Venovenous extracorporeal membrane oxygenation for acute respiratory distress syndrome: a systematic review and meta-analysis

Articles

Venovenous extracorporeal membrane oxygenation for acute respiratory distress syndrome: a systematic review and meta-analysis Laveena Munshi, Allan Walkey, Ewan Goligher, Tai Pham, Elizabeth M Uleryk, Eddy Fan

Summary

Background Use of extracorporeal membrane oxygenation (ECMO) in adults with severe acute respiratory distress syndrome has increased in the past 10 years. However, the efficacy of venovenous ECMO in people with acute respiratory distress syndrome is uncertain according to the most recent data. We aimed to estimate the effect of venovenous ECMO on mortality from acute respiratory distress syndrome. Methods In this systematic review and meta-analysis, we searched MEDLINE (including MEDLINE In-Process and Epub Ahead of Print), Embase and the Wiley search platform in the Cochrane database for randomised controlled trials and observational studies with matching of conventional mechanical ventilation with and without venovenous ECMO in adults with acute respiratory distress syndrome. Titles, abstracts, and full-text articles were screened in duplicate by two investigators. Data for study design, patient characteristics, interventions, and study outcomes were abstracted independently and in duplicate. Studies were weighted with the inverse variance method and data were pooled via random-effects modelling. We calculated risk ratios (RRs) and 95% CIs to summarise results. The primary outcome was 60-day mortality across randomised controlled trials. The Grading of Recommendations Assessment, Development and Evaluation (GRADE) guidelines were used to rate the quality of evidence Findings We included five studies, two randomised controlled trials and three observational studies with matching techniques (total N=773 patients). In the primary analysis, which included two randomised controlled trials with a total population of 429 patients, 60-day mortality was significantly lower in the venovenous ECMO group than in the control group (73 [34%] of 214 vs 101 [47%] of 215; RR 0·73 [95% CI 0·58–0·92]; p=0·008; I² 0%). The GRADE level of evidence for this outcome was moderate. Three studies included data for the incidence of major haemorrhage in the ECMO group. 48 (19%) of the 251 patients in these three studies had major haemorrhages. Interpretation Compared with conventional mechanical ventilation, use of venovenous ECMO in adults with severe acute respiratory distress syndrome was associated with reduced 60-day mortality. However, venovenous ECMO was also associated with a moderate risk of major bleeding. These findings have important implications surrounding decision making for management of severe acute respiratory distress syndrome at centres providing venovenous ECMO. Funding None. Copyright © 2019 Elsevier Ltd. All rights reserved.

Introduction Use of extracorporeal life support for respiratory failure has garnered both enthusiasm and scepticism during the past five decades.1 Venovenous extracorporeal membrane oxygenation (ECMO) is used in patients with severe acute respiratory failure to facilitate gas exchange in the setting of refractory hypoxaemia or hypercapnic respiratory acidosis. It can also facilitate a reduction in the intensity of mechanical ventilation.2 Use of venovenous ECMO has increased substantially in the past 10 years after reports of successful use in several countries during the 2009 H1N1 influenza pandemic.3–9 However, most studies of venovenous ECMO were observational, and they often had conflicting results.6,9 The authors of guidelines published in 2017 concluded that evidence was insufficient to make a www.thelancet.com/respiratory Vol 7 February 2019

clinical recommendation for or against the use of venovenous ECMO in patients with severe acute respiratory distress syndrome (ARDS).1 Despite in­ creasing use of venovenous ECMO, rigorous evidence is required to establish the appropriate role for this treatment modality in the management of severe ARDS. Two randomised controlled trials3,10 have been done in the last 10 years to assess the potential efficacy of venovenous ECMO for severe ARDS. However, the results of both studies3,10 have been met with both enthusiasm and controversy. The Conventional Ventilator Support vs Extracorporeal Membrane Oxygen­ ation for Severe Acute Respiratory Failure (CESAR) trial3 showed that transfer to an ECMO centre was associated with a significant improvement in the composite endpoint of 6-month survival without disability

Lancet Respir Med 2019; 7: 163–72 Published Online January 11, 2019 http://dx.doi.org/10.1016/ S2213-2600(18)30452-1 See Comment page 106 Interdepartmental Division of Critical Care Medicine (L Munshi MD, E Goligher MD, T Pham MD, E Fan MD) and Institute of Health Policy, Management and Evaluation (E Fan), University of Toronto, Toronto, ON, Canada; University Health Network and Sinai Health System, Toronto, ON Canada (L Munshi, E Goligher, E Fan); Department of Medicine, Division of Pulmonary, Allergy, and Critical Care Medicine, The Pulmonary Center, Evans Center for Implementation and Improvement Sciences, and Department of Health Law, Policy and Management, Boston University School of Medicine, Boston, MA, USA (A Walkey MD); Keenan Research Centre for Biomedical Science, Li Ka Shing Knowledge Institute, St Michael’s Hospital, Toronto, ON, Canada (T Pham); and E M Uleryk Consulting, Mississauga, ON, Canada (E M Uleryk MLS) Correspondence to: Dr Laveena Munshi, Mount Sinai Hospital, 600 University Avenue, 18-206 Toronto, ON M5G 1X5, Canada laveena.munshi@ sinaihealthsystem.ca

