The “Open Lung Approach” (OLA) or “Protective Ventilation” are considered optimal strategies for patients with ARDS. This includes application of high levels of positive end-expiratory pressure (PEEP) to achieve lung recruitment coupled with limited tidal volumes (VT) or ventilator distending pressures. Occasionally, additional strategies such as prone positioning or recruitment manoeuvres (RM) are applied to help recruit the lung.

Based on the evidence of ARDSnet trial,1 strong agreement exists to support one component of the OLA, limiting VT and ventilator distending pressures. However, the evidence of benefit for high versus low PEEP is less clear. Many previous meta-analyses have evaluated the effect PEEP levels, with a suggestion of potential benefit of high PEEP strategies but notably with high statistical heterogeneity.2–4 The sources of heterogeneity are multiple and include factors related to the rules for PEEP management, levels of PEEP used to label an intervention as “High-PEEP”, differing inspiratory pressure and tidal volume limits, and patient related heterogeneity in ARDS severity and lung recruitabilty. As such, understanding the potential impact that these sources of heterogeneity have on the results of these trials is crucial before undertaking additional RCTs in this domain.

Since Amato MB et al.5 trial of 1995, PEEP is used more liberally as a routine strategy in mechanical ventilation (MV) of ARDS patients. Interestingly, this phenomenon was first observed in the control group of ALVEOLI trial,6 were the PEEP/FiO2 protocol for the intervention group was changed because there was little separation between intervention and control patients. Furthermore, after the ARDSnet trial1 was published in 2000, VT and distending pressures limitation became standard of care. We hypothesize that this evolving philosophy in the use of PEEP and VT in the control group is a significant source of heterogeneity among trials in evaluating the effect of OLA on mortality.

Also several modes of OLA strategy have been described, based on different forms of setting the high values of PEEP: above the lower inflection point of the static pressure-volume curve, using tables that fix mandatory PEEP/FiO2 scales, guided by the oesophageal pressure, due to the best compliance region, the effects of RMs, etc. This could also have had an influence on the heterogeneous effect of the OLA on ARDS mortality.

There is clear evidence that Driving Pressure (DP) is associated with excess mortality in ARDS patients.7 And recently Gattinoni et al.8 proposed that Mechanical Power (MP), a measure of the energy applied to the lung by the ventilator per unit of time, plays the main role in ventilator induced lung injury (VILI).9 Changes to PEEP can increase or decrease DP or MP. We hypothesize this is another major source of heterogeneity. Previous meta-analyses have incompletely explored all these sources of heterogeneity.

The main objective of this systematic review and meta-analysis was to determine the effect of OLA strategy on mortality of ventilated ARDS patients. We specifically sought to identify the most likely moderators of this effect, including variables such as modality of OLA strategy applied to the experimental group, the level of recruitment achieved in the control group, and the balance between MP or DP applied to both cohorts of patients in the randomized trials. We hypothesized that the most important moderators would relate to the degree of recruitment achieved in the control group, and differences in MP between control and intervention group.

This work was performed according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement,10 and the guidelines proposed by the Cochrane Collaboration in the Cochrane Handbook,11 was registered in the International Prospective Register of Systematic Reviews (PROSPERO, CRD number: CRD42020179778).

Of the articles examined, seven RCTs were excluded: one because it was duplicated publication,5 one because it was done on normal (not ARDS) lungs,18 one because experimental arm included low Vt without high PEEP,1 one because experimental arm included mandatory prone position,19 and three because experimental arm included only RMs.20–22

Fourteen RCTs met inclusion criteria (Fig. 1), including 4.237 ventilated adults with ARDS, of whom 1,570 (37.05%) died. Table 1 Apple6,23–35 summarizes baseline patient characteristics and the various OLA strategy protocols used among included studies. GRADE quality of evidence for mortality was downgraded (Table 2) because of imprecision, and because evidence of publication bias was detected. Fig. 2 shows the Funnel plot with a non-symmetric visual appearance. The Egger regression test was statistically significant (p=0.0405), and the Trim and Fill method detected that 2 studies could be missing.

Figure 1.

PRISMA 2009 (11) Flow diagram of the study selection process. PEEP: positive end-expiratory pressure. VT: tidal volume. RMs: recruitment manoeuvres. Exp.: experimental.

(0.3MB).

Table 2.

The GRADE quality of evidence for the outcome 28–30th day mortality.

Certainty assessment							Summary of findings
No of participants (studies) Follow-up	Risk of bias	Inconsistency	Indirectness	Imprecision	Publication bias	Overall certainty of evidence	Study event rates (%)		Relative effect (95% CI)	Anticipated absolute effects
							With control ventilation	With OLA strategy		Risk with control ventilation	Risk difference with OLA strategy
28–30th day mortality
4237 (14 RCTs)	Not serious	Not serious	Not serious	Seriousa	Publication bias strongly suspectedb	LOW	806/2134 (37.8%)	764/2103 (36.3%)	RR 0.90 (0.78 to 1.03)	378 per 1.000	38 fewer per 1.000 (from 83 fewer to 11 more)

CI: confidence interval; RR: risk ratio.

