Skip to main content

Foundational Trials

EOLIA

Image: Shutterstock / ChaNaWiT

Publication

  • Title: Extracorporeal Membrane Oxygenation for Severe Acute Respiratory Distress Syndrome
  • Acronym: EOLIA
  • Year: 2018
  • Journal published in: New England Journal of Medicine
  • Citation: Combes A, Hajage D, Capellier G, et al. Extracorporeal Membrane Oxygenation for Severe Acute Respiratory Distress Syndrome. N Engl J Med. 2018;378(21):1965-1975.

Context & Rationale

  • Background
    VV-ECMO was increasingly used as rescue support for refractory hypoxaemia/hypercapnia in ARDS, but contemporary randomised evidence was limited and confounded by centre effects, co-interventions, and evolving ventilatory standards.
    Equipoise persisted about whether “early” ECMO (before multi-organ injury accrues) improves survival compared with an optimised conventional strategy that includes prone positioning, neuromuscular blockade, higher PEEP, and rescue options.
  • Research Question/Hypothesis
    In adults with very severe ARDS, does early initiation of VV-ECMO reduce 60-day mortality compared with conventional lung-protective ventilation with standardised escalation and the option of rescue ECMO?
  • Why This Matters
    ECMO is resource-intensive and carries major haemorrhagic and neurological risks; a definitive estimate of benefit (and the impact of rescue crossover) is central to triage, service planning, and guideline recommendations for severe ARDS.

Design & Methods

  • Research Question: Whether early VV-ECMO reduces 60-day all-cause mortality versus conventional management in very severe ARDS.
  • Study Type: Randomised, multicentre, open-label, controlled, international trial with central randomisation (stratified by centre and by duration of mechanical ventilation before randomisation < 3 days vs ≥ 3 days) and a sequential (two-sided triangular) monitoring design.
  • Population:
    • Adult ICU patients with ARDS and very severe respiratory failure despite low tidal-volume ventilation (target 6 mL/kg predicted body weight) and PEEP ≥ 10 cmH2O, meeting at least one of: PaO2/FiO2 < 50 mmHg for > 3 h; PaO2/FiO2 < 80 mmHg for > 6 h; or pH < 7.25 with PaCO2 ≥ 60 mmHg for > 6 h despite RR increased to 35/min (maintaining Pplat ≤ 32 cmH2O).
    • Key exclusions: ≥ 7 days of mechanical ventilation before randomisation; age < 18; pregnancy; BMI > 45 kg/m2; irreversible comorbidity/expected death, contraindication to anticoagulation/ECMO, or decision to limit life-sustaining treatment.
  • Intervention:
    • Early VV-ECMO (goal initiation within 2 h after randomisation) using percutaneous cannulation; anticoagulation with unfractionated heparin targeting aPTT 40–55 s (or anti-Xa 0.2–0.3 IU/mL).
    • Post-cannulation “ultra-protective” ventilation targets: VT 3–4 mL/kg PBW; PEEP ≥ 10 cmH2O; Pplat ≤ 24 cmH2O; FiO2 0.30–0.60; RR 10–30/min.
  • Comparison:
    • Conventional lung-protective ventilation, including higher-PEEP strategy (target Pplat 28–30 cmH2O) with strong encouragement to use neuromuscular blockade and prone positioning.
    • Rescue therapies permitted (including inhaled nitric oxide/prostacyclin and recruitment manoeuvres); rescue ECMO allowed for refractory hypoxaemia (SaO2 < 80% for > 6 h despite recruitment manoeuvres, inhaled nitric oxide/prostacyclin, and prone positioning when feasible) and in the absence of irreversible multi-organ failure.
  • Blinding: Open-label; primary outcome (all-cause mortality) objective, but decisions around escalation and crossover were clinically mediated (with protocolised rescue criteria).
  • Statistics: Sequential two-sided triangular test; assumed 60-day mortality 60% (control) vs 40% (ECMO) (absolute reduction 20%); α=0.05; power 80%; maximum sample size 331; interim analyses planned every 60 patients; primary analysis by intention-to-treat.
  • Follow-Up Period: Primary endpoint at 60 days; additional follow-up through day 90 for mortality and selected secondary outcomes.

Key Results

This trial was stopped early. Recruitment was halted after the 4th planned interim analysis (triggered at 240 randomised participants) because the prespecified futility boundary was crossed; 249 participants had been randomised in total due to enrolment overrun.

