Twenty-five randomised controlled trials (RCTs) were included in the review investigating neurodevelopmental outcomes (n=3,509 children). Few trials described randomisation methods in detail, three reported clearly adequate allocation concealment; 21 of 25 blinded the outcome assessor. Four trials had 85% or more outcome data at the last follow-up point.
At six months, scores on mental scales (six trials) and performance scales (four trials) were not significantly different between the treatment groups.
At 12 months, scores on mental scales showed significant differences in favour of the intervention (WMD 5.57, 95% CI 2.29 to 8.86; 12 RCTs), as did scores on the performance scales (WMD 5.10, 95% CI 1.44 to 8.75; nine RCTs).
At 24 months mental scales showed significant differences in favour of the intervention (WMD 7.59, 95% CI 3.51 to 11.67; seven RCTs), but the difference was not statistically significant for performance scales (four trials).
At 36 months there was a statistically significant difference in favour of the intervention for mental scores (WMD 9.66, 95% CI 5.01 to 14.31; two RCTs) but performance scores were not evaluated.
At five years (three trials) there was no statistical difference between treatment groups.
Statistical heterogeneity was present at 12 and 24 months for mental scales, but not clearly present for other outcomes or time points.
Three trials specifically studied high-risk infants and a similar range of effects was observed for available data at six and 12 months. Analyses on intervention type had a range of effect sizes but broadly reflected the main results.
Funnel plots did not show evidence of publication bias.