Replies to post #4056 on Dendreon Corp fka DNDNQ

[iwfal]

Would one or some of you please step up and tell the rest of us if you think that we really stand a better than 50-50 chance at showing stat sig at the interim look? If not, please take a shot at the percentage.
Thanks again, I am just trying to stay focused on the prize and the known path to it.

You tell me what assumptions you would like to make:

a) In the end are you expecting curves like 9901? Or like 9902a? Or the blend? (This impacts the odds in obvious ways)

b) Do you believe that the curve shape will remain the same as in 9901 and 2a - where the HR in the early part of the curve is very significantly less than the HR in the later portion? (This very significantly impacts the chance of success for the interim look - note that I looked at the TAX327 trial and it too showed a very linear death curve which would tend to imply a big HR later and a lower HR earlier.)

[fordwill1953]

Sam...

In a way your post is perhaps the best answer as to why the FDA is requiring another set of data from DNDN regarding Provenge. You hit the ambivalence square in the head with your question.

We are all hoping for another 9901 set of survival data for 9902B. But remember, Sam, early on in 9902a, the curves actually inverted. Remember, there were more deaths in the Provenge arm early on and it caused all of us to scratch our heads to figure out how that could be until they figured out the problem with the level of illness in the Provenge arm and used the Cox Regression to adjust the data. The Board was collectively stunned for a few weeks over that phenomena.

Now that the Cox is offered as a legitimate tool on a prescribed basis, in the adjusted SPA, hopefully we don't get another set of wierd data. But, this has been the problem all along for DNDN. The trial sizes kept us all holding our breaths for the actual data thinking that only a few "bad" survival experiences could sink the ship.

I'm still not convinced that we are out of the woods on the interim look with 9902B even with the Cox helping us. I'm trying to remain optimistic with the data but I for one will exhale a very deep sigh when the 9902B interim is behind us and it supports the 9901 story and not 9902a.

I wish I could be my normal optimistic self, but when it comes to the data and Provenge, the story has been a mixed case.

But before everyone grabs the nearest rock and tosses it at me, yes, I acknowledge that even 9902a supported the survival stat, eventually. But please, everyone, the curves in 9902a didn't get supportive until the 29th/30th month or very late in the trial.

So, please be more circumspeck and careful about projecting the outcome. Everyone is saying this is a slam dunk. I hope so but I'm not in that camp just yet.

[rkcrules2001]

Sam – re: interim odds.

A poster called walkinizer has taken a stab at estimating interim chances. Courtesy of IV’s free search feature here are some of his posts:

http://www.investorvillage.com/smbd.asp?mb=971&mn=115125&pt=msg&mid=2154133

http://www.investorvillage.com/smbd.asp?mb=971&mn=117242&pt=msg&mid=2171711

http://www.investorvillage.com/smbd.asp?mb=971&mn=121547&pt=msg&mid=2215179

Here is an IHub post where I linked over to walkinizer’s IV post of 5/25. You will see a few informed comments in the threads from that IHub post.
http://www.investorshub.com/boards/read_msg.asp?message_id=19965691

There may well have been other related posts on IHub that I missed as I do not subscribe to the search feature.

[io_io]

Sam - I frankly think the odds are modest, certainly far less for the interim look than for the final look.


First of all, by far the most popular way of planning interim looks is the O'Brien-Fleming method (can you guess who Fleming is ? not sure, but I believe that's HIM - hey maybe this was his motive ?!? ha-ha). In that case, the p-value will be an awful lot closer to p=0.01 than to p=0.025, in fact use p=0.01 to be on the safe side.

Second, of any cancer therapy, this is not like chemo, whcih ahs generally rapid results, but it acts slow. Therefore initially the sickest patients will die in equal numbers on both arms - this not only prevents a good hazard ratio from developing, but will comprise a large portion of the 180 events.

Third, the censored numbers will be large and fairly equal (advantage given only by the corresponding survival advantage), again limiting chances of success.

At a rough guess, you would probably need 35%-40% more deaths (maybe 105 v 75 might be nice round numbers to compare with 9901 & 9902a) on the control arm to get your p-value here.


My conclusion is that it's a high hurdle (I mean that so few cancer drugs hit these interim looks anyway, and once again, this is not fast-acting chemo) and there will be easier meat between now and then.