Corticosteroid therapy for corona virus disease 2019-related acute respiratory distress syndrome: a cohort study with propensity score analysis

The impact of corticosteroid therapy on outcomes of patients with Coronavirus disease-2019 (COVID-19) is highly controversial. We aimed to compare the risk of death between COVID-19-related ARDS patients with corticosteroid treatment and those without. In this single-centre retrospective observational study, patients with ARDS caused by COVID-19 between 24 December 2019 and 24 February 2020 were enrolled. The primary outcome was 60-day in-hospital death. The exposure was prescribed systemic corticosteroids or not. Time-dependent Cox regression models were used to calculate hazard ratios (HRs) and 95% condence intervals (CIs) for 60-day in-hospital mortality. Estimated survival probability of multivariable Cox regression model with time-dependent corticosteroid treatment. Cox regression model with corticosteroid treatment was time-varying variable, adjusting for age, sex, SOFA score at admission, propensity score of corticosteroid treatment, and comorbidities : diabetes, hypertension, chronic pulmonary disease, chronic renal or liver disease, solid malignant tumor, hematologic malignancy, and immunosuppressive status. Propensity score was calculated by a non-parsimonious logistic regression model that included: age; sex; SOFA score at admission; temperature, respiratory rate, SpO2/FiO2 ratio, blood lymphocyte count, blood neutrophil count, and level of c-reactive protein at admission. ARDS, acute respiratory distress syndrome; SOFA, sequential Organ Failure Assessment.

Currently, there is no speci c treatment or vaccine for coronavirus disease-2019 (COVID-19), a clinical syndrome caused by SARS-CoV-2 infection. Corticosteroids has been widely used by clinicians among patients with COVID-19, especially those with critical illness, [3][4][5] albeit comprehensive controversy on its e cacy [6,7]. Lymphopenia, neutrophilia, and higher levels of multiple cytokines and chemokines were found in COVID-19 patients and were associated with disease severity [8,9], suggesting a role of dysregulated systemic in ammation leading to worse prognosis. Hyperin ammation lead to acute respiratory distress syndrome (ARDS) in 3-29% of the patients, [3,4,9] which was the main cause of death. Pathologically, interstitial mononuclear in ammatory in ltrates, dominated by lymphocytes, were reported in deceased patients [10,11]. Suppression of in ammation by adjunctive corticosteroids may be theoretically bene cial. [12] However, evidence on corticosteroid treatment among COVID-19 patients is scarce, and results from previous studies investigating corticosteroids in other coronavirus infections were inconsistent and inconclusive. Previous observational studies reported that corticosteroid treatment was related to no reduction or even increase of mortality in patients with SARS-CoV-2, severe acute respiratory syndrome (SARS), and Middle East respiratory syndrome patients (MERS) [13,14]. However, lower mortality was reported among the critically ill subgroup of SARS patients treated with corticosteroids in a retrospective study [15]. These ndings suggests a heterogeneity in e cacy of corticosteroids among different phenotypes. Recently, we reported that methylprednisolone might associated with lower risk of death in COVID-19 patients who developed ARDS [5]. However, due to the limited sample size at early stage of COVID-19 pandemic, the above study did not include the key determinants or potential confounding effects of corticosteroids use on clinical outcome. In this study, we examined the use of corticosteroid treatment in relation to hospital mortality among COVID-19 patients who have developed ARDS.

Study design and patients
This single-centre, retrospective, cohort study was conducted at the Jin Yin-tan Hospital, Wuhan, China. We identi ed all adult patients with con rmed COVID-19 according to WHO interim guidance [16] and patients were admitted between December 24, 2019 and February 24, 2020. Then we identi ed those who developed ARDS according to the WHO de nition for analysis. [16] To avoid the in uence of early mortality before cortical steroid presenting treatment e cacy, patients who died or discharged within 2 days on hospital admission were excluded. Other exclusion criteria were: (1) participating in any doubleblind clinical trial, (2) under long-term corticosteroid therapy, or (3) no valid medical history provided. The Jin Yin-tan Hospital Ethics Committee approved the study (No.KY-2020-44.01) and granted a waiver of informed consent from study participants.

