Overwhelming efficacy - the stats required, page-577

@kervio, thanks for your comments on this. I am also really shocked at the primary endpoint being set at a 43% improvement in mortality. I can't see how they would ever realistically have got overwhelming efficacy (e.g. using a sceptical prior) at any interim analysis if this was the goal.

I'm no expert at Bayesian analysis but I had a go at estimating some probability distributions for achieving various levels of mortality reduction for various observed deaths in the treatment arm at the 45% analysis stage (which obviously passed futility). I inferred some mean and standard deviations for both sceptical and enthusiastic priors under various scenarios. I have plotted these below for a particular scenario where 17 deaths were observed in the treatment arm and where there were theoretical deaths of 40.5 (i.e. mortality at 60%) in the control arm. This, I think, would be the 'worst case scenario' / maximum no of deaths that could have been observed at that the 45% stage for continuation of the trial. This also assumes NIH/MSB based futility analysis on an enthusiastic prior and selected 5% as the minimum required probability of achieving the endpoint.

If 17 deaths vs. 40.5 is in fact the worst case, that's still quite a treatment benefit!