163

Articles

Research in context Evidence before this study In the Conventional Ventilator Support vs Extracorporeal Membrane Oxygenation for Severe Acute Respiratory Failure (CESAR) trial, transfer to an extracorporeal membrane oxygenation (ECMO) centre was associated with a significant improvement in death or severe disability at 6 months compared with conventional mechanical ventilation. Some experts interpreted the results of CESAR with scepticism, citing the proportion of patients (24%) in the ECMO group who did not undergo ECMO and the absence of lung-protective ventilation in the control group. After CESAR, several observational studies of ECMO for acute respiratory distress syndrome were published, which typically showed improved outcomes. However, four systematic reviews and meta-analyses in which these data were pooled had conflicting results. Building on the knowledge and experience gained from CESAR, in the ECMO to Rescue Lung Injury in Severe ARDS (EOLIA) trial, ventilatory management in both the ECMO and conventional mechanical ventilation groups was included in the protocol, with particular emphasis on the use of prone positioning and neuromuscular blockading agents before and after enrolment. The data safety monitoring board stopped the trial early after 75% of the sample size had been achieved because the lower boundary of the predefined stopping rule for futility had been crossed. No significant difference was noted between the two groups in the primary outcome of 60-day mortality. However, a secondary outcome of death or treatment failure (which was defined as death or crossover to ECMO in the control group) was significantly improved in the

See Online for appendix

compared with treatment with conventional mechanical ventilation. However, only a subset of patients in the ECMO group underwent ECMO, and compliance with lung-protective ventilation was suboptimal in the control group.3,11,12 The ECMO to Rescue Lung Injury in Severe ARDS (EOLIA) trial10 did not show a significant reduction in mortality in the ECMO group compared with the control group, who received con­ventional mechanical ventilation in combination with alternative rescue strategies. However, the trial was stopped early for futility on the basis of pre-established stopping rules, which complicates interpretation of the seeming trend towards mortality benefit. In view of this uncertainty, we did a systematic review and meta-analysis of the available data to compare the efficacy of venovenous ECMO on mortality and the associated risk of adverse events compared with that of conventional mechanical ventilation in adults with severe ARDS.

Methods

Search strategy and selection criteria A medical information specialist (EMU) ran a search without language restrictions on the OvidSP search 164

ECMO group compared with the conventional mechanical ventilation group. Added value of this study To our knowledge, this systematic review is the first to pool mortality data from the two largest randomised controlled trials (CESAR and EOLIA) of venovenous ECMO for severe acute respiratory distress system and from three observational studies in which matching techniques were used. Across moderate-to-high quality studies, we noted a significant reduction in 60-day mortality with venovenous ECMO compared with mechanical ventilation. These results were largely consistent across various secondary analyses. In view of the challenges of doing large clinical trials in this population of critically ill patients, another randomised controlled trial of ECMO is unlikely in the near future. Therefore, the results of this meta-analysis represent the most comprehensive and up-to-date synthesis of the available evidence for clinicians on the use of venovenous ECMO in patients with severe acute respiratory distress syndrome. Implications of all the available evidence Our findings suggest that venovenous ECMO delivered at expert centres is beneficial for patients with severe acute respiratory distress syndrome. Our findings, which are based on evidence of moderate-to-high quality, have important implications supporting the use of venovenous ECMO at high-volume ECMO centres for select patients with severe acute respiratory distress syndrome.

platform of the MEDLINE (including MEDLINE InProcess and Epub ahead of print) and Embase databases, and on the Wiley search platform in the Cochrane database, from inception up to May 28, 2018, which was the date of our final search. We used both subject headings and text word terms to search for articles about ECMO in adults with ARDS, and applied Cochrane, McMaster, and Robinson Dickersin clinical trials filters. Our full search strategy is in the appendix. We included randomised controlled trials and obser­ vational studies with matching in which mechanical ventilation plus venovenous ECMO was compared with mechanical ventilation and the institution’s care algorithm for refractory hypoxia in adults with ARDS (including treatments such as inhaled nitric oxide or high-frequency oscillation), and in which mortality at any time was reported. When more than one extracorporeal life-support modality was used for respiratory support, we included the study if venovenous ECMO was used more than 70% of the time. We excluded studies in which the main focus was venoarterial ECMO, a modality that provides cardiopulmonary support and was used historically for ARDS, and those in which use of extracorporeal CO2 removal was assessed. LM and AW www.thelancet.com/respiratory Vol 7 February 2019

Articles

independently reviewed the titles and abstracts of all citations. They then independently reviewed the full text of studies that seemed potentially relevant to establish eligibility for inclusion. Disagreements were reviewed by a third reviewer (EF), who had a deciding vote.

Data analysis Data for study design, patient characteristics, interventions, and study outcomes were abstracted independently and in duplicate. Randomised controlled trials were assessed for evidence of bias with the Cochrane risk-of-bias tool.13 We judged a study’s overall risk of bias to be high if any domain was at high risk of bias, with the exception of caregiver blinding, for which we accepted standardisation of mechanical ventilation, sedation, and weaning in both study groups to mitigate performance bias in these necessarily unblinded trials. The Newcastle-Ottawa Scale was used to assess methodological strength in nonrandomised studies.14 The Grading of Recommendations Assessment, Development and Evaluation guidelines were used to rate the overall quality of evidence.15 A summary of findings table was prepared with GRADEpro software. The systematic review followed Preferred Reporting Items for Systematic Reviews and Meta-Analyses and Meta-analysis Of Observational Studies in Epidemiology guidelines (appendix). The primary outcome was 60-day mortality across the randomised controlled trials. If 60-day mortality was not reported, we extrapolated on the basis of available data (ie, survival curves, CONSORT diagrams) or substituted the closest common mortality timepoint. Secondary analyses, which included randomised controlled trials and observational studies, included treatment failure (death in the ECMO group and a composite outcome of crossover to ECMO or death in the control group), mortality at the last follow-up, mortality in patients receiving ECMO compared with that in patients who did not receive ECMO irrespective of group assignment (ie, as-treated analysis), mortality in patients (excluding crossover patients) who received their assigned treatment (ie, per-protocol analysis), pooled 30-day mortality across randomised controlled trials and observational studies (when unavailable, intensive-care unit mortality, hospital mortality, or the closest available mortality timepoint was used, in that order), and adverse events. The appendix contains a detailed breakdown of patient classification for each analysis. Adverse events included major haemorrhage (such as intracerebral, gastrointestinal, pulmonary, intraperitoneal, retroperitoneal, or haemo­ thoracical haemorrhage, or any haemorrhage requiring an intervention) and complications associated with cannulation or the ECMO circuit (vessel or cardiac perforation, cardiac arrest, circuit failure, membrane dysfunction, cannula dislodgement, or significant air in the circuit). Descriptive statistics are reported as medians and IQRs. Studies were weighted with the inverse variance www.thelancet.com/respiratory Vol 7 February 2019

