I was wondering if anyone could comment on the history of clinical trial design. It seems like correct setups are designed to be "foolproof" -- which is definitely a good thing. But is it necessary as it is?
Trial designs seem to be contrived from a patent's point of view. My sense is that the history of trial design is greatly influenced by frequentist statistics. Sadly, most patients don't think in these terms. Natural human learning is an evolution of beliefs about outcomes. This is more closely related to Bayesian statistics.
From a pure mathematical point of view, it is clear one not need to conduct "trials" to learn how to windsurf. One can learn as you go, developing beliefs about the wind and how board and sail position affect movement. A single person can do this and become an effective windsurfer, even in "random" environments where the wind and waves change. It would seem to me that adopting a Bayesian approach might allow patients to opt-in to trials at any given time -- assuming that conditional variable are chosen to have a homogeneous prior. A critical difference is that the signal to noise ratio of windsurfing is much larger than cancer.
I suspect that human behavior would lead to a significant subset of people -- "explorers" -- jumping to trials with fattest prior tails -- as opposed to largest median outcome. As approaches of interest became more well defined, patients would select other therapies with long odds.