Data extraction
For each trial, we extracted the journal name, dates of publication and extracted general characteristics reported in the trial (e.g., number of centres involved, geographic region, sample size, power calculations, reporting of primary outcomes, number of women randomised, number of women analysed, trial conclusions, P values reported, DSMB involvement, ethics approval and funding) and details on the study group (i.e., gestational age of the pregnant women), intervention, comparison and PTB outcome.
Three reviewers (JH, LD, CA) independently assessed the full text using a standardized data extraction form developed in a pilot of 50 RCTs. After duplicate extraction of 20 RCTs by reviewer JH and LD and 20 RCTs by reviewer JH and CA the overall inter-agreement was 98% and 95%, with the item ‘information gain’ scoring the lowest inter-agreement (93%). Any discrepancies were resolved by consensus and discussion with a fourth team member (JI). After a second round of 40 duplicate extractions the inter-agreement was stable at 98% with improvement for the item ‘information gain’ to 95%. Any ambiguous item during the remaining data extractions was discussed in detail within the group.