BackgroundThe aim of this investigation was to compare ventilation at different levels of positive end-expiratory pressure (PEEP) with regard to clinical important outcomes of intensive care unit (ICU) patients without acute respiratory distress syndrome (ARDS) at onset of ventilation.MethodsMeta-analysis of randomized controlled trials (RCTs) comparing a lower level of PEEP with a higher level of PEEP was performed. The primary outcome was in-hospital mortality.ResultsTwenty-one RCTs (1393 patients) were eligible. PEEP ranged from 0 to 10 cmH2O and from 5 to 30 cmH2O in the lower PEEP and the higher PEEP arms of included RCTs, respectively. In-hospital mortality was not different between the two PEEP arms in seven RCTs (risk ratio [RR] 0.87; 95% confidence interval [CI] 0.62–1.21; I
2 = 26%, low quality of evidence [QoE]), as was duration of mechanical ventilation in three RCTs (standardized mean difference [SMD] 0.68; 95% CI −0.24 to 1.61; I
2 = 82%, very low QoE). PaO2/FiO2 was higher in the higher PEEP arms in five RCTs (SMD 0.72; 95% CI 0.10–1.35; I
2 = 86%, very low QoE). Development of ARDS and the occurrence of hypoxemia (2 RCTs) were lower in the higher PEEP arms in four RCTs and two RCTs, respectively (RR 0.43; 95% CI 0.21–0.91; I
2 = 56%, low QoE; RR 0.42; 95%–CI 0.19–0.92; I
2 = 19%, low QoE). There was no association between the level of PEEP and any hemodynamic parameter (four RCTs).ConclusionVentilation with higher levels of PEEP in ICU patients without ARDS at onset of ventilation was not associated with lower in-hospital mortality or shorter duration of ventilation, but with a lower incidence of ARDS and hypoxemia, as well as higher PaO2/FiO2. These findings should be interpreted with caution, as heterogeneity was moderate to high, the QoE was low to very low, and the available studies prevented us from addressing the effects of moderate levels of PEEP.Electronic supplementary materialThe online version of this article (doi:10.1186/s13613-016-0208-7) contains supplementary material, which is available to authorized users.