The effects of adding glucocorticosteroids to standard care for children with sepsis. A systematic review of randomized clinical trials with meta-analysis and Trial Sequential Analysis

Background: Glucocorticosteroids are widely used to treat severe sepsis in pediatric intensive care units. However, the evidence on the clinical effects is unclear. Objective: To assess the benets and harms of glucocorticosteroids for children with sepsis. Data Sources: We conducted a systematic review of randomized clinical trials with meta-analysis and Trial Sequential Analysis (TSA) (PROSPERO CRD42017054341). We searched CENTRAL, MEDLINE, Embase, LILACS, SCI-Expanded, and more. Study Selection: Randomized clinical trials assessing the effects of adding glucocorticosteroids to standard care for children with sepsis. Data Extraction: Two independent reviewers screened studies and extracted data. Evidence was assessed by GRADE according to our published protocol. Data Synthesis: We included 24 trials randomizing 3073 participants. Meta-analyses showed no evidence of an effect of adding glucocorticosteroids for children with sepsis with a mixed focus for any of our outcomes. Meta-analyses suggested evidence of a benecial effect of dexamethasone for children with meningitis when assessing serious adverse events (risk ratio (RR) 0.68, 95% condence interval (CI) 0.53 to 0.86; P = 0.001, very low certainty of evidence) and ototoxicity (RR 0.63, 95% CI 0.45 to 0.88; P = 0.007, low certainty of evidence). TSAs showed that we did not have sucient data to conrm or reject these results. We found insucient evidence to conrm or reject an effect on mortality or our other outcomes. No trials reported quality of life or organ failure. Most trials were at high risks of bias. We found high clinical heterogeneity between participants. None of our TSAs showed benets, harms or futility. Conclusions: Generally, we found no evidence of an effect of glucocorticosteroids for children with sepsis without meningitis. Dexamethasone for sepsis in children due to meningitis may decrease serious adverse events and ototoxicity.


Background
Sepsis is a leading cause of death in infants and children worldwide (1). Guidelines suggest that the glucocorticosteroid, hydrocortisone, might be used for children with uid refractory and vasopressorresistant septic shock, but the recommendation is based on unclear evidence (2,3). The use of glucocorticosteroids for sepsis has been controversial for decades (4). A study from the UK suggested that 76% of pediatric intensive care units used steroids for septic shock (5). A worldwide cross-sectional study showed that the use of glucocorticosteroids was at 45% for children with severe sepsis (6).
Glucocorticosteroids seem to slightly reduce 28-day mortality in adults with sepsis (7). Important differences exist between children and adults with regard to sepsis and septic shock (8,9). There is, therefore, a need to conduct an up-to-date review to address the bene ts and harm of treatment with glucocorticosteroids in children with sepsis.

Methods
We detailed our prede ned methodology in our pre-published protocol (10,11) according to international guidelines (12). In accordance with our protocol, we conducted our systematic review based on the Preferred Reporting Items for Systematic Reviews and Meta-Analysis guidelines (PRISMA) (13), The Cochrane Handbook for Systematic Reviews of Interventions (12), Keus and colleagues (14), and the eight-step assessment suggested by Jakobsen and colleagues for better validation of meta-analytic results in systematic reviews (15). Review Manager 5.3 was used for all meta-analyses (16).
We searched for trials assessing the effects of adding any glucocorticosteroid to standard care versus standard care for hospitalized children (age < 18 years) with a diagnosis of sepsis based on the current international consensus (SIRS) or similar terms (as de ned by trialists) (17). We also included participants suspected of or diagnosed with severe/deep-seated infections such as meningitis, osteomyelitis, endocarditis, and necrotizing enterocolitis (11 Trials were included irrespective of trial design, setting, publication status, publication year, language, and the reporting of our outcomes. Two authors (SKK and SS) independently selected relevant trials, extracted data using a standardized data extraction sheet, and systematically assessed risks of bias (12). We contacted trial authors if relevant data were unclear or missing.
Our primary outcomes were all-cause mortality and serious adverse events (10,11). Our secondary outcomes were quality of life, shock reversal, organ failure, hearing loss or ototoxicity, and adverse events not considered serious. For all outcomes, we used the trial results reported at maximal follow-up.
We planned several subgroup analyses including subgroups based on risk of bias, type of glucocorticosteroids, dose, age, and presence of shock (10,11).
We used risk ratios (RR) for dichotomous outcomes. We performed both random-effects (Der Simonian-Laird) and xed-effect (Mantel-Haenszel) meta-analyses and chose the most conservative result as our primary result (11). We used Trial Sequential Analysis (TSA) to control random errors and reported TSAadjusted con dence interval (CI) if the cumulative Z-curves did not reach the futility area or passed the diversity-adjusted required information size (DARIS) (11,15,(18)(19)(20)(21)(22)(23)(24)(25). We assessed two primary outcomes and, hence, considered a P-value of 0.033 or less as the threshold for statistical signi cance for the primary outcomes to account for multiplicity (11,15). We assessed ve secondary outcomes and considered a P-value of 0.05 as the threshold for statistical signi cance for the secondary outcomes. We used 'best-worst' and 'worst-best' case analyses to assess the potential impact of missing data (15). We calculated Bayes factor to quantify the likelihood of the meta-analysis results being more or less compatible with either the null hypothesis or the anticipated intervention effects (15). We used GRADE to assess the certainty of the body of evidence (26).

