Key takeaways
The study isn’t perfect. In their methods section they use terms like “ensemble machine learner (super learner)” and “immortal time bias.” The first is a fancy way of saying they did a form of propensity scoring, which in turn is a fancy way of saying they tried to control for confounding. The second is a way to adjust for time delays between drug administration. Both are attempts to compensate for the observational design, as is their argument for biologic plausibility. Here they’re on particularly thin ice when trying to prove causal inference. Biologic plausibility is never hard to find; after all, what compound doesn’t have pleiotropic effects? Furthermore, the analysis lacks any data to support their biologic plausibility hypothesis that fludrocortisone’s effect on mortality is mediated via activation of innate immunity and/or clearance of alveolar fluid.
The editorial accompanying this Big Data study endorsed adding fludrocortisone. We have very little that reduces ICU mortality so the low number needed to treat is enticing, especially in light of the low risk from adverse events, so I’m going to start using it. Do I think I’ll save one life for every 28 patients with septic shock to whom I give hydro/fludro instead of hydrocortisone alone? I sure don’t. No way an oral mineralocorticoid at that dose has that type of impact on top of hydrocortisone alone. I still believe that the Annane studies are positive because of the mortality rate in the population enrolled and not because fludrocortisone was added. It all comes full circle, though – 20 years after I abandoned hydro/fludro, I’m going back to it.
Aaron B. Holley, MD, is a professor of medicine at Uniformed Services University in Bethesda, Md., and a pulmonary/critical care and sleep medicine physician at MedStar Washington Hospital Center in Washington, D.C.
A version of this article first appeared on Medscape.com.