My question is this--If we have drunk the Kool aid and really believe Provenge works as the explanation for the results of 9901 and pooled data of 9901 and 9902a, why should we not believe that the interim--with more patients and more death events than the pooled groups--will be stat sig--and even gloriously stat sig after Cox? Am I missing something? Do we need to wait for 3 year survival to expect the best possible results? Is there something about "curve separation" (which has never been explained to me by the way and I don't really understand--I'd appreciate an explanation)?
a) You should not expect Cox Regression to generally get the kind of improvements in p value that you saw in 9901 and 9902a. A factor of 2 is probably more 'normal' - but note that on occasion it can actually hurt your p value.
b) The instantaneous HR is not constant through time in either 9901 or 9902a. Both have very low HRs early on and much higher HRs later. I could take 3 pages to explain this, but to keep it brief you should just understand that if you have two straight lines on a survival curve that start at the same point that the HR will start at 1.00 near the point where the lines intersect and move to infinity as the lower line approaches 0% still alive. And the 9901 curves are more like straight lines than they are exponential curves (which maintain a constant HR). Thus, by getting more events in the early part of the curve you will be bringing down the HR and thus the p value. Will this characteristic repeat in 9902b? I believe so since I believe it is consistent with how cancer vaccines work - they work better on patients that are less ill.
Seems like if there are 180 deaths--and again the vast majority are placebo guys, the interim should look as good as the pooled data did
Most deaths will be treated because it is a 2:1 randomization. FWIW.
Clark
News 
Market Data 
Discover 
