Use of corticosteroids in influenza-associated acute respiratory distress syndrome and severe pneumonia: a systemic review and meta-analysis

Influenza-related severe pneumonia and acute respiratory distress syndrome (ARDS) are severe threats to human health. The objective of this study was to assess the effects of systematic corticosteroid therapy in patients with pneumonia or ARDS. The PubMed, EMBASE, Web of Science and SCOPUS databases were searched up to July, 2019. Nineteen studies including 6637 individuals were identified, and fifteen studies (6427 patients) were included in the meta-analysis of mortality. Eighteen were observational studies and one was a randomized controlled trial (RCT). The meta-analysis results showed that corticosteroid therapy was associated with significantly higher mortality (OR 1.53, 95% CI [1.16, 2.01]) and incidence of nosocomial infection (OR 3.15, 95% CI [1.54, 6.45]). Subgroup analysis showed that among patients with unadjusted estimates, the odds of mortality were higher in patients receiving corticosteroid treatment (OR 1.98, 95% CI [1.23, 3.17]), however, among patients with adjusted estimates, the result showed no statistically significant difference between corticosteroid group and control group (OR 1.31, 95% CI [0.95, 1.80]). Current data do not support the routine use of corticosteroids in patients with influenza severe pneumonia or ARDS. RCTs are needed to provide more robust evidence.

Influenza is a viral infection that attacks the respiratory system. Rapidly progressing viral pneumonia and acute respiratory distress syndrome are pulmonary manifestations that are commonly observed in patients with influenza and are associated with considerable mortality 1-3 , representing a severe threat and imparting a substantial financial burden worldwide 4 .
Individuals with community-acquired pneumonia may benefit from systematic corticosteroid therapy, which may block the inflammatory cascade reaction 5 . Corticosteroids could improve the lung tissue damage induced by influenza pneumonia and decrease the risk of mortality in animal models with influenza infections 6,7 . Many clinicians administer corticosteroids as an anti-inflammatory treatment for patients with severe influenza-related pneumonia to stop disease progression and improve clinical outcomes. A large cohort study of patients admitted to 148 ICUs in Spain found that the frequency of corticosteroid treatment by study period was 34.9% in 2009, 39.6% in 2010, 29% in 2013, and 31.4% in 2014 8 . Recently, some studies have shown that corticosteroids may not be beneficial for patients with severe influenza and may even increase mortality [9][10][11] . However, there is considerable uncertainty regarding whether patients with influenza-related ARDS or severe pneumonia can benefit from adjuvant corticosteroid therapy.
We aimed to systematically review all experimental and observational studies on corticosteroid use in patients with influenza-related ARDS and severe pneumonia. The effect of corticosteroid treatment on clinical outcomes was investigated.
Search strategy and study selection. We comprehensively searched the PubMed, EMBASE, Web of Science and SCOPUS databases from inception to July 2019. The core search terms were defined as those related to influenza-related pneumonia, ARDS, acute respiratory failure and corticosteroid use (for details on the search strategy in EMBASE, refer to Supplementary Table S1). The references of eligible studies were screened, and two authors independently reviewed all citations that met the inclusion criteria. Study selection was performed in 2 stages: first, study title and abstract screening; second, full text examination.
Data extraction and quality assessment. Outcome data were independently extracted from the included studies by two investigators using a previously piloted standardized pro forma. We obtained the following data: (a) characteristics of studies (design, setting, country, period, methodological details for quality assessment); (b) characteristics of participants (demographics, co-morbid illnesses, disease severity, numbers in each group, influenza virus type); (c) characteristics of interventions (type, dose, timing and duration of corticosteroid use); and (d) outcomes.