Explanations:

a

95% confidence intervals (CIs) around effect estimates crossed the threshold between recommending and not recommending a treatment (which we a priori defined as a relative risk (RR) of 1.0 for the outcome of mortality).

b

We did Egger regression and funnel plot analysis, and publication bias was suspected. With the fill trim method, we expected 2 trials missing.

Figure 2.

The Funnel–Plot. There is visual evidence of non-symmetric appearance. Egger regression test for funnel plot asymmetry: t=-2.2960, df=12, p=0.0405. Trim and Fill method: Estimated number of missing studies on the right side: 2 studies (SE=2.5634).

(0.1MB).

In ventilated adult patients with ARDS, pooled analysis in the Random Effect Model (REM) did not show a significant difference in the 28-day mortality between OLA strategy and control groups: RR=0.90; 95% CI=0.78–1.03; z-val=−1.513; p=0.1303 (DerSimonian–Laird method) or RR=0.89; 95% CI=0.77–1.05; t-val=−1.466; p=0.1664 (Hartung–Knapp–Sidik–Jonkman method). The analysis detected statistical evidence for heterogeneity among the studies (chi-square Q(df=13)=24.8607, p-val=0.0241, I2=48%), graphically represented in L’Abbé plot (Fig. S1, online supplement). The details are shown in Fig. 3, that includes risk of bias evaluation. Overall risk of bias was considered to be low. Fig. S2, online supplement shows the cumulative meta-analysis, highlighting that the beneficial effects of the OLA appear to be diminishing over time.

Figure 3.

(A) Forest plot of comparison: OLA strategy vs. control ventilation, outcome: 28–30th day mortality. Includes risk of bias summary for each included study. (B) Risk of bias graph: review authors’ judgements about each risk of bias item presented as percentages across all included studies.

(0.38MB).

Table 3 shows the results of two prespecified subgroup sensitivity analysis. In the first, mortality was significantly lower in the OLA group compared to the control group only when the OLA strategy was based on setting PEEP based on the lower inflection point of the P-V curve. In the second, mortality was significantly lower in the OLA group only when VT was not limited in control group.

Meta-regression

Other possible causes of heterogeneity between studies were explored by meta-regression. We had complete data for all necessary variables in twelve RCTs. Seven meta-regression models were fit. The first six models used one of the above-mentioned variables as the hypothesized moderator, representing possible candidate theories to explain the relationship between the OLA strategy and mortality. The seventh model combined the two more probable candidates. Table 4 shows the results of model selection process.

Table 4.

Model selection process.

Model	Test for residual heterogeneity	Log likelihood	Deviance	AICc	AICc differences	Model likelihood	Akaike weight	WOE against (decibans)
M1:ln(RR)=0.1785−0.0213*expPEEP_1	QE(df=10)=24.0433; p-value=0.0075	−2.8910	5.7821	15.7821	3.3089	0.1912	0.0554	−7.19
M2:ln(RR)=−0.2105−0.0005*controlPF_0	QE(df=10)=22.3847; p-value=0.0133	−3.3029	6.6058	16.6058	4.1327	0.1266	0.0367	−8.97
M3:ln(RR)=−1.8348+0.0105*controlPF_3	QE(df=10)=13.5333; p-value=0.1954	−1.2366	24731	12.4371	0	1	0.2901	0
M4:ln(RR)=0.1134−0.0049*gradPF_3	QE(df=10)=24.2398; p-value=0.0070	−2.5170	5.0340	12.4731	2.5609	0.2779	0.0806	−5.56
M5:ln(RR)=−0.8806+0.8909*RelativeDP_1:	QE(df=10)=21.5973; p-value=0.0173	−1.3468	2.6937	12.6937	0.2205	0.8956	0.2599	−0.48
M6:ln(RR)=−0.8119+0.5952*RelativeMP_1	QE(df=10)=21.3908; p-value=0.0185	−1.3041	2.6083	12.6083	0.1351	0.9346	0.2712	−0.29
M7:ln(RR)=−1.6211+0.0062*controlPF_3+0.4046*RelativeMP_1	QE(df=9)=12.7636; p-value=0.1736	−1.1320	2.2640	20.2640	7.7908	0.0203	0.0059	−16.92

AICc: Second-Order Bias Corrected Akaike Information Criterion. AICc differences: AICc−minimum AICc of the set. Akaike Weight: Probability that every model is the “best” model (in the expected Kullback–Leibler discrepancy sense). WOE against: Weight of Evidence (in decibans) against every model, relative to the best model.

