Fifteen studies met the inclusion criteria, but only 12 adequate-quality RCTs (n=4,030) were included in the final analysis.
Four studies excluded a significant number of women after randomisation. Randomisation was performed with sealed envelopes or computer-generation. In 9 studies the outcome assessors were blinded.
All studies.
Across all setting types there were significant differences in favour of AI for MAS (RR 0.47, 95% CI: 0.22, 0.99; 12 studies) and frequency of meconium below the vocal cords (RR 0.31, 95% CI: 0.17, 0.55; 11 studies). However, in both cases there was evidence of significant statistical heterogeneity. No significant differences were found between AI and control for the risk of a 5-minute Apgar score <7 and the risk of Caesarean delivery, although for the latter there was evidence of significant statistical heterogeneity.
Standard peripartum surveillance settings.
There were no significant differences between AI and control groups for MAS (10 studies; significant heterogeneity), the incidence of a 5-minute Apgar score <7 (7 studies; no heterogeneity), and Caesarean delivery (10 studies; significant heterogeneity). However, in comparison with control, AI did significantly reduce the frequency of meconium below the vocal cords (RR 0.29, 95% CI: 0.14, 0.57; 9 studies) and neonatal acidosis (RR 0.62, 95% CI: 0.40, 0.96; 7 studies). However, in both cases there was evidence of significant statistical heterogeneity.
Limited peripartum surveillance settings.
Compared with control, AI was associated with a significant reduction in the risk of MAS (RR 0.25, 95% CI: 0.13, 0.47; 2 studies), the frequency of meconium below the vocal cords (RR 0.42, 95% CI: 0.21, 0.83; 1 study), and the incidence of a 5-minute Apgar score <7 (RR 0.36, 95% CI: 0.18, 0.72; 2 studies); there was no evidence of statistical heterogeneity. No limited peripartum surveillance studies assessed the risk of neonatal acidosis. There was no significant difference in the risk of Caesarean delivery between AI and control groups.
There was no evidence of publication bias from the funnel plot (Egger’s test, p=0.137).