Extracorporeal Membrane Oxygenation for Severe Acute Respiratory Distress Syndrome

Combes et al. N Engl J Med 2018; 378:1965-1975. DOI: 10.1056/NEJMoa1800385

Clinical Question

  • In patients with severe ARDS, does the early initiation of Extracorporeal Membrane Oxygenation (ECMO), compared to standard care, improve mortality at day 60?


  • Severe ARDS continues to have very high mortality despite improvements in mechanical ventilation and rescue therapies
  • Venovenous (VV) ECMO provides oxygenation and carbon dioxide removal and may enable improved lung rest and reduced ventilator induced lung injury (VILI)
  • However, ECMO is a highly invasive and complex procedure which may result in significant complications
  • The CESAR trial (Lancet 2009) showed that referral to an ECMO center for consideration ECMO improved outcomes, but many patients did not receive ECMO and the treatments in the control arm were not standardized
  • It remains unclear if ECMO improves outcomes compared to standard treatments, including protective lung ventilation, proning, and recruitment manoeuvres


  • Prospective, multicenter, randomized controlled trial
  • Maquet (an ECMO equipment provider) provided cannulas, circuits, and devices, but were not involved in trial design, data analysis or interpretation of manuscript
  • Research staff members responsible for the primary outcome were blinded, while participants, medical and nursing staff were not blinded
  • Randomization was stratified according to center and the duration of ventilation before randomization (<72 hours vs ≥72 hours)
  • Allocation concealment was ensured using a centralized, secure, web­based randomization system
  • Health care workers had protocol training prior to the trial’s commencement
  • Target sample size was 331 patients
    • Based an expected 60 day mortality rate of 60%, and an estimated 40% mortality in the ECMO group (Absolute risk reduction (ARR) 20%)
    • For 80% power at an alpha level of 5% and with a group­ sequential analysis occurring after the randomization of every 60 participants
  • Stopping rules, using the two ­sided triangular design methodology, were defined prior to the commencement of the trial. The trial could be stopped due to:
    • Safety (due to excess mortality in the ECMO arm)
    • Efficacy
    • Futility (ie unlikely to reach a definitive result)
  • The data safety monitoring board (DSMB) recommended stopping the trial after the 4th interim analysis 249/331 (75% of planned recruitment) when it met the predefined stopping rules for futility
  • A post hoc rank preserving structural failure time analysis was performed to adjust for high rate of crossover in the control group in the estimation of survival


  • 64 centres, international, though predominantly in France
  • Non ECMO centers were included if they had extensive expertise in treating patients with ARDS and ECMO could be established within 2 hours of randomization. ECMO centers were not defined


  • Inclusion:
    • ARDS (American-European Consensus Conference definition, 1994)
    • Endotracheal intubation of <7 days, AND had one of the following disease-severity criteria despite optimisation of mechanical ventilation (MV) (FIO2>80%, tidal volume (VT) 6 ml per kg, predicted body weight (PBW) and positive end-expiratory pressure (PEEP) >10 cm of water) and despite potential use of various usual adjunctive therapies (inhaled nitric oxide, recruitment manoeuvres, prone position, high frequency oscillation (HFO) ventilation, almitrine infusion):
      • PaO2:FIO2 ratio <50 mmHg for >3 hours; or,
      • PaO2:FIO2 <80 mmHg for >6 hours; or,
      • Arterial blood pH <7.25 with a partial pressure of arterial carbon dioxide (PaCO2) >60 mmHg for >6 hours (with respiratory rate increased to 35/min) resulting from MV settings adjusted to keep Pplat ≤32 cm H2O (first, tidal volume reduction by 1-mL/kg decrements to 4 mL/kg, then PEEP reduction to a minimum of 8 cm H2O)
  • Exclusion:
    • <18 years old
    • MV >7 days
    • Pregnancy
    • Overweight (a weight of more than 1 kg per centimeter of height or a BMI of more than 45)
    • Long-term chronic respiratory insufficiency treated with oxygen therapy or noninvasive ventilation
    • Cardiac failure resulting in venoarterial ECMO
    • A history of heparin­induced thrombocytopenia
    • Cancer with a life expectancy of less than 5 years
    • A moribund condition or a Simplified Acute Physiology Score (SAPS­II) value of more than 90
    • A current non–drug ­induced coma after cardiac arrest
    • Irreversible neurologic injury
    • A decision to withhold or withdraw life­sustaining therapies
    • An expected difficulty in obtaining vascular access for ECMO in the femoral or jugular vein, or a situation in which the ECMO device was not immediately available
  • 1015 patients were assessed for eligibility
  • 249 patients were randomized
    • Mean age range 51-54 years-old
    • Mean SOFA score range 10.6-10.8
    • Mean PF ratio range was 72-73
    • Septic shock range 65-67%
    • Bacterial pneumonia was commonest diagnosis, followed by viral pneumonia
    • Recruitment manoeuvers were more common in control group 34/125 (27%) vs 22/125 (18%), otherwise there were no obvious differences in baseline characteristics between groups at baseline


