Abstract
In recent years, the use of adaptive design methods in clinical trials has attracted much attention due to its flexibility in identifying the best clinical benefit of the test treatment under investigation. The flexibility, however, comes at the price of decreasing the accuracy and reliability of the statistical inference drawn. In addition, it is susceptible to abuse. The Food and Drug Administration (FDA) draft guidance justifiably distinguishes between well-understood and less well-understood adaptive designs and suggests the use of the latter with caution. In this discussion paper, we further classify the less well-understood adaptive designs into the categories of flexible and wildly flexible ones and recommend the latter not be used. In addition, we suggest a set of performance characteristics as criteria for choosing a good design from a pool of flexible adaptive designs and group sequential designs.