Model M3, using PaO2/FiO2 in the control group on day 3 of ventilation, had a probability of 0.290 of being the “best” one. Model M6, which represents the mechanical power theory, was the second-best model with a probability of 0.271 and a WOE against of −0.29 decibans. Model M5, representing the driving pressure theory, was the third best model (probability=0.260 and WOE against=−0.45 decibans). RelativeDP_1 and RelativeMP_1 were highly correlated variables (R2==0.709, 95% CI=0.228–0.912, p.value=0.0098; see Fig. S3, online supplement) probably because they are mathematically coupled (DP intervenes in MP computation). For the rest of the models, the evidence weights against them. Model M7, with two moderators, scores the worst.

Figs. 4 and 5 represent the two best models (both with WOE=0). In Fig. 4 (and Fig. S4, online supplement), it is clear that the benefit of the OLA over the control group is seen when the PF ratio on day 3 in the control group is lower, with a suggestion of potential for harm when PF ratio climbs above 180. In Fig. 5 (and Fig. S5, online supplement), the benefit of the OLA over the control group is seen only when the MP in the OLA group is less than the MP in the control group, with a suggestion for potential harm when the MP in the OLA group exceeds 1.4 times the MP in the control group.

Figure 4.

Effect of PaO2/FiO2 in the control group (3rd day) on RR of mortality. Each trial is represented by a symbol of area proportional to its precision (inverse variance of RR).

(0.1MB).

Figure 5.

Effect of relative MP (1st day) on RR of mortality. Each trial is represented by a symbol of area proportional to its precision (inverse variance of RR).

(0.11MB).

We were not able to show that OLA strategy reduces mortality in ventilated adults with ARDS, although there was significant statistical heterogeneity. We found that sources that account for this heterogeneity include the method of PEEP management in the intervention group, the lack of use of lung protective tidal volume in the control group, the level of recruitment achieved in the control group (PF ratio on day 3 of ventilation), and the relative difference in mechanical power or driving pressure between intervention and control patients. This highlights that interpretation of the benefits of the OLA must consider these factors, and future randomized trials should specifically try to address these sources of heterogeneity to truly understand the combination of interventions and patients who are likely to benefit from a particular OLA strategy.

Our results differ from previous meta-analysis on the OLA and/or high-PEEP strategies. Lu et al.13 included 15 RCTs focused on an OLA strategy which included high PEEP (with or without prone position) and/or RMs. High PEEP was therefore not a mandatory constituent of the experimental arm. Using a subgroup of 9 studies the found lower 28-day mortality with the OLA, with no substantial heterogeneity. Their analysis included 2 trials in which the experimental treatment consisted only of RMs (one of them with mandatory prone position) that were excluded in our analysis. However, they did not include 3 big RCTS published since 2017, which are included in our analysis, and likely contribute to the different results (note the influence of these 3 recent trials on Fig. S2, online supplement).

Our meta-analysis is congruent with the results of Walkey et al.12 and Santa Cruz et al.2 These studies showed no beneficial effects of High PEEP strategies on mortality. Both authors made attempts to restrict inclusion of studies to minimize heterogeneity related to aspects of the OLA outside PEEP management, with focused meta-analysis on PEEP. We choose to take a different approach because when ventilating ARDS patients, ICU clinicians tend to use the whole protective ventilation strategy. Therefore, the relevant clinical question is the effect of the whole strategy. If the price to pay is heterogeneity, statistical techniques can be used to explore the sources of heterogeneity.

Statistical heterogeneity was detected in our study. Fig. 3 shows substantial quantitative differences in the results from the different RCTs, and it is probable that there were substantial clinical differences between studies and the patients in them. An analysis which ignores this heterogeneity is a missing opportunity for investigation.36 For this reason, we estimated the global effect using classical Der Simonian–Laird and a Hartung–Knapp–Sidik–Lonkman37 REMs. We could have used another Likelihood based (or Bayesian) method.38 However, when heterogeneity among RCTs is present, if the results are to affect future clinical practice a single overall summary estimate of treatment benefit has little practical applicability. So, it is of clinical and scientific importance to investigate potential sources of that heterogeneity.

We performed two sensitivity analysis to explore heterogeneity. The first looked at modality of achieving the OLA. We found evidence of heterogeneity coming from studies in which PEEP was adjusted based on a lower inflection point of the static Pressure-Volume curve. However, this was applied in only 101 of the 2103 patients (less than 5%) of the experimental arm, in 3 early trials, and the OLA resulted in lower mortality as was seen in previous meta-analysis.39 The second sensitivity analysis focused on limited VT in the control group. Higher VT was applied in the control group in the same subgroup of 3 early trials, and again the OLA resulted in lower mortality only when limited VT was not applied. Given this small number, we cannot exclude residual confounding (or regression to the mean) so we cannot assure that therapeutic modality in the way the OLA strategy was prescribed, limiting VT or both were the primary source of heterogeneity.