  • Venovenous ECMO
    • Percutaneous cannulas, with the access cannula inserted into the femoral vein and return cannula inserted into the SVC via the jugular vein
    • The Maquet CARDIOHELPTM device and MAQUET HLS Set Advanced 7.0 TM ECMO circuits
    • The pump outflow and the fraction of oxygen delivered (FDO2) by the gas blender (an air-oxygen mixture) were adjusted to obtain a PaO2 between 65 and 90 mmHg or an arterial oxygen saturation (SaO2) >90%.
    • The membrane ventilation was adjusted to maintain a partial pressure of carbon dioxide (PaCO2) <45 mmHg
  • Ventilator setting during ECMO:
    • Volume-assist control mechanical ventilation mode with following settings
      • FIO2: 30-50%;
      • PEEP at least 10 cmH2O;
      • VT was lowered to obtain a Pplat of 24 cmH2O or less;
      • Respiratory rate (RR): 10–30 breaths/min or
    • Bilevel positive airway pressure-release ventilation (APRV) with following settings
      • FiO2: 30-50%
      • PEEP at least 10 cmH2O
      • High pressure level 24 cmH2O or less
      • RR: 10–30 breaths/min
  • Anticoagulation with heparin targetting APTT 40-55 or anti Xa 0.2-0.3IU


  • Volume-assist-controlled ventilation
    • FIO2 (0.21–1) was adjusted to obtain SaO2 between 88-95% and a PaO2 55-80 mmHg
    • VT set at 6 ml per kg PBW
    • PEEP set so as not to exceed a Pplat of 28–30 cmH2
    • Neuromuscular blocking agents (NMBAs) and prone positioning were strongly encouraged for control group patients
    • When the oxygenation objectives were not met during at least 1 hour, despite an FIO2 of at least 0.8, the following adjuvant therapies could be used at the discretion of the treating clinicians:
      • Recruitment manoeuvres, inhaled nitric oxide (iNO) or inhaled prostacyclin, or intravenous almitrine
      • The RR could be set up to 35 breaths/min and adapted so as to obtain an arterial pH between 7.30 and 7.45
        • If, despite a respiratory rate of 35 breaths/min, the pH remained less than 7.30, sodium bicarbonate could be infused
        • If the pH remained <7.15 with a PaCO2 >35 mmHg, the VT could be increased stepwise by 1 ml/kg up to 8 ml/kg to obtain a pH of at least 7.15, provided that the PaCO2 remained at 35 mmHg or higher and the Pplat remained <32 cmH2O, if necessary by decreasing the PEEP to 5 cmH2O
        • If, despite VT set at 6 ml/kg, Pplat was over 30 cmH2O, the PEEP could be lowered to 5 cmH2
        • If despite a VT set at 6 ml/kg and PEEP set at 5 cmH2O, the Pplat remained above 32 cmH2O, the VT could be decreased stepwise by 1 ml/kg down to 4 ml/kg, provided that the pH was maintained at 7.15 or above

Cross over criteria (from control to ECMO)

  • SaO2 <80% for over 6 hours, despite the use of:
    • recruitment maneuvers, and
    • iNO or inhaled prostacyclin, and
    • a test of prone positioning, and
    • only if the patient had no irreversible multiple organ failure and
    • if the physician in charge of the patient believed that this could actually change the outcome
  • For patients cared for at non-ECMO centers, the mobile ECMO retrieval team was alerted if the aforementioned criteria were met and the team initiated ECMO at the non-ECMO center, returning to their home center with the patient then receiving ECMO
Management common to both groups
  • Administration of corticosteroids, choice of sedation and analgesia, haemodynamic management, weaning from the ventilator and tracheostomy were left to the discretion of the treating physician


