Abstract:
|
The T-statistic design (Ivanova, 2008) for adaptive dose-finding is not typically used for dose escalation trials to find Maximum Tolerated Dose (MTD). It was compared to the mTPI, CRM, and BLRM adaptive designs based on probability of correctly estimating at or adjacent to MTD, minimizing Dose-Limiting Toxicities and the probability of assigning subjects to doses > TRUE MTD. Multiple versions of each design were chosen in an attempt to optimize the designs to improve the performance characteristics. 5 TRUE underlying dose-toxicity-response curves for simulations to compare the designs were chosen to span the range of potential TRUE underlying DR curves for a new drug with no prior human exposure to inform the Bayesian prior distributions for mTPI, CRM, and BLRM. Results of simulations suggest the T-statistic design performs favorably among mTPI, BLRM, and CRM. Each design performed best for at least 1 case. A scoring system based on proximity to best performance in relation to the range of values across all designs yielded similar findings; however, when that scoring was averaged over all cases, the T-statistic design performed best, although not much better than mTPI.
|