Types of studies

We will include RCTs in preterm infants at risk for BPD.

Types of participants

Preterm infants (< 37 weeks' gestation) with BPD risk, as determined by individual trials’ inclusion criteria. We assume that these infants would be equally eligible to be randomized to any of the below treatments.

Types of interventions

Treatment with systemic corticosteroids (high‐dose dexamethasone, low‐dose dexamethasone, or hydrocortisone) versus control (placebo or no treatment) or other corticosteroid preparation (specifically high‐dose dexamethasone, low‐dose dexamethasone, or hydrocortisone).

We will separately consider early (within the first seven days of life) versus late (after seven days of life) treatment. We will define high‐dose versus low‐dose dexamethasone relative to the median value of cumulative dexamethasone dose in our data set (a 3.0 mg/kg cut‐point was used in previous literature; Doyle 2005). We will evaluate the included hydrocortisone studies for potential for separation into similar high‐ and low‐dose nodes; given that we anticipate few studies, however, it is likely that such partitioning will not be appropriate. We will determine high versus low contamination, or the extent to which infants received corticosteroids outside of the RCT parameters, around the median (35% in prior literature; Doyle 2005).

Analysis will be based on the intended dosing for the trial, which, for the above large contamination rates, may vary from the doses received. We will attempt to account for this study characteristic as well, with the addition of subgroup analysis by dose contamination. In the network meta‐analysis, nodes will be defined by each of the treatments (specifically, high‐dose dexamethasone, low‐dose dexamethasone, hydrocortisone, and placebo/no treatment). We anticipate no splitting or lumping of nodes.

Types of outcome measures

We chose these for their anticipated clinical relevance around the decision for corticosteroid treatment.

Electronic searches

We will search the Cochrane Library for all systematic reviews of systemic corticosteroids for the prevention and treatment of bronchopulmonary dysplasia. We will perform an updating search and update each individual systematic review to reflect any new trials' data since review publication. The updated searches are based upon the searches in the individual reviews. These include Cochrane Neonatal's standard highly sensitive search strategies along with population‐ and intervention‐specific terms to search the Cochrane Central Register of Controlled Trials, PubMed, Embase, and the Cumulative Index to Nursing and Allied Health Literature (CINAHL); see Appendix 1.

The authors of the Cochrane Reviews additionally searched clinical trials databases (ClinicalTrials.gov; the World Health Organization International Clinical Trials Registry Platform (apps.who.int/trialsearch); and the ISRCTN Registry), conference proceedings, and reference lists of retrieved articles for RCTs. All systematic reviews were updated in 2017. We will test and update the searches to ensure all recently published trial data have been found. Any new information uncovered in our searches will be referred back to the original authors of the review for updating and incorporation in our overview.

We will consider adverse effects described in included studies only.

Searching other resources

We will also review the reference lists of all identified articles for relevant articles not located in the primary search.

Selection of studies

Two review authors will perform study selection. We will independently evaluate studies for inclusion in agreement with the exclusion criteria. Studies that do not satisfy the criteria will be systematically and sequentially excluded via the exclusion categories. We will record the reason for exclusion. We will resolve disagreements between the reviewers at any stage by discussion; and where we cannot reach a decision, a third reviewer will mediate and, when necessary, make the final decision on inclusion. We will create a flow chart to graphically depict the inclusion and exclusion of studies from the initial search, to those that satisfy all criteria and that we include in the review.

Data extraction and management

Our study will pay attention to meaningful clinical groupings by building separate networks for studies evaluating early (within the first seven days of life) prophylactic treatment of corticosteroids and for studies evaluating later, more targeted, application of corticosteroids (after the first seven days of life). Two review authors will independently perform data extraction for each study. This will be performed on the review level if possible, with reference to the original manuscript when items are missing. We will include details on intended corticosteroid dosing, day of life steroids were initiated, duration of therapy, and the inclusion of rescue‐therapy with systemic steroids; in addition to the assigned trial intervention. We will resolve disagreements between the reviewers at any stage by discussion; and where we cannot reach a decision a third reviewer will mediate and, when necessary, make the final decision on inclusion. We will enter data into Review Manager 5 (Review Manager 2014), export to R software (R Core Team 2013), and check it for accuracy. If information is unclear, we will attempt to contact authors of the original reports to provide further details.

Assessment of risk of bias in included studies

Risk of bias, or the extent to which a study's results may have deviated from the truth secondary to a systemic flaw in design or execution, has already been determined and published for most included studies by Cochrane Neonatal authors. Since we continue to evaluate the same or similar outcomes for which this risk of bias was assessed, we will incorporate these assessments into our analysis. For new studies identified and included in this review, we will follow the same process of risk of bias assessment, in accordance with the criteria outlined in the Cochrane Handbook for Systematic Reviewsof Interventions (Higgins 2011). This is based on domains anticipated to impact RCT outcomes; specifically, randomization process, deviations from intended interventions, missing outcome data, outcome measurement, and selection of the reported result.

For these new studies to be included, two review authors will independently assess the risk of bias (low, high, or unclear) of all included trials using the Cochrane ‘Risk of bias’ tool for the following domains (Higgins 2011).

1. Sequence generation (selection bias)

2. Allocation concealment (selection bias)

3. Blinding of participants and personnel (performance bias)

4. Blinding of outcome assessment (detection bias)

5. Incomplete outcome data (attrition bias)

6. Selective reporting (reporting bias)

7. Any other bias

We will resolve any disagreements by discussion or by consulting a third author. See Appendix 2 for a more detailed description of risk of bias for each domain.

