Seven RCTs (n=290) were included.
Allocation concealment was clearly adequate in 4 studies, but only one used double-blinding. The mean Jadad score was 3 (range: 2 to 5). The meta-analysis may have been underpowered to show a modest difference in effect between the interventions (e.g. 40% difference in risk of requiring intubation).
There was no statistically significant difference between the groups for hospital mortality, requirement for invasive ventilation, duration of noninvasive ventilation required until resolution of pulmonary oedema, length of hospital stay or new-onset myocardial infarction (7 RCTs, n=290). Based on limited data (4 RCTs, n=167), there was a non significant trend towards increased risk of new-onset myocardial infarction in the CPAP group (RR 2.10, 95% CI: 0.91, 4.84, p=0.08; I2=25.3%). Stratification by level of airway pressure had little effect on the results, nor did sensitivity analyses including only studies of patients with marked hypercapnia. There was no statistically significant heterogeneity in any of the analyses.
The funnel plot showed some potential for publication bias, but it was of a small degree deemed unlikely to affect the overall findings.