Another major source for heterogeneity relates to whether there is variation in the treatment benefit according to a patient's underlying risk of the event that the treatment is designed to decrease. We attempted to take a statistical approach39 to evaluate this, using MEM meta-regression to investigate the dependence of the treatment effect on a priori stated predictors in every trial. Each predictor represented a competitive hypothesis on the causal pathways of mortality through Ventilator Induced Lung Injury (VILI). Our results showed two theories were practically equal in terms of probability of being the “best” explicative model, given the data and relative to this set of models.

The most probable “best” model of our set represents heterogeneity in treatment effect based on the recruitment achieved in the control group, measured with the PaO2/FiO2 of the control group on the day 3 of ventilation. As the randomization has created equivalent cohorts at the beginning of the trial, this measure also represents the counterfactual evolution of experimental group if the OLA strategy had not been applied. Our analysis shows that OLA strategy is only effective when conventional strategies (control group) fail to yield a PaO2/FiO2≥170. This threshold of a potential differential treatment effect and/or outcome when the PF ratio passes an inflection point between 150 and 175 is reproducible in many areas of the ARDS literature, including Non-Invasive Ventilation (NIV),40 prone positioning,41 neuromuscular blockade,42 and High-Frequency Ventilation literature.43

The second most probable “best” model, represents heterogeneity in treatment effect based on the difference in Mechanical Power between experimental and control groups. We used Gattinoni's simplified formulae17 to compute this, making some adjustments in Pressure-Controlled ventilation cohorts because there is no generally accepted formula for MP in Pressure-Controlled modes. We found that the OLA strategy is no longer beneficial over control group when the OLA results in higher MP than the control group, the beneficial effects of OLA ventilation vanishes. MP is associated with mortality,44 and some authors are trying to explain the benefits of prone position45 and neuromuscular blockade46 in terms of MP. However, the subject is far from being completely understood.

Nearly identical results were seen when using Driving Pressure instead of MP to explain heterogeneity, which is expected given high mathematical coupling and high statistical significance in correlation between DP and MP. When the OLA strategy group receives a higher DP than the control group, the beneficial effects of OLA are not seen. The effects described for DP on mortality of ventilated patients7 mimic those produced by MP.9,47,48 Likely MP and DP are carrying the same clinical information and our results confirm that DP seems to be the most important contributor to MP. DP is therefore unlikely an independent source of heterogeneity. The evidence weights against the rest of the models of our set, including the two-moderator model (MP and PF ratio combined) that raises concern of overfitting.

This study has limitations. First, we identified publication bias. We estimated that 2 studies could be missing, and their results could change the conclusions. Second, our set of possible theories explaining heterogeneity is not exhaustive, it might be that there are other alternative explanations that have not been investigated in this study. Third, the heterogeneity analysis was done only in twelve of the fourteen RCTs. In this way, our analysis loses power with higher Type II error (false negative rate). Fourth, as a secondary analysis of RCTs the effect of our “best” candidate moderators is not as robust as would be generated by a new RCT specifically addressed to clarify those hypotheses. However, there may be challenges to conducting such, an RCT because of lack of equipoise. In fact, both recently published RCTs were planned with the highest PEEP levels ever in the control group.34,35 And both were inconclusive, probably because the control groups were also OLA strategies in and of themselves. We are unaware of any RCT underway or planned that specifically address the relationship between MP and mortality in ARDS patients. We hope our results will encourage new trials to clarify this likely relationship.

In conclusion, this systematic review and meta-analysis has not been able to show the benefit of OLA strategy on mortality during mechanical ventilation of ARDS patients. However, our study points in the direction that the mortality effect of OLA strategy depends on recruitment (PaO2/FiO2) and mechanical power. Taken together, these results highlight important physiologic concepts which will be crucial for future RCTs: attempts to further open the lung with an OLA strategy are likely only going to be beneficial in those who have not already been recruited with conventional strategies (i.e. PF<170), and in whom application of higher PEEP results in less amount of energy applied to lung (i.e. lowers DP or MP).

VM and AM conceived the presented idea. AB developed the theory and performed the computations. PV and SFU conducted an independent literature search to identify potentially relevant studies. PV and CC independently reviewed the search results to identify pertinent articles. VM, AM, PV, CC, SFU, FG and RK contributed to the interpretation of the results. VM, AM, RK took the lead in writing the manuscript. All authors provided critical feedback and helped shape the research, analysis and manuscript.

The authors declare that they have no conflict of interest.