Summary of findings

The 'Summary of findings' tables will present evidence comparing treatment with systemic corticosteroids (high‐dose dexamethasone, low‐dose dexamethasone, or hydrocortisone) versus control (placebo or no treatment) or other corticosteroid preparation (specifically high‐dose dexamethasone, low‐dose dexamethasone, or hydrocortisone). We will define high‐dose versus low‐dose dexamethasone relative to the median value of cumulative dexamethasone dose in our data set.

We will separately consider early (within the first seven days of life) versus late (after seven days of life) corticosteroid treatment.

We will create 'Summary of findings' tables for each primary and secondary outcome. Outcomes in the 'Summary of findings' table will include:

1. death, from any cause, or bronchopulmonary dysplasia;
2. bronchopulmonary dysplasia;
3. mortality (at 36 weeks' postmenstrual age and at latest follow‐up);
4. failure to extubate (within seven days of treatment initiation);
5. cerebral palsy (at 24 months of age); and
6. presence of major neurodevelopmental disability.

We will use GRADE (Guyatt 2008), a standardized and systematic tool to rate the certainty in the evidence and strength of recommendations. As GRADE assessments were designed for application to direct evidence, we will utilize relatively new methods to apply the GRADE framework to our network meta‐analysis, to optimize clinical applicability and interpretation (Khalifah 2018).

To do so, we will begin by assessing the certainty of available direct evidence for each outcome and rate the evidence using the standard GRADE approach (Higgins 2019). The assessment will be conducted based on consideration of:

1. study design limitations (risk of bias);

2. inconsistency;

3. imprecision;

4. indirectness;

5. publication bias.

In accordance with the GRADE approach, we will evaluate the risk of bias based on:

1. random sequence generation;

2. allocation concealment;

3. blinding of participants and personnel;

4. blind of outcome assessment;

5. completeness of outcome data;

6. selective reporting;

7. other sources of bias.

We will rate each of these factors as 'high', 'low', or 'unclear' risk of bias. Having rated the risk of bias, we will use CINeMA (Confidence in Network Meta‐Analysis)—a webtool developed to compute the percentage contribution of each direct contrast to each of the network estimates (CINeMA 2017). The CINeMA tool rates the evidence from each study based on six criteria (Salanti 2014).

1. Within‐study bias

2. Across‐studies bias

3. Indirectness

4. Imprecision

5. Heterogeneity

6. Incoherence

The results of the assessments for each of the four criteria are then mapped to a final rating following the usual GRADE scale as: 'high', 'moderate', 'low', and 'very low.'

At each stage, two review authors will independently evaluate the certainty rating for the evidence (direct and indirect). We will resolve disagreements through discussion and, where necessary, through consultation with a third review author.

For ease of comparison when interpreting the relative effects of all systemic corticosteroids, the 'Summary of findings' tables include the effect estimate and certainty judgements for direct evidence, indirect evidence and the network meta‐analysis. Given the potential complexity of 'Summary of findings' tables with multiple comparisons, we will have a single 'Summary of findings’ table for each of the outcomes listed below which will be structured based on recent recommendations (Yepes‐Nuñez 2019).

If potential indications of inconsistency are encountered, we will explore study characteristics which may explain their appearance and undertake sensitivity analyses to address the issue. To assess the impact of covariates on our findings, we will explore subgroup analyses or meta‐regression adjustments (or both) (Dias 2012; Salanti 2009). We will use comparison‐adjusted funnel plots to explore for the presence of publication bias (Chaimani 2013).

Measures of treatment effect

Unit of analysis issues

Dealing with missing data

For included studies, we will note the levels of attrition. We will employ sensitivity analysis to explore the impact of including studies with high levels of missing data in the overall assessment of treatment effect. We will define a high level of missing data as 30% or greater. For all outcomes, we will carry out analyses, as far as possible, on an intention‐to‐treat basis: that is, we will include all participants randomized to each group in the analyses, and all participants will be analyzed in the group to which they are allocated, regardless of whether or not they receive the allocated intervention. We will use the number randomized minus any participants whose outcomes are known to be missing as the denominator for each outcome in each trial.

Assessment of heterogeneity

Assessment of reporting biases

We will use comparison‐adjusted funnel plots to explore for the presence of publication bias in the primary outcomes (Salanti 2009).

Data synthesis

Methods for indirect and network comparisons

We will initially generate and assess the network diagrams to determine if a network meta‐analysis is feasible. Then we will perform the network meta‐analysis within a frequentist framework, using a random‐effects model. We will perform all analyses using R statistical software, using the igraph, and netmeta packages (R Core Team 2013). See Figure 1 for a draft network.

Figure 1

---

---

Anticipated network structure. Node size is proportional to estimated number of studies, and edge width is proportional to estimated number of comparisons. Blue dotted lines represent indirect comparisons to be made.

Subgroup analysis and investigation of heterogeneity

Sensitivity analysis

For the primary outcomes we plan to perform sensitivity analysis for the following.

1. Risk of bias (restricted to low risk of bias studies only, per GRADE determination).

2. Trial size (restricted to large studies, in recognition of the greater likelihood for small studies than large or multi‐center studies to suffer publication bias). The cut‐off for defining a small study can vary between research topics, and we will determine the specific cut‐off for this sensitivity analysis based upon the median study size of the included studies.

3. Removing trials with more than 20% missing data for any relevant outcome.

We will assess differences by evaluating the relative effects and assessment of model fit.