Based on the company's decision (with Teva's blessing) to announce the interim outcome (continue or stop for efficacy), and the way the CEO talked about it, I have to assume the interim boundary is not crazy stringent (e.g. like it was in the ONTY START trial). If it were crazy stringent, then I doubt he would be discussing it as if stopping were a real if unlikely possibility.
So that probably means its OBF (which would be around .024 two-sided at 80% of events), or I suppose it could be something more stringent like Haybittle-Peto, which would be .001 two-sided regardless of the event count percentage.
Given that this is an attempt at approval on a single registration trial, a more stringent boundary makes more sense, but I don't think it can be something borderline impossible like .0001
In my 24/31 model, at 400ish events, the p-value is .0015 two-sided, plenty good for OBF, not quite there for Haybittle-Peto.
“but I don't think it can be something borderline impossible like .0001”
Not sure why you say this. I happen to think it a darn good chance the p value hurdle is crazy low (or crazy high, depending on perspective. hmmm tough hurdles are high but tough p values are low, anyway....). An HR of about 0.70 observed from 400 events would get your p value down in the neighborhood of 0.0003. Now, a 30% reduction in the risk of death is not too crazy to hope for, is it? In contrast, a more standard interim hurdle (not for a registrational trial operating under a SPA in a still unapproved drug seeking approval from a single trial) of below about 0.02 would be hit w/ an HR in the neighborhood of 0.80. But an HR of 0.80 is not going to stop this trial early. The required p value is very low (IMO) but 0.0003 (or 0.0001) is not borderline impossible (also IMO).
“Who loves false precision??? Put your hand up!!”
LOL, reminds me of the Tom Fleming review where he ripped the sponsor apart and then teased them by noting that they needlessly specified the non-inferiority margin with 6 sig figs.
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.