Seven RCTs (n=1,650) were included. The number of patients ranged from 5 to 291.
Six of the 7 included studies reported methods of randomisation, including block randomisation, computer-generated randomisation methods, and consecutive randomisation based on medical record numbers. Blinding was reported in 6 studies. Intention-to-treat analysis was performed in 5 studies.
Overall, 9% of participants developed VAP in the treatment group compared with 12% in the comparison group (7 studies; RR 0.74, 95% CI: 0.56, 0.96, p=0.03). When a random-effects model was used, there was no significant difference between the groups (7 studies; RR 0.70, 95% CI: 0.48, 1.04, p=0.08). No significant heterogeneity was found when using the Cochran Q statistic, but I-squared was 43%. When the analysis was limited to cardiac surgery patients, the results were significant in favour of treatment (2 studies; RR 0.41, 5% CI: 0.17, 0.98, p=0.04). When the analysis was restricted to 5 published RCTs, the results of the random-effects analysis were also significant (RR 0.57, 95% CI: 0.39, 0.83).
Overall, 16% of participants died in the treatment group compared with 14% in the comparison group (6 studies; RR 1.07, 95% CI: 0.76, 1.51, p=0.69).
The funnel plot did not suggest the presence of publication bias.