6117 citations identified by electronic literature search

155 duplicates removed

5962 included in combined search

4408 excluded based on screening of titles

1554 included in abstract screening

1423 excluded 775 wrong design 587 no control group 28 animal studies 33 neonatal studies

131 included in full-text review

126 excluded 63 no control group 43 wrong design 20 wrong population

5 fulfilled inclusion criteria and were included in meta-analysis (two randomised controlled trials, three matched observational studies)

Figure 1: Flow diagram of study selection

method and data were pooled via random-effects modelling.16 We used a risk ratio (RR) and 95% CIs to summarise results. Clinical heterogeneity among studies was assessed qualitatively, and statistical heterogeneity was calculated with the I² measure.17 Publication bias was assessed by visual inspection of funnel plots (appendix). All statistical analyses were done in RevMan (version 5.3).

Role of the funding source There was no funding source for this study. The corresponding author had full access to all study data and had final responsibility for the decision to submit for publication.

Results Our electronic search retrieved 6117 citations, 131 of which were selected for full-text review (figure 1). Five studies with a combined population of 1055 patients fulfilled our inclusion criteria—two randomised controlled trials3,10 and three observational studies5,6,18 with matched controls. 773 patients were included in the secondary analysis. Median mortality in the included studies across both study groups was 44% (IQR 40–58). Studies varied with respect to criteria for initiation of ECMO (table 1). However, all criteria used reflected the 165

Articles

Study type

Peek et al (CESAR),3 2009

Noah et al,5 2011

Pham et al,6 2013

Tsai et al,18 2015

Combes et al (EOLIA),10 2018

Randomised controlled trial

Observational

Observational

Observational

Randomised controlled trial

n Overall

180

150

260 (104)*

216

249

ECMO group

90

75

103 (52)*

81 (45)†

124

Severe but potentially reversible ARDS‡; Murray score ≥3·0 or uncompensated hypercapnia with pH <7·2 despite optimal conventional treatment

Age 18–65 years; severe but potentially reversible ARDS; Murray score ≥3·0 or uncompensated hypercapnia with pH <7·2 despite optimal conventional treatment; infection with H1N1 influenza

Severe ARDS, which was defined as ARDS due to H1N1 influenza, and any one of a modified lung injury score ≥3·0, an arterial pH <7·21, PaO2:FiO2 <100 mm Hg, or arterial oxygen saturation <90%

Moderate-to-severe ARDS (more specific criteria for ECMO indications not described)

PaO2:FiO2 <50 for 3 h, PaO2:FiO2 <80 for 6 h, or PaCO2 >60 mm Hg and pH <7·25 for 6 h

Indication for ECMO

Mean age (SD), years ECMO

40 (13)

36 (11)

45 (15)

56 (2)

52 (14)

CMV

40 (13)

37 (12)

45 (13)

56 (2)

54 (13)

Mean (SD) or median (IQR) days of CMV before ECMO ECMO

1·5 (0·7-4·4)

4·4 (3·7)

2·0 (1·0–5·0)

3·0 (2·0)

1·4 (0·6–3·7)

CMV

1·5 (0·6–4·2)

4·2 (4·2)

··

··

1·4 (0·7–4·2)

Proportion of ECMO recipients who received venovenous ECMO

68 (100%)

··

45 (87%)

37 (82%)

121 (100%)

Cause of respiratory failure

Pneumonia or various causes of ARDS

H1N1 Influenza

H1N1 Influenza

Pneumonia or various causes of ARDS

Pneumonia or various causes of ARDS

ECMO

76 (30)

55 (14)

70 (26)

93 (13)

73 (30)

CMV

75 (36)

55 (12)

68 (20)

124 (12)

72 (24)

3·5 (0·6)

NA

3·3 (0·7)

NA

NA

Mean PaO2:FiO2 (SD)

Mean lung injury score in ECMO group (SD)

Pressure-limited and volume-limited ventilation ECMO

Yes

Yes

Yes

Yes

Yes

CMV

Yes

Yes

Yes

Yes

Yes

Mechanical ventilation protocol in ECMO group

Respiratory rate 10 breaths per min; peak inspiratory pressure <20–25 cm H2O; PEEP 10–15 cm H2O; FiO2 0·3

Respiratory rate 10 breaths per min; peak inspiratory pressure <30 cm H2O (ideally <25 cm H2O); PEEP 10–15 cm H2O; FiO2 0·3

Not in protocol but a median decrease in tidal volumes of 2·8 mL/kg predicted bodyweight; decrease in respiratory rate by 8 breaths per min; decrease in plateau pressure by 6 cm H2O

NA

Volume-assisted control mode; plateau pressure <25 cm H2O; FiO2 0·30-0·60; respiratory rate 10–30 breaths per min; or airway pressure release ventilation mode with high pressure level <25 cm H2O and low-pressure level ≥10 cm H2O

Adjuvants to mechanical ventilation

High-frequency oscillation, inhaled pulmonary vasodilators, prone positioning, and corticosteroids

Neuromuscular blockading agents, conservative fluid administration, and corticosteroids

Prone positioning and inhaled pulmonary vasodilators

NA

Increased recruitment strategy, neuromuscular blockading agents, and prolonged prone positioning; inhaled pulmonary vasodilators could be used if oxygenation objectives were not met

Crossover

No

NA

NA

NA

Yes: in patients with refractory hypoxia with saturation <80% for >6 h§

Primary outcome

Death or disability at 6 months

Hospital mortality

Intensive-care-unit mortality

Hospital mortality

60-day mortality

ECMO=extracorporeal membrane oxygenation. ARDS=acute respiratory distress syndrome. PaO2=partial pressure of arterial oxygen. FiO2=fractional concentration of oxygen in inspired air. CMV=conventional mechanical ventilation. LIS=lung injury score. NA=not applicable. PEEP=positive end expiratory pressure. *Data in parentheses are n after matching; results that follow are for the cohort after matching. †Only 45 patients who received ECMO were analysed. ‡Reversibility based on clinical opinion of ECMO consultant. §35 (28%) participants in the control group crossed over to ECMO.

