Interesting point. I was thinking about it in terms of dose, usually phase 1 trials seem to have significantly lower doses, to feel out any issues with safety. But perhaps that doesn't apply in this case, and you are left with the issues you mention about statistical significance, selection bias and regression to the mean.