Included trials
Our literature search identi ed a total of 9133 studies. 1929 duplicates were excluded. 7204 studies were excluded based on the title or abstract. 21 studies were excluded based on the full text assessment. 24 trials met our inclusion criteria randomizing 3073 participants , of which 20 trials randomizing 2866 participants provided data for our prede ned meta-analyses (27-43, 45, 47, 48). See PRISMA owchart ( Figure 1) for details regarding the literature search and the selection of trials.
The visual inspection of the forest plot and test for subgroup difference (P = 0.02) in the meta-analysis on our primary outcome serious adverse events showed that the effects of glucocorticosteroids seemed to differ between trials randomising participants with meningitis and trials randomising participants with sepsis of mixed focus (Figure 3 and 4). It was therefore not justi able to pool trials including only children with meningitis with trials including children with difference underlying infections. Hence, we chose to report results separately for each group of trials (children with mixed focus of infection and children with meningitis). We have attached the results of the overall analyses in Appendix 1.
Trial Sequential Analysis showed that we did not have su cient data to con rm or reject that adding glucocorticosteroids to standard care reduced the risk of death by 20% and that the accrued information was compatible with either a reduced risk of death by 79% or an increased risk of death by 639% (TSAadjusted CI 0.21 to 7.39) ( Figure 6). Bayes factor (4.31) was above the Bayes factor threshold for signi cance of 0.1. Hence, the result con rmed the meta-analysis result showing no difference. We assessed the risk of bias of this outcome as high risk of bias. There were no missing data, so we did not perform 'best-worst' and 'worst-best' case meta-analyses on this outcome. As we only included ve trials, no funnel plot was constructed.

Subgroup analyses
None of the planned subgroup analyses assessing risk of bias, age, type of steroids, and presence of shock showed evidence of a difference (Figure 7-10).

Serious adverse events
A total of 5/9 trials (55.5%), randomizing 358 participants, reported serious adverse events. In the glucocorticosteroid group, 37/184 (20.1%) participants experienced one or more serious adverse events compared with 30/174 (17.2%) participants in the control group. The trials including children with sepsis and mixed focus did not report any neurological events. The majority (80%) of these trials administered hydrocortisone (See Table 2). Meta-analysis showed no evidence of a difference when assessing serious adverse events (RR 1.24, 95% CI 0.82 to 1.87; P = 0.31; I 2 = 0%; 358 participants; 5 trials; very low certainty of evidence; Figure 11). Neither visual inspection of the forest plot nor tests for statistical heterogeneity (I 2 = 0%; P = 0.96) showed clear signs of heterogeneity. Trial Sequential Analysis showed that we did not have su cient data to con rm or reject that adding glucocorticosteroids to standard care reduced serious adverse events by 20% and that the accrued information was compatible with either a decrease of serious adverse events by 77% or an increase of serious adverse events by 562% (TSA-adjusted CI 0.23 to 6.62) ( Figure 12). Bayes factor (5.20) was above the Bayes factor threshold for signi cance of 0.1. Hence, the result con rmed the meta-analysis result showing no difference. We assessed the risk of bias of this outcome as high risk of bias. There were no dropouts, so we did not perform 'best-worst' and 'worst-best' case meta-analyses on this outcome. As we only included ve trials, no funnel plot was constructed.

Subgroup analyses
None of the planned subgroup analyses assessing risk of bias, age, type of steroids, and presence of shock showed evidence of a difference (Figure 13-16).
No trials assessed quality of life, organ failure, or ototoxicity. Hence, no meta-analysis was performed.

