Replies to post #3515 on Dendreon Corp fka DNDNQ

[iwfal]

This is, of course true. But I am assuming that the efficacy effect is not small. Also Cox only corrects for imbalances that we know about, it cannot correct for factors that are not recorded. In theory at least, a reason for a large trial is to minimize the chance of missing some factor that might turn out to be significant and imbalanced.

This is true. But also true in a small trial.

In any case, how do you run MC with Cox without data on individual patients?

Generally I am actually doing the inverse. Using a Cox model to add noise to a system and see what happens to p value. I did check once to ensure that the inverse does work the same as the forward regression - but I have to run the forward regression by hand so it is prohibitive to run it routinely.

[p3analyze]

Question to Clark regarding the MC Cox exercise: what covariates did you use in the model?

I am slightly confused by the discrepancy between what Gold said on the CC, and what Mark did on the Adcom.

Both said Gleason score (<=7 vs >=8) is a stratification factor, so it can't be used as a covariate.

In the AdCom, Mark said in addition, they stratify for bisphosphate use and bony metastases was another stratification factor, but in Gold's response to Duncan, he seem to indicate that the number of bone mets will be used as a covariate in the model.

"
DR. HUSSAIN: And are you somehow
2 doing any kind of stratification to account
3 for potential prognostic variables?
4 DR. FROHLICH: We are stratifying
5 for Gleason score bisphosphonate use and
6 study center.
7 DR. HUSSAIN: Thank you.
8 DR. FROHLICH: I'm sorry, number
9 of bony metastases as well.
"