First author [Ref.] | Favours bilevel n | Favours control n | MD±se | Weight % | MD (random) 95% CI |

Ambrosino 25 | 7 | 7 | 5.0000±2.4201 | 18.22 | 5.00 (0.26–9.74) |

Krachman 20 | 6 | 6 | 3.6500±10.0177 | 2.26 | 3.65 (−15.98–23.28) |

Lin 27 | 10 | 12 | 0.3000±1.3508 | 26.08 | 0.30 (−2.35–2.95) |

Maragoni 28 | 14 | 14 | 13.2000±4.3177 | 9.32 | 13.20 (4.74–21.66) |

Meecham Jones 29 | 14 | 14 | 5.9000±2.7030 | 16.44 | 5.90 (0.60–11.20) |

Nava 17 | 6 | 7 | 5.1000±1.9522 | 21.51 | 5.10 (1.27–8.93) |

Strumpf 30 | 7 | 7 | 2.0000±5.6552 | 6.17 | 2.00 (−9.08–13.08) |

Total | 64 | 67 | 100 | 4.49 (1.43–7.55) |

MD: mean difference; CI: confidence interval. Comparison: crossover trials of bilevel noninvasive positive pressure ventilation

*versus*all modalities. Outcome: arterial oxygen tension mmHg. Test for heterogeneity: Chi squared = 12.51, df = 6 (p = 0.05), I^{2}= 52.0%. Test for overall effect: Z = 2.88 (p = 0.004).