Table 1: Baseline characteristics of included studies

characteristics of patients with severe ARDS. The pre­ dominant cause of ARDS was pneumonia. Two of the observational studies5,6 with matching focused mainly 166

on pandemic H1N1 influenza. ECMO was initiated within 72 h of the onset of mechanical ventilation in four of the five included studies.3,6,10,18 The mean ratio of www.thelancet.com/respiratory Vol 7 February 2019

Articles

ECMO

Weight (%)

CMV

Risk ratio (95% CI)

Events

Total

Events

Total

Peek et al (2009)3 Combes et al (2018)10

29 44

90 124

44 57

90 125

40·9% 59·1%

0·66 (0·46–0·95) 0·78 (0·57–1·06)

Combined Heterogeneity: τ²=0·00; χ²=0·47, df=1, (p=0·49); I²=0% Test for overall effect: Z=2·66 (p=0·008)

73

214

101

215

100·0%

0·73 (0·58–0·92)

0·5

0·7

1

Favours ECMO

1·5

2

Favours CMV

Figure 2: Forest plot of 60-day mortality in randomised controlled trials of ECMO vs CMV in adults with severe acute respiratory distress syndrome 50-day mortality was the closest timepoint available to our primary endpoint of 60-day mortality in the Peek et al’s trial (appendix). Risk ratios were calculated with a random-effects model. ECMO=extracorporeal membrane oxygenation. CMV=conventional mechanical ventilation. df=degree of freedom.

the partial pressure of arterial oxygen to the fractional concentration of oxygen in inspired air (PaO2:FiO2) was 80 or less in the ECMO groups in all but one study (table 1).18 In observational studies with matching techniques, patients who did not undergo ECMO were similar at baseline to those who underwent ECMO with respect to age, severity of illness scores, and several other clinical variables. However the variables used for matching differed between the studies. Pressurelimited and volume-limited ventilation were used in all studies (table 1). Risk of bias was low across both randomised controlled trials and moderate to low in the observational studies with matching (appendix). Visual inspection of a funnel plot did not suggest publication bias (appendix). Use of ECMO in randomised controlled trials was associated with a significant reduction in 60-day mortality (73 [34%] deaths among 214 patients in the ECMO group vs 101 [47%] deaths among 215 patients in the control group; RR 0·73 [95% CI 0·58–0·92]; I² 0%; p=0·008; figure 2; table 2). The results were similar when we assessed mortality at the latest available followup timepoint (figure 3). The risk of treatment failure was lower in patients randomly assigned to ECMO (77 [36%] of 214 patients) than in those assigned to control treatments (138 [64%] of 214; RR 0·58 [95% CI 0·39–0·84]; I² 68%; p=0·004; figure 4A). In the astreated analysis, mortality at the longest available followup did not differ significantly between those who underwent ECMO (87 [39%] of 225) and those who received conventional mechanical ventilation only (92 [45%] of 204; RR 0·87 [95% CI 0·68–1·10; I² 0%; p=0·23; figure 4B). The as-treated analysis included 35 patients in the EOLIA trial (comprising 28% of the control group) who crossed over into the ECMO group because of refractory hypoxaemia, 20 of whom (57%) had died at the longest available follow-up. In the perprotocol analysis, mortality at the longest available follow-up did not differ significantly between the ECMO group (67 [35%] of 189) and the control group (82 [46%] of 179; RR 0·79 [95% CI 0·61–1·02; I² 0%; p=0·07; figure 4C). When data from the two randomised controlled trials and three observational studies with matching were pooled, 30-day mortality (or the closest www.thelancet.com/respiratory Vol 7 February 2019

CMV risk anticipated absolute effects per 1000

Risk ratio ECMO (95% CI) anticipated absolute effects* per 1000 (95% CI)

n (studies)

Certainty of evidence†

60-day mortality

470

343 (272–432) 0·73 (0·58–0·92)

429 (from two randomised controlled trials)

Moderate‡

Mortality at longest available follow-up

474

361 (285–451) 0·76 (0·60–0·95)

429 (from two randomised controlled trials)

Moderate‡

Treatment failure

642

372 (250–546) 0·58 (0·39–0·85)

429 (from two randomised controlled trials)

High

Mortality at longest available follow-up (astreated analysis)

451

392 (307–496) 0·87 (0·68–1·10)

429 (from two randomised controlled trials)

Moderate§

Mortality at longest available follow-up (perprotocol analysis)

458

362 (279–467) 0·79 (0·61–1·02)

429 (from two randomised controlled trials)

Moderate§

30-day mortality¶

455

314 (227–432) 0·69 (0·50–0·95)

773 (from two randomised controlled trials and three observational studies)

Moderate§

All outcomes are based on 429 participants from two randomised controlled trials, except for 30-day mortality, which is based on 773 participants from two randomised controlled trials and three observational studies. Randomised controlled trials were thought to have a low risk of bias. ECMO=extracorporeal membrane oxygenation. CMV=conventional mechanical ventilation. *Risk in the intervention group (and its 95% CI) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). †Certainty of evidence was assessed with the Grading of Recommendations Assessment, Development and Evaluation criteria. ‡Downgraded because of imprecision based upon optimal information size and because RRs of 0·5–2·0 were not judged to be large effect sizes. §Downgraded because of plausible confounding. ¶30 days or closest available timepoint.

