy and toxicity outcomes.
The research is motivated by our application of the TC method to the BMT CTN protocol 0301, a multi-center trial in bone marrow transplantation for severe aplastic anemia. In this trial, enrollment is periodically suspended while awaiting 100-day post-transplant efficacy and toxicity outcomes. If potential recruits cannot wait for a transplant, they must be turned-away and treated off-study. Since severe aplastic anemia is a very rare disease (< 5 cases/million), this is very problematic.
As a result, we began studying how to modify the TC algorithm to incorporate incomplete subject follow-up data, with the goal of facilitating enrollment without sacrificing design performance. The first modification uses a bivariate time-to-event model for the efficacy and toxicity outcomes which allows for censoring. The second modification is our adaptation of the look-ahead rule, introduced by Cook (2004), to bivariate binary and time-to-event outcomes. We examine the performance of both modifications in simulation studies.