Cortisol concentrations and mortality from COVID-19 – Authors' reply

We are grateful to Rimesh Pal and colleagues and Kay Choy for their interest in our work¹ and their useful comments. As Choy correctly points out, pulsatility could affect cortisol levels. Gibbison and colleagues² showed concordance of adrenocorticotropic hormone and cortisol pulses, crucially with significantly less pulsatility in critical illness compared with healthy volunteers. This finding suggests that pulsatility might not have as great an effect on cortisol levels in critically ill patients as they would in healthy patients.

With regard to the comment made by Pal and colleagues on the diagnosis of critical illness-related corticosteroid insufficiency, we note that the consensus statement of Annane and colleagues³ on diagnostic criteria could not recommend the use of a short Synacthen (tetracosactide) test because the evidence was of low quality. An alternative definition of critical illness-related corticosteroid insufficiency is a cortisol concentration of less than 276 nmol/L (10 μg/dL). Of the 403 patients with COVID-19 in our cohort, we found that only 18 had a cortisol level below this cut-off (compared with 13 of 132 in the patients without COVID-19).¹ This result suggests that critical illness-related corticosteroid insufficiency is not a widespread problem in the context of COVID-19 in a non-intensive care unit setting, but further data are needed. Pal and colleagues also comment that clinical practice for septic shock involves the use of hydrocortisone and other glucocorticoids. Of specific relevance, glucocorticoids (eg, dexamethasone) might confer benefit in severe COVID-19.⁴ However, this benefit is less likely to be driven by any treatment of critical illness-related corticosteroid insufficiency per se, than by the anti-inflammatory and immunomodulatory effects of dexamethasone. Indeed, no benefit of dexamethasone was seen in patients with COVID-19 not requiring oxygen in the RECOVERY trial.⁴

Choy points out that our analysis did not examine the effect of binding proteins (eg, cortisol-binding globulin). These proteins do indeed decrease in the context of physiological stress as we have previously documented,⁵ and this probably occurs in COVID-19. The net effect of an elevation in total cortisol and a reduction in cortisol-binding globulin will be to increase free cortisol levels.⁵ This supports our main finding, which is that COVID-19 is associated with a marked elevation in cortisol. Some immunoassays do indeed exhibit a positive bias versus gold-standard assays due to cross-reaction with other steroids in samples derived from critically ill patients. We used an Abbott immunoassay and we note that Dodd and colleagues⁶ showed that this assay exhibits no significant bias in comparison with the gold-standard gas chromatography–mass spectrometry assay in the context of critical illness.⁶ Therefore, assay interference does not plausibly explain the markedly elevated cortisol levels in our study.

Both Pal and colleagues and Choy caution against the routine use of serum cortisol as a prognostic biomarker in the context of COVID-19. Cortisol is likely to co-vary with disease severity as a marker of physiological stress. Our dataset does not include all the variables necessary for construction of APACHE-II or SOFA scores to verify this hypothesis. However, we believe that the correspondents' comments support our contention that the potential use of cortisol levels as a prognostic marker in COVID-19 will require validation in a prospective study incorporating validated measures of disease severity.