Guerin 2004.

Study characteristics
Methods	Prospective, unblinded, multi‐centre, randomized controlled trial over 48 consecutive months. Randomization was computer‐generated and was done separately for each ICU, with participants to supine (n = 378) or prone (n = 413) position ventilation
Participants	791 participants 593 males, 198 females Inclusion: mechanical ventilation (oral or nasal tracheal intubation or tracheostomy), PaO2/FIO2 ≤ 300, ≥ 18 years of age, expected duration of mechanical ventilation > 48 hours, written informed consent from next of kin Exclusion: prone position for ≥ 6 hours per day in the 4 days preceding enrolment, contraindications to proning (ICP > 30 mmHg, cerebral perfusion < 60 mmHg, massive haemoptysis, bronchopleural fistula, tracheal surgery or sternotomy in the past 15 days, MAP < 65 with or without vasopressors, DVT, pacemaker inserted for fewer than 2 days, unstable fracture), therapeutic limitation indicated in the first 24 hours of ICU admission, high risk of death in the next 48 hours, chronic respiratory failure requiring mechanical ventilation and inclusion in another protocol with mortality as a primary endpoint
Interventions	Participants were randomly assigned to the supine or the prone group, in which they were placed in a prone position for ≥ 8 hours per day Tidal volume: 10.1 mL/kg IBW* (~ 95% CI 5.5 to 14.7 mL/kg IBW). *Imputed from measured body weight data (Bloomfield 2006)
Outcomes	Primary: 28‐day mortality Secondary: 90‐day mortality, duration of mechanical ventilation, rate of ventilator‐associated pneumonia and oxygenation
Notes	VAP was well defined; 11 of 802 participants (1.4%) recruited, lost from final analysis Trial commenced in 1998, was completed in 2002 and was published in 2004 Mean of 36.9 hours prone per participant
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Detailed methods provided
Allocation concealment (selection bias)	Low risk	Detailed methods provided
Blinding of participants and personnel (performance bias) All outcomes	High risk	Not blinded, and assigned treatment readily identified
Blinding of outcome assessment (detection bias) All outcomes	Unclear risk	Risk of assessment bias different for different outcomes. Mortality has low risk of bias; other less well‐defined outcomes have higher risk of outcome assessment bias
Incomplete outcome data (attrition bias) All outcomes	Low risk	11 of 802 participants (1.4%) recruited, lost from final analysis
Selective reporting (reporting bias)	Low risk	Multi‐centre trial ‐ conduct and outcome measure likely agreed in advance
Other bias	High risk	Very high cross‐over rates reported: "At day 28, 83 (27.9%) of 297 patients in the supine group died, 36 (44.4%) of the 81 patients who had crossed over from the supine group died, 76 (31.3%) of 243 patients in the prone group died, and 58 (34.1%) of 170 patients who crossed over from the prone group died (P value = .85)". Overall, 32% of participants in the trial were crossed over to the opposite limb of the study. This level of cross‐over events makes reported effects difficult to interpret. This level of selective cross‐over of participants impairs the statistical power of the study and leads to bias against a positive result (Lipsey 1990; Porta 2008)