Key points are not available for this paper at this time.
For life-threatening diseases (HIV/AIDS, oncology, etc.), doses studied in the Phase IIB stage are usually chosen to be high enough to achieve at least a required therapeutic effect. These doses are likely located at or close to the plateau of a dose–response curve and have a similar treatment effect. In this article, we consider a combined Phase IIB/III design with multiple doses being tested in the Phase IIB portion and one selected for Phase III. The final test statistic to compare the experimental treatment at the selected dose to the control is a weighted sum of a “MiniPool” statistic for the Phase IIB portion, which is the minimum of potential pooled statistics based upon pooled data from the selected dose and any potential combinations of other doses, and a conventional statistic for the Phase III portion. Intuitively, the MiniPool statistic for the Phase IIB portion includes data from less efficacious doses and excludes data from toxic doses. The MiniPool approach may be considered as an application of the closed testing principle. A clinical trial in HIV-infected patients is used to illustrate this method. A simulation study shows that the MiniPool method controls Type I error rate and preserves power compared to the naïve approach. In scenarios where doses studied in Phase IIB have similarly efficacy, the proposed MiniPool method performs well as expected. The performance of the MiniPool approach is reasonably robust even when the studied doses are substantially different.
Chen et al. (2009) studied this question.