We report a multicenter observational study involving 11 French ICUs in the Outcomerea© network that assessed the efficacy and safety of early CS therapy in patients admitted to the ICU for COVID-19 pneumonia.
Early administration of CS during the first two days after ICU admission was not significantly associated with 60-day mortality. Similar results were observed in patients who received high doses of corticosteroids. Importantly, early CS administration was beneficial in older ICU patients but not in patients younger than 60. Early CS seemed to be potentially disadvantageous in younger patients without inflammation on admission. It was also associated with a significant increase in the risk of hyperglycemia and insulin requirement and did not affect significantly the incidence of HAP-VAP and/or ICU-BSI.
The benefit in patients older than 65 had already been suggested by a subgroup analysis of the MetCOVID trial . One explanation given by the authors was that older patients had more inflammation on admission. In our cohort, CRP and ferritin levels were similar in the two age groups. However, older patients had a lower lymphocyte count, a higher DDimer level and had a higher severity score (table S5 in additional file 1). Another hypothesis could be therefore that the more severe the disease, the more effective are steroids. These results need confirmation, since the Recovery trial found a protective effect of steroids for patients younger than 70 years.
Our results suggested that steroids should be avoided in younger patients in the absence of inflammation. Such results could also be related to the severity of the pneumonia. Indeed, in our cohort, patients without inflammation had less severe pneumonia symptoms (table S5 in additional file 1). Our results are consistent with those of other studies which also found that steroids were beneficial in the subgroup of patients with inflammation [15, 19, 40]. Furthermore, in the Recovery trial and another meta-analysis, the less severe patients, i.e. those without oxygen, did not benefit from CS [11, 41]. The beneficial effects of steroids could be explained by their potential role in suppressing inflammatory storms, reducing inflammatory exudation, and preventing multiple organ injuries [3, 4, 42]. However, further studies of COVID-19 ARDS are needed to better understand the direct effect of steroids on this particular immune response , which is different from that of other bacterial sepsis [44, 45]. In contrast, steroids could have a deleterious effect in the absence of inflammation because they induce immunosuppression .
One of the potential consequences of immunosuppression is delayed SARS-CoV-2 RNA clearance , which had already been observed in SARS and MERS  but not observed in SARS-COV 2 patients. In addition, the immunosuppression induced by steroids could also lead to a higher risk of superinfections [23, 45]. Several studies have already reported a high rate of ICU-acquired pneumonia in mechanically ventilated COVID-19 patients .
There are several reasons why our results are at variance with those of the Recovery trial, which support the use of corticosteroids to reduce death rates. First, our patients received high doses of steroids (20 mg of dexamethasone for 5 days and then 10 mg of dexamethasone for 5 days), which could have been more harmful than lower doses. To date, only a few studies have assessed high doses of steroids. One observational study reported that a higher dose was associated with harmful effects . Only the CoDEX trial has assessed dexamethasone at a higher dose (up to 20 mg per day), reporting a positive effect measured as a composite of days alive and free of mechanical ventilation . However, in the CoDEX trial, 28-day mortality was not different between high dose of steroids and placebo. Results from other clinical trials are pending before definitive conclusions can be drawn. Second, our patients received steroids from symptom onset, later than in the Recovery trials (10 days versus 8 days), which could have been too late to prevent or reverse the damage caused by extensive inflammation. Third, one third of the patients in the non-Early-CS group finally received steroids, which could have minimized differences between the two groups. Forth, the absence of benefit of CS in our study may be related to the significant increase in the rate of hyperglycemia and in the insulin use. Indeed the absence of glycemic control in critically ill patients is associated with a demonstrated increase risk of death. Finally, other immunomodulatory treatments could have interfered with the effects of steroids. In our cohort, some patients also received tocilizumab or Anakinra. Such treatments are under evaluation and preliminary results of interleukin-6 or interleukin-1 blockade and/or anti-TNF are varying. Most studies of tocilizumab were gathered in a meta-analysis which showed that it did not reduce short-term mortality [52, 53]. Anakinra could reduce the risk of invasive mechanical ventilation and death  but results from randomized controlled trials are still pending.
The strength of our study resides in our subgroup analyses, the 60-day endpoint and the use of weighted models that minimize the weight of patients unlikely to have received corticosteroids. We also excluded all patients previously exposed to steroids to reduce immortality time bias. Our study has several limitations. First, despite the use of propensity score analyses to draw causal inferences the study was observational, and potential unmeasured confounders may still have biased our results. Second, our study dealt with heterogeneity in the prescription of corticosteroids in terms of drugs, doses, and duration but also in that of other immunomodulatory treatments. Furthermore, a part of the patients in the non-Early-CS received finally corticosteroids. We also only considered the first episode of BSI or HAP-VAP.