Effect of positive end-expiratory pressure on intracranial pressure, cerebral perfusion pressure and brain-tissue oxygenation in acute brain injury: a systematic review and dose–response meta-analysis

The effect of positive end-expiratory pressure (PEEP) on the injured brain is uncertain, and existing evidence is usually reduced to a single high-versus-low contrast. We quantified the dose–response relationship between PEEP and intracranial pressure (ICP), cerebral perfusion pressure (CPP) and brain-tissue oxygenation (PbtO₂) in adults with acute brain injury (ABI). We systematically searched MEDLINE, Embase, Scopus, Cochrane CENTRAL and Web of Science to 5 July 2026 (PRISMA 2020). Within-patient PEEP-titration studies formed a primary layer; between-patient and observational studies formed a separate secondary layer. Overlapping cohorts were resolved by a pre-specified anti-duplication rule. Study-clustered multilevel meta-regression with cluster-robust (CR2) variance estimated the slope of each outcome per cmH₂O of PEEP; restricted cubic splines tested non-linearity. Robustness was examined with a one-stage dose–response model reconstructing the within-study covariance, prediction intervals, leave-one-out, leave-cluster-out analyses, publication-bias tests and pre-specified effect modifiers. Certainty was rated with GRADE against pre-anchored minimal important differences (MID; ICP 2 mmHg, CPP 5 mmHg). Of 4471 records, 37 studies were included. In the primary layer (410 patients), ICP rose + 0.118 mmHg per cmH₂O (95% CI 0.046 to 0.189; p = 0.006; 12 studies, 31 contrasts) — +0.59 mmHg per 5 cmH₂O, entirely below the 2 mmHg MID — with no departure from linearity across the sampled range of 3–21 cmH₂O (non-linearity p = 0.82), corroborated by the one-stage model (+ 0.133; 0.065 to 0.202). Evidence above 15 cmH₂O was confined to two contrasts from one study; restricted to ≤ 15 cmH₂O the slope was + 0.134 (0.071 to 0.197). CPP fell − 0.177 mmHg per cmH₂O (− 0.339 to − 0.015; p = 0.038; 10 studies, 379 patients), less than one fifth of the 5 mmHg MID; the estimate was of similar magnitude (− 0.168) when ICP, MAP and CPP were modelled in an identical set of contrasts. In that common set the ICP component was estimated with precision and the arterial component was not. Effect modifiers derived from individual studies suggested larger effects with non-recruitable lungs, lung injury/obesity and low intracranial compliance; prediction intervals crossed zero. Small-study asymmetry was not detectable. Certainty was low for ICP and low for CPP; PbtO₂ was not estimable. Across the available within-patient evidence, raising PEEP is associated with a small, linear increase in ICP that is, on average, well below accepted minimal important differences over the PEEP range studied, with a correspondingly small change in CPP. The convergence of exploratory modifier analyses generates the hypothesis that the effect is concentrated in patients with non-recruitable lungs, a stiff chest wall or reduced intracranial compliance, in whom PEEP should be titrated under multimodal neuromonitoring. This hypothesis requires prospective, phenotype-stratified testing. PROSPERO CRD420261442522 (registered 5 July 2026).