The quality of each study was independently assessed by two individuals according to the Cochrane Risk of Bias tool for RCTs and the Newcastle-Ottawa Scale for nonrandomized trials and comparative observational studies. Three domains are assessed on the NOS for observational studies 12 : (1) "selection bias", (2) "comparability bias", and (3) "outcome bias". Disagreements at any stage were resolved through discussion with the other authors until consensus was reached. Sensitivity analysis. We performed sensitivity analysis to assess the effect of the study design on clinical outcomes using stratification if the number of studies was sufficient. Data analysis. Odds ratios (ORs) and their corresponding 95% confidence intervals (CIs) were generated during the analysis of dichotomous outcome data, and mean differences or standardized mean differences and their corresponding 95% CIs were generated during the analysis of normally distributed continuous data. ORs or hazard ratios (HRs) for adjusted outcome estimates and their corresponding 95% CIs were obtained and are presented in the pooled analyses. Medians and interquartile ranges were generated in the analysis of continuous data that were not normally distributed.
The I² test for inconsistency was used to analyse heterogeneity. If I² > 50%, the heterogeneity across studies was significant, and a random-effects model was used in the meta-analysis; otherwise, a fixed-effects model was used. Subgroup analysis was performed in the following areas where possible: adult population versus child population; seasonal influenza versus outbreak influenza or pandemic influenza; ICU versus inpatient; adjusted estimates versus unadjusted estimates; and corticosteroid dose, timing and duration. All statistical analyses were performed using Cochrane systematic review software Review Manager (RevMan; Version 5.3.5; The Nordic Cochrane Centre, The Cochrane Collaboration, Copenhagen, 2014).

Main results.
A total of 7771 relevant articles were identified during the initial search. After the removal of 2469 duplicates, 5302 articles remained. After screening the titles and abstracts of those articles, 5204 articles were excluded because of irrelevance. Of the 98 full-text articles reviewed, 81 were excluded for various reasons, and 19 articles remained. Details regarding the reasons for the exclusion of these studies are shown in Fig. 1 and Supplementary Table S2. Ultimately, 15 studies were included in the meta-analysis of mortality, while 4 studies reported outcomes other than mortality in association with corticosteroid use.
The characteristics of the participants in the included studies are summarized in Table 1 and Supplementary  Table S3. The studies were published between 2010 and 2018. Eighteen of the studies had an observational design, while one had a randomized controlled trial design 13 . Outcome data were reported in 19 studies (6637 individuals, including 2675 in corticosteroid group and 3962 in the non-corticosteroid group), while mortality data were reported in 15 studies (6427 individuals). Eight studies (n = 2558) included only ICU patients. Fourteen studies assessed individuals with H1N1pdm09 virus infection, 1 study assessed individuals with H7N9 virus infection, and 4 study assessed individuals with inter-pandemic influenza virus infection. Eight studies (1956 individuals) had useable data related to patients with ARDS 9,10, [14][15][16][17][18][19] . Fourteen studies (n = 6335) reported mortality associated with adults only.
The median ages varied from 2.5 to 60.1 years in all patients included. The proportion of male participants was higher than that of females (58.9% versus 41.1%) and range varied from 44.8% to 78.0% (15 studies, 2969 individuals). Obesity (BMI ≥ 30 kg/m 2 ) was common (31.7%, 779/2454) in the included studies (7 studies, 2454 individuals), and the proportion of obese individuals ranged from 2.1% to 41.4%. Disease severity at baseline was reported in seven studies and in 4 studies (2166 individuals), the baseline disease severity was higher in individuals in the corticosteroid group than in those in the non-corticosteroid group [19][20][21] . Methylprednisolone (88.7%, 809/912) was the most common steroid used in the corticosteroid group (4 studies), and the median duration varied from 5.1 to 11.0 days. Almost all patients (97.0%, 5056/5211) received antiviral therapy (ranging from 96.0% to 100%, 12 studies), and 96.5% (2579/2673) of participants received antibiotic therapy (5 studies). The details of the therapies are described in Table 1.
Because all studies included in the meta-analysis of mortality were observational cohort studies, selection bias was inevitable. The risk of bias identified in the 19 included studies is shown in Supplementary Tables S4a,b. The studies' NOS scores varied from 6 to 9, indicating that the quality of the included studies was high 22 . However, most included studies had substantial comparability bias because we could not adequately adjust for disease severity, and individuals with greater diseases severity tended to use corticosteroids.
