Our search yielded 788 unique studies (Figure 1). During the conceptualization of this manuscript, the preliminary results of the RECOVERY trial in the United Kingdom were published, which led to amendments in CDC guidelines and to consideration of amendments by WHO on corticosteroid therapy. Therefore, this trial was judged to be so relevant that it was included after the original database search. After qualification of title and abstract, 30 studies were selected for full review. After exclusion based on low NOS score or the inability to extract risk estimates, 22 studies, studying 9,760 subjects, were included (Table 1, 2).
Study characteristics (Table 1)
Seventeen of the 22 studies originated from China. The inclusion period of patients ranged from late December 2019 until June 2020.
Most studies (20/22) were retrospective observational studies, except for Fadel et al., who conducted a quasi-experimental study with a historical control group and Horby et al., who performed a large randomized controlled trial (RCT). The study populations varied from hospitalized patients (19/22) to patients admitted to the Intensive Care Unit (ICU) (2/22), and one study included discharged patients for viral clearance assessment. The median age of patients ranged from 34 to 72 years. The severity of illness was generally assessed through the guideline of the National Health Commission of the People's Republic of China and ranged from mild to critically ill.[34,35] The median NOS score of the included studies was 6 (5-8) points. The risk of bias table is depicted in Figure 2. All observational studies were prone for selection bias (e.g. indication bias) and in three studies follow up time was too short to allow all patients to have reached their endpoint with 6.5% to 75.6% of patients remaining hospitalized.[33,37,42]
Corticosteroid therapy (see also Table 2)
In the 22 included studies, corticosteroid regimen was very diverse with respect to type, dose, duration, and median time of initiation of corticosteroid therapy after admission. In all studies, except two,[23,33] more severely ill patients were more likely to receive corticosteroids.
The indication for corticosteroids was reported in 7/22 studies and varied: most frequently it comprised (a combination of) worsening oxygenation, severe radiological findings, ICU admission, or persistence of inflammatory symptoms.[33,38-43] The median time of initiation after admission was reported in 8/22 studies and varied from 24 hours after hospital admission, immediately after ICU admission  to approximately eight to fourteen days after onset of symptoms.[23,41,42,45-47] One study compared early (average 2 days after hospital admission) to late corticosteroid administration (average 5 days after hospital admission).
Methylprednisolone was the most frequently used corticosteroid (13/22 studies).[33,37-40,42-44,46,48-51] In two studies methylprednisolone, hydrocortisone, or dexamethasone was used.[36,41] Ling et al. used prednisolone or dexamethasone, and Horby et al. reported on dexamethasone use. The type of steroid was not reported in the remaining five studies.[45,47,53-55]. In most studies a low dose methylprednisolone was administered, ranging from 0.5 to 2 mg/kg body weight per 24 hours.[33,38,40,42,43,51] Horby et al. administered 6 mg dexamethasone per 24 hours (equivalent of 30 mg methylprednisolone per 24 hours). Only Liu et al. reported a single high dose up to 500 mg methylprednisolone. The duration of corticosteroid therapy varied from 3 to 10 days except for Liu et al., who administered a single pulse dose.
Mortality (see also Table 2)
Fourteen studies reported mortality, including in-hospital and 28-day mortality. In four of these, mortality was only described in the overall study population and not specified for steroid or non-steroid study groups.[41,51,54,55] Two studies reported that corticosteroid treatment did not significantly affect mortality, without the possibility to quantify the effect.[42,43]
In the remaining eight studies, the risk of corticosteroid use was quantified, and these were included to calculate the pooled estimate.[23,33,36,37,40,46,48,50] The overall pooled estimate (observational studies and the RCT) showed a favorable outcome in the corticosteroid group (relative risks RR 0.55, 95% CI 0.27-0.83, random effects) (Figure 2).
In four of these eight studies, a beneficial effect on mortality was demonstrated.[23,33,37,46] Horby et al. demonstrated in a RCT of 6,425 patients that dexamethasone reduced mortality among patients receiving oxygen or invasive mechanical ventilation at allocation of treatment: dexamethasone reduced mortality by one-third in patients who received invasive mechanical ventilation (RR 0.64, 95% CI 0.51 to 0.81) and by one-fifth in patients receiving oxygen (RR 0.82, 95% CI 0.72 to 0.94). No positive effect was found in patients who did not receive respiratory support. Wu et al. found in a study of 201 patients with ARDS that administration of methylprednisolone (dose and duration not reported) reduced the risk of death (HR 0.38, 95%CI 0.20-0.72). Patients developing ARDS were more likely to receive methylprednisolone compared to patients who did not develop ARDS (59.5% vs. 10.3%). Fadel et al. studied 213 patients with moderate to severe COVID-19 and found that early corticosteroid treatment, consisting of three days of systemic methylprednisolone 0.5-1 mg/kg/day, reduced mortality from 26.3% to 13.6% when compared to standard of care in moderate to severe patients (p=0.024). Fernandez-Cruz et al. reported lower mortality in patients with COVID-19 pneumonia complicated with ARDS and/or a hyperinflammatory syndrome treated with corticosteroids (HR 0.36, 95% CI 0.14 to 0.93, p=0.04).
