COVID STEROID 2 Trial Investigates Dexamethasone 6 mg or 12 mg in COVID-19
Jon-Emile S. Kenny MD [@heart_lung]
“If you’re afraid of butter, use cream.”
Early observational studies and the RECOVERY collaborative group initially demonstrated the benefit of corticosteroid therapy in oxygen-requiring, COVID-19-patients; additional randomized studies as well as a meta-analysis of these reports corroborated this.
Yet the dosing of corticosteroids across the above studies was inconsistent. In the largest, RECOVERY, the dose was 6 mg of dexamethasone every 24 hours for at least 10 days. In REMAP-CAP, the dose was 200 mg of IV hydrocortisone [i.e., 7.5 mg of dexamethasone] for at least 7 days while CAPE-COVID used a taper of IV hydrocortisone that also started at 200 mg per day for at least 8 days.
By contrast, 3 smaller trials administered much higher doses of corticosteroids. The CoDEX trial investigated 20 mg of dexamethasone for 5 days followed by 10 mg for at least another 5 days. DEXA-COVID 19 also gave 20 mg of dexamethasone per day and Steroids-SARI administered 80 mg of IV methylprednisolone [i.e., 15 mg of dexamethasone] per 24 hours.
The REACT meta-analysis of all of the aforementioned trials divided them into those that administered ‘high dose’ – defined as at least 15 mg of dexamethasone equivalent per 24 hours – or ‘low dose’ and found that only the latter statistically-favoured corticosteroids relative to placebo. However, the number of patients enrolled in the ‘high dose’ trials was comparatively small and may have represented a type II error. Importantly, there was no difference is steroid-related adverse events in either group.
Finally, and as mentioned by the authors, studies of additional immunomodulation by IL-6 receptor antagonists showed clinical benefit beyond that of the 6 mg of dexamethasone afforded to patients in the ‘usual care’ arm of said investigations. Therefore, giving a higher dose of corticosteroids might, similarly, offer additional benefit. Accordingly, in the COVID STEROID 2 trial, 6 mg of daily dexamethasone was compared to the median dexamethasone dose [i.e., 12 mg] administered in the non-RECOVERY corticosteroid studies included in the REACT meta-analysis.
What they did
Between August 2020 and May 2021, patients were screened in over two-dozen hospitals in India, Denmark, Sweden and Switzerland. Inclusion criteria were adult hospitalized patients requiring: 1.] supplemental oxygen flowing at least at 10 L/min or 2.] non-invasive ventilation or continuous positive airway pressure for hypoxemia or 3.] invasive mechanical ventilation.
Patients were excluded if they were treated with glucocorticoids greater than 6 mg per day of dexamethasone, or its equivalent, for indications other than COVID-19 or were treated with systemic glucocorticoids for COVID-19 for 5 days or longer, had invasive fungal disease, active tuberculosis, had known hypersensitivity to dexamethasone or were pregnant.
Patients were randomized by central allocation to either 6 mg or 12 mg of dexamethasone per day for up to 10 days post randomization. Other immunosuppressives were recommended against, until January 2021 when the REMAP-CAP trial investigators found benefit to the IL-6 receptor antagonist tocilizumab.
The primary outcome was the number of days alive without life-support – defined as invasive mechanical ventilation, circulatory support or renal replacement therapy – at 28 days post randomization.
What they found
503 patients were randomized to 12 mg per day of dexamethasone and 497 to 6 mg per day. 982 were included in the full analysis because 18 did not consent to data use. The groups were well-balanced at baseline though, by chance, there were 7% more diabetics in the 6 mg group. The two groups were equally-likely to receive IL-6 receptor antagonists.
At 28 days post randomization, those who received 12 mg of dexamethasone were alive and without life support for an additional 1.5 days [p = 0.07] in absolute terms. For each component of the primary outcome, there was no difference in days alive without renal replacement therapy, whereas for both invasive mechanical ventilation and circulatory support, there was additional day alive for those randomized to 12 mg.
Within the secondary outcomes, there were no statistically-significant differences between groups for days alive without life support at 90 days, days alive out of the hospital at 90 days, or pure mortality at 90 or 28 days. Nevertheless, there were clinically-significant trends favouring 12 mg of dexamethasone. For example, 28-day and 90-day mortality rates for 6 mg versus 12 mg were 32.3% versus 27.1% and 37.7% versus 32%, respectively.
After reading the COVID STEROID 2 trial, I was immediately reminded of the ANDROMEDA-SHOCK investigation from a few years ago. While there was no statistically-significant difference between the groups based on frequentist methods, there was a distinct trend favouring one group versus the other with the lack of statistical-significance, potentially, reflecting a type II error due to inadequate sample size.
Of interest, the authors of COVID STEROID 2 committed to a pre-planned Bayesian analysis of their results to address the shortcomings of traditional, frequentist models, especially when the null hypothesis fails to reject – as happened in COVID STEROID 2.
The traditional, frequentist, paradigm, posits a world in which the null hypothesis [i.e., no difference] is assumed to be true and then this is tested probabilistically. Based on an arbitrary threshold – usually 5% – the outcome is dichotomized into either accepting the null hypothesis [i.e., no difference] or rejecting it [i.e., there is a difference]. Crucially, as the null hypothesis is assumed to be true, this necessitates certainty about the population value being measured. For example, if you are studying 28-day mortality, you must know, and accept as true, the value of 28-day mortality to set the null; this is the world in which the frequentist approach grounds itself.
On the other hand, the Bayesian approach makes assumptions about prior probability and refines this with collected data to give posterior probability. This avenue is analogous to diagnostic tests where the clinician sets pre-test odds, multiplies this by the likelihood ratio of a diagnostic test and is left with post-test odds. Because Bayesian ontology does not dichotomize the result of the study into ‘acceptance’ or ‘rejection’ of the null – but rather gives posterior probabilities – the output is, arguably, more nuanced. As nicely argued by Bendtsen, the frequentist attitude lets the statistics make the scientific inferences for us, however, with the Bayesian stance, inferences are made after the statistical output.
In their Bayesian analysis of COVID STEROID 2, Granholm and colleagues defined a clinically-important difference for days alive without life support as at least 1 day. There was a 64% chance that this was true for those who received 12 mg of dexamethasone. Further, there was a 94% chance that there was ‘any’ benefit for being alive and without life support at 28 days for those who received 12 mg of dexamethasone. Additionally, they defined a clinically important difference in 28-day mortality as at least 2% absolute difference [i.e., a number needed to treat of, at most, 50]. There was an 81% probability that those who received 12 mg of dexamethasone met this threshold. Importantly, the probability of significant harm was no different than those who received 6 mg of dexamethasone.
It has been nicely articulated that lack of statistical significance does not preclude clinical action. Based on the traditional and Bayesian analyses, it would not be incorrect to interpret 12 mg of dexamethasone in severely-ill COVID-19 patients as an appropriate path. This might be especially true in settings where expensive immunomodulators are not available. As pointed out by the authors, the absolute, short-term mortality benefit for tocilizumab in addition to 6 mg of dexamethasone was 4% in the RECOVERY collaboration. In COVID STEROID 2, the benefit for 12 mg of dexamethasone relative to 6 mg was almost identical [i.e., 4.5%], but this was not statistically-significantly different based on a frequentist model.
A way a lone a last a loved a long the ...
Dr. Kenny is the cofounder and Chief Medical Officer of Flosonics Medical; he is also the creator and author of a free hemodynamic curriculum at heart-lung.org. Download his free textbook here.