An Application of Adaptive Randomization to Guide Dose-Finding in a Large Multicenter Phase I Clinical Trial

• Activity Number: 522
• Type: Contributed
• Date/Time: Wednesday, August 1, 2012 : 10:30 AM to 12:20 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Alison Marie Graham*+ and Wijith Prasantha Munasinghe and James Bolognese and Yi-Lin Chiu
• Companies: Abbott and Abbott Laboratories and Cytel Inc. and Abbott Laboratories
• Keywords: phase I clinical trial ; dose-finding ; adaptive randomization

Abstract

A multi-center phase I clinical trial was conducted in more than 100 healthy subjects to determine a regimen which provided an acceptable low rate of a binary outcome event. For cost and time efficiency, an adaptive randomization plan was implemented to use accumulating data from prior regimens to guide the choice of subsequent regimens. Simulations under an assumed dose response (DR) suggested the frequentist T-statistic design (TSD) provided better model fit than the Bayesian four parameter logistic model (4PL) with non-informative priors. Actual trial data did not follow the assumed DR, and so both TSD and 4PL were applied, with both identifying the maximum allowed regimen, suggesting a higher dose range was needed. Based on observed rates of the binary outcome, the trial was extended to include additional regimens. A normal dynamic linear model was found to best fit the accumulated data, and predictions for the extension regimens were made. Unique challenges of this trial included a long event observation period and fast enrollment; thus predictions for the full observation period were made based on available treatment data using predictive probabilities.