|
Abstract:
|
Optimum m-stage designs for single arm studies are commonly used in oncology phase II programs. Existing methods select designs with minimum expected sample size under null hypothesis (E(n|H0)), which is denoted by minN. These approaches miss designs with better properties when corresponding E(n|H0) is slightly larger than minN. This issue has been addressed by introducing a concept of family of optimum designs. For a given hypothesis testing setup and a non-negative real number D, a family of optimum m-stage designs consists of all m-stage designs with E(n|H0) less than or equal to minN + D. One can chose a design from this family based on his/her selection criteria. For example, a design with maximum probability of early termination when H0 is true, smallest sample size in stage 1, minimax, etc. Methodologies associated with this approach along with examples will be presented.
|
