I typed up a long reply on the phone and lost it before sending so here goes again...
"Among the secondary efficacy endpoints evaluated in patients treated with High Flow Nasal Cannula (HFNC) therapy and with Mechanical Ventilation, ZYESAMI showed an advantage in 15 of 16 comparisons and demonstrated a 40% or better advantage (hazard ratio <0.7). The likelihood of this magnitude of advantage being seen by chance alone is about 1 in 2,000 (P=.0005). This difference includes at least a five-day median reduction in hospital stay. (P=.043). The largest difference observed was among those treated with HFNC who experienced a median of 11 fewer days in hospital (15 vs. 26)."
It seems we are both technically wrong. I interpreted "This difference includes at least a five-day median reduction in hospital stay. (P=.043)" too broadly, and I think you interpreted it too narrowly. I think my mistake is larger, though. I'm wrong in that it does not include all patients, but it is also not just HFNC patients. It also includes mechanically ventilated patients. Based on previous date/trials the percentage of the trial population meeting this criteria will likely fall into the 25 to 50% range. Edit: I actually see this sentence in the study summary "Patients with Critical COVID-19 and respiratory failure, currently treated with high flow nasal oxygen, non-invasive ventilation or mechanical ventilation will be treated with ZYESAMI..." which seems to imply that it will be a high percentage then that, but that is still just speculation. It makes no mention of patient on lower levels of oxygen support like standard nasal cannula/ventimask/non-rebreather mask.
You are, of course, correct that slicing up the data into more and more subsets increases the chances that a random/chance occurrence will be labeled as statistically significant(1 in 20 by definition if using the standard p value of 0.05), but that is more relevant for subsets that do not correlate. We do not know the subsets yet, and we do not know how many of them trend towards significance suggesting correlation. Only full data release will help us there. They do tease 15 out of 16 comparisons showing favorably, but we don't know what they are. They also claim a 40% or better advantage in these comparisons. Depending on the subsets, that could definitely imply correlation. Only time will tell.
I want to reiterate, again, that these result are in addition to current standard of care which is presumably steroids and Remdesivir. That is not a trivial point. Additional therapies to current multi-treatment standard of care are at a disadvantage in terms of raw absolute risk reductions because part of the total risk reduction is already being obtained from previous standard of care. Companies power large studies in the thousand to tens of thousands of patients just to eke out 1% or less absolute risk reductions and are approved by the FDA.
Lastly, I'm not trying to give the impression that I am married to this drug. Actual garbage trial results would flip me instantly, but I remain excited about it's prospects with what is currently known from the press releases. A cheap, benign(so far as know yet), and potentially broadly useful treatment for all ARDS...not just covid? I'll continue to pay attention, and I'll continue to hold a small stock position.
