A one-two punch could still knock us out. Could deaths in control be under, say 30, which would definitely mean no significant benefit?
The board has talked continuously about the death rate in the "control group," but we should have been more careful to say the "available-for-control population." Consider as follows:
We've discussed population severity comparability, changes in SOC, stress of high load periods and the need to consider the results from the specific hospitals in the trial. I stand by my comment that 28% to 35% (OK, 36%) reflects the sense of this board on that number. Of course, the board could be wrong.
But what if we knew the perfectly-case-comparable death rate at our trial hospitals was at the low end at 28%? Unfortunately, that's not the relevant comparison. We might have also randomly drawn a really healthy 131 patients from the pool of potential controls at those hospitals - there is a 20% chance you would draw 30 or less (23%) deaths. That's the one-two punch we can't handle.
Of course it's equally likely such randomness would work in our favor. I don't think I've been fully recognizing the size of that risk, though. However, it in no way changes prior "significance boundary" calcs, as they refer to observed results.