Hi Cato!This is the post with the best modelling in it -...

Hi Cato!

This is the post with the best modelling in it - https://hotcopper.com.au/threads/overwhelming-efficacy-the-stats-required.5579134
There's some chance at the 60% readout of overwhelming efficacy but it mostly depends on the mortality rate of the control group. If lots of them died then it's easier to show efficacy.

I'll reproduce the 60% table here for those that don't like clicking:

60% Readout

			35%	40%	50%	60%	65%	70%	75%	80%	85%
		15% skeptical	0.373	0.511	0.742	0.895	0.942	0.973	0.990	0.997	0.999
		15% uninformed	0.932	0.985	0.999	0.999	0.999	0.999	0.999	0.999	0.999
		20% skeptical	0.188	0.303	0.545	0.757	0.841	0.907	0.953	0.981	0.994
		20% uninformed	0.708	0.890	0.992	0.999	0.999	0.999	0.999	0.999	0.999

In my best guess, we should be reading the row for "15% skeptical", so a control mortality of 70% or over may end the trial early.

biggest caveats:
- we don't know the prior in use (which does affect the result a lot)
- we don't know the treatment effect rate, though I assumed here it's the same as the EUA data)

Enjoy!