Skip to main content

Foundational Trials

CESAR

Image: Shutterstock / ChaNaWiT

Publication

  • Title: Efficacy and economic assessment of conventional ventilatory support versus extracorporeal membrane oxygenation for severe adult respiratory failure (CESAR): a multicentre randomised controlled trial
  • Acronym: CESAR
  • Year: 2009
  • Journal published in: The Lancet
  • Citation: Peek GJ, Mugford M, Tiruvoipati R, et al; CESAR trial collaboration. 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(9698):1351-1363.

Context & Rationale

  • Background
    • Severe adult respiratory failure (including ARDS) carried high mortality despite “best” conventional ventilation, and ventilator-induced lung injury was increasingly recognised as an iatrogenic contributor to poor outcomes.
    • Modern extracorporeal membrane oxygenation (ECMO) offered a physiological rationale: provide gas exchange, facilitate “lung rest”, and potentially avoid injurious ventilator settings.
    • Earlier adult extracorporeal support trials were conducted with older technology and different critical care standards, limiting inference for contemporary practice.
    • In the UK, neonatal ECMO had already been evaluated in a pragmatic referral-based randomised trial, providing a potential template for adult evaluation.
  • Research Question/Hypothesis
    • Whether referral and transfer of adults with severe but potentially reversible respiratory failure to a specialist ECMO centre (for protocolised management and ECMO when indicated) improved survival without severe disability at 6 months compared with continued conventional management in existing centres.
    • Whether such a strategy was economically justifiable within the NHS (short-term within-trial and longer-term modelled cost-utility).
  • Why This Matters
    • ECMO is resource-intensive, logistically complex, and carries transport, anticoagulation, and circuit-related risks; therefore, an RCT-level estimate of benefit (and cost-effectiveness) was crucial for service planning.
    • A positive result would justify centralised ECMO retrieval networks and influence escalation thresholds for refractory hypoxaemia/hypercapnia.
    • A neutral/negative result would argue for maximising conventional lung-protective strategies and adjuncts rather than expanding ECMO services.

Design & Methods

  • Research Question: In adults aged 18–65 years with severe but potentially reversible respiratory failure, does randomisation to referral/transfer for consideration of ECMO (with management at a specialist ECMO centre) increase survival without severe disability at 6 months compared with continued conventional management at the referring centre?
  • Study Type: Multicentre, pragmatic, randomised controlled trial; UK NHS setting; many referring ICUs with a single designated ECMO centre (Glenfield Hospital, Leicester) delivering the intervention strategy; unblinded clinical management; follow-up assessors blinded.
  • Population:
    • Setting: Adults in critical care units referred for severe respiratory failure.
    • Key inclusion (threshold-based): Age 18–65 years; severe but potentially reversible respiratory failure defined by either (i) Murray lung injury score ≥3.0, or (ii) uncompensated hypercapnia with arterial pH <7.20, despite “optimal conventional treatment”.
    • Acceleration criterion: Considered for inclusion if Murray score ≥2.5 when deterioration was expected with ongoing conventional management (to enable earlier trial entry).
    • Key exclusions: High-pressure (peak inspiratory pressure >30 cmH2O) or high FiO2 (>0.8) ventilation for >7 days (168 h); signs of intracranial bleeding; other contraindication to limited heparinisation; contraindication to continuation of active treatment.
  • Intervention:
    • Randomised strategy: Referral/transfer to a specialist ECMO centre for an ECMO-capable management protocol (not “ECMO for all”).
    • At ECMO centre (first 12 h): Protocolised optimisation of conventional management, including lung-protective limits (pressure-restricted ventilation), diuresis, nutritional support, and adjuncts where appropriate; ECMO initiation if severe hypoxaemia persisted (FiO2 >0.9 to maintain SaO2 >90%) and/or persistent acidosis (pH <7.20), or if haemodynamic instability prevented further optimisation.
    • ECMO delivery (when used): Predominantly venovenous ECMO with percutaneous cannulation; “lung rest” settings reported as peak inspiratory pressure 20–25 cmH2O, PEEP 10–15 cmH2O, ventilator rate 10 breaths/min, FiO2 0.3.
    • Duration: Median ECMO duration 9.0 days (IQR 6.0–16.0) among those receiving ECMO.
    • Transport: Dedicated ECMO team supported inward transport (air or ground); transport was not undertaken on ECMO.
  • Comparison:
    • Conventional management in the existing centre (including potential transfer to other non-ECMO centres as per usual practice); no mandated protocol for ventilation or adjunct therapies.
    • Recommended (not enforced) approach: “Low-volume low-pressure ventilation strategy” (tidal volumes 4–8 mL/kg bodyweight with plateau pressure <30 cmH2O).
    • Contamination control: Crossover to ECMO was not permitted by protocol.
  • Blinding: Treatment allocation could not be blinded; 6-month follow-up assessments were undertaken by researchers blinded to group allocation.
  • Statistics: Planned sample size 240 to detect a reduction in the primary composite endpoint (death or severe disability) from 73% to 55% (absolute 18%), with 80% power (β=0.20) at two-sided α=0.05; primary analysis by intention-to-treat; sample size revised to 180 after observed control event rate of 67% at n=63 (recalculated to detect an approximate one-third reduction). 12
  • Follow-Up Period: Primary clinical endpoint at 6 months after randomisation (or earlier at discharge if discharged after 6 months); within-trial economic evaluation to 6 months plus modelled lifetime cost-utility.