Dexamethasone for meningitis
Primary outcomes All-cause mortality A total of 14/14 trials (100%), randomizing 2449 participants, reported all-cause mortality. In the dexamethasone group, 193/1243 (15.5%) participants died compared with 191/1206 (15.8%) participants in the control group. Meta-analysis showed no evidence of a difference when assessing allcause mortality (RR 0.97, 95% CI 0.78 to 1.21; P = 0.77; I 2 = 7%; 2449 participants; 14 trials; low certainty of evidence; Figure 19). Neither visual inspection of the forest plot nor tests for statistical heterogeneity (I 2 = 0 %; P = 0.58) showed signs of heterogeneity. Trial Sequential Analysis showed that we had did not have su cient data to con rm or reject that adding glucocorticosteroids to standard care reduced the risk of death by 20% and that the accrued information was compatible with either a reduced risk of death by 41% or an increased risk of death by 58% (TSA-adjusted CI 0.59 to 1.58) (Figure 20). Bayes factor (4.23) was above the Bayes factor threshold for signi cance of 0.1. Hence, the Bayes factor result con rmed the meta-analysis result showing no difference. We assessed the risk of bias of this outcome as high risk of bias. The 'best-worst' and 'worst-best' case meta-analyses showed that incomplete outcome data bias did not have the potential to in uence the results (Figure 21 and Figure 22). Visual inspection of the funnel plots showed no clear signs of asymmetry ( Figure 23).

Subgroup analyses
None of the planned subgroup analyses assessing risk of bias, age, and dose showed evidence of a difference (Figure 24-26).

Serious adverse events
A total of 14/14 trials (100%), randomizing 2379 participants, assessed serious adverse events. In the dexamethasone group, 370/1210 (30.6%) participants experienced one or more serious adverse events compared with 435/1169 (37.2%) participants in the control group. The trials primarily reported neurological complications, hearing loss/ ototoxicity, or a combination of both (see Table 1). Metaanalysis showed evidence of a difference when assessing serious adverse events (RR 0.68, 95% CI 0.53 to 0.86; P = 0.001; I 2 = 64%; 2379 participants; 14 trials; very low certainty of evidence; Figure 27). There were signs of statistical heterogeneity (I 2 = 64%; P = 0.0006), however, visual inspection of the forest plot did not show clear signs of heterogeneity. Trial Sequential Analysis showed that we did not have su cient data to con rm or reject that adding glucocorticosteroids to standard care reduced serious adverse events by 20% and that the accrued information was compatible with either a decrease of serious adverse events by 75% or an increase of serious adverse events by 80% (TSA-adjusted CI 0.25 to 1.80) ( Figure 28). Bayes factor (0.02) was under the Bayes factor threshold for signi cance of 0.1. Hence, the result con rmed the meta-analysis result suggesting a difference. We assessed the risk of bias of this outcome as high risk of bias. The 'best-worst' and 'worst-best' case meta-analyses showed that incomplete outcome data bias did not have the potential to in uence the results (Figure 29 and Figure  30). Visual inspection of the funnel plots showed clear signs of asymmetry ( Figure 31) con rmed by Harbord test (P=0.0009).

Subgroup analyses
None of the planned subgroup analyses assessing risk of bias, age, and dose showed evidence of a difference in intervention effects (Figure 32-34).

Secondary outcomes
Hearing loss or ototoxicity A total of 11/14 (78.6%), randomizing 1825 participants, reported hearing loss or ototoxicity. In the dexamethasone group 130/941 (13.8%) participants experienced ototoxicity compared with 174/884 (19.7%) participants in the control group. Meta-analysis showed evidence of a bene cial effect of adding dexamethasone to standard care (RR 0.63, 95% CI 0.45 to 0.88; P = 0.007; I 2 = 44%; 1825 participants; 11 trials; low certainty of evidence; Figure 35). Trial Sequential Analysis showed that we did not have su cient data to con rm or reject that adding glucocorticosteroids to standard care reduced serious adverse events by 20% and that the accrued information was compatible with either a reduced the risk of ototoxicity by 84% or an increased the risk of ototoxicity by 148% (TSA-adjusted CI 0.16 to 2.48) ( Figure   36).
No trials assessed quality of life, organ failure or shock reversal. Hence, no meta-analysis was performed.

Post-hoc analysis of neurological complications
The trials including children with meningitis reported many neurological complications as serious adverse events. We therefore decided to analyze that outcome separately as well.

