American Statistical Association
Minimisation and methods that make use of optimum design theory have been suggested to balance treatment groups across prognostic factors. Although the problem of analysing a trial when one of these methods has been used has been looked at in the fixed-sample case, it has so far not been considered in the group sequential setting. In this talk, simulation is used to explore the consequences of adapting for prognostic factors in a group sequential trial. Both Pocock's test and the O'Brien and Fleming test are considered and three methods of adjusting for covariates are studied. When the variance of the response variables is unknown, the critical values are obtained using those in the known variance case and the significance level approach. The resulting tests have approximately the required type I error probability. To maintain the desired power, sample size re-estimation is incorporated. Simulation shows that the tests satisfy the power requirement for moderate sample sizes, with complete randomisation being less powerful than the adaptive methods.
Steve Coad is a Reader in Statistics at Queen Mary, University of London, where he has worked since 2005, having previously held appointments at the Universities of Sussex, Michigan and Newcastle upon Tyne. He obtained his DPh from the University of Oxford. He is currently an Associate Editor for Sequential Analysis and the Journal of Statistical Planning and Inference. His research interests are mainly in the area of sequential design and analysis, with particular emphasis on estimation problems, asymptotic approximations and adaptive treatment allocation. For more information, or to arrange a meeting with Dr. Coad, please contact Cheng-Shiun directly.
|Date:||Thursday, June 23, 2011|
|Time:||2:00 - 3:00 P.M.|
Mailman School of Public Health
Department of Biostatistics
722 West 168th Street
Biostatistics Conference Room
6th Floor - Room 627
New York, New York