We use a Bayesian decision framework to construct Adaptive Enrichment trials that choose the optimal trial population at an interim analysis. The possible decisions at the interim analysis must be pre-planned, for example continuing to recruit the same population as the first part of the trial, recruit only from a sub-population, or stopping the trial early for futility.
We ensure strong control of the Familywise Error Rate using well known hypothesis testing methods. The properties of these tests are not impacted by using a Bayesian decision rule at the interim analysis, due to the pre-defined decisions.
These Adaptive Enrichment trials may be optimised to any given scenario, however this does not mean that an adaptive design is always the best choice. The Bayesian decision framework allows us to compare the effectiveness of Adaptive Enrichment and fixed sampling alternatives, showing which design is most suitable. We conduct simulation studies to demonstrate scenarios where Adaptive Enrichment trials can offer an improvement over fixed sampling designs.
These simulation studies make use of several simplifying assumptions, however the method of optimisation is very flexible. The key requirement is to evaluate the expected future behaviour of the trial. In practice this allows for trials that incorporate: multiple sub-populations (these must be pre-identified for the hypothesis testing), multiple interim analyses or the use of more complex data types such as survival endpoints or longitudinal observations.