Table 2: Summary of findings for ECMO vs CMV in patients with severe acute respiratory distress syndrome

available timepoint) was significantly lower with ECMO (120 [31%] of 386) than with conventional mechanical ventilation (176 [45%] of 387; RR 0·69 [95% CI 0·50–0·95]; I² 66%; p=0·02; figure 5). The results were similar in a post-hoc analysis in which one outlier study5 was excluded (appendix). Three studies5,6,10 included data for major haemorrhage, and four studies3,5,6,10 included data for circuit-associated or 167

Articles

ECMO

Weight (%)

CMV

Risk ratio (95% CI)

Events

Total

Events

Total

Peek et al (2009)3 Combes et al (2018)10

33 44

90 124

45 57

90 125

44·4% 55·6%

0·73 (0·52–1·03) 0·78 (0·57–1·06)

Combined Heterogeneity: τ²=0·00; χ²=0·06, df=1, (p=0·80); I²=0% Test for overall effect: Z=2·39 (p=0·02)

77

214

102

215

100·0%

0·76 (0·60–0·95)

0·5

0·7

1

Favours ECMO

1·5

2

Favours CMV

Figure 3: Forest plot of mortality at latest follow-up in randomised controlled trials of ECMO vs CMV in adults with severe acute respiratory distress syndrome 6-month mortality or death before discharge was the latest follow-up timepoint in Peek et al’s trial, whereas 60-day mortality was the latest timepoint in Combes et al’s trial. Risk ratios were calculated with a random-effects model. ECMO=extracorporeal membrane oxygenation. CMV=conventional mechanical ventilation. df=degree of freedom.

A ECMO

Weight (%)

CMV

Risk ratio (95% CI)

Events

Total

Peek et al (2009)3 Combes et al (2018)10

33 44

90 124

46 92

89 125

45·9% 54·1%

0·71 (0·51–0·99) 0·48 (0·37–0·62)

Combined Heterogeneity: τ²=0·05; χ²=3·16, df=1, (p=0·08); I²=68% Test for overall effect: Z=2·87 (p=0·004)

77

214

138

214

100·0%

0·58 (0·39–0·84)

Peek et al (2009)3 Combes et al (2018)10

25 62

69 156

53 39

111 93

41·0% 59·0%

0·76 (0·52–1·10) 0·95 (0·70–1·29)

Combined Heterogeneity: τ²=0·00; χ²=0·82, df=1, (p=0·36); I²=0% Test for overall effect: Z=1·20 (p=0·23)

87

225

92

204

100·0%

0·87 (0·68–1·10)

Peek et al (2009)3 Combes et al (2018)10

25 42

68 121

45 37

89 90

46·4% 53·6%

0·73 (0·50–1·06) 0·84 (0·60–1·20)

Combined Heterogeneity: τ²=0·00; χ²=0·33, df=1, (p=0·57); I²=0% Test for overall effect: Z=1·84 (p=0·07)

67

189

82

179

100·0%

0·79 (0·61–1·02)

Events

Total

B

C

0·5

0·7

Favours ECMO

1

1·5

2

Favours CMV

Figure 4: Forest plots of treatment failure (A), mortality at longest available follow-up in the as-treated analysis (B), and mortality at longest available follow-up in per-protocol analysis (C) in randomised controlled trials of ECMO vs CMV in adults with severe acute respiratory distress syndrome Risk ratios were calculated with a random-effects model. The mortality timepoint used in all three forest plots was the latest available timepoint in both studies (6 months or death before discharge in Peek et al’s study, and 60-day mortality in Combes et al’s study). Further information is available in the appendix. (A) Treatment failure was defined as death in the ECMO group or the composite endpoint of death or crossover to ECMO given risk of imminent death in the CMV group. One patient in the control group of Peek et al’s trial underwent arteriovenous CO2 removal ECMO (survived) and was therefore analysed as treatment failure. 35 patients in the control group in Combes et al’s trial underwent ECMO (20 of whom died) and were therefore analysed as treatment failure. (B) The ECMO group includes one patient in Peek et al’s trial who was in the control group but received arteriovenous CO2 removal ECMO (survived) and 35 patients in the control group of Combes et al’s trial who received ECMO (20 of whom died). The conventional mechanical ventilation group, includes three patients from Combes et al’s trial (two of whom died) and 22 patients from Peek et al’s study (eight of whom died) who were in the ECMO group but did not receive ECMO. (C) The ECMO group excludes 22 patients in Peek et al’s study (eight of whom died) and three patients in Combes et al’s study (two of whom died) who were in the ECMO group who did not receive ECMO. The CMV group excludes one patient in the control group of Peek et al’s who underwent arterio-venous extracorporeal CO2 removal (who survived) and 35 patients in the control arm of Combes et al’s study who underwent ECMO (20 of whom died). ECMO=extracorporeal membrane oxygenation. CMV=conventional mechanical ventilation. df=degree of freedom.

cannula-associated complications (table 3). In view of the inconsistent reporting of these outcomes across control group, the adverse events were not pooled. 48 (19%) of 251  patients in three studies had major haemorrhages, including 16 (6%) intracranial haemorrhage, one (<1%) fatal pulmonary haemorrhage, and eight (3%) received massive transfusions or went into haemorrhagic shock. Six (2%) of 341 patients had circuit-associated or cannulaassociated major compli­cations (table 3). EOLIA10 was the 168

only study to include the proportion of adverse events in the control group and a comparison between the ECMO and con­ ventional mechanical ventilation group. The proportion of patients with bleeding events necessitating transfusion was substantially higher in the ECMO group than in the control group (46% vs 28%; absolute risk difference 18% [95% CI –6 to 30]). The proportion of patients with ischaemic stroke was lower in the ECMO group than in the control group (0% vs 5%; absolute risk www.thelancet.com/respiratory Vol 7 February 2019

