Replies to post #143541 on Biotech Values

[iwfal]

ARQL -

If what Roche found to be "most predictive" of MET levels is the same test that ARQL used to arrive at its numbers, I don't understand why that would be not comforting.

Because it is clear that the efficacy of the drug is very strongly correlated to MET levels - so any post hoc subgroup they threw out to bring the remainder to stat sig would, almost by definition, have a low MET level. The only question is whether there is evidence that that removed post hoc subgroup is behaving as you would expect it to behave from an external validation standpoint. And the answer appears to be no - squamous should NOT have hugely lower MET levels.

For example, I'll bet that I could construct a subgroup composed of some combination: A) of hair color, B) year-of-birth-mod-5, and C) location birth (by US quadrant and world area) and that subgroup, when pulled out leaves the remainder stat sig. Now I'll bet you 10:1 that that removed subgroup will have low MET levels.

When it does indeed have low MET levels have I given myself more confidence that the ph iii (run without blond haired people born in 1963, 68, 73, ...) will actually be stat sig? Ans: no, because there is every to believe that blond haired people born in 63, 68, 73 etc should have the same MET levels as everyone else. By the same token the data I cited says that generally squamous does NOT have low MET levels and thus the squamous MET data from ARQL is actually evidence of post hoc contamination. The only question in my mind is the degree of contamination.


All that said, I am not saying the drug is inevitably going to fail in the on-going ph iii - only that it suffers from the same failures as most other first ph iii's of what eventually turns out to be a drug efficacious in a subgroup. And my guess is that is a >50% failure rate. (Further note - I think it is an even better bet that even if it passes the ph iii it will be with weak efficacy numbers. E.g. HR>=0.8).