This paper is a mess in my view. https://www.thelancet.com/action/showPdf?pii=S0140-6736%2820%2931180-6 The propensity model has very low likelihood of capturing the indication bias, and the fidelity of the data (huge registry with important reliance on administrative codes). The huge numbers don't give new information,

they just give lower p values. Do we really and truly believe that nothing else guided the decision to administer experimental therapies other than SpO2<94% and an mSOFA (plus some typical comorbidities)? Experimental therapies get administered when patients are

deteriorating, which isn't controlled in this propensity analysis. To my eye, this paper says that deteriorating patients do worse. That seems unequivocal. The work of answering these questions is RCTs. So we should continue to support RCTs with careful safety monitoring plans.