Premium
A varying‐stage adaptive phase II/III clinical trial design
Author(s) -
Dong Gaohong
Publication year - 2013
Publication title -
statistics in medicine
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.996
H-Index - 183
eISSN - 1097-0258
pISSN - 0277-6715
DOI - 10.1002/sim.6036
Subject(s) - stage (stratigraphy) , interim , computer science , interim analysis , computerized adaptive testing , phase (matter) , sample size determination , early stopping , adaptive design , statistics , clinical trial , artificial intelligence , mathematics , medicine , paleontology , chemistry , archaeology , organic chemistry , biology , artificial neural network , history , psychometrics
Currently, adaptive phase II/III clinical trials are typically carried out with a strict two‐stage design. The first stage is a learning stage called phase II, and the second stage is a confirmatory stage called phase III. Following phase II analysis, inefficacious or harmful dose arms are dropped, then one or two promising dose arms are selected for the second stage. However, there are often situations in which researchers are in dilemma to make ‘go or no‐go’ decision and/or to select ‘best’ dose arm(s), as data from the first stage may not provide sufficient information for their decision making. In this case, it is challenging to follow a strict two‐stage plan. Therefore, we propose a varying‐stage adaptive phase II/III clinical trial design, in which we consider whether there is a need to have an intermediate stage to obtain more data, so that a more informative decision could be made. Hence, the number of further investigational stages in our design is determined on the basis of data accumulated to the interim analysis. With respect to adaptations, we consider dropping dose arm(s), switching another plausible endpoint as the primary study endpoint, re‐estimating sample size, and early stopping for futility. We use an adaptive combination test to perform final analyses. By applying closed testing procedure, we control family‐wise type I error rate at the nominal level of α in the strong sense. We delineate other essential design considerations including the threshold parameters and the proportion of alpha allocated in the two‐stage versus three‐stage setting. Copyright © 2013 John Wiley & Sons, Ltd.