Skip to main content

Foundational Trials

EUPHRATES

Image: Shutterstock / ChaNaWiT

Publication

  • Title: Effect of Targeted Polymyxin B Hemoperfusion on 28-Day Mortality in Patients With Septic Shock and Elevated Endotoxin Level: the EUPHRATES Randomized Clinical Trial
  • Acronym: EUPHRATES
  • Year: 2018
  • Journal published in: JAMA
  • Citation: Dellinger RP, Bagshaw SM, Antonelli M, Foster DM, Klein DJ, Marshall JC, et al; EUPHRATES Trial Investigators. Effect of targeted polymyxin B hemoperfusion on 28-day mortality in patients with septic shock and elevated endotoxin level: the EUPHRATES randomized clinical trial. JAMA. 2018;320(14):1455-1463.

Context & Rationale

  • Background
    • Endotoxin (lipopolysaccharide) is a potent mediator of inflammation, vasoplegia, and organ dysfunction in sepsis, particularly in Gram-negative infection.
    • Polymyxin B–immobilised fibre haemoperfusion (PMX-HP/PMX-DHP) is an extracorporeal endotoxin adsorption strategy widely used in Japan, but global uptake has been limited by uncertainty about efficacy and reproducibility.
    • Earlier clinical trials produced mixed signals, including an early positive RCT (EUPHAS) and later neutral/negative RCTs in similar populations (ABDOMIX).12
    • A major methodological gap was the absence (until EUPHRATES) of a large, blinded, sham-controlled trial targeting therapy to patients with demonstrable endotoxaemia.
  • Research Question/Hypothesis
    • In adults with septic shock and elevated endotoxin activity (endotoxin activity assay [EAA] ≥0.60), does adding 2 sessions of polymyxin B haemoperfusion reduce 28-day all-cause mortality compared with a sham procedure?
    • After prespecified interim monitoring, the trial additionally evaluated whether any effect would be clearer in patients at higher baseline risk (Multiple Organ Dysfunction Score [MODS] >9).
  • Why This Matters
    • PMX haemoperfusion is resource-intensive (vascular access, extracorporeal circuit, anticoagulation, specialist staff) and carries device/circuit risks; robust evidence is required before widespread adoption.
    • The trial tested a “theragnostic” concept: using a biomarker (EAA) to identify patients most likely to benefit from a targeted extracorporeal therapy.

Design & Methods

  • Research Question: Whether polymyxin B haemoperfusion, targeted to endotoxaemic septic shock (EAA ≥0.60), reduces 28-day all-cause mortality versus a sham procedure.
  • Study Type: Multicentre, randomised, double-blind, sham-controlled clinical trial conducted in 55 tertiary hospitals (United States and Canada) in ICU settings.
  • Population:
    • Adults with septic shock and elevated endotoxin activity (EAA ≥0.60) assessed within 24 hours of shock onset.
    • Septic shock severity was high (median MODS 10 [IQR 9–12] in both groups at baseline; mechanical ventilation in 201/224 vs 194/226).
    • After the second interim analysis, subsequent enrolment was restricted to higher-risk patients (MODS >9), with MODS ≤9 becoming an exclusion criterion.
    • Key exclusions included inability to receive systemic anticoagulation with unfractionated heparin, pregnancy, and conditions expected to preclude meaningful 28-day survival (e.g., anticipated death within 28 days from pre-existing disease).
  • Intervention:
    • Two sessions of polymyxin B–immobilised fibre column haemoperfusion (Toraymyxin), each intended for 2 hours, performed approximately 24 hours apart.
    • Haemoperfusion delivered via a 12 Fr double-lumen dialysis catheter (internal jugular or femoral) and standard haemoperfusion/dialysis platform.
    • Anticoagulation with intravenous unfractionated heparin (citrate anticoagulation prohibited).
    • First session initiated promptly after randomisation (median 3 h 30 min from randomisation to initiation).
  • Comparison:
    • Sham perfusion designed to mimic treatment duration and workflow while avoiding intravascular catheterisation and blood exposure to an adsorption cartridge.
    • Sham catheter simulated (catheter cut and taped under an opaque dressing); a saline circuit was recirculated to simulate haemoperfusion for a matched duration and schedule.
  • Blinding: Participants, treating clinicians, investigators, and outcome assessors were blinded; designated unblinded staff delivered the assigned procedure. Blinding was reported as maintained through day 28 in 97.5% of participants.
  • Statistics: A total of 360 participants was initially estimated to detect a 15% absolute reduction in 28-day mortality (from an anticipated 35% in control) with 80% power at the 5% two-sided significance level; after prespecified interim monitoring, sample size was adjusted to 450 using a group sequential design (O’Brien-Fleming–type boundaries), alongside enrichment to MODS >9. Primary analysis was intention-to-treat; per-protocol analyses were prespecified.
  • Follow-Up Period: 28 days for the primary endpoint; additional physiologic/organ dysfunction outcomes were prespecified but not formally tested/reported in the main manuscript because the primary endpoint was not significant.

Key Results

This trial was not stopped early. Enrolment completed after a prespecified sample-size recalculation (final n=450) and an enrichment strategy restricting subsequent recruitment to MODS >9.

