Figure 1 shows the flow of studies through the selection process. Forty eight studies totalling 4897 patients were included (see web extra appendix C). Table 1 shows the comparisons made, the number of studies for each comparison by inpatient and outpatient population, the number of studies that provided data for our primary outcomes, the years of publication, and country of study. The drugs were administered in a variety of ways and varied across studies and interventions: corticosteroids—systemic (oral, intravenous, or intramuscular) or nebulised; adrenaline—nebulised; and bronchodilators—mostly nebulised. The risk of bias was low for eight studies (17%), unclear for 25 (52%), and high for 15 (31%). Twenty four studies only included infants aged less than 1 year.
Fig 1Flow diagram for study selection
Table 1 Overview of studies included in systematic review
Figure 2 displays the effect estimates for the primary outcome of admission rates from the emergency department (day 1) for the different direct comparisons. The results were statistically significant for only one comparison, showing a reduction of 33% for adrenaline compared with placebo (pooled risk ratio 0.67, 95% confidence interval 0.50 to 0.89). The strength of evidence for this finding was considered moderate owing to lack of precision (see web extra appendix B). The number needed to treat based on the average baseline risk of admission from all studies (20%) was 15 (95% confidence interval 10 to 45). The number needed to treat ranged from 4 (95% confidence interval 3 to 12, baseline risk 75%) to 20 (13 to 59, baseline risk 15%). The results were sensitive to risk of bias: when studies with an unclear risk of bias were removed, the pooled estimate for the two studies (n=842) at low risk of bias was no longer statistically significant (pooled risk ratio 0.77, 0.56 to 1.07). Subgroup analyses showed non-statistically significant differences between studies combining adrenaline with steroids that followed a protocol (pooled risk ratio 0.74, 0.45 to 1.23; one study, n=400) compared with those that did not follow a protocol (0.62, 0.40 to 0.94; four studies, n=520; ratio of risk ratios 1.19, 0.61 to 2.33). An effect of a similar magnitude was shown with combined adrenaline and dexamethasone compared with placebo (35%), but this did not reach significance (P=0.07) (pooled risk ratio 0.65, 0.4 to 1.05; one study, n=400).
Fig 2Results from meta-analysis of direct comparisons for admission rates from emergency department (day 1 and day 7) in outpatients. Only comparisons with quantitative results are shown
Figure 3 illustrates the comparisons and number of studies for each that were examined in the mixed treatment comparison for admissions at day 1. Cross validation showed that the results from the mixed treatment analysis were consistent with direct evidence not differing significantly from indirect evidence for any of the paired comparisons where direct evidence was available. The mixed treatment comparison identified adrenaline alone and combined adrenaline and dexamethasone as the interventions with the highest probability of being most effective, with about half the odds of being admitted from the emergency department compared with placebo (fig 4). The odds ratios were 0.48 (95% credibility interval 0.18 to 1.01) for adrenaline alone and 0.52 (0.15 to 1.57) for combined adrenaline and dexamethasone. Although this provides evidence on the relative efficacy of the different interventions, none of the interventions compared with placebo was statistically significant in this analysis.
Fig 3Comparisons (14 studies) contributing to mixed treatment analysis for admissions at day 1. Numerals within figure are studies at low risk of bias (four in total)
Fig 4Results of mixed treatment analysis for admissions at day 1, showing probability ranking and probability of being best statistic
Admission rates were also examined up to seven days after the emergency department visit using direct comparisons (fig 2). One large study with low risk of bias showed a statistically significant result for combined adrenaline and dexamethasone, with a 35% reduction compared with placebo (pooled risk ratio 0.65, 95% confidence interval 0.44 to 0.95); number needed to treat 11 (95% confidence interval 7 to 76). This result was based on a single large trial at low risk of bias; however, the study was factorial and the authors did not anticipate or hypothesise an effect for the combined adrenaline and dexamethasone group a priori.16
The strength of evidence for this finding, based on the modified GRADE system, is considered low, as evidence came from only one study with relatively few events. The overall results for steroids compared with placebo and for adrenaline compared with placebo were not statistically significant; however, subgroup analyses examining use of bronchodilators or steroids that followed a protocol showed some important effects (data not shown). Specifically, adrenaline along with steroids that followed a protocol compared with placebo and steroids showed a statistically significant reduction of 33% (pooled risk ratio 0.67, 95% confidence interval 0.45 to 0.98). Also, steroids with use of bronchodilators (adrenaline or salbutamol) that followed a protocol compared with placebo and bronchodilators showed a similar magnitude of effect (32%) but did not reach statistical significance (pooled risk ratio 0.68, 95% confidence interval 0.44 to 1.05; P=0.08).
