Does a Higher Positive End Expiratory Pressure Decrease Mortality in Acute Respiratory Distress Syndrome?: A Systematic Review and Meta-analysis

All relevant randomized controlled trials of adults with ALI or ARDS receiving mechanical ventilation using two levels of PEEP with or without other interventions were considered eligible for inclusion. Trials were identified by computerized searches of the Cochrane Controlled Trials Register, EMBASE (1980 to 2008 week 20), MEDLINE (1950 to May week 1 2008), and PubMed using combinations of the following terms:

• MeSH Term-Respiratory Distress Syndrome, Adult

• MeSH term-POSITIVE PRESSURE VENTILATION

• “Adult Respiratory Distress Syndrome”

• “Acute respiratory distress syndrome” or “ARDS”

• “Acute Lung Injury” or “ALI”

• “Sepsis”

• “Positive End Expiratory Pressure” or “PEEP”

• “Airway Pressure”

In addition, the references of relevant articles were read, and backward chaining of references was used to identify additional trials. Reports not including data, nonpublished studies, reports of earlier stages of studies (where the complete trial is subsequently published), nonhuman participants, and pediatric studies were excluded.

The primary outcome measures sought were mortality and the incidence of baro-trauma. For all trials, other relevant data, including trial design, setting, numbers of patients, interventions, withdrawals, and patients lost to follow-up were collected.

Trials were assessed for the quality of allocation concealment and methodological quality. The quality of allocation concealment was rated using the method proposed by Schulz et al. ,10and methodological quality was assessed using the scoring system developed by Jadad et al.  11 

The Mantel-Haenszel method was used to calculate the relative risks and 95% confidence intervals for mortality and the incidence of baro-trauma for each trial using StatsDirect statistical software (version 2.6.7, StatsDirect Ltd., Altrincham, UK). We tested for heterogeneity between the trials using the chi-square test (P ≤ 0.05 indicating significant heterogeneity). P ≤ 0.10 as suggested by Khan et al.  12and the I²index were also used as more stringent tests for heterogeneity. I²≥ 25% is indicative of low heterogeneity, I²≥ 50% indicates moderate heterogeneity, and I²≥ 75% indicates high levels of heterogeneity.13Even though it is theoretically feasible to apply a fixed effect model as long as the statistical heterogeneity is low, it is extremely difficult in practice to interpret even the most stringent heterogeneity tests when only a small number of studies are available for analysis. We have therefore used a random effect model in the present study.12–14Combined odds ratios were also obtained. The role of publication and selection bias was estimated by visual inspection of the funnel plot for asymmetry. In addition, the data were formally tested for publication bias using Eggers regression approach15and the Begg-Mazumdar rank correlation test.16An Eggers P  value ≤ 0.10 was considered to indicate significant asymmetry and therefore possible publication bias. For the Begg-Mazumdar rank correlation test, P ≤ 0.10 was considered indicative of asymmetry and publication bias.15,16 

Database searches and backward chaining of references initially identified 328 potentially relevant articles, and the abstracts were obtained for all of these (fig. 1). After application of the inclusion and exclusion criteria, there were six randomized controlled trials4–9that met the inclusion criteria. The details of all the studies identified through this process are summarized in table 1.

The six studies included a total of 2,484 patients (1,233 in the higher PEEP level group and 1,251 in the lower level PEEP group) obtained from 102 intensive care units in nine countries. Although the causes of lung injury varied slightly between the trials, pneumonia, sepsis, trauma, acute pancreatitis, and multiple blood transfusions accounted for the vast majority of patients. The mean ages of patients included in the trials were also largely similar and ranged from 48 to 60 yr. Patients in the lower PEEP group were significantly younger in one of the trials,6and the mean age in both groups was considerably lower in another trial.7 

The methodological quality of the included studies, as assessed by the Jadad and Schulz criteria, was high. All of the trials except one had a Jadad score of 3 out of a possible 5 points (table 1). Due to the nature of the interventions, it was not possible for them to be double-blinded, which limits the total score to a maximum of 3. The lack of blinding means that it is not possible to completely eliminate detection and performance bias. The technical/practical difficulty inherent in undertaking double-blinded trials in this area should also be borne in mind in interpreting this meta-analysis. The trial that scored only 2 points on the Jadad score,6lost a point because no mention was made about whether any withdrawals occurred during the study period. Three of the trials used intention-to-treat analysis of their data.4,5,7The other three trials6,8,9either excluded withdrawn patients from their analysis or made no mention of whether data obtained in withdrawn patients were included in the analysis. Five of the trials included a sample size power calculation in the design of the study.4–7,9The study by Ranieri et al.  was not intended to assess mortality; therefore, it did not include power calculations relevant to mortality.8All of the included trials received a Schulz score of A, which shows that suitable randomization protocols were employed. Three studies used additional interventions besides the high and low levels of PEEP, which are likely to have affected the degree to which PEEP levels were responsible for any observed mortality benefits.7–9Amato et al.,  7Ranieri et al. ,8and Villar et al.  9used lower tidal volumes in the higher PEEP groups in keeping with a protective ventilatory strategy. Therefore, we performed an additional subanalysis limited to the three larger trials4–6only to determine the effect of PEEP alone on observed mortality. The effect of tidal volume could be an important confounding factor, so the main conclusions drawn from the current meta-analysis are based on the latter subanalysis only. The Begg-Mazumdar rank correlation test for all six studies demonstrated significant evidence of publication bias, with a Kendalls tau value of −0.6, (P = 0.0556). This was verified by the Eggers regression approach, which had an intercept value of −1.71 (P = 0.024). Thus, both these results indicate statistical evidence of publication bias. This was confirmed by visual inspection of a funnel plot.