The pooled analysis of the crude results of the 15 included studies (6427 individuals) suggested that those who used corticosteroids had a significantly higher mortality rate (OR 2.30, 95% CI [1.68, 3.16], p < 0.01), and a moderate level of heterogeneity was observed (I² = 66%; Supplementary Fig. S1). Five studies 8,10,14,15,26 reported adjusted effect estimates of 30-day or inhospital mortality (adjusted OR or adjusted hazard ratio (aHR), Table 2). The pooled analysis of the crude results of ten included studies and five adjusted effect estimates suggested that those who used corticosteroids had a significantly higher mortality rate (OR 1.53, 95% CI [1.16, 2.01], I 2 = 53%, p < 0.002, Fig. 2). The subgroup analysis of unadjusted effect estimates showed a similar result (OR 1.98, 95% CI [1.23, 3.17], I 2 = 32%, p = 0.005). However, the subgroup analysis of the five adjusted estimates showed no association between mortality and corticosteroid use (HR 1.31, 95% CI [0.95, 1.80], I 2 = 70%, p = 0.1). The test for subgroup differences between adjusted and unadjusted mortality was not statistically significant (p = 0.06). There was no clear indication of publication bias in the funnel plot analysis ( Supplementary Fig. S2).

Mortality in ARDS.
Eight studies reported the mortality of patients with ARDS due to influenza (n = 1956, Supplementary Table S5). Meta-analysis of unadjusted and adjusted estimates suggested that the difference in mortality between the corticosteroid and control groups was not statistically significant (OR 1.68, 95% CI [0.94, 3.01], I 2 = 69%, p = 0.08, Fig. 3). Li et al. reported that corticosteroid use was associated with a decreased risk of mortality in patients with ARDS (PaO 2 /FiO 2 < 300 mmHg) (aHR 0.67, 95% CI [0.46, 0.98]) 15 . However, Brun-Buisson et al. 14 and Chawla et al. 18 suggested that the risk of mortality was higher in corticosteroid group in the patients with ARDS. Cao et al. reported that a low-to-moderate dose of corticosteroids had no statistically significant association with the risk of mortality in patients with ARDS (OR 1.69, 95% CI [0.78, 3.64]), whereas a high dose was associated with greater mortality (OR 2.89, 95% CI [1.10, 7.56]) 10 . The other four studies 9,16,17,19 showed that corticosteroid use had no association with mortality ( Fig. 3).
Three studies (n = 3473) have reported the results of mortality excluding patients who had potential indications (e.g., asthma, COPD exacerbation, pregnancy/post-partum, shock or immunosuppressive conditions) for corticosteroid treatment that may have skewed the results 8,14,15 . A pooled analysis of these three studies showed no statistically significant association between corticosteroid use and mortality (OR 1.37, 95% CI [0.86, 2.18]), with a high level of heterogeneity (I² = 77%, P = 0.01).    The number of studies was insufficient to perform subgroup analysis according to the various reported regimens. Two studies compared early versus later/no corticosteroid treatment; one defined early treatment as within three days of mechanical ventilation 14 , and the result suggested that early treatment with corticosteroids was associated with greater mortality (aHR 3.42, 95% CI [1.73, 6.75]); the other defined early treatment as within 72 h of NAIs 17 , and the result showed no statistically significant difference in mortality between patients receiving corticosteroids within 72 h of NAIs and those who did.
Three studies categorized corticosteroid dose as low/low-to-moderate and high 10,15 . A large retrospective cohort study reported that a low-to-moderate dose (25- 10 . Another study reporting the result of 62 patients with acute respiratory failure due to influenza showed no statistically significant difference between low dose and high dose corticosteroid therapy (8/19 versus 7/19, p > 0.05) 16 .
Two of the included studies reported outcomes related to children, but only one reported the risk of mortality (OR 8.12, 95% CI [2.44, 27.02]) related to corticosteroid use 27 . However, in that study, all children who received corticosteroids had ARDS, while the patients in the non-corticosteroid group had less severe disease conditions. Another retrospective cohort study of children with pneumonia caused by the 2009 H1N1 influenza virus only reported length of hospital stay and duration of fever 28 and found a shorter length of stay and duration of fever in corticosteroid group (Table 3).