Articles

ECMO Events

Weight (%)

CMV Total

Events

Total

Risk ratio (95% CI)

29 11 26 22 32

90 75 52 45 124

44 31 21 34 46

90 75 52 45 125

21·8% 14·4% 19·7% 22·7% 21·4%

0·66 (0·46–0·95) 0·35 (0·19–0·65) 1·24 (0·81–1·90) 0·65 (0·46–0·91) 0·70 (0·48–1·02)

Combined 120 Heterogeneity: τ²=0·08; χ²=11·92, df=4, (p=0·02); I²=66% Test for overall effect: Z=2·31 (p=0·02)

386

176

387

100·0%

0·69 (0·50–0·95)

Peek et al (2009)3 Noah et al (2011)5 Pham et al (2013)6 Tsai et al (2015)18 Combes et al (2018)10

0·5

0·7

Favours ECMO

1

1·5

2

Favours CMV

Figure 5: Forest plot of 30-day mortality across all studies of ECMO vs CMV in adults with severe acute respiratory distress syndrome If 30-day mortality was not reported, the closest available timepoint or outcome was used, or data were extrapolated: 50 days in Peek et al’s study, 30 days in Noah et al (extrapolated from Kaplan-Meier curve), intensive-care unit mortality in Pham et al’s study, in-hospital mortaility in Tsai et al’s study, and 30 days in Combes et al’s study (extrapolated from Kaplan Meier curve; appendix). GenMatch matching technique was used for the results from Noah et al’s study, and matching without replacement was used by Pham and colleagues. Risk ratios were calculated with a random-effects model. ECMO=extracorporeal membrane oxygenation. CMV=conventional mechanical ventilation. df=degree of freedom.

Major haemorrhage n/N (%)

Major haemorrhage type

Complications associated with ECMO circuit or cannulation n/N (%)

Peek et al (CESAR),3 2009

Not reported

NA

1/90 (1%)

Noah et al,5 2011

18/75 (24%)

Eight intracranial haemorrhages, five gastrointestinal or haemoperitoneal haemorrhages, four haemothoraxes, and one fatal pulmonary haemorrhage

1/75 (1%)

Pham et al,6 2013

24/52 (46%)

Seven haemothoraxes, seven gastrointestinal or haemoperitoneal haemorrhages, five intracranial haemorrhages, and five cases of haemorrhagic shock

3/52 (6%)

Three intracranial haemorrhages and three participants received massive transfusions

1/124 (1%)

Combes et al (EOLIA),10 2018

6/124 (5%)

No control group adverse events were reported in the studies by Peek et al,3 Noah et al,5 or Pham et al.6 Adverse events in the control group of Combes et al10 are detailed in the text. Tsai et al18 is not included because adverse events were not reported. ECMO=extracorporeal membrane oxygenation. NA=not applicable.

Table 3: Adverse events across ECMO groups

difference –5% [95% CI –10 to –2]). However, the proportion of haemorrhagic stroke was similar in both groups (2% vs 4%, absolute risk difference –2% [95% CI –7 to 30]).

Discussion Our systematic review and meta-analysis of two randomised controlled trials and three observational studies with matching (total N=1055) showed that venovenous ECMO was associated with a significant reduction in 60-day mortality compared with conventional mechanical ventilation in patients with severe ARDS. Improved outcomes in the ECMO group were noted in most of our secondary and sensitivity analyses. However, venovenous ECMO was associated with a moderate risk of major haemorrhage and a low risk of circuit-associated or cannula-associated complications. To our knowledge, our systematic review is the first in which the results of the two randomised controlled trials of venovenous ECMO in patients with severe ARDS were pooled.10 In CESAR,3 a high proportion of patients in the ECMO group did not receive ECMO, and low compliance with lung-protective ventilation in the control group meant that attribution of the www.thelancet.com/respiratory Vol 7 February 2019

improvement in 6-month survival without disability to use of venovenous ECMO was difficult. Conversely, in EOLIA,10 compliance with ECMO in the ECMO group was 98% (ie, 121 of 124 patients received ECMO), the mechanical ventilation strategy in the control group was specified in the protocol, and compliance with the best evidence-based approach to management of severe ARDS was high before randomisation (113 (90%) of the 125 patients in the control group underwent prone positioning and all 125 patients in the control group received neuromuscular blocking agents between random­isation and day 50). In our systematic review, ECMO was associated with significant reductions in mortality in all analyses except for the as-treated and per-protocol analyses. These analyses probably represent the limits of available estimates of treatment effect in view of the bias present in each analysis method. The cohort of patients in the control group of EOLIA who crossed over to ECMO (ie, who were included in the ECMO group in the as-treated analysis and excluded from the control group of the per-protocol analysis) represent a subset of critically ill patients in whom late rescue ECMO is necessary. These patients had the greatest severity of illness in the study, and, 169