The meta-analysis for mortality is shown in table 2and figure 2. Five of the studies contained data on in-hospital mortality,4–7,9but the study by Ranieri et al.  8only included data on 28-day mortality. In the current context, the two mortality figures are closely linked; therefore, we performed an initial analysis combining the two sets of mortality figures into a single meta-analysis (table 2and fig. 2), which provides a measure of “early mortality” associated with the disease/treatment. In all trials, the early mortality was lower in the group treated with higher levels of PEEP. This combined analysis of 2,484 patients with 1,233 in the higher PEEP group and 1,251 in the lower PEEP group shows that the higher PEEP group had a significantly lower early mortality than the group that received lower PEEP with a pooled relative risk of 0.87 (95% confidence interval [CI] 0.78–0.96, P = 0.007). The pooled odds ratio was 0.79 (95% CI 0.65–0.96, P = 0.0199). Exclusion of the 28-day mortality obtained from the study by Ranieri et al.  8did not make any substantial difference to the findings; relative risk for in-hospital mortality was 0.87 (95% CI 0.77–0.97; P = 0.0199), and pooled odds ratio was 0.80 (95% CI 0.65–0.98, P = 0.033). This statistically significant benefit is attributable to the disproportionate effect of the three smaller trials7–9(where high PEEP was used in conjunction with lower tidal volumes) that collectively account for less than 12% of the weighting (table 2) and may not represent the true picture. A meta-analysis restricted to the three larger studies4–6that included PEEP level as the main variable investigated was therefore then undertaken, and the results are summarized in figure 3. The pooled relative risk for in-hospital mortality of these studies was 0.90 (95% CI 0.81–1.01, P = 0.077), with a pooled odds ratio of 0.86 (95% CI 0.72–1.02, P = 0.077) in favor of the higher PEEP group. Even though this difference is not statistically significant, the Forest plot (fig. 3) shows a consistent trend towards a mortality benefit, with a 3.6% reduction in absolute risk of death. Assuming that this is reproducible, one additional life may be saved for every 28 patients treated using high PEEP.

Five studies included data on the incidence of baro-traumas. (table 3and fig. 4).4–7,9The pooled relative risk of baro-trauma was 0.95 (95% CI 0.62–1.45, P = 0.81), with a pooled odds ratio for baro-trauma of 0.91 (95% CI 0.55–1.51, P = 0.72). However, there was a degree of heterogeneity present between these trials (chi square = 8.28, df = 4, P = 0.08), with an I²value of 51.7%, indicating that there was a moderate level of heterogeneity present between these trials. Visual inspection of the Forest plot (fig. 4) indicates an overlap of all of the confidence intervals, except the study by Amato et al.  7again confirming heterogeneity. It was therefore necessary to exclude this study7and the relatively smaller study by Villar et al.  9(both trials have wide confidence intervals and account for less than 15% of the weighting; table 3), and we performed a further meta-analysis that involved the three larger trials4–6only; the results are summarized in figure 5. Although this analysis also failed to provide statistically significant evidence of increased risk of baro-trauma (relative risk 1.17, 95% CI 0.90–1.52, P = 0.25), visual inspection of the Forest plot (fig. 5) indicates a possible trend towards increased risk.

Protective ventilation strategies, which include low tidal volumes (approximately 6 ml/kg), high PEEP (>10 cm H²O or 1–2 cm H²O above the lower inflection point on the pressure-volume loop), and a plateau airway pressure of approximately 28–30 cm H²O, are currently accepted as desired end points for ventilating patients with ALI/ARDS. Dissecting out the relative merits of the individual components of this combined approach, however, is fraught with difficulties. The present meta-analysis shows that the reduction in absolute mortality risk with high PEEP alone is approximately 4%; as such, one could expect to save one additional life for every 25–30 patients treated with this strategy. The absolute risk reduction is small, so any definitive study would need to recruit approximately 3,000 patients to demonstrate statistical significance. This prospect, in our view, would pose considerable financial and ethical burdens in undertaking such a study. Even though the reduction in risk of death is small, when considered in the light of high incidence and the undisputed biologic/physiologic benefits, the use of high PEEP strategy should be considered the default option in treating patients with ARDS/ALI.

