Data synthesis and statistical analysis
We conducted the statistical analyses in accordance with Cochrane
guidelines.19
We estimated odds ratio (OR) for dichotomous outcomes and mean
difference (MD) for continuous outcomes, along with their 95%
confidence intervals (CI) trough pairwise meta-analysis for direct
comparisons. Heterogeneity was quantified with the I2statistic20 (30-60%
was considered ’moderate’ heterogeneity). We used a random-effects model
and tested subgroup differences (P< 0.05 or I² >
30%).
We performed a Bayesian random-effects NMA to estimate treatment effects
and 95% credible intervals (CrI), if the between-study homogeneity,
transitivity and coherence assumption across treatment comparisons were
judged to be
justifiable.21,22,23,24We explored the network geometry and connectivity using network
diagrams.
We assessed the statistical heterogeneity of the entire network by the
heterogeneity variance (τ2) considering the empirical
distribution.25
Our prespecified subgroup analyses included gestational age at trial
entry (24-28 , 29-34 , 35-37 weeks); intact vs ruptured membranes and
country income level: LMIC vs
HIC26. We
performed sensitivity analysis by low-moderate overall quality of the
studies and by using masked treatments. We performed network
meta-regression based on gestational age at entry, GNI per capita and
the year of publication.
We assessed small-study effects and publication
bias,27 We estimated
SUCRA values with their
CrIs28,29in a rank-heat plot.30NMA were conducted in OpenBugs (version
3.2.3)31 and pairwise
meta-analysis in RevMan
5.3.32.
We assessed the confidence in the estimates by outcome using the GRADE
approach and specific criteria for intransitivity (based on potential
effect modifiers) and incoherence (based on the statistical
consistency).33,34Two authors (AC, IDF) independently graded the certainty of the evidence
(CE), and differences were resolved by consensus.
Additionally, we conducted a focus group to reflect patients’
perspectives in the discussion (Appendix S3 ).