Nineteen studies (n=4,810 live births, range 31 to 2,415 live births) were included in the review: 18 RCTs (n=4,585) and one observational study (n=225). For most of the included RCTs, methods of randomisation, allocation concealment, assessor blinding and use of intention-to-treat analysis were assessed as adequate.
No statistically significant effect of antibiotics was identified for mortality outcomes, although the trend favoured antibiotics over controls (RR 0.90, 95% CI 0.72 to 1.12; 15 studies, n=4,265 live births).
For morbidity outcomes, statistically significant effects in favour of the intervention were identified for the outcomes of respiratory distress syndrome (RR 0.88, 95% CI 0.80 to 0.97; 13 studies, n=4,104 live births), intra-ventricular haemorrhage (RR 0.67, 95% CI 0.49 to 0.92; 12 studies, n=1,702 live births) and confirmed sepsis (RR 0.61, 95% CI 0.48 to 0.77; 13 studies, n=3,693 live births). There was a statistically non significant favourable effect for necrotizing enterocolitis.
Statistical heterogeneity was non significant for all outcomes listed (I2=0%) and so a fixed-effect model was used throughout.
The authors stated that very few data were available from low- and middle-income countries.