Replies to post #3537 on Dendreon Corp fka DNDNQ

[gofishmarko]

>> I believe you are omitting the fact that in the interim analysis the p value will be heavily influenced by the 180 or so patients who are much earlier in their course where the curve separation is much less. The time points of these alive patients would be censored but still contribute detrimentally I believe to the p value evaluation. <<

I did omit that from consideration because I assumed , probably incorrectly , that any such effect would be minimal after 180 deaths.

I agree that modeling can lead you astray when assumptions start to pile up. I would make the fewest and simplest assumptions -- that the population of 02B will look like 01/02A , that early vs. late enrollee relative health will not be a factor ( if it is , it would seem to be in favor of Provenge ) , and that the true HR is 1.5 , as by log-rank in the integrated set. I should have looked at the alternate Cox models , which I believe were in the AC docs , which could provide a sort of " average fudge-factor " that could be applied to the log-rank to arrive at an expected Cox result , though I don't recall if they looked at just 01 or both trials.

No worries , though. Clark will sort this all out for us soon. :)

Thanks to all for the input.

[iwfal]

Clark has suggested some modelling in earlier posts I am catching up on, but I do not know how accurate these are and how much they rely on a priori assumptions that may not hold true.

You are completely correct that the assumptions, and their effects, are the interesting part:

a) I assumed standard exponential survival curves, which have the same instantaneous HR at any point in time - when in fact the 9901/2a curve says that the HR increases substantially over time. Don't know what effect this difference will have.

b) I assumed constant enrollment curves, instead of the back end loaded curves that we will inevitably have.

c) I assumed a TRUE HR of 1.40. Who knows what it is in this population?

...

All in all there is a lot of noise. All you can do is get a rough idea of the playing field.

Clark

[iwfal]

I believe you are omitting the fact that in the interim analysis the p value will be heavily influenced by the 180 or so patients who are much earlier in their course where the curve separation is much less.

A interesting comment. IF(!) the 9902b patient population is the same as the 9901/2a population then HR in the early parts of the curve is pretty low - by 2008 you might expect perhaps 20-25% of the population to have made it to 3 years or more. And 50% to be 2 years or less - where the HR is 1.2 to 1.3 in the 9901/2a combo.

Given that I suspect that the patient population is actually sicker in 9902b (albeit perhaps not too much given the company's assertion that the nomogram survival was similar) I would actually expect lower HRs than in blended 9901/2a?