  • Primary outcome: No statistical difference in mortality at day 60
    • ECMO group 44/124 (35%) vs Control group 57/125 (46%)
    • Relative risk, 0.76; 95% confidence interval [CI], 0.55 to 1.04; P=0.09
  • Secondary outcomes:
    • Compared to control group, the ECMO group had:
      • Lower relative risk of treatment failure 0.62 (95% CI, 0.47 to 0.82; P<0.001)
        • Treatment failure was defined as death by day 60 in patients in the ECMO group, and as crossover to ECMO or death in patients in the control group
      • Lower risk of Renal replacement therapy (RRT) at day 60 (50 vs. 32 days; median difference, 18 days; 95% CI, 0 to 51)
      • Underwent less proning (59 vs. 46 days; median difference, 13 days; 95% CI, 5 to 59)
    • Cross over to ECMO occurred in 35/125 (28%) of control patients
      • Occurred at median 4 days (IQR 1 to 7)
      • Median PaO2:FiO2 51 mmHg (IQR 46 to 61), median SaO2 77% (IQR 74 to 87), median lactate was 3.2 mmol/L (1.5-6.2)
      • 9 patients had cardiac arrest, 7 had right heart failure, 11 received RRT
      • 7 underwent venoarterial (VA) ECMO (and 6 had E-CPR)
      • 60 day mortality was higher in cross over patients than rest of controls: 20/35 (57%) in cross over patients vs 90 (41%) in remaining cross over patients
        • The hazard ratio for death within 60 days, for the ECMO group compared to controls after adjusting for selective crossover with the rank-preserving structural failure time analysis, was 0.51 (95% CI, 0.24 to 1.02; P=0.055) (supplemental analysis)
    • Adverse Events. Compared to control patients, ECMO patients had
      • More severe thrombocytopenia (<20,000 platelets per cubic millimeter; 27% vs. 16%; ARR 11 %; 95% CI, 0 to 21)
      • More bleeding events (defined as ≥1 RBC transfusion: 46% vs. 28%; ARR 18%; 95% CI, 6 to 30
      • Less ischaemic stroke (no patients vs. 5%; ARR -5%; 95% CI, −10 to −2)

Authors’ Conclusions

  • ECMO for severe ARDS showed no significant benefit of mortality at day 60 as compared with a strategy of conventional mechanical ventilation, which included crossover to ECMO (used by 28% of the patients in the control group)


  • Important and relevant research question about the effectiveness of ECMO in severe ARDS
  • Largest multicenter ECMO RCT to date
  • Standardised criteria for entry into the trial, and for the initiation of rescue ECMO
  • Almost all patients who were randomized to ECMO received it (unlike in CESAR)
  • Allocation concealment
  • Protocolized delivery of ECMO and MV resulted in standardized treatments in both intervention and control arms


  • Underpowered to answer the trial question
    • Trial was stopped early at 249/331 (75% of recruitment) due to predefined futility rules (ie unlikely to get a definitive result), leading to a high risk the study was underpowered.
    • Implausible power calculation
      • Initial power calculation was based on a 60% mortality in the control group, which became clear was inflated compared to the actual mortality rate in the controls of 46%
      • A 20% absolute risk reduction sets a very large requirement for a single ICU intervention, increasing the risk of a falsely negative trial  (CESAR, published in 2009, had a ARR of 16%)
  • High cross over rate of controls
    • The 28% cross over rate resulted a reduction of separation between the two arms, and diluted the ECMO treatment effect. This potentially impacted the trial in the following ways:
      • The cross over complicates the interpretation of the two arms in an intention to treat analysis. It also introduces a potential bias against the ECMO group in the secondary risk of treatment failure analysis, as ECMO was initiated much later and in sicker patients than the rest of the controls
      • Despite the use of objective criteria, the decision to crossover controls to ECMO was ultimately at the discretion of unblinded treating clinicians
      • Equipoise – high cross over rate raises the question whether some clinicians had clinical equipoise
  • Lack of blinding of clinicians and patients/families (however difficult in such a trial)
  • Slow recruitment of 249 patients over 6 years, leading to potential trial fatigue and/or change of practice
  • Threats to external validity
    • Expertise of ECMO centers not clearly defined
    • Importance of VV ECMO configuration is uncertain (i.e. femoral-jugular versus femoral-femoral approach)
    • Majority of ARDS patients in this trial were male and had pneumonia and septic shock. Extrapolation to other ARDS patient groups is uncertain
  • The trial does not distinguish between the use of VV ECMO and VA ECMO. In particular, 7 cross over control patients underwent VA ECMO for cardiac arrest
  • Some inclusion criteria may have been too liberal (e.g. PaO2 <80mmHg for 6 hours), which may dilute the benefit of VV ECMO and expose patients who might otherwise have lived to the complications of ECMO

The Bottom Line

  • The EOLIA trial provides inconclusive support for the benefit of ECMO in severe ARDS
    • The early initiation of ECMO did not improve 60 day mortality (with ARR 20%) compared to standard care in patients
    • However the data also suggest a possible clinical benefit from ECMO over standard care, especially if used early rather than late
  • Standard therapy, including proning and NMB, had a high rate of failure requiring rescue with ECMO

External Links


Summary author: Aidan Burrell
Summary date: 28th May 2018
Peer-review editor: Chris Nickson

Leave a Reply