In the other four studies, no significant difference in mortality rates between steroid and non-steroid groups,[36,40,48] or even an increased risk in the group that received corticosteroids was demonstrated. However, Lu et al. did note that every increase of 10 mg in hydrocortisone equivalent dosage was associated with a 4% elevation of the mortality risk (HR 1.04, 95% CI 1.01-1.07).
Nine of 22 studies reported the effect of corticosteroids on viral RNA clearance. In 5/22 studies, time to viral clearance was not influenced by methylprednisolone treatment: Xu et al. (adjusted OR 1.38, 95% CI 0.52-3.65, p=0.519), Fang et al. (18.8 ± 5.3 vs. 18.3 ± 4.2 days, p=0.84), Shi et al. (adjusted HR 1.00 (0.53-1.89), p=0.990), Liu et al. (10.0 ± 5.3 days vs. 10.0 ± 7.9 days, p>0.05), and Zha et al. (15 (14-16) days vs. 14 (11-17) days, p=0.87).[39,44,47,49,51]
Four studies found that corticosteroids delayed viral clearance, i.e. Ling et al. (15 days vs. 8.0 days in pharyngeal swab, p= 0.013), Gong et al. (29.11 ± 6.61 days vs. 24.44 ± 5.21 days, p=0.03), Chen et al. (adjusted HR 0.60 (0.39-0.94), p=0.024), and Shen et al. (18 days vs. 13 days, p=0.003), although Shen et al. only observed delay in blood samples, not in oropharyngeal swabs.[38,50,52,53] Three did not adjust for confounders.[38,50,52]
The additional information possibly affecting viral clearance, i.e. timing, dose, and duration of corticosteroid therapy, and study population, was not consistently reported nor did show a consistent trend (Table 1 and 2).
Secondary outcomes (see also table 2)
The most frequently reported secondary outcomes were oxygenation (n=7 studies), use of mechanical ventilation (n=6 studies), and length of hospital stay (n=7 studies).
Concerning oxygenation, large heterogeneity between study results was observed. Oxygenation was described as SpO2, SaO2, or SpO2/FiO2 ratio, and different associations of corticosteroid use and oxygenation were reported, without providing exact numbers.[39-41,43] Lu et al. and Fernandez-Cruz et al. reported significant favorable effects of corticosteroid therapy on oxygenation.[36,46]
In four of six studies, positive effects of corticosteroids on the need for mechanical ventilation were reported. Horby et al. described that need for invasive mechanical ventilation was reduced from 7.8% to 5.7% (RR 0.77, 95% CI 0.62-0.95) by dexamethasone treatment. Wang et al. described that patients who had received corticosteroids less often required mechanical ventilation (11.5% of patients with methylprednisolone vs. 35% of patients without methylprednisolone, p=0.05). Similarly, Fadel et al. described that the development of respiratory failure requiring mechanical ventilation decreased from 36.6% in the standard care group to 21.7% in the early corticosteroid group (p=0.025). Chroboczek et al. reported a risk reduction of 47.1% (95% CI -71.8 to -22.5) of intubation after unspecified corticosteroid therapy in a cohort of 70 patients.
Regarding hospital length of stay (LOS), various effects of corticosteroids were reported. Fadel et al. found a significant decrease in the early corticosteroids group compared to standard of care (5 (3-7) vs. 8 (5-14) days, p<0.001). Horby et al. described that the chance of discharge from hospital within 28 days was higher in the dexamethasone group, RR 1.10, 96% CI 1.03-1.17. In contrast, Feng et al. reported that patients who received corticosteroids generally had a longer hospital stay. However, severe and critically ill patients were treated with corticosteroids significantly more frequently (moderate 13.4% vs. severe 51.9% vs. critical 74.3%, p<0.001). Zha et al. found no significant difference in hospital LOS between groups (p=0.14).