what could go wrong?, page-81

Yes impressive work Viperx1/Tewks

Without having done anywhere near enough homework on this trial (in other words may make a big blooper) a few random thoughts to add to the mix.

If I understand the mystery it is why is the trial taking longer to complete than was first planned,

And a whole raft of possibilities come to mind – some we might say are associated with a better prognosis for outcome .. but others maybe the opposite.

I suppose the negative is that you don’t really want to see significant disjointedness occurring in the up scaling from P2 to P3 in terms of assumptions not being right. As others have pointed out effect sizes just with gravity seem to decrease and companies being wrong footed just doesn’t bode well (usually).

The decision to recruit an additional 25 subjects from blinded data to improve statistical power implies some deterioration in variability estimates on PFS (assuming that was what the study was powered on) somewhere along the line. There is a remote possibility that power deteriorated because there was a change in the statistical analysis techniques to be used (ie a less powerful method) but this seems pretty unlikely. But a change to the HR profiles from interim data to what was assumed could lead to a change in what you thought you might be doing.

In terms of the additional 25 you really would like to know whether this was planned that they would have a look and decide or whether this just came out of the blue. Whether it was performed in the context of an interim analysis or not. Presumably not.

I suppose your nose would also head you to methodological issues either with say a very different patient population (which seems unlikely here to the extent it is causing the mystery) or around the timing, method, or criteria used in PFS calculation. PFS has lots of advantages over OS but is more wobbly and highly dependent of the timing of assessments which I guess could be more likely. It is a subjective endpoint that is most often looked at just a proxy to the “real deal’ OS. And because the criteria for PFS are often complex there have been quite few instances of the FDA picking up companies making errors here.

The timing of assessments particularly interacts with PFS estimates in complex ways that make it difficult to compare results between studies even within the same cancer group let alone between them. Those criteria keep changing and evolving and studies take so long to do. So what a bio-statistician works out on a nice sheet of paper on the basis of previous trials goes skew-wiff when someone in the real world makes what they think is “minor” adjustment to the measurement of PFS (which they don’t believe can alter significantly no/yes equation) but which can dramatically affect the timing of events. So the assumption of a “stochastic time to progression” I’m not sure about other than its certain to be wrong (exactly) ... but it might not matter much?

The other potential place for the mystery is in the up scaling between P2 and P3 typically the number of sites involved (and countries) increases and this can significantly increase the variability in your earlier estimates. Particularly if the intra class correlation between sites from your original P2 studies was very low – say because they were all in the same country.

All this probably just adds to at the complexity here ... and provides very little insight into the question of what does it mean for the results. I suspect that without the protocols for this study (and the earlier ones) there is only so far members of “uniformed market” like us can go. Compounding all this is the problem that many design aspects depend on the regulatory pathway and just what was and wasn’t agreed with the FDA.

This lack of transparency (does anyone actually know what the study was powered on in the first place?) is a massive problem (as QRX has just shown) and is a deterrent to investment because risk is unanalysable. But it still remains good fun .. and its the first time I think have seen bootstrapped estimates for anything on HC.

Regards Southoz

Ps Just as aside ... the N on the ACL website show 415 but the clinical trials registration still 390?