Our decision to include the three clinical trials7–9in which higher levels of PEEP were combined with variable tidal volumes in our initial meta-analysis (fig. 2), although controversial, is useful in demonstrating the fact that similar beneficial trends have been observed in very diverse populations and geographical locations. This is crucial in addressing concerns over possible adverse consequences of a high mean intrathoracic pressure on clinical outcome. Furthermore, the mortality benefits seen in these three studies7–9cannot be automatically attributed to the use of low tidal volumes alone. For example, the selection of PEEP in the study by Amato et al.  7was based on the lower inflection point of the pressure-volume curve. Patients who did not show an inflection point were also treated with a PEEP of approximately 15 cm H²O when they were randomized into the treatment group, and the pooled retrospective analysis showed that mean PEEP and driving pressures (P^Plat-PEEP) during the first 36 h, rather than low tidal volumes, were the main independent ventilator-associated variables associated with mortality benefits.7In this respect, two additional studies by Stewart et al.  17and Brochard et al.  18require further consideration. In these two studies, involving a total of 236 patients with ALI/ARDS, a low tidal volume (approximately 7 ml/kg) did not have any beneficial effects on mortality in patients who were receiving comparable levels of PEEP.17,18These three trials collectively suggest that a higher level of PEEP, which minimizes cyclical opening and collapse of alveolar units and the associated atelec-trauma, is in the very least an equally important component of the protective ventilator strategy.7,17,18For this reason, we believe that the mortality figures from our meta-analysis of all six trials (fig. 2) is relevant despite some theoretical limitations.

Though widely considered to be a distinct clinical entity, patients with a diagnosis of ARDS/ALI represent a very heterogeneous group. It is therefore relevant to consider the subgroups that may receive maximum benefit through a high PEEP strategy. The beneficial effects of PEEP are related to the prevention of atelectasis, recruitment of already collapsed alveolar units, and avoiding the cyclical opening/collapse of alveoli.2,3,19These conditions are maximal in patients with a greater lung injury score and severe lung edema.2The maximal benefit of the high PEEP strategy in patients with more severe lung injury is best evident in the study by Villar et al. ,9in which patients were recruited 24 h after meeting the ARDS criteria; as such, the study group represented a more severely ill cohort. Gattinoni et al.  2have suggested that patients with a Pao²less than 60 mmHg for longer than 1 h while being ventilated on 100% oxygen may represent this more severe end of the spectrum.2Such patients are most likely to receive an independent benefit with high PEEP; in this group, it may even be necessary to set the PEEP at approximately 15 cm H²O until the lung inflammation begins to resolve.2It is in this group of patients that the biologic/physiologic benefits achieved through the high PEEP strategy is likely to translate into mortality benefits.

Another important finding is the lack of significant differences in the incidence of baro-trauma between the two groups when all five trials were considered together (fig. 4). However, the three larger trials4–6(fig. 5) do show a nonsignificant but consistent trend towards a higher incidence. In all of the above trials, high PEEP was applied in the context of a protective ventilatory strategy, which limits the plateau airway pressure to less than 28–30 cm H²O. The definition of the term baro-trauma was variable, and only Brower et al.  6and Meade et al.  4provided an adequate description of the term. Mercat et al.  5only included patients suffering from pneumothorax in their data, which may account for the relatively wide confidence intervals and heterogeneity between trials. When considered together, current evidence suggests that the high PEEP strategy may, as expected, be associated with a higher incidence of baro-trauma. However, the benefits would far outweigh any potential disadvantages, particularly in patients with severe ARDS, as evidenced by the trend towards lower mortality (figs. 2 and 3).

The considerable overlap between the confidence intervals of the individual trials in figures 2 and 3makes it likely that the differences is the study populations are due to chance. Although all six trials in figure 2demonstrate a reduction in the mean mortality in the high PEEP group, the differences were statistically significant in only one trial, and it is relevant to note that this was the only trial that was restricted to patients with the more severe form of the disease.9Furthermore, inspection of the forest plots demonstrates that the point estimates of the relative risks of all the trials occur on the same side of the “line of no effect.” This indicates that future trials are likely to show a similar effect. It is widely recognized that, despite applying stringent protocols and statistical tests, studies included in a meta-analysis of this nature will necessarily include heterogeneous groups of patients.14In this review, heterogeneity may have arisen from the differences in the baseline characteristics of included patients, underlying causes of lung injury, and differences in many other practices that are likely to exist between different institutions. The lack of a standard definition for the term baro-trauma across the different trials may also have contributed to the moderate heterogeneity found in the baro-trauma data. Despite these differences, objective statistical tests show that these factors are unlikely to have influenced our findings.

The data included in this meta-analysis come from 102 different intensive care units in nine different countries. This study therefore shows that results from a variety of patients and backgrounds/practices follow a consistent pattern. In conclusion, the current meta-analysis suggests that the use of high PEEP may have an independent beneficial effect on mortality. Even though the effect size is small (less than 5%), the high incidence/prevalence of ALI/ARDS implies that this relatively simple and probably cost neutral intervention would save a large number of lives when translated globally. Any definitive study aimed at demonstrating statistical significance will require a sample size of approximately 3,000 patients; as such, it will pose considerable financial and ethical burdens. We consider that current evidence supports the use of high PEEP in unselected groups of patients with ALI/ARDS in general and those at the more severe end of the spectrum in particular in whom levels up to 15 cm H²O may be appropriate.