Outcome Polymyxin B haemoperfusion Sham perfusion Effect p value / 95% CI Notes
28-day all-cause mortality (primary; participants with known 28-day vital status) 84/223 (37.7%) 78/226 (34.5%) RR 1.09 95% CI 0.85 to 1.39; P=0.49 Risk difference 3.15%; 95% CI −5.73 to 12.04.
28-day all-cause mortality (MODS >9 subgroup; prespecified high-risk population) 65/146 (44.5%) 65/148 (43.9%) RR 1.01 95% CI 0.78 to 1.31; P=0.92 Risk difference 0.60%; 95% CI −10.75 to 11.97.
28-day all-cause mortality (per-protocol: completed 2 assigned treatments) 50/173 (28.9%) 59/202 (29.2%) Risk difference −0.3% 95% CI −9.5 to 8.9; P=0.94 Per-protocol denominators differed (173 vs 202), reflecting incomplete delivery of 2 sessions in the intervention arm.
28-day all-cause mortality (per-protocol; MODS >9) 38/115 (33.0%) 47/129 (36.4%) Risk difference −3.1% 95% CI −15.2 to 9.0; P=0.58 Not statistically significant; direction favoured haemoperfusion in per-protocol analysis.
Participants with ≥1 serious adverse event 65.1% 57.3% Not reported Not reported Most frequent: worsening sepsis (10.8% vs 9.1%) and worsening septic shock (6.6% vs 7.7%).
Device-related adverse events 11 events 5 events Not reported Not reported 2 serious adverse events judged probably related (deep vein thrombosis; air embolism).
Circuit clotting 17/212 circuits (8.0%) 5/220 circuits (2.3%) Not reported Not reported Led to premature discontinuation of treatment in 4 polymyxin B participants.
  • No mortality benefit was demonstrated for polymyxin B haemoperfusion, either overall (37.7% vs 34.5%) or in the enriched high-risk MODS >9 population (44.5% vs 43.9%).
  • Intervention delivery was imperfect: 212/224 received assigned haemoperfusion (vs 220/226 receiving sham), and fewer intervention participants completed both sessions (per-protocol n=173 vs 202), yet per-protocol mortality remained similar.
  • Safety signals included more circuit clotting and a numerically higher proportion of participants experiencing serious adverse events (65.1% vs 57.3%).

Internal Validity

  • Randomisation and allocation: Central randomisation with allocation concealment; stratified by site, using variable block sizes (2 and 4).
  • Baseline characteristics: Groups were broadly comparable: age 60.9 (14.9) vs 58.8 (15.2) years; APACHE II 29.4 (8.5) vs 28.1 (8.6); median MODS 10 (IQR 9–12) vs 10 (IQR 9–12); baseline EAA 0.77 (0.1) vs 0.78 (0.1).
  • Blinding/performance bias: Sham control and reporting of blinding maintenance (97.5%) reduced risk of differential co-interventions, though unblinded procedural staff were necessary and the sham avoided intravascular catheterisation/blood exposure (potential for subtle unblinding).
  • Dropout/missing data: One participant in the polymyxin B arm lacked 28-day survival status (primary analysis denominator 223 vs 226).
  • Protocol adherence and separation: Assigned procedure was delivered to 212/224 vs 220/226; per-protocol population was smaller in the intervention arm (173 vs 202), reflecting incomplete delivery of the full 2-session “dose”.
  • Timing: Median time from randomisation to initiation of haemoperfusion was 3 h 30 min (95% CI 3:15 to 3:45), supporting relatively early delivery once randomised.
  • Outcome assessment: Primary outcome (28-day all-cause mortality) is objective and unlikely to be biased by ascertainment.
  • Statistical rigour: Prespecified interim monitoring and hierarchical testing plan limited multiplicity; the primary endpoint was neutral, and secondary outcomes were not formally tested/reported in the main manuscript as a consequence.

Conclusion on Internal Validity: Overall, internal validity appears moderate-to-strong, supported by concealed randomisation, objective primary endpoint, and a sham-controlled blinded design; key constraints are the mid-trial enrichment strategy and incomplete delivery of two full sessions in the intervention arm.

External Validity

  • Population representativeness: Participants were very unwell septic shock patients in high-resource ICUs (high ventilator use; high organ dysfunction), enriched for endotoxaemia (EAA ≥0.60) and later restricted to MODS >9.
  • Applicability: Generalisability depends on access to EAA testing and extracorporeal therapy infrastructure; many ICUs do not have rapid endotoxin assay pathways or established PMX haemoperfusion programmes.
  • Intervention feasibility: Real-world implementation requires vascular access, anticoagulation, and specialist staff; circuit clotting and catheter-related complications are relevant barriers.
  • Case-mix: Infections were heterogeneous (abdominal ~47–50%, respiratory ~19–20%, urinary ~9%), which supports breadth but may dilute any effect if benefit is phenotype-specific.

Conclusion on External Validity: Findings are most generalisable to high-resource ICUs treating severe endotoxaemic septic shock; they are less directly applicable to broader sepsis populations, settings without EAA testing, or where extracorporeal therapy capacity is limited.

