题目: Bayesian Adaptive Randomization for Oncology roof-Of-Concept Trials

时间：2011年3月23日星期三，下午3:00-4:00报告, 4:00-5:00座谈交流

报告人：Pantelis Vlachos（Merck Serono 制药公司）

Pantelis Vlachos is a Principal Biostatistician in the Global Processes, Standards, Research and Systems group (GPSRS) of Merck Serono. He joined the company two years ago and has been involved in projects and studies spanning all therapeutic areas of Merck Serono. Before that, for 12 years he was a faculty member at the Statistics department at Carnegie Mellon University. He has worked and published on Bayesian Designs of Clinical trials, as well as Bayesian Hierarchical Model Checking and Text Analysis. He is the Editor of StatLib, and the Managing editor of the Current Index of Statistics and the open access online journal "Bayesian Analysis".

Learning - Review the way proof of concept trials in oncology

Objectives: are conducted - Understand the concept and the workflow of adaptive randomization - Evaluate Bayesian methodology in proof-of-concept trials

Abstract Summary: Understand the Adaptive Randomization design and how this fits into early development. Evaluate through simulations the operating characteristics of Bayesian Adaptive Randomization (BAR) in proof-of-concept dose selection studies in oncology.

Abstract Text: Objectives. Evaluate through simulations the operating characteristics of Bayesian Adaptive Randomization (BAR) in proof-of-concept dose selection studies in oncology. Methods. A proof-of-concept study (placebo, low and high dose of experimental treatment) was used to evaluate the BAR characteristics. The statistical design links treatment assignment probabilities to performance of respective arms. Smooth adaptation of assignment probabilities is guaranteed by a tuning parameter controlling how the randomization is influenced by data. A standard block randomization with equal assignment probabilities is used in the first batch of subjects. A Bayesian model summarizing prior information has been implemented with priors assuming no activity in order not to unbalance the randomization. Higher prior variance was given to the experimental arms reflecting uncertainty on the drug activity. The posterior probability that placebo has better performance than experimental treatment arms will be used to reject a “null hypothesis” of no drug activity. A screening design was used to calculate the maximum sample size and to compare operating characteristics of the two methods. Results. Simulations evaluated different scenarios of activity, priors, and rejection regions, showing consistent allocation of subjects with the performance observed in the study. Final Bayesian analyses were more powerful while controlling the same alpha level (90% vs. 85%) Conclusions. BAR is a flexible tool which may fit easily to the needs of early development and may be ethically appealing as more subjects are assigned to more active treatment arms. In addition futility stopping can be easily implemented in case of no activity.