Key Results

This trial was not stopped early. Recruitment target was revised (240 to 180) following independent data monitoring committee review of the observed control-group event rate, before completion of recruitment.

Outcome Referral to ECMO centre Conventional management Effect p value / 95% CI Notes
Primary outcome: death or severe disability at 6 months 33/90 (37%) 46/87 (53%) RR 0.69 95% CI 0.05 to 0.97; P=0.03 Primary outcome known for 177/180 patients (3 survivors in conventional group had unknown disability status).
Sensitivity in report: if all 3 were severely disabled, RR 0.67; 95% CI 0.47 to 0.97; P=0.03; if none were severely disabled, RR 0.72; 95% CI 0.51 to 1.01; P=0.051.
Death ≤6 months or before discharge 33/90 (37%) 45/90 (50%) RR 0.73 95% CI 0.52 to 1.03; P=0.07 Conventional-group percentage corrected to 50% in published Department of Error. 3
Time from randomisation to death (median, IQR): 15 (3–41) days vs 5 (2–14) days.
Severe disability at 6 months (among those with known status) 0 reported 1 patient reported Not reported Not reported In the trial report, only one patient receiving conventional management was known to be severely disabled at 6 months; severe disability defined as “confined to bed” and unable to wash/dress alone.
Critical care length of stay (all patients) 24.0 (13.0–40.5) days 13.0 (11.0–16.0) days Not reported Not reported Median (IQR); longer length of stay in the ECMO-referral group was a major driver of cost differences.
Hospital length of stay (all patients) 35.0 (15.6–74.0) days 17.0 (4.8–45.3) days Not reported Not reported Median (IQR).
Low-volume low-pressure ventilation strategy at any time after randomisation 84/90 (93%) 63/90 (70%) Not reported P<0.0001 Time under strategy (mean, SD): 23.9 (20.4) days vs 15.0 (21.1) days; P<0.0001.
Serious adverse events (reported) 2 events Not reported Not reported Not reported Both events were fatal: ambulance oxygen supply failure during transfer; and vessel perforation during cannulation. A further death during transfer (pulmonary haemorrhage) occurred but was not classified as a serious adverse event in the report.
Within-trial (6 months) mean health-care cost per patient (base case) £73,979 £33,435 Difference £40,544 95% CI £24,799 to £56,288 Costs in 2005 prices (NHS perspective); complete-case costing reported for 90 vs 87 patients.
Incremental cost-effectiveness (base case): cost per additional survivor without severe disability at 6 months Mean cost £73,979; probability 0.63 Mean cost £33,435; probability 0.47 £250,162 Not reported Base-case ICER reported in trial economic analysis table.
Modelled lifetime cost-utility (gain in QALYs extended over lifetime) £19,252 per QALY Not reported Not reported 95% CI £7,622 to £59,200 Value stated in published Department of Error. 3
  • Primary endpoint difference reflected mainly mortality (severe disability was rare: 0 vs 1 reported), while mortality alone did not reach conventional statistical significance (RR 0.73; 95% CI 0.52 to 1.03; P=0.07).
  • Intervention delivery was heterogeneous by design: 68/90 (76%) in the referral group received ECMO; 22/90 (24%) did not receive ECMO (including clinical improvement on protocolised conventional management and deaths before/during transfer).
  • Referral strategy was associated with substantially longer critical care and hospital stay (median ICU 24.0 vs 13.0 days; median hospital 35.0 vs 17.0 days) and higher short-term costs.