Nosocomial infection. Six studies reported an association between corticosteroid use and nosocomial infection.
In four of these studies, corticosteroid use was associated with an increased risk of developing a nosocomial infection 15,17,20,21 , while the remaining two studies did not show a statistically significantly increased odds of developing infection 10,14 . Overall, the pooled results revealed that the odds of nosocomial infection were significantly higher in patients who were administered corticosteroids than in those who were not (OR 3.15, 95% CI [1.54, 6.45], p < 0.0001), but a high level of heterogeneity was observed (I² = 82%) (Fig. 4).
Three studies reported the common pathogens isolated from patients with nosocomial infection. One study reported that the most common bacteria isolated was Acinetobacter baumannii (28.2%) 17 . In a study of 2141 patients with severe influenza pneumonia 15 , 245 patients had nosocomial infection, and the most commonly isolated pathogens were Acinetobacter baumannii (35.0%), Pseudomonas aeruginosa (13.5%), and Staphylococcus aureus (11.0%), while in another cohort study of 1846 patients with severe influenza pneumonia 8 , Streptococcus pneumoniae (49.1%), Pseudomonas aeruginosa (10.1%), and Staphylococcus aureus (7.5%) were the most frequently isolated microorganisms. www.nature.com/scientificreports www.nature.com/scientificreports/ Length of stay and length of MV. Seven studies reported length of stay according to corticosteroid use; all were unadjusted for disease severity (Table 3). Six studies found no statistically significant difference between the groups. One study 19 showed a longer length of ICU stay associated with corticosteroid use, while the total length of hospital stays was not significantly different between the groups. Notably, one of the five studies analysed the duration of hospital stay in people with influenza pneumonia treated with corticosteroid versus those receiving placebo, and found no significant difference between the groups (adjusted difference −2.24 days, 95% CI [−9.61, 5.12]) 13 .
Linko et al. 19 reported a longer duration of mechanical ventilation in the group treated with corticosteroid therapy while Brun-Buisson et al. 14 and Moreno et al. 8 found no statistically significant difference between the groups. (Table 3).

Time to fever alleviation, time to clinical stability and viral shedding.
Two studies reported the time to fever alleviation according to corticosteroid use 28,29 . Kudo et al. 29 found no statistically significant difference between the groups, while another cohort study of children 28 with severe influenza pneumonia reported a shorter time to fever alleviation. Notably, two studies found a shorter time to clinical stability in the corticosteroid group. The study of influenza A/H7N9 found a significantly longer duration of viral shedding associated with corticosteroid treatment 10 . The details of these outcomes are described in Table 3.

Discussion
The overall findings of this meta-analysis indicated that patients with pneumonia or acute respiratory distress syndrome who were administered corticosteroids had significantly higher mortality and incidence of nosocomial infection but the use of corticosteroids did not influence the length of hospital stay.
Our studies suggested a deleterious effect of steroids on mortality and nosocomial infection. Several factors need to be accounted for in interpreting these findings. First, most studies did not adjust the clinical outcomes for potential confounding factors. Clinically, more severe cases tended to be treated with corticosteroids, which may obscure the real value of this treatment regarding mortality 30,31 . Therefore, in this study, we preferred the use of adjusted estimates of the effect to minimize potential confounding between the treatment groups. However, five studies reported adjusted estimates of mortality, and their inclusion in the meta-analysis still revealed a higher odds of mortality related to steroids use. Good evidences indicated that secondary bacterial pneumonia is an  www.nature.com/scientificreports www.nature.com/scientificreports/ important cause of mortality related to influenza 32,33 . Therefore, increasing risk of nosocomial infection due to corticosteroid treatment may partly account for the potential harm from corticosteroid use. Two included studies 8,15 found that secondary bacterial pneumonia such as due to Acinetobacter baumannii, Pseudomonas aeruginosa, Streptococcus pneumoniae, Staphylococcus aureus or invasive fungal infection, were more common in corticosteroid-treated patients. Several studies showed that prolonged viral shedding and delayed viral clearance were noted in corticosteroid-treated patients 10,34 , whereas slower clearance of virus loads was associated with higher mortality in patients with ARDS due to H1N1pdm09 virus infection 35 . Thus, prolonged viral shedding and delayed viral clearance may also contribute to higher mortality.