Data collection
Administration of corticosteroids was de ned as systemic use (oral or intravenous) of corticosteroids, including methylprednisolone, dexamethasone, hydrocortisone, and prednisone. The primary outcome was 60-day in-hospital mortality. Patients were followed to death or discharge from hospital up to 60 days since hospital admission (last clinical outcome was observed on March 21, 2020). Data on demographics, medical history, laboratory ndings, chest radiology, medication use, and clinical outcomes were extracted retrospectively from electronic medical records using a standardized data collection form. All data were checked independently by two physicians (DH and XC). To monitor the clearance of viral RNA, SARS-CoV-2 RNA were tested using polymerase chain reaction from throat-swab specimens for every other day after clinical remission of symptoms, including fever, cough, and dyspnea.
[5] Viral clearance analysis was performed excluding patients who never cleared SARS-CoV-2 RNA before discharge or death. The de nitions of ARDS and other diseases were described in eMethods (Additional le 1). To reduce the effect of steroids treatment bias and potential confounding factors, we performed propensity score analysis [17] to adjust the differences in baseline characteristics. For each patient, a propensity score indicating the likelihood of receiving systemic corticosteroid treatment was calculated by a logistic regression model that included 10 pre-selected baseline variables: age; sex; SOFA score at admission; temperature, respiratory rate, heart rate, SpO 2 /FiO 2 ratio, blood lymphocyte count, blood neutrophil count, and level of CRP at hospital admission. The outcome variable was whether or not the patient received corticosteroid therapy in current hospital stay. Goodness of t was evaluated by the cstatistic and the Hosmer-Lemeshow test.
The effect of corticosteroid treatment on risk of 60-day in-hospital all-cause death were analyzed using a series of Cox proportional-hazard regression models. First, we constructed a univariable Cox regression models on hospital death by 60 days since admission with corticosteroid treatment treated as a timevarying covariate. Then we constructed a multivariable Cox model of 60-day hospital death with corticosteroid treatment as time-varying covariate and incorporated the individual propensity score into the model as a covariable to calculate the propensity adjusted hazard ratio (HR). In the nal model, the effects of corticosteroids on 60-day in-hospital death were adjusted for propensity score of corticosteroid treatment, as well as the following pre-selected covariates: age, sex, Sequential Organ Failure Assessment (SOFA) score at hospital admission,, and comorbidities: diabetes, hypertension, chronic pulmonary disease, chronic renal or liver disease, solid malignant tumor, hematologic malignancy, and immunosuppressive status. [6] Furthermore, we performed the same models described above in the subgroups strati ed by days from ARDS onset to the initiation of corticosteroid treatment (≤ 1 day vs. > 1 day) and days from symptom onset to the initiation of corticosteroid treatment (≤ 14 days vs. > 14 days). Several sensitivity analyses were performed to assess the robustness of our ndings. To test whether the ndings were in uenced by the time point of baseline the time-dependent Cox regression analysis were repeated using the values of previously mentioned variables on the date of ARDS onset. To ensure that treatment-related survival differences were not confounded by prescription of lopinavir-ritonavir, which has been associated with shorter median time to clinical improvement in patients with COVID-19 [18], the effects of corticosteroids were evaluated among patients without lopinavir-ritonavir treatment. To test whether the ndings might be in uenced by ARDS de nition, we conducted survival analysis using the same model among patients diagnosed with ARDS by Berlin de nition.
Results were analyzed with SAS (version 9.4, SAS Institute, Cary, NC). Unadjusted and adjusted hazard ratios and their 95% con dence intervals (CIs) were reported. Two-sided P values less than 0.05 were considered statistically signi cant.

