Replies to post #123892 on Biotech Values

[iwfal]

Wrong. I would say the HR is not much worse than 0.9 at the interim already, at 20-25% conditional power.

I will not have the tools to settle this for another week or so, and of course I have not run this particular case before, but I'll make you a wager that assuming we agree on how futility is typically adjudged (see me previous post or below assumption) and we agree on interim alpha assignment then the HR at the interim would have to be HR>=0.95 in order to trigger Futility. It will be an interesting test case.

Assumptions:

1) Futility is triggered when conditional power (aka chance of being stat sig given the HR and number of events seen at the interim point) on the final is less than 25% assuming the same HR used initially to size the trial.

2) The trial used O'Brien-Flemming as the alpha assignment. (If it had, for example, assigned a high alpha to the interim, and a low alpha to the final then even for very good interim HRs the final would be unlikely to be stat sig)

Martin says 5% chance for second and final, he essentially is claiming the true hazard ratio is 1 or at the null hypothesis, when the overall power will equal to the significance level remaining at the second and final analysis.

I agree that Martin is trying to say that it is a null treatment. But of regardless his intent, he is actually saying that it is NOT a null treatment - 5% chance of success does not happen with alpha=0.05 and a null treatment.