Fifteen RCTs (636 patients with COPD) were included.
The studies scored from 4 to 10 out of 11 points. None of the studies used blinding. Most studies explicitly described cointerventions (9 RCTs), specified criteria for intubation and outcome assessment (8 of 11 RCTs) and explicitly defined outcomes (8 of 11 RCTs).
Main analyses.
NPPV significantly reduced in-hospital mortality in comparison with standard therapy alone (RD 10%, 95% CI: 5, 15), based on 11 RCTs with 629 patients. No statistically significant heterogeneity was detected (P>0.2).
NPPV significantly reduced the rate of endotracheal intubation in comparison with standard therapy alone (RD 28%, 95% CI: 15, 40), based on 13 RCTs with 654 patients. Statistically significant heterogeneity was detected (P<0.001).
NPPV significantly reduced the length of hospital stay in comparison with standard therapy alone (WMD 4.57 days, 95% CI: 2.30, 6.83), based on 9 RCTs with 340 patients. Statistically significant heterogeneity was detected (P<0.001). Two RCTs that reported the results as medians found no significant difference between treatments in the length of hospital stay.
Planned subgroup analyses. In patients with severe COPD exacerbations, NPPV significantly reduced in-hospital mortality (RD 12%, 95% CI: 6, 18), rate of endotracheal intubation (RD 34, 95% CI: 22, 46) and length of hospital stay (WMD 5.59 days, 95% CI: 3.66, 7.52), compared with standard therapy alone, based on 9 RCTs. Statistical heterogeneity remained but the direction of effect was consistent among studies.
In patients with nonsevere exacerbations, there was no significant difference between treatments in hospital survival (RD 2%, 95% CI: -8, 12), intubation (RD 0, 95% CI: -11, 11) or hospital stay (WMD 0.82 days, 95% CI: -0.12, 1.77), based on 2 RCTs with 72 patients.
Post hoc subgroup analyses.
The results were similar when only studies that used endotracheal intubation or other pre-specified criteria to define NPPV treatment failure were analysed.
The proportion of people offered conventional mechanical ventilation did not influence the results.
No effect of publication status on the results was found.