Results
A total of 1147 patients with COVID-19 were screened for the study. 40 patients were excluded for participating in any double-blind clinical trial (n = 15), death or discharge from the hospital on the rst day of admission (n = 21), underwent long-term corticosteroid therapy (n = 3), or no valid medical history provided (n = 1). From 1107 patients remained, 382 patients were identi ed as ARDS (eFigure 1, Additional le 1). Among these patients, 84 have been described previously by Wu et al [5]. In addition, 91 patients with ARDS participated in the open-label trial of lopinavir -ritonavir. [18] Baseline characteristics of the ARDS patients at hospital admission by receiving systemic corticosteroid treatment are shown in Table 1. In the entire cohort, the mean age was 60.7 ± 14.1 years, and 234 (61.3%) patients were male. 147 (38.5%) were treated with NIMV, 94 (24.6%) with IMV, and 11 (2.9%) with ECMO.   Among patients prescribed corticosteroids, methylprednisolone was the most frequently administered corticosteroids (213, 94.2%) ( Table 2)    In the logistic regression model generating propensity score, the pre-selected variables most closely correlated with prescription of systemic corticosteroids included age, blood lymphocyte count, heart rate and CRP. (eTable 2, Additional le 1) The multivariable regression model of propensity for corticosteroid treatment had area under the receiver operating characteristic curve (ROC) of 0.71, indicating good discrimination.
In survival analysis, univariable time-dependent Cox regression model showed the prescription of corticosteroids was associated with a lower risk of death (HR: 0.48; CI: 0.25, 0.93; p = 0.0285) ( Table 3). When adjusted for propensity score, the estimated HR was 0.49 (CI: 0.25, 0.93; p = 0.0298). In full model adjusted for age, sex, SOFA score, propensity score, and comorbidities, the association remained signi cantly (HR: 0.51; CI: 0.27, 0.99 ; p = 0.0471) (Fig. 1). The association was found among patients treated within 1 day after ARDS onset (HR: 0.50; CI: 0.26, 0.98; p = 0.0439  In sensitivity analysis, narrowing to patients meet the Berlin de nition of ARDS did not alter the association between corticosteroids and lower risk of death. The HR of corticosteroids on mortality were negative albeit insigni cant when excluding patients treated with lopinavir or vital signs, laboratory ndings and SOFA score on the ARDS onset date were used in time-dependent Cox regression analysis adjusting for the same confounders. CRP level decreased among corticosteroids group on the rst 4 days after ARDS onset (Fig. 2), while an increase was found in non-corticosteroids group. CRP levels were signi cantly lower in corticosteroids treated group after 2 days of ARDS onset, indicating a suppression of in ammatory response due to corticosteroid treatment. The IL-6 level was comparable between the two groups(eFigure 2, Additional le 1).

