Replies to post #4316 on Dendreon Corp fka DNDNQ

[i760]

Cox vs. Log rank & IMPACT true HR

What is expected HR for IMPACT/9902b ? My WAG based on the analysis below is 1.60 for interim and 1.75 for final. This is why I think DNDN says reasonable power for interim and very well powered for final, even after allocating alpha. Comments welcomed…..

Cox approach used in 9901 / 9902a was not fully pre-specified and had a much larger benefit than typical, and thus it would be imprudent to assume that HR for IMPACT would be as good as the COX HR for the combined dataset (i.e. 1.86). That being said true HR for IMPACT should be better than the combined log rank of 1.5, since I do think that Cox was not a total sham analysis and we have other evidence that confirm that 9902 a was biased against treated. So split the difference -- 1.68 for Cox analysis of IMPACT.

Re the other factor, the "ego" effect (i.e. HR at interim will be lower than final), IMHO it will be less than most here think. Remember that if you estimate it by looking at the combined log rank curves, you are not taking into account that the combined dataset was imbalanced against the treated group (which is why Cox HR was so much better). Thus the lack of curve separation could have been because the sicker treated patients died soon. The analysis plan for IMPACT corrects for this in two way: Cox and stratification by mets. Thus if the placebo vs. treated are better balanced, perhaps there will be more separation sooner - how much is a WAG. Again split the difference: since using the combined log-rank curves seems to suggest two year HR goes down by about 0.30 when compared to the final (i.e. 2 yr of 1.3 vs. 1.5 for final), I would think that reducing it by half or by 0.15 is reasonable.

Lastly, there will be some benefit from the early enrollees as IMPACT is event driven for both interim and final analysis (call this the Mr. Garcia effect). My WAG about 0.05.
Put it all together:

For Interim: Cox HR 1.68, minus “ego” effect of 0.15, plus Garcia effect of 0.05 = 1.58

For Final: true cox of 1.68, plus Garcia effect of 0.02 = 1.70.