Second, most of the included observational studies did not explain why some patients received systemic corticosteroid therapy and others did not. The initial intentions of corticosteroid therapy were unclear (was it used as a rescue therapy or due to COPD/asthma exacerbation or due to pneumonia/ARDS?). Different indication may easily confound the effect of the corticosteroid. Some evidences supported the use of corticosteroids for asthma or COPD or septic shock in the context of influenza infection [36][37][38] . In order to minimize the influences of different indications, subgroup analysis of the mortality in three studies (n = 3347) was performed after excluding patients receiving corticosteroids as rescue therapy or due to COPD/asthma exacerbation, and found no statistically significant difference between the steroid therapy groups and control groups and the heterogeneity was high (I² = 77%). However, the high level of statistical heterogeneity may result in unstable estimates of the meta-analysis. Therefore, well-designed clinical trials should be conducted to decrease the heterogeneity of patients and to provide more robust evidence.
The results from clinical studies of corticosteroid therapy in patients with influenza are conflicting. Many studies have shown a significant association between corticosteroid treatment and mortality in patients with influenza; however, several studies have reported that corticosteroids can provide benefits to patients under certain conditions 15,28,39,40 . An RCT 13 included in this review noted an association between adjuvant corticosteroid therapy (50 mg of prednisone given orally for 7 days) and decreased time to clinical stability. Low-to-moderate doses of corticosteroids are beneficial in people with hypoxia ((PaO 2 /FiO 2 ) <300 mmHg), whereas high doses of corticosteroids showed no benefit in this group; however, low-to-moderate doses of corticosteroids may increase the 60-day mortality rate in those with PaO 2 /FiO 2 > 300 mmHg 15 . Kil et al. 28 reported that rapid (methylprednisolone, 10 mg/kg/d) and short-term (tapered off within a week) corticosteroid treatment for children with severe pneumonia halted clinical exacerbation and possibly prevented progression to ARDS. However, in another study, compared with no treatment, administration (steroid therapy was initiated at a median daily dose equivalent to 270 (IQR, 200-400) mg of hydrocortisone, and a median duration of 11 (IQR, 6-20) days within the first 3 days of MV was more strongly associated with an increased risk of death, whereas when administration was beyond the first 3 days of MV, the association was no longer significant 14 . Considering the findings of the aforementioned studies, the condition of the patients' respiratory system and the dose, timing and duration of corticosteroids could be contributing factors that affect the effects of corticosteroids.
Several recent systematic reviews and meta-analyses concluded that corticosteroid therapy is significantly associated with mortality [41][42][43] . However, in these studies, there were no special limitations on subject inclusion criteria, which means that the patients were very diverse. Additionally, there was no subgroup analysis for these patients under different disease conditions. Compared to patients in those previous studies, we focused only on patients with pneumonia or ARDS, which is more specific and makes the outcomes more targeted. Our study observed a different outcome according to corticosteroid use in patients with ARDS due to influenza.
This study has some limitations, including the lack of sufficient data on the dose, duration, timing and rationales of corticosteroid administration and the timing and duration of antiviral therapy. In addition, only one study included in this meta-analysis was an RCT, and 18 were observational in nature. Thus, it is possible that selection bias or comparability bias could have affected the quality of the analysed evidence. There is insufficient evidence in this meta-analysis to make a firm determination about the effectiveness of corticosteroids for people with influenza-related pneumonia or ARDS. The small number of included studies and the small number of patients in the included studies might also make the effect size of some outcome indicators insufficient, and we were unable to analyse the effect of some factors on the outcome indicators by meta-regression or subgroup analysis.

Conclusion
Current data do not support the routine use of corticosteroids in patients with influenza pneumonia or ARDS. However, the data assessed in this meta-analysis were extracted from 18 observational studies and only one RCT; therefore, the limitations associated with study design are important to consider. There is a need for more robust evidence on the role of corticosteroids in the treatment of influenza-related ARDS and severe pneumonia before a firm recommendation for clinical practice can be made.