Nineteen RCTs (1,548 children) were included.
Most studies were underpowered. Four studies had adequate concealment of randomisation, 7 studies used adequate sequence generation, and 15 studies reported or gave the impression that no exclusions had taken place.
Ondansetron was associated with a significantly lower risk of POV than droperidol or metoclopramide.
There was no evidence of statistical heterogeneity for ondansetron, compared with either metoclopramide (p=0.25) or droperidol (p=0.14 for POV and p=0.10 for PONV).
The drug dose, study quality and type of outcome (POV or PONV) were not associated with effect. It was unlikely that the results were influenced by publication bias.
Ondansetron versus droperidol (8 RCTs with 563 children): the pooled RR of POV using a random-effects model was 0.67 (95% confidence interval, CI: 0.49, 0.90, p=0.0092). The pooled RR of PONV was 0.70 (95% CI: 0.53, 0.92, p=0.010). The fail-safe N was 17. The number- needed-to-treat with ondansetron was estimated as 11, assuming the risk of emesis with droperidol was 30%.
Ondansetron versus metoclopramide (13 RCTs with 985 children): the pooled RR of POV using a random-effects model was 0.56 (95% CI: 0.44, 0.71, p<0.001). The fail-safe N was 19. The number-needed-to-treat with ondansetron was estimated as 8, assuming the risk of emesis with metoclopramide was 20%.