Some thoughts about the RECOVERY-DEX preprint. First, some more clarity about why it stopped--it hit the target sample size. Oddly, no discussion about alpha spend or what the DSMB reviewed. Second, post hoc analyses that control for age only in their primary analysis. This is
confused and confusing, since they should have had a prespecified model that included more factors (table S2). The primary analysis, flawed as it was, looked a bit worse for non-oxygen patients (unadjusted p = 0.05), less impressive for the oxygen group (unadjusted p=0.014), but
still seemed like they hit on their primary (albeit without any knowledge of their alpha spend). _Probably_ not an issue since a superior model would _likely_ look like their post hoc model, but it's not certain. (Depending on how conservative, the p=0.014 may be non-sig)
Third, if you follow what I think is my sensible rule about sub-groups (only trust the largest one that has physiological credibility and looks similar to the overall results), that would be the "supplemental oxygen" group, which is a wide array of patient severity. There, the
effect is modest once you pare it down to brass tacks--3–3.5% absolute risk reduction, or an NNT of about 33. This seems like a more credible result than the mechanical ventilation results, which are about 11% ARR (NNT ~9).
Fourth, they don't distinguish between duration of symptoms (most of the effect appears to be >7 days of sx) and severity of illness at presentation. This seems like it will be of at least some interest as we figure out what to do with these results.
Fifth, we don't get anything substantial on concomitant medications. Their table S1 just refers to pure experimental contamination (people getting one of the study drugs despite being in the usual care control arm) but doesn't give a sense for whether other therapies differed by
treatment arm.
Sixth, I think they're probably safe from this, but they don't have any post-day-28 f/u that I can see. One concern about steroids is that they increase the risk of late-onset hospital infections, which may only express their harm after 28 days.
So, based on the preprint, what do I think? Well, the data on steroids in ARDS has been a hot mess for a long time. The signal overall seems in the realm of credible but is happening in the setting of a non-blinded trial--NNT of 33. A subtle signal that could easily disappear
in a blinded trial. And then a prespecified subgroup that has a wildly positive finding, much better than anything anyone has ever seen before with this familiar drug in similar syndromes. On general principles, a single subgroup with a wildly positive result probably has
< 20% chance of being replicated. And it comes in the setting of a signal in the opposite direction in less sick patients. This sure seems like a question to answer with placebo control. In another setting, I think most of us would agree that this is excellent prelim data
that should motivate a definitive blinded trial. It sure seems to me like internationally everyone is so hungry for good news that they are likely to bypass definitive answers and go with these results, which may make it harder to understand the effects of other drugs that may be
more effective. What would be wonderful would be to know the enrollment status for other trials of dexamethasone or similar steroids and know whether confirmatory (or non-confirmatory) results are forthcoming soon. This is much better data than we've seen for many other drugs.
It's not obviously clear to me that this is better than remdesivir, for example (where we have blinded data with mortality benefit in the main subgroup). But it is better than we have for other agents. I think it's probably reasonable if a patient on oxygen requests it to give
dexamethasone. I'm not sure that in other circumstances we would say that it is standard of care. I'm interested to hear what others think. I'm not feeling partisan about this, mostly trying to think clearly.