Strengths & Limitations

  • Strengths:
    • Large multicentre North American RCT with a sham-controlled blinded design (uncommon in device trials for sepsis).
    • Biomarker-targeted enrolment (EAA ≥0.60) aligned the intervention with a plausible mechanism (endotoxin adsorption).
    • Objective primary outcome (28-day mortality) and prespecified interim monitoring/statistical hierarchy.
  • Limitations:
    • Mid-trial enrichment (restriction to MODS >9) created two enrolled eras and complicates interpretation as a single unified population.
    • Incomplete delivery of the intended “dose” in the intervention arm (fewer participants completing both sessions), which could dilute efficacy signals and impacts pragmatic feasibility.
    • Key prespecified secondary/exploratory outcomes were not formally tested/reported in the main manuscript due to the hierarchical approach after a neutral primary endpoint.
    • The sham avoided intravascular catheterisation and blood exposure, which may not fully mimic procedural physiology and could influence co-interventions despite reported blinding success.

Interpretation & Why It Matters

  • Clinical practice
    In a rigorously controlled sham-blinded trial, polymyxin B haemoperfusion did not reduce 28-day mortality in endotoxaemic septic shock; routine use as an adjunctive “sepsis rescue” strategy is not supported by the index trial result.
  • Precision phenotype question
    The neutral overall result does not exclude a smaller effect in a narrower biologic phenotype; this frames PMX haemoperfusion as a hypothesis-generating “theragnostic” strategy rather than a broadly effective intervention.
  • Trialist implications
    Future device trials in sepsis may require tighter biomarker windows, better demonstration of target engagement (endotoxin reduction), and strategies to maximise treatment completion.

Controversies & Subsequent Evidence

  • Conflicting pre-EUPHRATES trial signals: Early positive findings (EUPHAS) contrasted with later neutral/negative randomised evidence in peritonitis-associated septic shock (ABDOMIX), raising concern for small-study effects and risk-of-bias in earlier work.12
  • Post hoc “sweet spot” hypothesis (EAA 0.60–0.89): A prespecified post hoc analysis of EUPHRATES suggested a mortality signal when excluding very high EAA (≥0.90): 28-day mortality 26.1% (PMX) vs 36.8% (sham); OR 0.52; 95% CI 0.27 to 0.99; P=0.047, with signals in haemodynamics and ventilator-free days; this remains vulnerable to multiplicity and cut-point selection.3
  • Target engagement uncertainty: The index report noted no significant between-group reduction in EAA over the first days post-randomisation (detailed numerical data not presented in the main manuscript), raising mechanistic questions about adsorption efficacy and/or assay responsiveness at high endotoxin burdens.
  • Meta-analytic synthesis remains inconsistent: Modern systematic reviews (including trial sequential approaches) have generally attenuated early benefit estimates and highlight ongoing imprecision and heterogeneity across trials/designs.4567
  • Guideline positioning: Contemporary sepsis guidelines recommend against routine polymyxin B haemoperfusion outside research contexts, reflecting the EUPHRATES-neutral result and uncertainty in the aggregate evidence base.89
  • Real-world evidence continues but is confounded: Observational analyses and prospective cohorts have reported mixed associations (some suggesting benefit in selected phenotypes), but residual confounding by indication and co-interventions remains a major limitation for causal inference.10111213

Summary

  • EUPHRATES was a multicentre, randomised, sham-controlled, blinded device trial targeting endotoxaemic septic shock (EAA ≥0.60).
  • Polymyxin B haemoperfusion did not reduce 28-day mortality overall (37.7% vs 34.5%) or in the enriched MODS >9 population (44.5% vs 43.9%).
  • Fewer intervention participants completed both intended sessions (per-protocol n=173 vs 202), yet per-protocol mortality remained similar.
  • Serious adverse events were numerically higher in the intervention group (65.1% vs 57.3%) and circuit clotting was more frequent (8.0% vs 2.3%).
  • Subsequent analyses and meta-analyses support uncertainty overall, with a debated signal in moderate endotoxaemia (EAA 0.60–0.89) that remains hypothesis-generating.

Overall Takeaway

EUPHRATES is a landmark sepsis device trial because it brought sham-controlled blinding and biomarker-targeting to a field historically dominated by open-label studies and biologically appealing but uncertain extracorporeal therapies. Its neutral mortality result, alongside incomplete intervention delivery and persistent mechanistic uncertainty, shifts polymyxin B haemoperfusion away from routine care and towards narrower phenotype-driven research questions (notably moderate endotoxaemia) that require confirmatory testing.

Overall Summary

  • In endotoxaemic septic shock (EAA ≥0.60), polymyxin B haemoperfusion did not reduce 28-day mortality versus sham.
  • Sham-controlled blinding and multicentre design strengthen credibility; incomplete two-session delivery and enrichment strategy complicate pragmatics and interpretation.
  • Subsequent post hoc and meta-analytic work supports uncertainty overall, with an unresolved hypothesis of benefit in moderate (not extreme) endotoxaemia.

Bibliography