Discussion
In this observational study, prescription of low-to-moderate dose systemic corticosteroids was signi cantly associated with lower risk of 60-day in-hospital death among of COVID-19 patients who have developed ARDS. Early corticosteroid treatment before or within 1 day after ARDS is related with lower mortality. Reduction of CRP, as the marker for systemic in ammation responses, was found in patients with corticosteroid treatment. No associations between corticosteroid treatment with viral shedding were found in our study.
The interim guidance of WHO recommend against the routine use of systemic corticosteroids for treatment of viral pneumonia. [16] The recommendation was based on previous studies reported no bene t or possible harms of corticosteroids therapy among SARS and MERS patients [14,[19][20][21].
However, the e cacy of corticosteroids on the patients developed ARDS after coronavirus infection was not clear. Our previous study on COVID-19-related ARDS patients with a small sample size showed that methylprednisolone was associated with lower risk of death. [5] In this study of a larger population, we reported the detailed regimens of corticosteroids therapy and used rigorous statistical method to control for indication and survival bias. We found that corticosteroids therapy was associated with a lower risk for death. At the same time, CRP levels decreased in patients receiving corticosteroid treatment, which indicated suppressed in ammatory responses and was compatible with previous the randomized trials [22,23]. Our results were in line with several previous randomized controlled studies that showed administration of corticosteroids was associated with a survival bene t in patients with established ARDS. [12,22,24,25] Notably, the difference in mortality between groups was similar to the results reported in a recent randomized controlled study on moderate-to-severe ARDS [25]. These ndings suggest that some patients with ARDS due to COVID-19 might bene t from appropriate corticosteroids therapy.
Survivors-treated bias is common in observational studies that assess exposure after the start of followup, where only patients survived long enough had an opportunity to receive the intervention. Therefore, the patients died early are more likely to be misclassi ed to the no-treatment group, leading to overestimation of the effectiveness of medicine [26]. This study was speci cally designed to address survivors-treated bias of corticosteroid treatment, by using a time-dependent variable for corticosteroids initiation to de ne corticosteroids group and non-corticosteroids group. [27]In addition, there was a propensity of clinicians to give corticosteroids to patients who were critically ill in non-randomized clinical condition. The imbalance in baseline characteristics may introduce confounders in comparison of mortality between corticosteroids and non-corticosteroids group. In this cohort, lymphopenia and elevation of CRP and lactate dehydrogenase levels were more severe in patients who received corticosteroids therapy. Propensity score is a validated method to account for baseline confounding and control selection bias in this case. [28] We performed a rigorous propensity adjustment analysis in this study, and the results of time-varying Cox model was unchanged. These added to the strength of our results that found bene t of corticosteroids on mortality.
Timing, dosage, and duration of corticosteroids therapy were fundamental variables of corticosteroid treatment regimens. Bene t of early administration of corticosteroids has been reported by several randomized controlled trials in patients with ARDS. [22,25] Most of the trials showed corticosteroids initiated in early ARDS was associated with better outcome. [24,29] We also found survival bene t of corticosteroids therapy in the subgroup initiated within 1 day after ARDS onset, favoring administration of corticosteroids in the early phase of disease process. In addition, we found that administration of corticosteroids within 14 days after symptom onset were also related with better outcome. These results underlines early identi cation and intervention of ARDS in COVID-19 patients.
Low dose corticosteroid treatment (methylprednisolone of 1-2 mg/kg) was found to accelerates the resolution of ARDS, [25,29,30] while high dose treatment have risk of immune-suppression and corticosteroid-induced complications including diabetes, osteonecrosis, and psychosis in SARS. [19,31,32] The ranges of maximum dose of corticosteroids in this cohort were 40-160 mg of methylprednisolone equivalent, which was close to the recommended dose for ARDS patients. [30,33] Our results show that low-to-moderate dose of corticosteroids may bene t COVID-19 patients who developed ARDS, without a signi cant increase of hyperglycemia.
In our study, the course of corticosteroid therapy was short, which was in agreement with most previous studies which suggested short-course corticosteroids of low-to-moderate dose was effective and safe among patients with ARDS [34,35], sepsis [36], and H1N1 virus infection [37]. Tampering strategy were performed as has been suggested by the guidelines for the ARDS -related corticosteroid insu ciency (CIRCI) to reduce deterioration from the development of a reconstituted in ammatory response and febrile response. More research is needed to determine the best duration of corticosteroid therapy. A recent individual patient data analysis of four trials including 322 patients showed prolonged corticosteroids therapy reduced mortality [29]. However, chronic side effects of corticosteroids including secondary infection and osteoporosis may occur in prolonged course of treatment.
Delayed virus clearance was reported in corticosteroid-treated patients with both SARS and MERS. [21,38] In contrast, we found no difference in viral shedding duration from symptom onset between corticosteroid and non-corticosteroid groups from whom the data were available. Notably, positive SARS-CoV-2 test results has been reported after two consecutive negative results.
[39] Unfortunately, viral tests of throat swabs were not monitored after two consecutive negative tests in our cohort, thus more evidence is needed for assessing the effects of corticosteroids on viral shedding.
Our study had some limitations. First, unlike randomized controlled trials, the selection bias and potential confounding effects might exist. We used propensity analysis rather than standard multivariable analysis to rigorously adjust for selection bias, and time-dependent model to avoid survivors-treated bias.
Nonetheless, only measured factors were controlled for due to the nature of observational study design.
Second, secondary infections was not monitored in this study, because microbiological culture results needed for de nite diagnosis of secondary infection were possibly affected by antibiotic treatment the patients received simultaneously. To include the delayed effects of secondary infections on mortality, a longer the follow-up period of 60-day were used. Third, this study was single-centre and patients were sicker and transferred from other hospital, so might lacking of generality. Declarations Ethics approval and consent to participate

Conclusion
The Jin Yin-tan Hospital Ethics Committee approved the study (No.KY-2020-44.01) and granted a waiver of informed consent from study participants.

Consent for publication
Not applicable.

Availability of data and materials
The corresponding authors had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.
The data are available from the corresponding authors upon reasonable request.

Supplementary Information
Additional le 1.pdf: Include eMethods, eTables and eFigures. Figure 1 Estimated survival probability of multivariable Cox regression model with time-dependent corticosteroid treatment. Cox regression model with corticosteroid treatment was time-varying variable, adjusting for age, sex, SOFA score at admission, propensity score of corticosteroid treatment, and comorbidities : diabetes, hypertension, chronic pulmonary disease, chronic renal or liver disease, solid malignant tumor, hematologic malignancy, and immunosuppressive status. Propensity score was calculated by a nonparsimonious logistic regression model that included: age; sex; SOFA score at admission; temperature, respiratory rate, SpO2/FiO2 ratio, blood lymphocyte count, blood neutrophil count, and level of c-reactive protein at admission. ARDS, acute respiratory distress syndrome; SOFA, sequential Organ Failure Assessment. Figure 2