Nine RCTs (n=654) were included in the meta-analysis: all trials were randomised; three reported allocation concealment; none were blinded. Co-interventions were reported in four trials. Intention to treat analysis was used in two trials. Outcome assessment criteria were implemented in eight trials. And eight trials showed comparability of groups at baseline. Follow up of more than seven days was reported in two trials.
In the pooled analysis of nine trials (n=654), the addition of continuous positive airway pressure produced a statistically significant lower rate of postoperative pulmonary complications than standard treatment alone: risk reduction 0.34 (95% CI: 0.15, 0.48); the number needed to treat to benefit was 14.2 (95% CI: 9.9, 32.4). Significance remained when one highly weighted study was excluded, and when the potential bias of missing intention to treat analysis data in another study was explored. Heterogeneity was low ( I 2 = 14%). Publication bias was indicated in the funnel plot, but was not confirmed by Egger's test.
In the pooled analysis of five trials (n=175), the addition of continuous positive airway pressure to standard therapy produced a statistically significant lower rate of atelectasis: risk reduction 0.25 (95% CI: 0.03, 0.42); the number needed to treat to benefit was 7.3 (95% CI:4.4, 64.5). Significance remained when one one highly weighted study was excluded.
The pooled analysis of four trials (n=506) showed that the addition of continuous positive airway pressure to standard therapy produced a statistically significant lower rate of pneumonia: risk reduction 0.67 (95% CI: 0.25, 0.86); the number needed to treat to benefit was 18.3 (95% CI: 14.4, 48.8). Heterogeneity in both analyses was reported to be negligible, although this was based on a small number of studies. Publication bias was noted in the analysis of atelectasis (p=0.003).
A beneficial effect was reported for postoperative continuous positive airway pressure on intubation rate (risk reduction 0.85, 95% CI: 0.34, 0.97; two trials, n=413). The effect on mortality rate was not possible to determine due to the small number of events.
Fixed-effect analyses produced comparable results to the random-effects analyses.