Shirin Golchi - Snr Research StatisticiaShirin Golchi - Snr Research Statistician, MTEK Sciences, Vancouver BC.) - ( SPECIAL SEMINAR n, MTEK Sciences, Vancouver BC.) - ( SPECIAL SEMINAR

Computational Techniques for Bayesian Adaptive Randomized Trials

http://www.sfu.ca/~sgolchi/Bayesian adaptive clinical trials have gained popularity due to their flexibility in modifying the design and use of stopping rules according to the intermediate results throughout the trial. Decision criteria, including arm dropping, trial termination and adjusting the randomization probabilities, are commonly defined based on functionals of the posterior distribution of model parameters. In most interesting statistical models, however, the posterior distribution is analytically intractable and therefore repeatedly updating the posterior poses inferential and computational challenges. In addition, design of Bayesian adaptive trials requires comprehensive simulation studies since the operational characteristics of these designs (namely power and type I error rate) cannot be analytically obtained as a function of the decision rules. This talk will be focused on computational techniques and methodology to address the challenges that arise in design of Bayesian adaptive trials. A Sequential Monte Carlo sampler is used to efficiently update the posterior within each simulation run. A stochastic emulation technique is then proposed to optimize the design operational characteristics based on a number of simulation runs at selected decision criteria.