Custirsen is not an immunotherapy. It is anti-apoptetic. It reduces an overexpressed element that is reputed to prevent cancer cell death. It does not involve the immune system, at least not directly.
You will note that the survival curves in the Phase II trial separated early, not late. This could be evidence of a baseline imbalance in favor of the tx arm, but it the multivariate analysis argues against this, as I mentioned.
I think summer was a bit harsh in her comment (although she's an incredibly valuable poster here, IMO). I think your effort to provide your own precise evaluation is useful here.
For me, the big outstanding question that we can't address in the "confounding" dilemma is whether the new drugs are likely to even the playing field between custirsen and control arms (survival curves come back together) or go further apart because reduction of clusterin improves the performance of the new drugs.
If the former, your 29/25 projection may well be close. If not, we could be more like low-30's vs low-20's.
I'm hoping for synergy :-), and the latter outcome.
That's my background too, Dr. Kuvasz.
I was a Wittgenstein fan back when most of my brain still worked.
He was right, you know. If lions actually could speak, we would not be able to understand them.
This is a thoughtful assessment. I do think it is worth noting that the multivariate analysis "disagrees" with your contention that the arms in the Phase 2 were unbalanced, against the control arm. The multivariate HR of .49 vs the unadjusted HR of .61 would argue that there was an imbalance in the opposite direction.
Not trying to start an argument -- just noting that.
I guess I'm not super worried about limited posters. The small cost of membership is a good seriousness filter.
I'm happy to go over to InvestorVillage.... nobody is over there using the OGXI board these days...
(I bought a lifetime membership there. It has cost me mega multiples of the $200 I paid in investment losses.)
Is ~.02 at a 75% of events interim Considerably More Stringent than ~.05 at 100%.
For this trial (my model -- near linear curves), the difference between those two measures is about 1.5 months (call it 5.5 month MST spread at the former, 4 month MST spread at the latter -- HR of ~.79 vs HR of ~.84).
Is that Considerably More Stringent (CMS)? I dunno. Doesn't feel like it.....
At the Avii interim level (~.0003), we're talking a not quite 9 month MST spread and an HR of ~.695. That is CLEARLY CMS.
One thing I've learned NOT to do is take a CEO's adjectives at face value. We can't have any idea how to translate those three words into thresholds.
Avii, if they had a p-value higher than the threshold, whatever it was, what other data would they have needed?
I think we have to keep in mind here that the "confounding" effect of these treatments is going to occur, in this trial, in a post-chemo context. It may be the case that they do even more for OS pre-chemo, but none of the pts in SYNERGY will have gotten either xtandi or zytiga pre-chemo.
So I think a median of 24 or 25 is in play for the control arm in SYNERGY, but anything above that would a really big surprise. VENICE and READY are good proxies for little xtandi or zytiga add-on, and the pre-chemo results for their usage showed a 4 to 5 month add-on.
Or am I missing something?
I have carefully considered both Summer's and Avii's opinion on this issue (p-value required for single registrational trial to yield approval).
I consider Summer's opinion to be superior, because it increases my chances of winning.
Thanks to both of you for your insights.
Captain, USS Confirmation of Bias
Every time you post, the stock goes up a dime. If you take off work through Turkey Day, you should be able to get the stock back up to solid double digits.
Luckily I ran a model here, and the p-value came in less than .001, so we're in good shape.
There's no way a spreadsheet can be wrong. Microsoft wrote the freaking software, you know.
Hey Avii, I totally agree with this except for the following problem -- if this methodology is always used, including when OS is the primary endpoint, then it leads to curtailment of incremental gains in a particular indication.
As an "unmet need" gets met, the difficulty in producing low p-values on the margin gets increasingly difficult, as expected placebo OS time goes up, and the reasonably expected benefit of the incremental treatment goes down.
3 months improvement in an indication where 30 months is the median OS seems worthwhile, but that trial would produce an HR of around .9, presumably. You'd have to run a HUGE trial to get a p-value under .001 with an HR of .9.
I wonder if the FDA thinks about these things, or they're simply as dumb as they look from the outside.
It's really equivalent, in this case, to saying that the probabilities of (a) the drug working, or (b) the control group having outsized survival, are roughly equal.
I think Summer's contention is generally true, but there are exceptions (as always). I would not classify this situation as an exception, because of the "confounding" issues with the other recent treatments available. The trial going long could easily be explained by much better general overall survival in this indication due to Xtandi and Zytiga being out there.
Summer, I think he wanted to know what the p-value would be at that many events with a four month difference (28 vs 24), which is different from asking a powering question. The p-value is tied to HR, not medians, of course, and curve shapes matter -- in that model of mine I told you about, a reasonably shaped 28 vs 24 gives me an HR right around .85 and a two-side p-value of just over .05. So my answer would be "not quite".
Your powering calculation is consistent with that, of course. It's saying a trial with 500 events is a coin-flip to show stat sig, so a deterministically modeled trial with those medians should show a p-value right around .05.