Publication

• Title: Polymyxin B haemoadsorption in endotoxic septic shock (Tigris): a multicentre, open-label, Bayesian, randomised, controlled, phase 3 trial
• Acronym: Tigris
• Year: 2026
• Journal published in: The Lancet Respiratory Medicine
• Citation: Neyra JA, Legrand M, Tidswell MA, et al. Polymyxin B haemoadsorption in endotoxic septic shock (Tigris): a multicentre, open-label, Bayesian, randomised, controlled, phase 3 trial. Lancet Respir Med. 2026 Mar 23. Online ahead of print.

Context & Rationale

• Background

  • Septic shock remains a heterogeneous syndrome with persistently high mortality, and repeated attempts to improve outcomes with broad, non-enriched adjunctive therapies have largely failed.
  • Endotoxic septic shock was conceptualised as a biologically defined endotype in which high circulating endotoxin activity, together with advanced organ dysfunction, identifies a subgroup at particularly high risk and potentially susceptible to endotoxin removal.
  • Prior polymyxin B haemoperfusion evidence was discordant: the overall EUPHRATES RCT was neutral, a post hoc EUPHRATES subgroup restricted to endotoxin activity assay (EAA) 0.60–0.89 with substantial organ failure suggested benefit, and a 2023 network meta-analysis still judged the totality of evidence low certainty because of heterogeneity and trial quality concerns. ¹²³
• Research Question/Hypothesis

  • Whether, in adults with vasopressor-dependent septic shock, multiorgan failure, and EAA 0.60–0.89, two sessions of polymyxin B haemoadsorption added to standard care would reduce 28-day mortality compared with standard care alone.
  • Whether prespecified Bayesian borrowing from the closely matched EUPHRATES “treatable” cohort was a valid and efficient way to test a rare, mechanistically enriched sepsis phenotype. ⁴
• Why This Matters

  • TIGRIS is important less as a generic “blood purification” trial than as a test of precision medicine in sepsis: predictive enrichment (endotoxin activity) plus prognostic enrichment (high organ failure burden).
  • If valid, the signal would convert a broadly negative programme into a clinically actionable endotype-specific strategy and provide a high-profile example of Bayesian historical borrowing in critical care trials.

Design & Methods

• Research Question: In ICU adults with endotoxic septic shock defined by EAA 0.60–0.89 and multiorgan dysfunction, does polymyxin B haemoadsorption plus standard care improve survival versus standard care alone?
• Study Type: Phase 3, multicentre, US-only, ICU-based, randomised, controlled, open-label, Bayesian trial with 2:1 allocation and site-stratified blocked randomisation.
• Population:
  • Setting: 19 academic and community hospital ICUs in the USA.
  • Inclusion criteria: Age ≥18 years; suspected or documented infection on intravenous antibiotics; intravenous fluids >30 mL/kg within 24 h; vasopressor requirement for at least 2 h and up to 30 h before randomisation (norepinephrine >0.05 μg/kg/min or equivalent); MODS >9 or, if MODS incomplete, SOFA >11; EAA 0.60–0.89; and at least one new organ dysfunction feature (mechanical ventilation, thrombocytopenia, or oliguria despite fluids).
  • Key exclusions: Inability to achieve or maintain MAP ≥65 mm Hg; end-stage kidney disease on chronic dialysis; clinically supported non-septic shock; CPR without immediate neurological recovery; recent myocardial infarction; uncontrolled haemorrhage; major trauma; severe granulocytopenia or thrombocytopenia; heparin hypersensitivity/HIT; polymyxin B allergy; or lack of commitment to full medical support.
• Intervention:
  • TORAYMYXIN PMX-20R polymyxin B haemoadsorption added to standard care.
  • Two sessions, each 90–120 min (minimum 90 min), approximately 20–22 h apart.
  • Blood flow 80–120 mL/min through a haemodialysis or CRRT machine.
  • 12F or 14F femoral or internal jugular double-lumen catheter when needed.
  • Systemic unfractionated heparin was recommended: circuit priming with heparinised saline, 3000-U bolus, then 20 U/kg over 2 h.
  • If CRRT was already in use, it was interrupted during PMX sessions rather than run in tandem.
• Comparison:
  • Standard medical care for septic shock alone, in line with contemporary Surviving Sepsis Campaign practice.
  • No sham haemoperfusion procedure was used.
  • For control patients, baseline and post-randomisation time anchoring used a proxy treatment start time of randomisation + 2 h.
• Blinding: Open-label at the bedside. Allocation concealment was centralised through a web-based randomisation system held by an independent CRO. No sham cartridge was used, so treating teams could not be blinded. The independent safety committee and the study statistician were masked, and the primary endpoint was objective all-cause mortality.
• Statistics: Primary analysis was Bayesian APACHE-II-adjusted logistic regression in the intention-to-treat cohort. A 75% weighted informative prior was derived from the US EUPHRATES treatable cohort (EAA 0.60–0.89, MODS >9). Success was prespecified as a posterior probability of benefit ≥95% for 28-day mortality. With 150 evaluable patients and a baseline 28-day mortality risk of 48%, the design had 80% Bayesian power to achieve that threshold if the true marginal absolute risk reduction was 13.5%; there was no conventional frequentist α-based primary hypothesis test. Supportive analyses used modified intention-to-treat and per-protocol cohorts, with extensive sensitivity analyses across prior weights and model forms. ⁴
• Follow-Up Period: Mortality assessed at day 28 and day 90; 12-month mortality collection was ongoing for future analysis.