Internal Validity

  • Randomisation and allocation concealment: Central telephone randomisation using minimisation with a random element; minimisation factors included age, hours of high-pressure/high FiO2 ventilation, hypoxaemia vs hypercapnia, diagnostic group, number of organs failed, and type of centre (ECMO centre vs non-ECMO centre), supporting concealment at the point of assignment.
  • Post-randomisation exclusions and missingness: Primary outcome known for 177/180; 3 survivors in conventional management had unknown severe disability status at 6 months (sensitivity analyses provided in report).
  • Attrition for health-related quality-of-life measurement: At 6 months, 52 patients in the ECMO-referral group and 32 in the conventional group completed formal assessment; additional patients had “restricted” follow-up information (e.g., vital status only), which limits precision and may introduce bias for secondary QoL and economic outcomes.
  • Performance and detection bias: Trial was necessarily open-label for clinicians; only follow-up researchers were blinded, reducing bias for the disability component of the primary endpoint but not for co-interventions, ventilator management, discharge timing, or resource utilisation.
  • Protocol adherence (intervention fidelity): Of 90 allocated to ECMO referral, 68 (76%) received ECMO with median duration 9.0 days (IQR 6.0–16.0); 22 (24%) received conventional management instead (including 16 judged to have improved and 5 deaths before/during transfer).
  • Baseline comparability: Groups were broadly similar on reported demographics and severity indices (e.g., mean age ~40 years; Murray score mean 3.5 vs 3.4; organ failure median 2 vs 2; duration of IPPV before trial entry median 35 vs 37 h), supporting comparability, though some baseline variables had missingness (e.g., APACHE II not recorded in 37% vs 32%).
  • Timing and logistics (risk of delay/transport harms): Median time from randomisation to start of treatment at the ECMO centre was 6.1 h (IQR 4.0–7.1); 2 patients died during transfer and 3 died before transfer, directly embedding real-world retrieval risk into the intervention strategy.
  • Separation of the variable of interest (centre strategy vs ECMO per se): The trial compared “referral to a specialist ECMO centre” versus continued conventional care, rather than “ECMO vs no ECMO” in the same setting; 24% of the referral group did not receive ECMO, and both ventilation practices and adjunct therapies differed between groups.
  • Ventilator practice separation (reported): Low-volume low-pressure ventilation strategy at any time: 93% (84/90) vs 70% (63/90); P<0.0001; time under strategy (mean, SD): 23.9 (20.4) days vs 15.0 (21.1) days; P<0.0001.
  • Adjunctive therapy imbalance (reported): Steroids 84% (76/90) vs 64% (58/90); P=0.001; MARS liver support 17% (15/90) vs 0%; P<0.0001; prone positioning 36% (32/90) vs 42% (38/90); P=0.58; nitric oxide 10% vs 7%; P=0.60.
  • Outcome definition and assessment: Severe disability was defined using EQ-5D domains (mobility/self-care) as confinement to bed and inability to wash/dress; ascertainment at 6 months was performed by blinded assessors, supporting objectivity for disability classification, though missingness existed.
  • Statistical rigour: Intention-to-treat framework preserved; primary endpoint statistically significant; sample size was revised mid-trial based on observed control event rate, which is methodologically defensible under independent monitoring but can raise interpretive concerns regarding power and estimation stability for secondary outcomes (notably mortality alone).

