Replies to post #138553 on Biotech Values

[iwfal]

DNDN -

What I get though is an alpha of .060. The difference is more than can be accounted for by rounding. As an additional check prior to the SPA amendment the protocol called for 360 events, 90% power, overall alpha .05, and HR 1.45 (1/.6897). When I run that through the formula I get alpha .041 (note Pocock boundary, interim at 50% info fraction).

I.e. are you saying they were disingenuous when they amended the SPA? Hush now. How could that be? -g-

I publicly noted from day 1 that the power they were claiming after the change in SPA was extreme spin - I hadn't done the particular calculation you just did, but had done more sophisticated calculations (modeling) and it was clear they were not being forthright. Also note that prior to the SPA change they were clear that the HR assumed for powering calcs was 1.45 - but I don't believe they ever quoted an HR associated with the post-SPA powering.

As for the check that the pre-change numbers were realistic - I had not done that calc before. So that was interesting in an of itself. Tx

Historical note for those who weren't following DNDN at the time. They amended the SPA mid-trial to reduce total events from 360 to 304 (and changed interim point and alpha) and claimed it reduced powering from 90% to 88%. It was dead obvious at the time that they were either idiots or being disingenuous to claim only a 2% reduction in power. But it was also controversial with a variety of defenders who should have known better claiming the company must be right.

[iwfal]

Hirogen - Thanks for this dialog. I know what they did (and it is always useful to figure out how companies game their PRs):

a) The interim analysis was done with exactly 228 events - exactly at the tigger number. No patients found during clean up were added. (you can tell this from the HR uncertainty boundaries released at the interim - when you reverse engineer them it is obvious no additional patients were added.)

b) But for the final they actually used 331 (not the 304 in the PR about the trigger) - and per their own AC material
it sounds like that was actually the design number. And THAT number correctly correlates to 88% power they touted with the correct final alpha.

I.e. the trigger for the final was 304 events but they had FDA permission to use patients events found or occurring during database clean up (i.e. prior to unblinding) up to 331 and that ups the power. Why they felt it necessary to play hide-the-truth I do not know - but based on the AC material, the document you posted earlier (tx - cool document), and simple math, it is virtually certain that is what they did.