Replies to post #3631 on Dendreon Corp fka DNDNQ

[walldiver]

Let's assume that the 180th event occurs on July 1, 2008 and the DSMB unblinds for the first interim look. Let's also assume that 120 patients had enrolled by July 1, 2005, approximately two years after the first patient enrolled. Let's further assume that enrollment for the first year was linear, at 4 patients per month. The second year enrollment assumption is 6 patients per month. I wonder how much the extra time beyond three years the long-term survivors are followed would help the p value.

Looking at some of the simulations, I'm starting to wonder if it would be better for the company if the interim unblinding occurs in the 210-240 event range instead, and reserved alpha is 0.02...

[nmstav]

interim stat sig but just barely...

iwfal,You say your guess is that interim is stat sig - but just barely given the < than 2-yr randomizaation. If you had to guess at stat sig for final look...how much more than just barely would you guess?

Walldiver re your Q/A...does the "final" look alleviate a lot of your own concerns about interim data?

[gofishmarko]

>>> How much of the shortfall in log rank p-value and HR is attributable to smaller size? <<<

"Very little. Trial size was only a small contributor."

Clark ,

Can you elaborate on this some ?

Say a large trial achieved the same 3-yr. survival figures as 02A , 32% vs 21%. That's a 52% relative improvement in 3-yr. survival. Surely that would yield high statistical significance at some reasonable trial size , at least by landmark-type tests , no ? If so , and log-rank K-M would fail to yield a similar outcome , then there would seem to be some problem in Statisticsland.

My tendency is to look at the last standing data as being most predictive , and that's the 3-yr. survival rates. 01 and 02B numbers seem to reflect a range that's within the variation expected in small trials , and the integrated figure is what I'd place my bet around for the outcome of a large trial ( 33% vs. 15% , or a 120% relative improvement ). The K-M curve has to get to that 3-yr. point in some manner that makes sense in a larger trial , i.e. , the zigs and zags in the curves will smooth out , and the associated stats will have to make sense as well.

TIA for your comments.