Thinking about PROSEVA ( nejm.org/doi/full/10.10… ) with the wise and thoughtful @andrewmhersh got me wanting to be a tiny bit clearer about mortality as an endpoint. I'm actually opposite many in this regard--I think mortality is the most important endpoint and we should
continue to find treatments that improve survival. In my experience (and my review of literature and first principles) the talk about non-mortality endpoints is really conditional in people's minds on _having survived_. So at that level, we should work to improve survival AND
improve outcomes among survivors. Here I want to point out a problem with the notion that mortality is a hard outcome so blinding clinicians (as opposed to assessors) is not crucial. The problems comes with the frequency of mortality in both arms and the exquisite dependence of
mortality on pessimism among clinicians and families (who depend almost exclusively for their sense about "how things are going" on clinicians, even when they are routinely more optimistic than clinicians). The data I recall suggest that for decades about 90% of death in ARDS
comes after a "negotiation" or decision to transition to comfort care. These decisions, while they are correlated with physiology, are also driven by non-physiological considerations, as dozens of papers have suggested over the years. We also know from various sources that
clinicians are going to push harder and longer when they have optimism associated with their treatment (or in a murkier mode, with external metrics like survival after transplant). The same is true of patients. If we're actively doing something we believe in, we have reasons to
continue the current course. So, take PROSEVA. These are ICUs in a traditionally fairly hierarchical culture, and the study ICUs are experienced with an believe in proning. You randomize half of patients not to receive the treatment that the ICU is committed to.
And their mortality is double that of the patients randomized to the treatment that the ICU is committed to. Double. Which is an extremely rare effect size in physiological critical care. Which makes me wonder whether, to some degree, PROSEVA was randomizing people to
clinician optimism vs. pessimism. And there, given the strong correlation between that and stopping treatment in ARDS (the more dominant role of clinicians in France vs. USA may have played a minor role too but we shouldn't be distracted by this) may be the secret to the
doubling of mortality among patients left supine in ICUs committed to proning. Note that this involves no malice or intentionality. I don't for a moment believe that the local investigators were intentionally sabotaging supine positioning. I think they are human and were
intrinsically optimistic about proning so they were more willing to stick with the difficult course of ARDS with patients where they felt more optimistic. Nothing nefarious at all. But still a profound methodological limitation. Hence my skepticism about routine proning even with
PF < 150. Hence my interest in a US trial that includes ICUs who don't prone before the trial (but who nevertheless get trained up). And my interest in a post hoc exploration of PROSEVA (assuming there was variation) that looked at correlation between proning effect estimate and
pre-trial proning rates (or some other metric of how much optimism ICUs and clinicians associated with proning.)