Outcome Early VV-ECMO (n=124) Conventional (n=125) Effect p value / 95% CI Notes
All-cause mortality at day 60 (primary) 44/124 (35%) 57/125 (46%) RR 0.76 95% CI 0.55 to 1.04; P=0.09 Frequentist primary endpoint did not cross prespecified significance boundary.
Treatment failure by day 60 (death in ECMO group; death or crossover to ECMO in control) 44/124 (35%) 72/125 (58%) RR 0.62 95% CI 0.47 to 0.82; P<0.001 Key secondary endpoint; structurally penalises rescue ECMO crossover in control.
All-cause mortality at day 90 50/124 (40%) 59/125 (47%) Difference −7% 95% CI −19 to 5 Between-group difference not statistically tested in the reported table.
Rescue ECMO crossover (control group) — 35/125 (28%) — Not reported Mean time to ECMO crossover: 6.5 ± 9.7 days after randomisation; mortality among crossover patients: 20/35 (57%).
Median days free from mechanical ventilation (inclusion to day 60) 23 (0 to 40) 3 (0 to 36) Median difference 20 days 95% CI −5 to 32 Composite of survival and time off invasive ventilation; highly influenced by early deaths.
Median days free from renal failure (inclusion to day 60) 46 (0 to 56) 21 (0 to 53) Median difference 25 days 95% CI 6 to 53 Supports a signal toward reduced multi-organ failure burden.
Prone positioning (any use) 82/124 (66%) 112/125 (90%) Difference −24% 95% CI −35 to −14 Co-intervention use was greater in control (as intended by protocolised escalation).
Bleeding leading to transfusion (safety) 57/124 (46%) 35/125 (28%) Difference 18% 95% CI 6 to 30 Clinically important harm signal consistent with ECMO anticoagulation and cannulation risks.
Severe thrombocytopenia (safety) 34/124 (27%) 20/125 (16%) Difference 12% 95% CI 1 to 22 Potentially reflects circuit-related platelet consumption and/or anticoagulation effects.
Ischaemic stroke (safety) 0/124 (0%) 6/125 (5%) Difference −5% 95% CI −10 to −2 Direction unexpected; small numbers; interpret cautiously.
  • Early VV-ECMO produced a numerically lower 60-day mortality (35% vs 46%; RR 0.76; 95% CI 0.55 to 1.04; P=0.09) but the trial stopped early for futility and did not meet the prespecified frequentist threshold.
  • High rescue ECMO crossover in the control group (28%) likely attenuated contrast in mortality while remaining clinically and ethically important.
  • Safety profile differed: transfusion-requiring bleeding was higher with ECMO (46% vs 28%; difference 18%; 95% CI 6 to 30), alongside more severe thrombocytopenia (27% vs 16%; difference 12%; 95% CI 1 to 22).

Internal Validity

  • Randomisation and allocation: Central randomisation stratified by centre and pre-randomisation duration of mechanical ventilation (< 3 days vs ≥ 3 days), supporting allocation concealment and balance.
  • Baseline comparability and severity: Groups were similar at enrolment (e.g., age 51.9 ± 12.5 vs 54.4 ± 13.4 years; SOFA 10.8 ± 3.4 vs 10.6 ± 3.3; PaO2/FiO2 73 ± 30 vs 72 ± 24 mmHg).
  • Protocol adherence / separation: 121/124 (97.6%) assigned to ECMO received ECMO; control “contamination” occurred via rescue crossover in 35/125 (28%).
  • Co-interventions: Control received more prone positioning (90% vs 66%; difference −24%; 95% CI −35 to −14) and more inhaled vasodilators (62% vs 46%; difference −16%; 95% CI −29 to −3), which could reduce apparent incremental benefit of ECMO (but reflects real-world escalation pathways).
  • Performance/detection bias: Open-label design; primary outcome objective. However, decisions to cross over to rescue ECMO (and timing) were clinically mediated, even with protocol criteria.
  • Early stopping and power: Designed for up to 331 patients to detect a 20% absolute mortality reduction; stopped at 249, leaving wider uncertainty around mortality effects (95% CI includes clinically meaningful benefit and no effect).
  • Selection into randomisation: Of 1015 eligible patients, 249 were randomised, implying substantial screening/selection that may enrich for patients meeting logistical and equipoise criteria (potential selection bias away from “all comers”).
  • Subgroup consistency (exploratory): No strong interaction signals were reported across prespecified subgroups; examples include PaO2/FiO2 ≥ 66 mmHg (RR 0.56; 95% CI 0.34 to 0.91) vs < 66 mmHg (RR 1.04; 95% CI 0.68 to 1.58), interaction P=0.173; SOFA < 11 (RR 0.55; 95% CI 0.31 to 0.99) vs ≥ 11 (RR 0.95; 95% CI 0.67 to 1.35), interaction P=0.105.

Conclusion on Internal Validity: Moderate: randomisation and objective primary outcome support internal validity, but early stopping and substantial rescue ECMO crossover plausibly diluted mortality differences and complicate interpretation of effect size.

External Validity

  • Population representativeness: The cohort represents a highly selected, very severe ARDS population (PaO2/FiO2 ~72 mmHg) meeting strict physiological thresholds; patients ventilated ≥ 7 days, those with BMI > 45, pregnancy, and those with major contraindications were excluded.
  • Centre and system requirements: Implementation requires mature ECMO programmes (and, in this trial, a mobile ECMO retrieval model for non-ECMO centres), limiting immediate generalisability to low-volume or resource-limited settings.
  • Applicability: Most applicable to high-income ICUs that can deliver contemporary lung-protective ventilation, prone positioning, and timely VV-ECMO with complication management; less applicable where ECMO access is delayed or prone/NMBA practice differs substantially.