Key Results

This trial was not stopped early.

• The primary 28-day conclusion was materially strengthened by prior borrowing: using Tigris data alone, the APACHE-II-adjusted estimate was OR 0.80 with 95% CrI 0.38 to 1.49 and posterior probability of benefit 79.7%; at 90 days, the Tigris-only estimate remained favourable at OR 0.50 with 95% CrI 0.23 to 0.94 and posterior probability 98.4%.
• Among patients still in ICU on day 28, 6/9 (67%) in the polymyxin B arm versus 1/6 (17%) in the standard-care arm were alive at day 90; none of the six control patients still in ICU on day 28 were discharged alive by day 90, versus 5/9 in the intervention arm.
• Early haemodynamic separation was modest despite later mortality divergence: mean change in MAP from day 0 to day 3 was 6.46 (SE 1.53) mm Hg versus 6.19 (SE 2.17); change in vasopressor dose was −0.16 (SE 0.02) versus −0.14 (SE 0.03) μg/kg/min; vasopressor duration was 52.3 (IQR 25.7–69.5) h versus 39.6 (22.6–68.1) h.

Internal Validity

• Randomisation and allocation:
  • Allocation was central, web-based, blocked, and stratified by site.
  • The sequence was held by an independent CRO, which makes pre-randomisation selection bias unlikely.
• Drop out or exclusions:
  • Follow-up to 90 days was complete for all 157 randomised patients.
  • Six of 106 patients (5.7%) allocated to polymyxin B never received treatment, versus 0/51 in control; reasons were deterioration before initiation (n=3), family withdrawal (n=1), and staffing issues (n=2).
  • The modified ITT cohort was therefore 100 versus 51; the per-protocol cohort was 77 versus 44.
• Performance/Detection Bias:
  • No sham procedure was used, so bedside performance bias cannot be excluded.
  • This matters less for mortality than for softer endpoints; the key outcomes were hard, objective, and time-fixed.
• Protocol Adherence:
  • Exposure separation at the treatment level was clear: 100/106 intervention patients received PMX, whereas control patients received no haemoadsorption.
  • Protocol adherence was less secure for full-dose delivery: only 77/106 intervention patients entered the per-protocol cohort, implying meaningful treatment-delivery fragility in real-world ICU conditions.
• Baseline Characteristics:
  • Groups were well balanced and appropriately high risk: APACHE II 32.8 ± 8.5 versus 31.7 ± 7.3; SOFA 14.3 ± 2.0 versus 14.6 ± 2.1; MAP 74.0 ± 12.8 versus 74.1 ± 11.6 mm Hg; cumulative vasopressor index 8.1 ± 3.3 versus 7.4 ± 3.6.
  • Norepinephrine >0.1 μg/kg/min was present in 97/106 (92%) versus 47/51 (92%); baseline EAA was 0.74 ± 0.07 versus 0.73 ± 0.13.
  • The control event rate was high enough to make treatment benefit plausible: 45.1% mortality at 28 days and 60.8% at 90 days.
• Heterogeneity:
  • Biological enrichment reduced syndrome-level heterogeneity by requiring both endotoxaemia and severe organ failure.
  • Microbiological heterogeneity persisted: no growth 39% versus 37%; Gram-negative organisms 27% versus 29%; Gram-positive organisms 20% versus 22%; bacteraemia 33% versus 35%.
  • This supports the concept that “endotoxic septic shock” is not synonymous with culture-proven Gram-negative sepsis.
• Timing:
  • Patients had to be on qualifying vasopressors for 2–30 h before randomisation, and the second session occurred about 20–22 h after the first.
  • This is reasonably early, but six intervention-assigned patients could not start treatment in time, showing how narrow enrolment windows can erode deliverability.
• Dose:
  • The regimen was the standard historical prescription: two sessions of at least 90 min each at 80–120 mL/min.
  • It was intentionally restricted to EAA 0.60–0.89 because the trialists believed fixed two-cartridge therapy was inadequate for EAA ≥0.90.
  • Serial quantitative between-arm EAA clearance data were not reported, so dose-to-target biological separation cannot be directly verified in the published paper.
• Separation of the Variable of Interest:
  • Treatment exposure clearly differed, but short-term physiology separated only minimally: MAP change by day 3 was 6.46 (SE 1.53) versus 6.19 (SE 2.17) mm Hg; vasopressor dose change was −0.16 (SE 0.02) versus −0.14 (SE 0.03) μg/kg/min.
  • Total vasopressor duration from day 0 to day 3 was 52.3 (IQR 25.7–69.5) h versus 39.6 (22.6–68.1) h.
  • This weak early physiological separation makes the later mortality divergence clinically interesting but mechanistically less transparent in the published report.
• Outcome Assessment:
  • The primary and key secondary outcomes were all-cause mortality at 28 and 90 days, which are robust and clinically meaningful.
  • Safety outcomes were investigator-adjudicated and therefore more exposed to open-label bias than mortality endpoints.
• Statistical Rigor:
  • The Bayesian framework, priors, and sensitivity analyses were prespecified and published in advance; site-adjusted analyses changed little.
  • Prior–data conflict checks were negative.
  • However, the 28-day result crossed the prespecified success threshold only when the prior weight was 72% or greater, whereas 90-day benefit remained robust across all examined prior weights.

Conclusion on Internal Validity: Overall, internal validity is moderate. Randomisation, allocation concealment, complete follow-up, and objective mortality endpoints are major strengths, but open-label delivery, asymmetric non-treatment in the intervention arm, modest sample size, and a 28-day primary result that depends on informative prior borrowing materially temper certainty.

External Validity

• Population Representativeness:
  • The enrolled patients were very sick and clinically recognisable, but highly selected: 157 of 14,890 screened patients (about 1.1%) were randomised.
  • Among 328 clinically eligible patients who consented to EAA measurement, 171 were excluded before randomisation; 87 had EAA <0.60, 42 had EAA ≥0.90, 5 were EAA non-responders, and the remainder were lost for consent or evolving eligibility reasons.
  • The findings therefore apply to a narrow endotype, not to unselected septic shock.
• Applicability:
  • Translation requires rapid bedside EAA testing, haemoperfusion infrastructure, dialysis-trained staff, large-bore central venous access, and the ability to deliver two extracorporeal sessions in a narrow time window.
  • Applicability is strongest in tertiary ICUs in high-resource systems and substantially weaker in low-resource settings or units without extracorporeal capability.
  • The data do not apply to patients with low endotoxin activity, extreme endotoxaemia (EAA ≥0.90), chronic dialysis dependence, refractory inability to maintain MAP ≥65 mm Hg, major bleeding risk, or non-septic shock phenotypes.
• Clinical Scope:
  • Despite narrow eligibility, the studied phenotype is clinically important because untreated risk was very high.
  • The microbiological profile was broader than Gram-negative sepsis alone, which modestly widens conceptual applicability within the EAA-defined phenotype.