Articles

when analysed in the ECMO group or excluded from the control group, could bias the study towards the null. Ideally, one would compare patients with very severe ARDS who crossed over to undergo rescue ECMO with patients with similar disease severity who were assigned to the ECMO group in the EOLIA trial. However, we could not identify that subset of patients in the ECMO group. Furthermore, the data safety monitoring board stopped the EOLIA trial early after 75% of the sample size had been achieved because the lower boundary of the predefined stopping rule for futility (defined as <20% absolute risk reduction in mortality) had been crossed. Some experts have argued that the 20% absolute risk reduction was an unreasonably large effect size with which to establish the sample size of the study.19 In retrospect, stopping for futility might have been likely in view of the large effect size. Although the EOLIA investigators’ simulations suggested that the study would have failed to reach significance in terms of its primary endpoint, these simulations were centred around the high absolute risk reduction. Had the study continued to full enrolment and had less stringent stopping rules, EOLIA might have reached the effect estimate to meet significance. Specifically, simulations suggested a 10% chance of stopping for efficacy at the next interim analysis. Four systematic reviews20–23 of extracorporeal life support compared with conventional mechanical ventilation for respiratory failure have been published in the past 5 years. The first21 included one randomised controlled trial and two observational cohort studies with matching (total N=353) and showed no effect on hospital mortality with ECMO (odds ratio 0·71 [95% CI 0·34–1·47]) in the primary analysis. However, ECMO was associated with significant benefits when the analysis was restricted to a propensity score analysis with matching with replacement in the two observ­ ational studies and when only patients who actually received ECMO in randomised controlled trials were included (odds ratio 0·52 [95% CI 0·35–0·76).21 A subsequent review20 from our group included studies of any type of ECMO for respiratory failure. Meta-analysis of ten clinically heterogeneous studies (four randomised controlled trials and six observational studies with a total population of 1248 patients) showed no difference in hospital mortality between patients who underwent ECMO and those who received conventional mechanical ventilation (RR 1·02 [95% CI 0·79–1·33]). In a sensitivity analysis limited to the one randomised controlled trial and observational studies that used matching techniques for the control group, mortality was significantly reduced in the ECMO group compared with the mechanical ventilation control group (RR 0·64 [95% CI 0·51–0·79]). The cohorts included in this matched analysis were composed of younger patients (ie, aged 30–40 years) and patients with higher severity 170

of illness at baseline compared with the other systematic reviews. Another systematic review22 assessed the use of extracorporeal CO2 removal with mechanical ventilation in severe ARDS. The authors identified 14 studies (including randomised controlled trials, observational studies, and case series) and found no difference in mortality with extracorporeal CO2 removal compared with conventional mechanical ventilation. However, in the subgroup of patients with more severe disease, those who underwent extracorporeal CO2 removal had significantly more ventilator-free days than those who received conventional mechanical ventilation. Finally, the most recent23 of the four systematic reviews was a Cochrane review that included only randomised controlled trials but identified four trials of any extracorporeal support modality. In view of the clinical heterogeneity, the authors decided not to pool the data and concluded that more randomised controlled trials are needed. In the appendix, we outline the randomised controlled trials included in previous systematic reviews that we excluded from this review. The excluded trials focused mainly on venoarterial ECMO and extracorporeal CO2 removal. Historically, venoarterial ECMO was the more common modality of ECMO used for all forms of cardiopulmonary failure, including ARDS. Venovenous ECMO is now the most commonly used modality in patients with ARDS, in view of its ability to provide gas-exchange support without the need for arterial cannulation and partial cardiopulmonary bypass. Extra­ corporeal CO2 removal is used only in refractory hypercapnia and provides minimal oxy­genation support. Our population of interest was patients with severe ARDS—which is characterised by severe hypoxaemia— corporeal CO2 removal would not be for which extra­ indicated. We thus limited our analysis to studies of venovenous ECMO. Design and execution of randomised controlled trials of this intervention are difficult and take a long time (ie, 9 years for CESAR,3 6 years for EOLIA10). Challenges include randomisation of very sick patients early with a risk of imminent death, engagement of many centres in different countries, management of mobile ECMO teams and remote ECMO initiation and transfer, and establishment of a manage­ment protocol for the control group. As a result, another large study of ECMO in patients with severe ARDS in the near future is unlikely. Thus, our meta-analysis provides clinicians with the most comprehensive synthesis of available evidence for the efficacy of venovenous ECMO in adult patients with severe ARDS. Importantly, all the study centres where patients were enrolled in both CESAR3 and EOLIA10 were expert, high-volume ECMO centres, and whether these results can be generalised to other centres is unclear. Since the publication of the CESAR trial,3 observational studies1,8,24 have provided further insights www.thelancet.com/respiratory Vol 7 February 2019

Articles

into optimal candidates, timing of initiation, mechanical ventilation strategies during ECMO, and the effect of centre experience, and the results of one study8 support regionalised care of ECMO patients. Therefore, intensivists need to interpret these findings in the context of their particular population and their centres’ experience with ECMO. Our study has several important limitations. Our primary results are drawn from only two randomised controlled trials. In an attempt to minimise clinical heterogeneity, and in view of the changes in management of ARDS and extracorporeal life support care, we designed this systematic review to include only the most up-to-date management approaches and focused specifically on venovenous ECMO. Second, the intervention group of the CESAR trial were all treated at one ECMO centre. A centre effect could have inflated the effect size because the centre specialised in ECMO manage­ment, and because a subset of patients did not receive ECMO. We attempted to address the fact that some patients in the ECMO group of CESAR did not undergo ECMO with our as-treated and perprotocol analyses. However, these analyses were subject to bias in view of the likelihood of increased severity of illness in the ECMO group after crossover patients were accounted for. A meta-analysis of individual patients’ data is needed to further assess whether early ECMO or rescue ECMO in a particularly sick subgroup might be beneficial. Third, the treatment failure composite outcome and secondary analyses of mortality that included observational studies were prone to bias. We aimed to include high-quality observational studies with matched controls to further assess our findings across the totality of the literature and enable assessment of adverse events in more depth. These studies were also included because they implemented more modern venovenous ECMO techniques. Fourth, although visual inspection of funnel plots did not suggest publication bias, definitive confidence in excluding bias was limited by the small number of studies included in our plot. Finally, we could not analyse risks of ECMO such as off-site cannulation, transportation on ECMO, and cannula-associated colonisation or infection, or outcomes across smaller-volume centres compared with largervolume centres because of the lack of data across studies. The use of venovenous ECMO in adults with severe ARDS is associated with reduced mortality across our meta-analysis. ECMO should be considered in adults with severe ARDS at expert, high-volume centres. Contributors LM and EF conceived the study, which was overseen by EF. LM, EMU, and EF devised the search strategy. EMU executed the search, and LM, AW, and EF selected the studies for inclusion. LM and AW extracted data, which were analysed by LM, AW, EG, TP, and EF. All authors were involved in development of the methodological approach. LM and EF wrote the Article, which was reviewed and revised by all other authors. Declaration of interests EG reports personal fees from Getinge. AW discloses payment for an UpToDate chapter. EF reports personal fees from MC3 Cardiopulmonary