Conclusion on Internal Validity: Moderate—allocation concealment and intention-to-treat analysis support credibility, but the pragmatic “referral strategy” intervention, lack of blinded co-interventions, differential ventilator practice/adjunctive therapy use, and missing disability/QoL data constrain attribution of benefit specifically to ECMO rather than to centre-based care processes.

External Validity

  • Population representativeness: Adults aged 18–65 with severe but potentially reversible respiratory failure; applicability to older patients, those with major comorbidities, prolonged high-intensity ventilation (>7 days), or contraindications to anticoagulation is limited by design.
  • Health system dependence: The strategy relied on a centralised specialist ECMO centre with a dedicated retrieval team and transport infrastructure; generalisability depends on availability of similar regionalised ECMO networks and safe inter-hospital transfer capability.
  • Comparator realism: Conventional management varied across referring centres and was not protocolised, resembling real-world heterogeneity but complicating extrapolation to units with consistently high adherence to lung-protective ventilation and standardised ARDS adjunct algorithms.
  • Intervention scalability: The trial tested a single-centre ECMO service model; scaling to multiple centres may alter performance due to variable expertise, case mix, and volume-outcome effects.

Conclusion on External Validity: Findings generalise best to high-resource systems able to deliver timely referral and safe retrieval to experienced ECMO centres for carefully selected, potentially reversible severe respiratory failure; extrapolation beyond these constraints (older age groups, longer prior ventilation, limited retrieval capacity) is uncertain.

Strengths & Limitations

  • Strengths:
    • Pragmatic multicentre randomised evaluation of a complex, service-level intervention (including real transport risks and systems constraints).
    • Central randomisation with minimisation and random element; intention-to-treat analysis.
    • Primary outcome incorporated patient-centred disability status with blinded follow-up assessors.
    • Embedded economic evaluation (within-trial costing and modelled lifetime cost-utility) supporting policy relevance.
  • Limitations:
    • Intervention was “referral to ECMO centre” rather than ECMO itself; single specialist centre delivered the core intervention, limiting ability to isolate ECMO effect from centre expertise and protocolised care.
    • Unblinded management with measurable between-group differences in ventilation strategy uptake and adjunct therapies, introducing risk of performance bias and confounding by co-intervention.
    • Primary outcome statistical significance depended on a composite outcome in which severe disability was rare and disability status was unknown for 3 conventional-group survivors.
    • Mortality alone was not statistically significant and the trial was not powered for mortality as a standalone endpoint after sample size revision.
    • Substantial missingness for detailed 6-month QoL assessment (52 vs 32 assessed), affecting precision and robustness of secondary QoL and economic endpoints.

Interpretation & Why It Matters

  • Clinical implication
    For selected adults with severe, potentially reversible respiratory failure, a strategy of early referral and transfer to an experienced ECMO centre can increase the likelihood of survival without severe disability at 6 months (63% vs 47% in CESAR).
  • Service delivery implication
    The core “dose” tested was specialist-centre care (including protocolised lung-protective ventilation and escalation pathways) plus ECMO when indicated, embedding a real-world argument for regionalised networks and retrieval capability.
  • Cost and capacity implication
    Short-term resource use was markedly higher (ICU median 24 vs 13 days; hospital median 35 vs 17 days), with higher 6-month costs; economic acceptability depends on longer-term valuation of survivors and assumptions about durable quality-of-life benefit.

