Wall Diver

The Short Case on Dendreon Corporation

Thank you for your civil reply.

- "First, it's not an inaccuracy if I don't mention the p-value. They missed plain and simple. The fact that it was close matters to bulls, not to the FDA. Also, even if they hit and it was 0.04 I would question it. Think about it, with a p-value of 0.04 there is still a significant chance that this data is a fake. Coupled with the fact that 9901 was never meant to be a registration trial and it was very small and I just don't see how anyone at the FDA can have confidence that Provenge is definitely a real drug. They will simply make them run another trial, as one is ongoing already. And I wouldnt expect patient groups to be up in arms about Provenge if it is delayed. And as you mention the 9902A trial, I've never seen the TTP data for that one, was it ever disclosed? I'm thinking it will be disclosed tomorrow in the briefing docs and the DNDN bulls may be in for a shock."

Not sure what your point is...even if the p value was 0.04, that's a 96% chance that the results were not random. Such a tiny miss of 0.002 in TTP is not going to be lightly discarded by FDA biostatisticians. I don't buy the argument that the 95% cutoff point is completely arbitrary, and 94.8% means nothing. In addition, there have been a fair amount of oncology approvals on trials of this size, plus some that weren't randomized and were approved on surrogate endpoints. As for 9902A TTP, the p value was in the 0.7 range, as I recall from the ECCO 2005 presentation. Most of the bulls know this already. Obviously not close, but the story ever since 2004 has been all about survival, especially after the FDA did away with time to progression as a valid endpoint for hormone-refractory prostate cancer patients.

-"The Taxotere statistical analysis is very questionable. First, it was not an analysis that was planned on being done before the trial and therefore is akin to data mining. Second, the problem with subsets is that the baselines may be even less balanced than in the trial as a whole. We are talking about only around 80 patients among the two trials after all, so there can be a lot of statistical hand waving going on."

The docetaxel analysis was released over 18 months after the bears first argued that 9901's survival benefit was due to more patients in the Provenge arm receiving Taxotere, in addition to receiving earlier Taxotere. This analysis successfully refutes that bear argument. The common statistically significant prognostic factors for each trial in overall survival were baseline PSA and number of bone metastases, with disease localization being a stat sig prog factor in 9901 and extremely close to stat sig in 9902A. Gleason score was not a stat sig prog factor in either trial. We haven't even discussed cause-specific survival for the entire intent-to-treat group in 9901. Basically, 13 patients in the provenge arm died of causes unrelated to their prostate cancer, while only 4 patients in the placebo arm did. This improved the p value from 0.01 to 0.002.

-"What I would have really liked to see is the TTP and survival for every Gleason Score integer in the trial. Theyre bundling of so many Gleasons together I think makes the data confusing."

They did that for the three-year survival data in 9901. Obviously, since there are so few HRPC patients with Gleason scores of less than 6, you can count that cohort as Gleason less than or equal to 6.

So, here is that data:
GS less than or equal to 6 Provenge arm: 41% or 9/22 survived for 3 yrs after randomization
GS less than or equal to 6 placebo arm: 28.6% or 2/7 survived
----------------------...
GS equal to 7 Provenge arm: 32.1% or 9/28
GS equal to 7 placebo arm: 11.1% or 2/18
----------------------...
GS equal to 8 Provenge arm: 40% or 4/10
GS equal to 8 placebo arm: 12.5% or 1/8
----------------------...
GS equal to 9 or 10 Provenge arm: 27.3% or 6/22
GS equal to 9 or 10 placebo arm: 0% or 0/12
----------------------...
All GS Provenge arm: 34% or 28/82
All GS placebo arm: 11% or 5/45)
P = 0.0046

I should add that the Cox regression analysis showed that there was an imbalance favoring the placebo arm in the trial...the Provenge patients were sicker on average.