Airway administration of corticosteroids for prevention of bronchopulmonary dysplasia in premature infants: a meta-analysis with trial sequential analysis

This systematic review was conducted and is reported according to the recommendations of the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement [26]. Electronic searches were carried out in multiple databases, including PubMed, Web of Science, Embase, Cochrane Library, Clinicaltrials.gov, Controlled-trials.com, Google scholar, VIP, WangFang and proceedings of the Pediatric Academic Society meetings (from 1980) on December 31, 2016 for relevant studies. Search terms included: preterm infant, premature infant, infant-newborn, bronchopulmonary dysplasia, chronic lung diseases, anti-inflammatory agents, neurodevelopmental outcomes, steroids, glucocorticoids, corticosteroids; administration, inhalation; aerosols, budesonide, beclomethasone, dipropionate, flunisolide, and fluticasone propionate. No language restriction was applied. The protocol of this systematic review was registered before the literature search in PROSPERO (Prospero2016 CRD42016054098) [27].

Studies had to meet the following criteria: 1) randomized controlled trials; 2) infants were randomized to receive treatment with an airway administration (inhalation or instillation) of corticosteroid vs placebo or systemic corticosteroids; 3) reported more than one of the following outcome parameters: primary outcome of BPD (defined by the need for supplemental oxygen or positive pressure support at 36 weeks PMA) or/and death; secondary outcomes of the use of systemic corticosteroids, effect on extubation and not extubate (including extubated within 14 days and duration of mechanical ventilation), adverse outcomes of sepsis, hyperglycemia requiring treatment, intraventricular hemorrhage (IVH), periventricular leukomalacia (PVL), necrotizing enterocolitis (NEC) Bell’s stage > or = II, retinopathy of prematurity (ROP), PDA requiring drug treatment or surgery, and neurodevelopmental outcomes of neurodevelopmental impairment and cerebral palsy. Exclusion criteria were: a) non-clinical studies (experimental and basic studies); b) observational or retrospective studies; c) duplicate reports or secondary or post hoc analyses of the same study population; d) lack of sufficient information on baseline or primary or secondary outcome data; and e) therapeutic study.

Two reviewers (Zhang and Zhong) independently assessed the risk of bias of individual studies and the bias domains across studies using the Cochrane collaboration tool [28]. All discrepancies were resolved by discussion and consensus. The studies were rated to be at high risk of bias, low risk of bias, or unclear risk of bias based on sequence generation, concealment of allocation, blinding of participants/parents and personnel, blinding of outcome assessment, incomplete outcome data, and selective outcome reporting.

The statistical analyses were performed by the DerSimonian and Laird method (random-effect model) using the Review Manager meta-analysis software (version 5.3, 2012; The Cochrane Collaboration, Copenhagen, Denmark). TSA viewer version 0.9 β was used for trial sequential analysis. Treatment effect estimates for all trials were calculated, expressed as typical relative risk (RR) for dichotomous outcomes and weighted mean difference (WMD) for continuous outcomes, all with a 95% confidence interval. If the continuous measures were reported in median and inter-quartile range, mean and standard SD values were estimated using the method described by Wan et al. [29]. P values of ≤0.05 and RR point estimate 95% CIs that excluded the null (<1.00 or >1.00) were considered statistically significant. Where the pooled estimates of relative risk were statistically significant. We calculated the numbers needed to treat (NNT) for all outcomes. The heterogeneity between-trial regarding treatment effects was analyzed by the χ² test for heterogeneity and I ² statistics. Heterogeneity was deemed significant when the corresponding p-value was <0.1 or when the I ² percentage was >50 [30]. Subgroup analyses were carried out to assess the source of heterogeneity. When more than 10 articles were included, the presence of publication bias was assessed and displayed through a funnel plot. Analysis by excluding studies at high risk of bias or abstract forms was part of a predefined subgroup analysis [28]. Sensitivity analysis with trial sequential analysis was performed to correct for random error and repetitive testing of accumulating and sparse data; meta-analysis monitoring boundaries and required information size (meta-analysis sample size) were quantified, along with D2 (diversity adjusted information size) and adjusted 95% CIs [31‐34]. Risk of type 1 error was considered 5% with a power of 80%. A clinically meaningful anticipated relative risk reduction was used based on the low bias trial [16]. Trial sequential analysis 95% CI boundaries that excluded the null (<1.00 or >1.00) were considered statistically significant.