Controversies & Subsequent Evidence

  • What was actually randomised: The study tested referral/transfer to a single specialist ECMO centre rather than ECMO versus no ECMO; 24% of the referral group did not receive ECMO, so improved outcomes could plausibly reflect specialist-centre processes and protocolised conventional care as much as extracorporeal support. 4
  • Comparator care quality and separation: Conventional management was not protocolised, and uptake of the “low-volume low-pressure” strategy differed (93% vs 70%), raising concern that part of the observed effect might be attributable to differential adherence to lung-protective ventilation rather than ECMO itself. 4
  • Primary outcome interpretability: Severe disability was rare (0 vs 1), making the composite endpoint predominantly a mortality signal; disability status was unknown for 3 conventional-group survivors and the primary relative risk confidence interval appears typographically inconsistent in the published table (no corresponding correction identified), underscoring fragility of inference from small numbers.
  • Short-term mortality signal: Mortality at ≤6 months favoured referral (37% vs 50%) but did not reach conventional statistical significance (P=0.07), emphasising limited power for mortality alone and the importance of subsequent trials and meta-analyses.
  • Subsequent RCT evidence: EOLIA (2018) evaluated early venovenous ECMO for severe ARDS with protocolised conventional management and allowed crossover to ECMO; intention-to-treat mortality was not statistically different, but crossover was substantial, complicating causal interpretation. 5
  • Evidence syntheses: Systematic reviews and individual patient data meta-analyses incorporating CESAR and EOLIA generally support a survival benefit for venovenous ECMO in severe ARDS when delivered in experienced centres, while highlighting residual uncertainty from small numbers of RCTs and trial design heterogeneity. 67
  • Observational signals in pandemics (selection and system effects): During the 2009 H1N1 pandemic, observational UK data associated referral to an ECMO centre with lower mortality, supporting a “centre strategy” effect consistent with CESAR’s design. 89
  • Guideline incorporation: Modern ARDS and ECMO guidelines incorporate CESAR (and later evidence), recommending consideration of venovenous ECMO as rescue therapy in severe ARDS in experienced centres with attention to patient selection, timing, and transport logistics. 101112

Summary

  • CESAR randomised 180 adults with severe, potentially reversible respiratory failure to referral/transfer for consideration of ECMO at a specialist centre versus continued conventional management.
  • Primary outcome (death or severe disability at 6 months) favoured referral (37% vs 53%; RR 0.69; 95% CI 0.05 to 0.97; P=0.03), but severe disability was rare and mortality alone did not reach conventional statistical significance.
  • The intervention tested a centre-based strategy; only 68/90 (76%) actually received ECMO, and between-group differences in lung-protective ventilation uptake and adjunct therapies were substantial.
  • Referral was associated with longer ICU/hospital stay (median ICU 24 vs 13 days; hospital 35 vs 17 days) and higher short-term costs; modelled lifetime cost per QALY was reported as £19,252 (95% CI £7,622 to £59,200).
  • CESAR catalysed development of adult ECMO referral pathways and retrieval networks, while leaving an enduring methodological debate about ECMO versus “specialist centre care”.

Further Reading

Other Trials

Systematic Review & Meta Analysis

Observational Studies

Guidelines

Notes

  • CESAR is best interpreted as a trial of a regionalised referral-and-retrieval ECMO service model; it cannot fully disentangle ECMO’s causal effect from specialist-centre care processes.
  • Within-trial disability outcomes were sparse; mortality dominated the primary composite endpoint, and missing disability status in 3 survivors in the control group required sensitivity analysis.

Overall Takeaway

CESAR was a landmark because it provided pragmatic randomised evidence that, in a real-world health system, early referral and transfer of carefully selected severe respiratory failure patients to a specialist ECMO centre improved 6-month survival without severe disability compared with continued conventional care. Its enduring significance lies as much in its service-model implications (regionalisation, retrieval, and protocolised lung-protective management) as in its direct evaluation of ECMO as a technology.

Overall Summary

  • Referral to a specialist ECMO centre improved the primary composite endpoint (death/severe disability) at 6 months (37% vs 53%).
  • Only 76% of the referral arm received ECMO; the trial tested a centre strategy, not “ECMO for all”.
  • Ventilation practice differed materially between groups (93% vs 70% received low-volume low-pressure ventilation at any time).
  • Short-term costs and length of stay were higher; lifetime modelled cost per QALY was reported as £19,252 (95% CI £7,622 to £59,200).

Bibliography