Conclusion on External Validity: External validity is limited. TIGRIS is generalisable chiefly to ICU patients with a very specific, biomarker-defined, high-severity endotoxic septic shock phenotype in centres able to perform rapid EAA-guided extracorporeal treatment.

Strengths & Limitations

• Strengths:
  • Prospective endotype enrichment using both a biological marker (EAA 0.60–0.89) and severe organ dysfunction.
  • Randomised multicentre ICU design spanning academic and community sites.
  • Complete 90-day follow-up with hard mortality endpoints.
  • Prespecified Bayesian framework, prior-data conflict checks, and extensive sensitivity analyses.
  • Clinically meaningful control mortality, indicating the population was sufficiently high risk to plausibly benefit.
• Limitations:
  • Small total sample, reflecting rarity of the phenotype but still limiting precision.
  • Open-label treatment without sham control.
  • Primary 28-day conclusion depended materially on informative prior borrowing and APACHE-II adjustment.
  • Substantial attrition from intervention assignment to full per-protocol treatment delivery.
  • US-only setting and highly selected phenotype constrain generalisability.
  • Industry funding was substantial; the sponsor participated in design, conduct management, interpretation, and report writing, and multiple authors had commercial relationships relevant to the device programme.
  • Serial biological confirmation of endotoxin removal was not presented in the published efficacy report.

Interpretation & Why It Matters

• Clinical practice

  TIGRIS does not justify routine use of polymyxin B haemoadsorption in unselected septic shock. It does, however, provide the strongest prospective evidence yet that a narrowly selected endotoxic septic shock subgroup may benefit.
• Biological signal

  The trial supports the view that endotoxin burden and organ failure can define a therapeutically relevant sepsis endotype. The most persuasive signal is the 90-day mortality result, not the more prior-sensitive 28-day primary endpoint.
• Methodological significance

  TIGRIS is likely to be cited as an important example of precision-enriched critical care trial design using prespecified Bayesian borrowing from a closely matched precursor trial.

Controversies & Subsequent Evidence

• The central controversy is whether a 28-day primary endpoint that reaches success only after informative borrowing from EUPHRATES is sufficient to change practice. The counterargument is that the 90-day mortality signal was stronger and much more robust than the 28-day result.
• The Bayesian result requires careful interpretation: a prespecified one-sided posterior probability threshold can be met even when the central 95% credible interval still crosses 1. That is not a statistical error; it is a property of how Bayesian posterior probabilities and central credible intervals answer different questions. ⁶
• The accompanying editorial emphasised implementation barriers: only about 1% of screened patients were randomised, optimal timing of EAA measurement and PMX initiation remains unsettled, and real-world patient selection will probably require integration of biomarkers, haemodynamics, and severity scores rather than EAA alone. It also highlighted that fewer than 30% of TIGRIS patients had Gram-negative organisms isolated, reinforcing that ESS is a host-pathobiology construct rather than a simple microbiological label. ⁵
• The broader evidence base remains mixed. Overall EUPHRATES was neutral, the EUPHRATES EAA 0.60–0.89 subgroup was favourable after adjustment, and meta-analytic syntheses have suggested possible mortality benefit but with persistent concerns about heterogeneity, trial quality, and the inclusion of non-randomised data. ¹²³⁷
• Subsequent non-randomised evidence has been directionally informative rather than definitive. A 2025 prospective observational study showed that combining EAA ≥0.6 with SOFA >11 or MODS >9 identified a subgroup with roughly 60% mortality, supporting the TIGRIS enrichment strategy, whereas a separate 2025 Japanese multicentre prospective cohort in high-dose norepinephrine shock showed preserved blood pressure and lower vasopressor exposure with PMX but no 28-day or 90-day mortality benefit. ⁸⁹
• A 2026 scoping review of EAA use in sepsis underscored how underdeveloped the field still is: serial EAA testing is common, but optimal cut-offs, timing, and integration with other biomarkers remain insufficiently standardised for broad routine implementation. ¹⁰
• Guideline uptake has not followed the TIGRIS signal. The contemporaneous 2026 Surviving Sepsis Campaign adult guideline still suggests against polymyxin B hemoperfusion in adults with sepsis or septic shock, with low certainty of evidence. Accordingly, routine adoption remains unsupported pending independent confirmation and guideline reassessment. ¹¹

