This study provides evidence for real-world effectiveness of steroids in reducing either in-hospital mortality or a composite endpoint of in-hospital mortality/ICU admission amongst severe COVID-19 patients. The protective effect-size of treatment with steroids was similar to that reported in the RECOVERY clinical trial3 for a comparable group of patients defined by receipt of oxygen therapy. This adds to the evidence base for a clinical benefit of steroid treatment in COVID-19.
Our study shows longer steroid treatment is associated with lower risk of in-hospital mortality or ICU admission. We found a significant decreased hazard of both death and death/ICU admission using the continuous steroid-days in the multivariate model. Further increasing confidence in this finding, hazards when associated with longer steroids (> 10 days) were lower than with treatment captured by categories of shorter duration: <3 days and 3–10 days. Whilst the group treated with longer durations of steroids had the highest overall crude mortality, suggesting the group treated with longer duration of steroids had more severe disease than others treated with shorter courses, steroids were protective once analysis adjusted for other factors affecting severity. This finding supports additional assessment of the association between duration of steroid treatment and severe outcomes, and to our knowledge is the first study to address this question.
Originally, we set out to confirm the RECOVERY trial in a real-world setting, also finding a surprising effect of longer duration of steroids. We adjusted for potential confounders (e.g. age, sex, ethnicity, comorbidities, BMI and IMD) with the effect of steroids remaining statistically significant and a clear trend towards lower hazard ratio with longer durations. To avoid the non-consistent treatment of COVID-19 patients before and after the steroid WHO treatment guideline, only patients after the guideline were included in the study to ensure as few confounders as possible. Undoubtedly, we are unable to adjust for all confounders, including other improvements introduced around the time of steroids e.g. thromboprophylaxis and proning. However, it is reassuring that steroid-days (which also includes no steroid treatment) as a continuous variable are significant, suggesting the associations found are robust.
It is notable the study was done in a centre that had good overall comparative NHS outcomes and an SMR of 0.5 in ICU and had guidelines and practice recommending longer courses of steroids for severe patients. Over 80% of the > 10 steroid-days group were treated deliberately with long steroids and the remaining were on long term steroids as therapeutic immunomodulation for other conditions. We hypothesise from the analysis that there is a sub-population of COVID-19 patients that have prolonged severe high-inflammatory disease that benefit from longer course of steroids treatment, that reduces their mortality. This hypothesis needs formal trial analysis to resolve. Causality between longer steroid treatment and improved outcomers should not be inferred before interventional trials are conducted.
Clinical trials will also need to assess whether longer duration of steroids increase the risk/rate of adverse events, which includes delayed viral clearance19. Some studies are identifying other potential adverse events associated with steroids such as invasive mould infections including aspergillosis and mucormycosis20, with work ongoing to assess the effect of steroids on risk of bloodstream infection 21.
In this study we investigated the association of steroid duration with outcomes, however our analyses are agnostic to the dose of steroids used. There may be reasons why duration of steroid treatment mediate effects on outcomes independently of cumulative dose, for instance if a sustained period of immunosuppression is needed to prevent immune-mediated inflammation.
Many other studies on the real-world effectiveness of steroids have failed to reproduce the findings of clinical trials. Partly, this may be due to small sample size, heterogeneity of treatment and non-treatment groups, and incorrectly testing associations on individuals not expected to benefit, i.e. cases without evidence of hypoxia. Additionally our adjustment accounts for many baseline variables which have previously been associated with severe outcomes. The validity of our analyses is supported by the findings that variables previously associated with severity, such as age and cardiovascular comorbidity retain significance in our modelling.
Other studies have found the alpha variant of SARS-CoV-2 to be associated with severe disease, especially mortality22–24 and hypoxia on admission25. However, another study in hospitalised patients did not find such an association26. To our knowledge, no studies on alpha variants adjusted for newly introduced therapeutics. Interestingly, the association of alpha variant with severe disease as measured by mortality and ICU admission was not found in this study. This is in contrast to our initial findings in the same dataset that the Alpha variant was associated with severity as measured by hypoxia on admission25. It may be that severity of the alpha variant is ameliorated by efficacious treatment of hospitalised patients. This may be especially true as during the second wave steroid treatment had been introduced and standardised as the alpha variant emerged. This would also explain the disparity between findings of other published studies, with the only study of variant status and death in hospitalised patients not finding an association.
Further limitations might include potential bias for patients who did not have a chance to receive steroids or received very short steroids because they were very severe and died, or went to ICU soon after admission. This is an issue that is intractable with retrospective study, and we attempt to address this by excluding those who died (or went to ICU for the secondary outcome) in the first 24 hours after admission.