Did the same calcs as you. This PR is a bit disappointing, because it shows that our trial results were actually pretty muddy.
The q-3wk seems to have strong signal, but the flip side of that coin is that the q-1wk is no signal at all, and together make little physiological sense. (You could even claim B made SOM worse in the q-1wk, which you can dismiss but then you compromise the sub-grouping itself.)
The company is putting a spin on this, but this is not really what you want to see. They're even fibbing a bit: the Q-3wk was not the "more effective" sub, it was the only effective sub. This is going to attract unfortunate accusations of data mining.