Summary

• TIGRIS prospectively tested polymyxin B haemoadsorption in a tightly enriched septic shock endotype: EAA 0.60–0.89 plus severe multiorgan dysfunction.
• The primary 28-day Bayesian analysis was favourable but sensitive to informative prior borrowing; the 90-day mortality signal was stronger and more robust.
• Randomisation, allocation concealment, complete follow-up, and hard mortality endpoints strengthen credibility, but open-label delivery, a small sample, and treatment-delivery fragility limit confidence.
• External validity is narrow: only about 1% of screened patients were randomised, and implementation requires rapid EAA testing plus extracorporeal capability.
• TIGRIS is more convincingly a landmark in precision-enriched sepsis trial methodology than a definitive mandate for routine PMX use in current practice.

Overall Takeaway

TIGRIS is an important modern sepsis trial because it prospectively validates a biomarker-defined, high-risk endotype that emerged from earlier negative-to-mixed PMX literature. Its most compelling contribution is the combination of precision enrichment and prespecified Bayesian borrowing, which yielded a credible 90-day survival signal, but the narrow eligibility, modest sample, and unresolved guideline scepticism mean it should currently shape how we study and select patients for endotoxin-directed therapy more than it dictates universal bedside practice.

Bibliography

• 1.Dellinger RP, Bagshaw SM, Antonelli M, Foster DM, Klein DJ, Marshall JC, et al. 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.
• 2.Klein DJ, Foster D, Walker PM, Bagshaw SM, Mekonnen H, Antonelli M. Polymyxin B hemoperfusion in endotoxemic septic shock patients without extreme endotoxemia: a post hoc analysis of the EUPHRATES trial. Intensive Care Med. 2018;44(12):2205-2212.
• 3.Chen JJ, Lai PC, Lee TH, Huang YT. Blood Purification for Adult Patients With Severe Infection or Sepsis/Septic Shock: A Network Meta-Analysis of Randomized Controlled Trials. Crit Care Med. 2023;51(12):1777-1789.
• 4.Tomlinson G, Al-Khafaji A, Conrad SA, Factor NF, Foster DM, Galphin C, et al. Bayesian methods: a potential path forward for sepsis trials. Crit Care. 2023;27(1):432.
• 5.Schupp T, Behnes M, Akin I. Implementing endotoxin adsorption for septic shock into clinical practice. Lancet Respir Med. 2026 Mar 23. Online ahead of print.
• 6.Yarnell CJ, Goligher EC. Interpreting posterior probabilities in Bayesian analyses of clinical trials. Lancet Respir Med. 2024;12(3):188-190.
• 7.Li C, Zhang J, Yang P, Wang R, Chen T, Li L. The role of polymyxin B-immobilized hemoperfusion in reducing mortality and enhancing hemodynamics in patients with sepsis and septic shock: a systematic review and meta-analysis. Heliyon. 2024;10(13):e33735.
• 8.Molinari L, Tidswell MA, Al-Khafaji A, Davison D, Galphin C, Kamaluddin E, et al. Organ Failure, Endotoxin Activity, and Mortality in Septic Shock. Crit Care Explor. 2025;7(9):e1308.
• 9.Kawazoe Y, Miyamoto K, Miyagawa N, Ohta Y, Yamamura H, Kimura T, et al. Polymyxin B Hemoperfusion for Patients With Septic Shock Requiring High-Dose Norepinephrine: A Multicenter Prospective Cohort Study. Crit Care Explor. 2025;7(10):e1320.
• 10.Komaru Y, Maeda A, Inokuchi R, Foster DM, Kellum JA, Doi K. Clinical applications of the endotoxin activity assay in sepsis: a scoping review. Crit Care. 2026;30(1):97.
• 11.Prescott HC, Antonelli M, Alhazzani W, Møller MH, Alshamsi F, Azevedo LCPA, et al. Surviving Sepsis Campaign: International Guidelines for Management of Sepsis and Septic Shock 2026. Crit Care Med. 2026 Mar 23. Online ahead of print.