gfp - another explanation might involve the fact that GOSE is a finer scale than the earlier GOS scale, with 8 categories vrs 5 categories
Thanks. That is certainly true with the interim lumps that Pharmos reported at the lower GOSE levels. But it just happens that they reported the GOSE 7-8 together and GOSE 7-8 is defined to be identically equal to GOS 5. The nomenclature is even the same "Upper Good Outcome", ... . (I have no idea if the mechanics of scoring the test actually rigidly make this so, but I would assume even at a worst case it should only be a small change.)
BTW - One more thing that makes me very comfortable that Dex actually has efficacy inre GOS is that the probabilities that they reported all had to do with only GOS 5. Is the rest of the scale chopped liver? No, but there isn't any (easy?) way to numerically assess the rest of the scale in traditional statistics that looks at the entire curve. But I would argue that a null hypothosis that allows you to test the shape of the rest of the curve is that if the GOS 5 treated got better by chance it must have done so at the expense of the the GOS 4 group. But the GOS 4 treated group should have a hole in it. It doesn't. Presumably because it, in turn, got filled from GOS 3. And that appears to be what happened in the 1 month GOS chart which is the cleanest (least ceiling'd but otherwise pretty much the same as the 3 and 6 months charts). With a few Bayesian assumptions it is actually possible to calculate odds - and the odds that the treated and untreated curves would look the same but with treated slipped to the left are fairly low. Of course the Bayesian assumptions are not allowed by the FDA, but they are, I think, fairly good ones (e.g. I think that the chance that someone will randomly 'show up' as a GOS 5 when they 'should have been' a GOS 3 or less is, say, less than 10%). I haven't done the full calculation (and I may never do it since my goal was just to get ballpark odds from the shape of the curve) but it provides additional confirmation. It also gives some comfort that the culled data will not kill us, since it is unlikely that they culled the data set just to get the right curve shape.
Note that there is a difference between saying that Dex has an effect (as per above) and saying that the Phase III will get positive primary endpoints (e.g. the entry criteria changed quite a bit between II and III).
Clark |
