where is Skint?, page-36

Well to be honest Interestingtome I disagree with this line you are arguing.

PBT thought that by using Pet scans at baseline it could select a more homogenous patient group that were most likely to respond to PBT2 because in prodomal stage a high proportion of false positives had been incorrectly recruited to previous trials that couldn’t benefit from the treatment.

This meant that patients were randomised following a PET brain scan to confirm high levels of beta amyloid deposit.

The control group were always go to improve simply from regression to the mean. The key question was whether the PBT2 group could be shown to have improved more. It couldn’t.

So nothing is a fluke here. But yes the design was not optimal (or most efficient) for efficacy because safety was prioritised in the 1:2 allocation. And the small N was less than ideal.

But the whole idea of the R bit of RCT is to control for the bias (along with many other biases) that occurs when selection takes place on high scores. This is the inherent problem of the “sweet spot” idea. If you could side-step this like your suggesting ... well no trial would ever fail again.

Regards Southoz