Discussion
We included 24 trials randomizing a total of 3073 infants or children below 12 years. Six trials were assessed at overall 'low risk of bias', and 18 trials were assessed at overall 'high risk of bias'. The certainty of evidence according to GRADE ranged from very low to low. The trials included a heterogeneous group of children with different underlying infections such as pneumonia, meningitis, and a mix of different foci; the trials were conducted in both high-income countries and low-income countries. The types of glucocorticosteroids were hydrocortisone, dexamethasone, or methylprednisolone. Eighteen trials used placebo and six trials only used standard care as control intervention.
When meta-analyzing the trial results, visual inspection of the forest plots and test for subgroup differences showed that the effects of glucocorticosteroids seemed to differ between trials randomising participants with meningitis and trials randomising participants with sepsis of mixed focus. Hence, we chose to report results separately for each group of trials.
Meta-analysis showed no evidence of an effect of adding glucocorticosteroids to standard care for children with sepsis with a mixed focus when assessing all-cause mortality, serious adverse events, shock reversal, or adverse events. None of the trials assessed quality of life, ototoxicity, or organ failure for children with sepsis with mixed focus.
Meta-analyses suggested evidence of a bene cial effect of adding dexamethasone to standard care for children with meningitis on serious adverse events and ototoxicity. Bayes factor supported these ndings. However, Trial Sequential Analysis showed that we did not have su cient evidence to con rm that dexamethasone reduced serious adverse events by 20% or more and GRADE assessment indicated of low certainty of evidence. Meta-analyses showed no evidence of an effect of adding dexamethasone to standard care for children with meningitis when assessing all-cause mortality, adverse events, and neurological complications. No trials assessed quality of life, organ failure, or shock reversal for children with meningitis.
Dexamethasone is thought to suppress crucial in ammatory pathways responsible for meningitis (51). Accordingly, it was the glucocorticosteroid chosen in all the trials including only children with meningitis.
Our review has several strengths. Our methodology was described in detail in a protocol that was published before the literature search was initiated (10,11). We systematically assessed the risks of systematic errors through bias risk assessments, we conducted Trial Sequential Analyses to guide our GRADE assessments of levels of downgrade for imprecision, and we adjusted our thresholds for statistical signi cance to control the risks of random errors (15). We systematically used our eight-step procedure to assess if the thresholds for statistical and clinical signi cance were crossed (15). This added further robustness to our results and conclusions. Furthermore, we included a larger number of both trials and participants than any previous review (52), which gives us increased precision and power. We included enough participants to reject that adding glucocorticosteroids to standard care would reduce the risk of death by 20% or more. Moreover, the two most recent systematic reviews assessing the use of corticosteroids for sepsis among adults and children did not identify enough pediatric trials to perform meta-analysis for the pediatric population (7,52). One review included participants with community acquired pneumonia that might not have sepsis (53) and excluded trials assessing children with meningitis (7).
Our review also has several limitations. First, we chose to both include participants with sepsis and meningitis because we hypothesized that the effects of glucocorticosteroids might be similar in these two types of patients (5,6). However, based on the present results, we reached to the conclusion that pooling trials randomizing children with sepsis and children with meningitis would not be valid since the effects seem to differ. Another limitation is that most trials were at 'high risk of bias'. For all outcomes, a varying proportion of trials did not report on the patient-relevant outcomes we had prespeci ed in our protocol (5,6 (54). The types of participants and choice of glucocorticosteroids differed between the included trials, which leads to a certain degree of clinical heterogeneity. Neither did we distinct between children with different degrees of severity (e.g. PRISM, PIM, PELOD, SOFA scores). We did not reach a su cient information size for most of our outcomes to con rm or reject a bene cial or harmful effect of glucocorticosteroids. A large ongoing multicenter trial, that is planning to randomize 1032 participants, will likely be an important contribution to the assessment of the effects of glucocorticosteroids, but results are not expected before 2024 (55).
Guidelines suggest that one might use hydrocortisone for children with uid refractory and vasopressorresistant septic shock (3), but we found no evidence from randomized clinical trials to support this recommendation. No bene cial effects of glucocorticosteroids were found in children with septic shock, but only few children were randomized. Dexamethasone was the only glucocorticosteroids that seemed to show a bene cial effect for children with meningitis, however, the evidence was of low certainty.

Conclusions
Generally, we found no evidence of an effect of glucocorticosteroids for children with sepsis without meningitis.
Glucocorticosteroids (dexamethasone) seems to reduce serious adverse events and ototoxicity for children with meningitis but does not seem to have any effect on all-cause mortality. The clinical effects of glucocorticosteroids on shock reversal, and adverse events considered non-serious are unclear based on current evidence. No trials assessed quality of life and organ failure.
Based on our results, the guidelines need updating and the use of glucocorticosteroids for sepsis in children should be examined in randomized placebo-controlled clinical trials conducted at low risk of bias and low risk of systematic errors. Such trials ought to be designed according to the SPIRIT statement (56) and reported according to the CONSORT statement (57