Our search identified 121 potentially relevant articles during the initial electronic database search, of which 35 RCTs were involving the use of AACs in premature infants. However, only 25 (27 articles) of 35 trials presented appropriate and sufficient data for further analysis [16, 20, 24, 25, 35‐57], and the remaining eight were excluded because of therapeutic study of inhaled steroids for BPD [18, 19, 58‐61] and when the raw data in Ref. [62, 63] is lacking due to unable to reach the original investigators. Overall, the 20 trials of airway administration of corticosteroid (16 trials of inhaled corticosteroid and 4 trials of instillation of steroids) vs placebo included 2484 infants, and 5 trials of inhaled corticosteroid vs systemic corticosteroids included 765 infants. Two studies were published in abstract form [20, 47]. Two articles [50, 55] were follow-up studies of two included trials [49, 54]. Fig. 1 shows the details of the selection process.

The 25 RCTs (27 articles) selected for analysis included a total of 3249 participants (Tables 1 and 2, Table S1-S2 in Additional file 1: Table S1 and Additional file 2: Table S2) [16, 20, 24, 25, 35‐57]. The publication dates of the RCTs ranged from 1993 to 2016. The AACs group vs the placebo group or systemic corticosteroids group were well matched; birth weight and gestational age did not differ significantly significance. Other aspects of respiratory treatment, including the resuscitation devices used and the criteria for using antenatal glucocorticoids as well as surfactant, were adequately described in the studies and conformed to current international guidelines [64‐66]. The incidence of neonatal respiratory distress syndrome (was diagnosed based on respiratory symptoms and corresponding X-ray changes) was comparable among the airway administration of corticosteroid group and comparative groups.

Although all studies attempted to include infants at risk of developing BPD, the inclusion criteria, the intervention (type of AACs), and duration of therapy varied between studies. The intervention regimens,including the prescribed dosages and administration schedules, varied considerably between the RCTs. Overall, the duration of AACs treatment ranged from 3 to 28 days, and the types of AACs included beclomethasone, budesonide, fluticasone, and dexamethasone. Delivery systems included metered dose inhaler (MDI) with a spacer device, nebulization, and airway instillation. Four studies used airway instillation of budesonide using surfactant as a vehicle [25, 36, 48, 49].

Three studies were deemed to have a low risk of bias [16, 24, 25], seven studies were deemed to have a high risk of bias [36, 38, 53‐57] due to selection bias, detection bias, attrition bias or performance bias, and 17 studies were classified as having an unclear risk of bias [20, 35, 37, 39‐52]. Although all studies were presented as randomized trials, the method of randomization was determined to be inadequate in 17 studies [20, 35, 37, 39‐48, 51, 53, 56, 57]. Sixteen studies were found to have adequate concealment of allocation and clearly described blinding for the intervention method [16, 20, 24, 25, 38, 39, 41, 42, 44, 45, 47, 49‐51, 56]. Thirteen studies were found to have adequate blinding of outcome assessment [16, 24, 25, 35, 37, 38, 40, 44, 46, 49, 50, 52, 56]. Follow-up data were reported in five studies [24, 25, 41, 50, 56]. Fig. 2 and 3 summarise the risks of bias.

Our primary outcomes, were robust to sensitivity and trial sequential analyses. When evaluating the incidence of BPD, death, and death or BPD between AACs and placebo, consistent results was observed in trial sequential analyses and after exclusion of studies at high risk of bias, small sample, non-English literature, abstract, or studies of instillation of steroids using surfactants as carriers. In addition, sensitivity analysis showed that NNT of 5 and 4 in subgroup of instillation of steroids using surfactant as a vehicle was lower than the NNT in the inhaled corticosteroids group (NNT = 14 and 20) for preventing one case of BPD and BPD or death (Table 3). Using as a vehicle surfactant may also enhance the solubility of budesonide, increase budesonide absorption, and strengthen the function of anti-inflammatory effects [25, 67]. Trial sequential analysis corrected the 95% CIs of the already statistically significant point estimates for main comparison of an airway administration of corticosteroid with the placebo through accounting for random error and repetitive testing of accumulating sparse data. In our analysis of incidence of BPD and death or BPD between AACs and placebo, the cumulative z score entered the futility area, indicating further trials are not required. Considering the overall morbidity, no association with benefit or harm can be found between groups, but the trial sequential analysis suggested it would be unnecessary to carry out more trials on this outcome. In our analysis of neurodevelopmental outcome between AACs and placebo, AACs did not increase neurodevelopmental impairment. Futhermore, inhaled corticosteroid also did not increase mortality and is similar in preventing BPD compared with systemic corticosteroids. The information size for the comparison of inhaled corticosteroids with systemic corticosteroids was too low to require futility boundaries in trial sequential analysis. European Consensus Guidelines on the Management of Respiratory Distress Syndrome do not recommend the use of inhaled corticosteroid for preventing BPD until further safety data become available [68]. The benefit of preventing BPD and death or BPD associated with AACs found in this systematic review could inform future updates of these and other clinical guidelines.

