Ann: Remestemcel-L FDA Fast Track Designation for COVID-19 ARDS, page-558

Well yep Bedger … I can see where you are coming from. But try this though:

If you reverse engineer things …

Start with a assumed mortality rate of 42% in the Control group (or a survival rate of 58% - S1 in the photo).

What would represent a clinically significant improvement (absolute)? Lets say 15%. This corresponds to a hazard rate of 0.57 – or the 43% improvement SI claims was the target.

This configuration gives you 80% power at p<0.05 two sided with a sample size of 300.

This all looks good on paper …. but what happens if SOC improves during your trial and so the control group mortality rate drops?

Lets say mortality dramatically improves to only 10% and not the original 42% you assumed.

Your power now drops down to 25%.

So you only have a 25% chance of detecting your posited 15% improvement in mortality with this sample size (300).

Would anyone consider it was futile to continue? No.

What would you do then? The whole idea of using the Bayesian futility approach is to cater for exactly this circumstance – so just increase your sample size.

And therefore maybe SI didn’t mislead you at all; and it is Left-e who is unintentionally misleading you because basically he doesn't have a clue what he is talking about.