Keywords: trial simulation, net present value, decision theory
The job of a Phase II trial is to help us decide whether to run a Phase III trial and to help us design ha trial. So the usual benchmarks of trial design - type-1 and type-2 error are only part of the story for a Phase II trial. A Phase II trial that leads us to design the wrong Phase III trial commits an error that has both the cost of a type-1 error and the lost opportunity of the type-2 error. But there is next to no help with this problem for the statistician designing a Phase II trial.
Perhaps if we were able to simulate both the different Phase II designs and what happened next, so we could quantify the expected overall outcomes of the whole development program, it would help us design Phase II trials that we knew were fit for purpose. But is such a thing practicable, and what would it need to take into account?