Mixed treatment comparison for admissions up to day 7 identified steroids with bronchodilators (adrenaline or salbutamol) as the interventions with the highest probability of being most effective, although the credibility intervals were wide and do not rule out the possibility of no effect (see web extra appendix D).
Table 2 presents the results from pairwise meta-analysis for change in clinical score. Only nine of the 25 comparisons were statistically significant, and in six of these adrenaline or adrenaline and dexamethasone was the preferred treatment. Compared with placebo, significant benefits were observed for adrenaline at 60 and 120 minutes, combined adrenaline and dexamethasone at 60 minutes, and salbutamol at 60 minutes. Adrenaline showed significant benefits compared with steroids at 60 minutes and salbutamol at 3-10 days. Combined adrenaline and dexamethasone was also superior to salbutamol at 3-10 days. The other two significant results were from one small study at unclear risk of bias showing benefits of salbutamol compared with steroids. The results for other clinical variables were consistent with the findings of admission rates and clinical score or provided little additional information (data available from authors). The incidence of return visits did not differ for any of the five comparisons where data were available (steroid v placebo, steroid v adrenaline, adrenaline v placebo, adrenaline v salbutamol, combined adrenaline and dexamethasone v placebo), although there was only one or two studies within each comparison for this outcome.
Table 2 Results of direct comparisons for change in clinical score among outpatients
Figure 5 displays the effect estimates for the primary outcome of length of stay for the different direct comparisons. Only one comparison was statistically significant, showing a shorter length of stay for adrenaline compared with salbutamol (mean difference −0.28 days, 95% confidence interval −0.46 to −0.09). The strength of evidence for this comparison is considered moderate. However, the practical implications of this result need to be considered alongside the finding that adrenaline showed no significant benefit compared with placebo. Furthermore, this finding was sensitive to risk of bias: only one study for this comparison was at low risk of bias and the result was not significant (mean difference −0.07 days, 95% confidence interval −1.01 to 0.88). The results showed high strength of evidence of no difference for steroid compared with placebo overall. Subgroup analyses showed a significant difference for steroids with use of bronchodilators that followed a protocol (mean difference −0.12 days, −0.23 to −0.00); however, the magnitude of effect is not considered clinically important.
Fig 5Results from meta-analysis of direct comparisons for length of stay in inpatients. Only comparisons with quantitative results are shown
Mixed treatment comparison identified combined adrenaline and dexamethasone as the preferred treatment (figs 6 and 7). This finding was driven by one small study at high risk of bias. The confidence interval was wide and did not rule out the potential for no effect. Moreover, the mixed treatment comparison shows that none of the interventions examined show clear efficacy in terms of length of stay among the inpatient population.
Fig 6Comparisons (19 studies) contributing to mixed treatment analysis for length of stay. Numerals within figure are studies at low risk of bias (two in total)
Fig 7Results of mixed treatment analysis for length of stay, showing probability ranking and probability of being best statistic
In terms of change in clinical score (table 3), few differed statistically significantly (5/23 comparisons). Significant benefits were observed for adrenaline compared with salbutamol at 60 and 120 minutes, steroids compared with placebo at 3-6 and 6-12 hours, and salbutamol or terbutaline compared with placebo at 6-12 hours. The strength of evidence for these findings is limited by risk of bias, inconsistency (or unknown consistency owing to limited numbers of studies within individual comparisons), and lack of precision. The results for other clinical symptoms provided little additional or inconsistent information (data available from authors). Data for return visits and readmissions were available for four comparisons (steroid v placebo, adrenaline v placebo, adrenaline v salbutamol, and salbutamol v placebo). No significant differences were found although only one or two studies were available for each comparison.
Table 3 Results of direct comparisons for change in clinical score among inpatients by comparison
Sixteen studies provided data on short term adverse effects (see web extra appendix E). No studies examined, or were necessarily designed to examine, long term adverse effects, such as cognitive injury. The types of adverse effects that were most commonly searched for (or reported on) included pallor, vomiting, tremors, hypertension, tachycardia, and infections. In general, the incidence of adverse effects was low and no important differences were observed between groups across the studies.