Conclusion on External Validity: Good for patients with very severe ARDS in experienced ECMO networks; limited for broader ARDS populations and health systems without rapid VV-ECMO capability.

Strengths & Limitations

  • Strengths: Enrolled a clearly defined, very severe ARDS population; pragmatic network design including non-ECMO centres with retrieval; protocolised conventional management with explicit rescue ECMO criteria; objective primary endpoint; high adherence to allocated ECMO strategy (97.6% received ECMO).
  • Limitations: Open-label; early stopping for futility with reduced sample size versus plan (249 vs 331 maximum); substantial rescue ECMO crossover (28%); large attrition from eligibility to randomisation (249/1015), potentially limiting representativeness; composite “treatment failure” endpoint embeds crossover in the control definition.

Interpretation & Why It Matters

  • Clinical interpretation
    EOLIA did not demonstrate a statistically significant reduction in 60-day mortality on the prespecified frequentist analysis (RR 0.76; 95% CI 0.55 to 1.04; P=0.09), but the point estimate and secondary outcomes are compatible with clinically meaningful benefit in a subset of patients with very severe ARDS.
    The high rescue ECMO crossover rate (28%) is central: it likely reduced the apparent mortality contrast while reflecting ethically necessary salvage practice, making EOLIA a trial of early ECMO strategy within a system that can deliver rescue ECMO.
    For modern practice, EOLIA supports VV-ECMO as part of a structured escalation pathway (optimised conventional care plus rescue ECMO capability), with careful balancing of potential benefit against haemorrhagic risk (transfusion-requiring bleeding 46% with ECMO).

Controversies & Subsequent Evidence

  • Early stopping and “negative” framing: Stopping for futility at 249 patients widened uncertainty, and the confidence interval remains consistent with meaningful mortality reduction; interpretation depends on philosophy of evidence thresholds and the weight given to rescue crossover.
  • Crossover as dilution vs ethical necessity: Rescue ECMO was frequent (28%) and associated with high mortality (57%), raising debates about whether earlier ECMO could be superior to delayed salvage in carefully selected patients.
  • Bayesian re-interpretation: A post hoc Bayesian analysis estimated a high posterior probability of mortality benefit across a range of priors, reframing EOLIA as “uncertain but likely beneficial” rather than “no effect”. 1
  • Meta-analytic synthesis: An individual patient data meta-analysis of CESAR and EOLIA reported lower 90-day mortality with ECMO (RR 0.75; 95% CI 0.60 to 0.94) and highlighted the extent of rescue ECMO in controls (17% overall; 35 in EOLIA). 23
  • Guidelines reflect conditional endorsement: Contemporary international guidelines recommend considering VV-ECMO in selected patients with severe ARDS at experienced centres, typically as escalation after evidence-based conventional strategies (including prone positioning). 456

Summary

  • EOLIA randomised 249 patients with very severe ARDS to early VV-ECMO vs optimised conventional ventilation with protocolised escalation and rescue ECMO.
  • The trial stopped early for futility; 60-day mortality was 35% with ECMO vs 46% with conventional care (RR 0.76; 95% CI 0.55 to 1.04; P=0.09).
  • Rescue ECMO crossover occurred in 28% of controls, likely attenuating mortality separation and shifting interpretation to an “early ECMO strategy” question.
  • ECMO increased transfusion-requiring bleeding (46% vs 28%) and severe thrombocytopenia (27% vs 16%), emphasising harm-benefit trade-offs and the need for experienced centres.
  • Subsequent Bayesian analyses, IPD meta-analysis, and guideline updates support conditional use of VV-ECMO in selected severe ARDS within established ECMO networks.

Further Reading

Other Trials

Systematic Review & Meta Analysis

Observational Studies

Guidelines

Notes

  • In ECMO trials where rescue crossover is ethically and clinically mandated, complementary estimands (e.g., Bayesian posterior probabilities, and meta-analytic approaches incorporating crossover) can be informative alongside prespecified frequentist primary analyses.

Overall Takeaway

EOLIA is the pivotal modern RCT of an early VV-ECMO strategy for very severe ARDS within a system that could deliver rescue ECMO. Although it stopped early and did not meet its frequentist primary endpoint, its point estimates, secondary outcomes, and subsequent syntheses underpin contemporary conditional guideline endorsement of VV-ECMO for carefully selected patients in experienced centres, with explicit attention to haemorrhagic risk.

Overall Summary

  • Early VV-ECMO in very severe ARDS reduced “treatment failure” and suggested (but did not prove) lower mortality; interpretation is dominated by early stopping and 28% rescue ECMO crossover.

Bibliography