Several systematic reviews have evaluated inhaled corticosteroids in the prevention of BPD [14, 15, 17, 21, 22]. A recent meta-analysis aimed to assess inhaled corticosteroids for BPD in preterm infants [17]. The evidence from this meta-analysis supported inhaled corticosteroids, especially budesonide, as a potentially efficacious and safe therapy for the prevention or treatment of BPD in preterm infants [17]. However they fail to implement sensitivity and subgroup analysis for the result of heterogeneity. In addition, they did not evaluate the advantage on inhaled corticosteroids compared with sysetmic corticosteroids. The previous Cochrane reviews examined either inhalation or systemic corticosteroids, further divided into early and late phases according to the time of administration. Hence, the analysis used 1 week after birth as a boundary [21, 22] However, in our meta-analysis, the duration of inhalation of corticosteroids ranged from 3 to 29 days, which was different from studies on intravenous administration, for which the course was about 1 week. For a long course of AACs, the boundary of 1 week after birth is sufficient to reflect the differences of AACs. Studies set the first delivery time from 12 h to 14 days after birth and the time span was long (Table 1). Therefore, we hypothesize that the subgroup analysis with 1 week after birth as the boundary is not reliable, and it is difficult to follow. The dose and duration of the AACs in each study were not the same, making it difficult to determine the exact dose and the course of corticosteroids that is supposed to be beneficial even via subgroup analysis. We believe that a unified drug delivery time and course of treatment are very important for developing precise treatment and firm recommendations.

The results of our analysis confirm those of the previous systematic reviews while improving their precision and further reducing the role of contingency. A major strength of present systematic review is the use of the robust Cochrane methodology [26, 28, 30], and the meta-analysis is further challenged with trial sequential analysis to correct for random error and repetitive testing, which often is biased towards an intervention [31‐33]. Our meta-analysis did focuse on bias and quality of evidence of included studies in accordance with GRADE [69]. Predefined sensitivity, subgroup analysis, and trial sequential analyses that included assessment of bias and clinically heterogenous groups were presented to aid healthcare professionals for clinical decisions. However, subtle underlying bias of the trials included in the review remains a possible limitation, as in any other systematic review although we excluded the studies that are at high risk of bias.

Admittedly, several limitations in our meta-analysis might have affected the interpretation of findings. First, the trials analyzed differed in their study design and clinical characteristics of the study participants. The effect on mortality and BPD might be quite different between very preterm infants (GA: 28~32 weeks) and extremely preterm infants (GA < 28 weeks), and subgroup analysis based on GA or BW could not be done due to the lack of individual patient data. For extremely preterm infants, the BPD definition is based on respiratory support and oxygen requirements at 36 weeks, which may be dependent on local practices and saturation targets. Despite attempts to standardize the definition of BPD, wide variation among centers has been reported with the diagnosis of BPD ranging from 6 to 57% depending on the definition chosen [70]. Hence the target population in each studies may be very different. The duration, dose, type, and inhalation or instillation of steroids using surfactants as carriers also were inconsistent across studies. Although we validated the stability of the result of our meta-analysis by subgroup analysis and sensitivity, the direct comparison between inhalation and instillation steroids using surfactant as a vehicle was lacking. Furthermore, there were few studies on AACs compared with systemic corticosteroids, and not all trials presented the primary and secondary outcomes, be especially neurodevelopmental outcomes. To some extent, the clinical diversity still potentially compromises the validity of the current results, even though we have pooled the data as if they were from clinically homogeneous studies. Third, formal tests for publication bias are still lacking. In principle, this should not interfere with the meta-analysis results, because publication bias generally results in an overestimation of the effect estimates.