Bayesian Adaptive Designs for Phase IIb Dose-ranging Study in Rheumatoid Arthritis (RA)
Shuying Yang, Pauline Lukey and Misba Beerahee
GSK
Objectives: A Bayesian adaptive design was proposed for the phase IIb dose ranging study to effectively and efficiently find the minimum efficacious dose (MED) of an anti-inflammatory drug for the treatment of Rheumatoid Arthritis (RA). The purpose of this work was to assess the possibility and validity of this method using clinical trial simulations.
Methods: With this design, the study started from a pre-defined initial dose (e.g. the most likely dose), the dose for the next cohort was determined by the posterior probability of the response based on the accrued data. An up-and-down allocation rule was proposed to allow progression towards the MED in either dose escalation or reduction. For the simulation, a logistic regression model of ACR20 response (American College of Rheumatology definition of response using a composite clinical improvement of 20%) on log transformed dose was used to illustrate the method.
The simulations were conducted based on three scenarios on dose response curve: slope=1, slope=0.5 and slope=0 (no difference from placebo). For illustration, 100 simulations per scenario were generated. R[1] and WinBUGS1.4[2] with R2WinBUGS library [3] was used to perform the simulations.
Results: The simulation results indicated that with scenarios where target dose was in the dose range, the design was able to identify the MED for all simulations. For the scenario with no difference from active to placebo, the design was able to stop the trail as early as possible. However the number of cohorts may depend on the initial dose selected and the allocation rules.
Conclusions: Bayesian adaptive design was appropriate in finding MED in the dose ranging studies. It minimised the number of patients assigned to least efficacious doses or administered with high doses. The sample size for each cohort and the randomisation ratio could be calculated by simulations as described. Finally it was flexible enough to incorporate different allocation and stopping criteria within the framework
References:
[1] The R project for statistical computing, http://www.r-project.org/
[2] Spiegelhalter D et.al. WinBUGS User Manual, 2003, http://www.mrc-bsu.cam.ac.uk/bugs
[3] Sturtz S et.al. R2WinBUGS: A package for running WinBUGS from R. Journal of statistical software, 2005, 12(3) 1-16.
