This systematic review of 10 RCTs comparing PVL strategies to ventilation strategies designed to approach more traditional ventilatory goals in ALI and ARDS suggests that PVL reduces mortality. However, this finding was not robust in sensitivity analyses and the confidence intervals include unity, so some uncertainty remains. We did not detect dose-response interactions between treatment effect and the magnitude the differences in tidal volumes or airway pressures. However, control group ventilation strategies did not achieve the full range of traditional tidal volumes; mean tidal volumes were consistently at the lower end of the traditional range. We found no effects of PVL ventilation on barotrauma, which was an anticipated benefit. We observed more acidosis with PVL strategies and a significant increase in the use of paralytic agents.
The analysis in which we pooled survival data from trials involving 1,749 patients may represent an overestimate of treatment effect. The summary estimate suggests a 16.0% reduction in the relative risk of mortality with PVL, and the confidence intervals suggest that the relative risk of mortality might be reduced as much as 30.0%, or not at all. While the ARDS Network trial 
contributed 19.9% of the weight toward this summary estimate, the trials of Amato 
and Villar 
contributed 8.8% and 9.1% weight (total 17.9% weight), respectively. Although 3 trials 
stopped early for benefit, the ARDS Network trial 
enrolled 861 patients and contributed a large number of events (>100) in each treatment arm, and its estimate of treatment effect is therefore unlikely to be biased. However, the primary analysis is strongly influenced by 2 small trials that employed additional open lung strategies and stopped early for benefit after only a small number of events (30 hospital deaths in one trial 
, and 42 in the other 
). One trial 
used a correction for multiplicity proposed by Peto 
et al and Geller 
et al with a significance level of <0.001 while the other trial 
used a two-step stopping rule when the between group difference in ICU mortality was ≥20% with at least 45 patients in each arm. Treatment effects from these trials are likely too optimistic. While the corrections proposed for multiple sequential analyses may control for type 1 error they cannot prevent associated changes in the magnitude of treatment effect caused by early termination of small trials. 
This issue persists in the sensitivity analysis that excluded a single randomized trial 
not reported as such in the manuscript, but stated to be randomized in discussions with the first author. The summary estimate, excluding this trial, suggests a nonsignificant 14.0% reduction in the relative risk of mortality with PVL, with confidence intervals suggesting that the relative risk of mortality might be reduced as much as 28.0% or increased by as much as 4.0%. While the ARDS Network trial 
contributed 21.4% of the weight toward the summary estimate of effect in this analysis, the trials of Amato 
and Villar 
contributed 9.49% and 9.8% weight (total 19.3% weight), respectively.
Whether or not open lung strategies improve survival is a subject of ongoing controversy. A recent meta-analysis of 6 RCTs involving 2,484 patients and comparing 2 different levels of PEEP (with or without other interventions) suggested that the use of high levels of PEEP may have an independent beneficial effect on mortality with an absolute risk reduction of approximately 5%. 
Meanwhile, empirical evidence shows that early stopping for perceived benefit, particularly after few events, results in inflated estimates of treatment benefits in RCTs and in subsequent meta-analyses. 
A sensitivity analysis excluding 2 stopped-early trials 
using PVL and an open lung approach (RR 0.90; 95% CI, 0.74, 1.09; p
0.27) was inconclusive and could not rule out the possibility of important benefit or harm with PVL. Our results are congruent with the lack of apparent dose-response interactions, the lack of effect on barotrauma, and the inconsistency of study findings with respect to rates of non-pulmonary organ dysfunction.
Historically, investigators using high tidal volumes reported high rates of barotrauma in clinical practice. 
The highest barotrauma rates in this review were noted in trials that either did not impose pressure constraints or permitted high airway pressures. Recent epidemiologic studies in ALI and ARDS have shown that the incidence of barotrauma is lower than in historical series where tidal volumes were much higher. 
However, our pooled analysis did not detect a reduction in barotrauma with PVL, nor did we detect an interaction with between-group gradients in airway pressures or tidal volumes.
A notable physiological effect of PVL strategies is respiratory acidosis. Among 6 trials reporting on the evolution of arterial carbon dioxide levels there were significantly higher arterial partial pressures of carbon dioxide and lower pH levels over the first week of study. Analyses exploring a possible interaction between tolerance for acidosis and survival effects of PVL were inconclusive.
The higher rate of paralysis with PVL strategies may be related to higher rates of respiratory acidosis and ventilator dysynchrony. While early observational studies suggested that neuromuscular blockade may increase rates of ICU polyneuropathy, a recent RCT suggested that neuromuscular blockade, itself, may improve gas exchange and biological markers of lung injury. 
A follow-up trial is presently underway to evaluate the effect of paralytic agents on ARDS mortality (NCT00299650).
Pooling results in a systematic review with meta-analysis implicitly assumes that the trials are sufficiently similar with respect to populations, study interventions, measurement of outcomes and methodologic quality that one could reasonably expect a similar underlying treatment effect. While this was our assumption in pooling data across trials, we launched this review with the explicit goal of testing hypotheses to explain the differences among study results. The most prominent of the 10 trials is the ARDS Network trial 
which enrolled more patients than all of the other trials combined and stopped early after a relatively large number of events, found a significant mortality reduction with PVL and contributed the largest weight to the pooled estimate of effect for mortality in this review. While this trial galvanized a change in the management of ALI and ARDS, we reviewed it in the context of all available RCT evidence comparing the alternative approaches to ventilator management and pooled it with other trials using conservative methods.
Strengths and Limitations
This review was strengthened by following a predetermined protocol for review methods and statistical analysis. Our extensive search strategy allowed us to identify an additional 341 patients from 3 trials 
not included in prior reviews. 
We used duplicate, independent citation screening and data abstraction. We corresponded with lead investigators for each trial. In addition to critically appraising usual methodologic quality of randomized trials, we also considered design characteristics specific to this field that might lead to biased estimates of treatment effect, most notably the confounding effects of open-lung ventilation. Finally, based on between-study variation in clinical protocols and statistically significant heterogeneity, we used random effects models which take into consideration both between-study and within-study variation for pooling data across studies. Random effects models typically generate wider confidence intervals than fixed effects models in the presence of appreciable between-study variability in results. 
Overall, most trials in this review included measures to reduce bias following randomization and were of moderate quality in reporting important outcomes including mortality, barotrauma, paralysis, and dialysis (see ). However, heterogeneity among trials in adopting these measures and in reporting their results may limit interpretation of the pooled results.
This systematic review suggests that PVL strategies for mechanical ventilation in ALI and ARDS may reduce mortality and, therefore, supports the current practice to ventilate these patients with low tidal volumes. However, we did not find a dose-response effect and this borderline significant finding was not robust in sensitivity analyses. Therefore, some uncertainty regarding the effect of PVL ventilation remains.