Background: Systemic corticosteroids are now recommended in many treatment guidelines, though supporting evidence is limited to one randomised controlled clinical trial (RECOVERY). 

Objective: To identify whether corticosteroids were beneficial to COVID-19 patients. 

Methods: 1514 severe and 249 critical hospitalized COVID-19 patients from two medical centers in Wuhan, China. Multivariable Cox models, Cox model with time-varying exposure and propensity score analysis (inverse-probability-of-treatment-weighting (IPTW) and propensity score matching (PSM)) to estimate the association of corticosteroid use with risk of in-hospital mortality in severe and critical cases.

Results: Corticosteroids were administered in 531 (35.1%) severe and 159 (63.9%) critical patients. Compared to non-corticosteroid group, systemic corticosteroid use was not associated with beneficial effect in reducing in-hospital mortality in both severe cases (HR=1.77, 95% CI: 1.08-2.89, p=0.023), and critical cases (HR=2.07, 95% CI: 1.08-3.98, p=0.028). Findings were similar in time-varying Cox analysis. For severe COVID-19 patients at admission, corticosteroid use was not associated with improved or harmful outcome in either PSM or IPTW analysis. For critical COVID-19 patients at admission, results were consistent with multivariable Cox model analysis.

Conclusion: Corticosteroid use was not associated with beneficial effect in reducing in-hospital mortality for severe or critical cases in Wuhan. Absence of the beneficial effect in our study in contrast to that was observed in the RECOVERY clinical trial may be due to biases in observational data, in particular prescription by indication bias, differences in clinical characteristics of patients, choice of corticosteroid used, timing of initiation of treatment and duration of treatment.  
 

Patients and methods

Study Population

Consecutive inpatients with laboratory confirmed or clinically diagnosed COVID-19 from Wuhan Hankou Hospital and No. Six Hospital of Wuhan between December 26th, 2019 and March 15th 2020 were collected in this study. The final follow up date was March 19th, 2020. Patients who met any of the following conditions were exclude from the study: 1. Non-severe or non-critical cases; 2. Not being diagnosed as severe cases within 24 hours since admission; 3. The time of being diagnosed of severe/critical cases were missing. Severe cases were defined as those who required oxygen therapy during hospital stay. Critical cases were defined based on the 7th trial version of Diagnosis and Treatment Scheme for Pneumonitis caused by COVID-19 Infection (16). Therefore, those who met any of the following conditions during the whole hospital stay were categorised as critical cases: 1. requiring mechanical ventilation; 2. requiring treatment in intensive care unit (ICU); 3. shock occurred in hospital (16).

Ethical considerations

This study was approved by the ethics committees of Wuhan Hankou Hospital, No. Six Hospital of Wuhan and the First Affiliated Hospital of Sun Yat-sen University, and the informed consent was waived. The study followed the tenets of the Declaration of Helsinki and is reported as per the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guideline.

Data Collection and Definition

We collected the patients’ clinical data, including demographic information, medical history, laboratory indexes, corticosteroid use and prognosis. The admission laboratory indexes were defined as the first records of laboratory indexes since admission. Corticosteroid use was defined as the use of intravenous systemic corticosteroids, including hydrocortisone, methylprednisolone, and dexamethasone. The dose of corticosteroids was converted to methylprednisolone-equivalent doses (1mg methylprednisolone = 0.1875mg dexamethasone = 5mg hydrocortisone) (20). At the time of the study limited guidance existed on when to use corticosteroids and therefore most decision were based on intuitiveness of the clinician following assessment of the clinical state and after excluding any contraindication for corticosteroid use.

Exposure and Outcome

We evaluated whether using corticosteroids could affect the in-hospital mortality of severe/critical cases. The time to death was defined as the time of being diagnosed of severe/critical cases to the date of death from any cause in hospital for the primary analysis.

Statistical Analysis

Data were expressed as mean and standard deviation (SD) if normally distributed, and median and interquartile range (IQR) with non-normal distribution, and frequency and percentages for categorical variables. We compared baseline characteristics and outcomes of patients who received corticosteroids and those who did not receive any corticosteroid using Students’ t test, Mann-Whitney test for continuous variables, and chi-square test or Fisher exact test for categorical variables. Survival with or without corticosteroid use was analysed using the Kaplan-Meier method and compared by log-rank test. The 28-day in-hospital mortality and its 95% confidence interval was reported for both corticosteroid use and non-corticosteroid use groups. We evaluated the associations between corticosteroid use and prognosis using two approaches. First, multivariable Cox regression models were applied to evaluate the association between corticosteroid use and mortality. Multivariable analysis was adjusted for age, gender, admission laboratory indexes (including lymphocyte, neutrophil granulocyte, platelet, haemoglobin, glucose, C-reaction protein (CRP), lactate dehydrogenase (LDH), creatinine, aspartate aminotransferase (AST), alanine aminotransferase (ALT), blood urea nitrogen (BUN), pulse oxygen saturation (SpO2), hypertension, diabetes, cancer, chronic obstructive pulmonary disease (COPD), chronic kidney disease (CKD) and smoking history. The proportion of missing values of baseline variables was 4.5%, ranging from 0% to 10.6%. The missing data were imputed by multiple imputation for ten times in the multivariable analysis. Second, in Cox regression models, corticosteroid use was accounted as a time-varying exposure to mitigate immortal time bias. In time-varying analysis, data was reconstructed according to the time of corticosteroid use. Therefore, time from a diagnosis of severe/critical case to time of corticosteroid use was categorized as unexposed. The hazards ratios (HRs) and 95% confidence intervals (CIs) of Cox regression models were reported.

We also performed propensity score driven analysis (inverse-probability-of-treatment-weighting (IPTW) and propensity score matching (PSM)) to account for confounding by indication bias in the time-dependant Cox regression analysis (21,22). Multivariable logistic regression was carried out with corticosteroid use as a binary outcome to obtain the predicted probability of corticosteroid use, which was taken as propensity score (PS). All baseline characteristics were used to construct PS for each three parts of patients. In IPTW, for the group with corticosteroids therapy, the weight was equal to 1/PS, while for the group without corticosteroids therapy, the weight was equal to 1/(1-PS). In PSM, calliper value was set as 20% of the standard deviation of propensity scores. We reported the effect sizes as hazard ratios in all four types of analysis 1) multivariable Cox regression; 2) Cox regression with time-varying exposure; 3) IPTW based analysis; and 4) PSM based analysis. Wilcoxon signed-rank test was used for pairwise comparison of mean blood glucose and lymphocytes before, during and after the use of corticosteroids, and method of Bonferroni was used for correction. We used Stata/MP 14.0 to conduct all data analyses, and statistical significance was set as p<0.05 bilaterally.