www.thelancet.com/respiratory Vol 7 February 2019

and ALung Technologies. LM, TP, and EMU declare no competing interests. Acknowledgments EF is supported by a New Investigator Award from the Canadian Institutes of Health Research. References 1 Fan E, Del Sorbo L, Goligher EC, et al. An Official American Thoracic Society/European Society of Intensive Care Medicine/Society of Critical Care Medicine clinical practice guideline: mechanical ventilation in adult patients with acute respiratory distress syndrome. Am J Respir Crit Care Med 2017; 195: 1253–63. 2 Marhong JD, Munshi L, Detsky M, Telesnicki T, Fan E. Mechanical ventilation during extracorporeal life support (ECLS): a systematic review. Intensive Care Med 2015; 41: 994–1003. 3 Peek GJ, Mugford M, Tiruvoipati R, et al. Efficacy and economic assessment of conventional ventilatory support versus extracorporeal membrane oxygenation for severe adult respiratory failure (CESAR): a multicentre randomised controlled trial. Lancet 2009; 374: 1351–63. 4 Davies A, Jones D, Bailey M, et al. Extracorporeal membrane oxygenation for 2009 influenza A(H1N1) acute respiratory distress syndrome. JAMA 2009; 302: 1888–95. 5 Noah MA, Peek GJ, Finney SJ, et al. Referral to an extracorporeal membrane oxygenation center and mortality among patients with severe 2009 influenza A(H1N1). JAMA 2011; 306: 1659–68. 6 Pham T, Combes A, Roze H, et al. Extracorporeal membrane oxygenation for pandemic influenza A(H1N1)-induced acute respiratory distress syndrome: a cohort study and propensity-matched analysis. Am J Respir Crit Care Med 2013; 187: 276–85. 7 Munshi L, Gershengorn HB, Fan E, et al. Adjuvants to mechanical ventilation for acute respiratory failure. Adoption, de-adoption, and factors associated with selection. Ann Am Thorac Soc 2017; 14: 94–102. 8 Barbaro RP, Odetola FO, Kidwell KM, et al. Association of hospital-level volume of extracorporeal membrane oxygenation cases and mortality. Analysis of the extracorporeal life support organization registry. Am J Respir Crit Care Med 2015; 191: 894–901. 9 Patroniti N, Zangrillo A, Pappalardo F, et al. The Italian ECMO network experience during the 2009 influenza A(H1N1) pandemic: preparation for severe respiratory emergency outbreaks. Intensive Care Med 2011; 37: 1447–57. 10 Combes A, Hajage D, Capellier G, et al. Extracorporeal membrane oxygenation for severe acute respiratory distress syndrome. N Engl J Med 2018; 378: 1965–75. 11 Jainandunsing JS, Ismael F. Ventilatory support versus ECMO for severe adult respiratory failure. Lancet 2010; 375: 549. 12 Zwischenberger JB, Lynch JE. Will CESAR answer the adult ECMO debate? Lancet 2009; 374: 1307–08. 13 Higgins JP, Altman DG, Gotzsche PC, et al. The Cochrane Collaboration’s tool for assessing risk of bias in randomised trials. BMJ 2011; 343: d5928. 14 Wells A, Shea B, O’Connell D, Peterson J, Welch V, Losos M. The Newcastle-Ottawa Scale (NOS) for assessing the quality of nonrandomised studies in meta-analysis. http://www.ohri.ca/ programs/clinical_epidemiology/oxford.asp (accessed May 1, 2018). 15 Balshem H, Helfand M, Schunemann H, et al. GRADE guidelines: 3. Rating the quality of evidence. J Clin Epidemiol 2011; 64: 401–06. 16 DerSimonian R, Laird N. Meta-analysis in clinical trials. Control Clin Trials 1986; 7: 177–88. 17 Higgins JP, Thompson SG, Deeks JJ, Altman DG. Measuring inconsistency in meta-analyses. BMJ 2003; 327: 557–60. 18 Tsai HC, Chang CH, Tsai FC, et al. Acute respiratory distress syndrome with and without extracorporeal membrane oxygenation: a score matched study. Ann Thorac Surg 2015; 100: 458–64. 19 Gattinoni L, Vasques F, Quintel M. Use of ECMO in ARDS: does the EOLIA trial really help? Crit Care 2018; 22: 171.

171

Articles

20 Munshi L, Telesnicki T, Walkey A, Fan E. Extracorporeal life support for acute respiratory failure. A systematic review and metaanalysis. Ann Am Thorac Soc 2014; 11: 802–10. 21 Zampieri FG, Mendes PV, Ranzani OT, et al. Extracorporeal membrane oxygenation for severe respiratory failure in adult patients: a systematic review and meta-analysis of current evidence. J Crit Care 2013; 28: 998–1005. 22 Fitzgerald M, Millar J, Blackwood B, et al. Extracorporeal carbon dioxide removal for patients with acute respiratory failure secondary to the acute respiratory distress syndrome: a systematic review. Crit Care 2014; 18: 222.

172

23 Tramm R, Ilic D, Davies AR, Pellegrino VA, Romero L, Hodgson C. Extracorporeal membrane oxygenation for critically ill adults. Cochrane Database Syst Rev 2015; 1: CD010381. 24 Schmidt M, Stewart C, Bailey M, et al. Mechanical ventilation management during extracorporeal membrane oxygenation for acute respiratory distress syndrome: a retrospective international multicenter study. Crit Care Med 2015; 43: 654–64.

www.thelancet.com/respiratory Vol 7 February 2019