Skip to main content

Foundational Trials

HEMOTION

Image: Shutterstock / ChaNaWiT

Turgeon AF, Fergusson DA, Clayton L, et al. Liberal or restrictive transfusion strategy in patients with traumatic brain injury. N Engl J Med 2024;391:722-735.

Publication

Context & Rationale

  • Background
    • Anaemia can reduce oxygen delivery to an injured brain with impaired autoregulation and limited capacity to compensate for reduced arterial oxygen content.
    • The widespread adoption of restrictive transfusion in general intensive care rested mainly on mortality outcomes and populations containing few patients with traumatic brain injury (TBI). Its neurological safety could not simply be assumed.
    • A 200-patient factorial trial found no neurological advantage from a 10 g/dL threshold and more thromboembolic events; a subsequent feasibility trial involving 44 analysed patients suggested possible benefit from a 9 g/dL threshold. Neither settled the question.12
  • Research Question/Hypothesis
    • Would a liberal red-cell transfusion threshold of haemoglobin ≤10 g/dL improve neurological recovery at six months compared with ≤7 g/dL in critically ill adults with moderate or severe TBI and anaemia?
  • Why This Matters
    • Even a modest improvement in independence may have lifelong consequences for patients and carers.
    • Any functional benefit must justify greater exposure to donor blood, transfusion complications and pressure on a finite resource.

Design & Methods

  • Research Question:
    • Superiority of a liberal over a restrictive transfusion strategy for reducing an unfavourable neurological outcome at six months.
  • Study Type:
    • Investigator-initiated, international, multicentre, pragmatic, parallel-group randomised superiority trial using a prospective randomised open blinded endpoint (PROBE) design.
    • Conducted in 34 trauma centres with specialised neurocritical care in Canada, the United Kingdom, France and Brazil; recruitment ran from 1 September 2017 to 13 April 2023.
    • Central computer-generated allocation in a 1:1 ratio, using concealed variable blocks of four and six, stratified by centre. Public funding included the Canadian Institutes of Health Research.3
  • Population:
    • Adults aged ≥18 years admitted to ICU with acute blunt TBI, a Glasgow Coma Scale (GCS) score of 3–12 and haemoglobin ≤10 g/dL. Acute injury required presentation to the first emergency department within 24 hours of injury; this was not a requirement to randomise within 24 hours.
    • GCS eligibility used the last emergency-department score, or the last score before intubation when the patient left the emergency department intubated. Patients were screened at ICU admission and reassessed daily.
    • Exclusions: red-cell transfusion after ICU admission but before randomisation; active life-threatening bleeding with haemorrhagic shock or requiring urgent surgery; contraindication or objection to transfusion; GCS 3 with bilaterally fixed dilated pupils; brain death; a decision to withhold or withdraw life-sustaining treatment; or no fixed address.
    • Transfusion before ICU admission was permitted. A do-not-attempt-resuscitation decision alone did not exclude a patient receiving otherwise unrestricted treatment.4
  • Intervention:
    • Liberal strategy: transfuse leucoreduced red cells when haemoglobin was ≤10 g/dL (100 g/L), one unit at a time, with reassessment before further units.
    • The aim was to transfuse within three hours of reaching the threshold. The assigned strategy continued until ICU discharge, death or a decision to withdraw life-sustaining treatment.
  • Comparison:
    • Restrictive strategy: transfuse at haemoglobin ≤7 g/dL (70 g/L), using the same single-unit approach and intended three-hour delivery window.
    • Other care remained at clinicians’ discretion, with adherence to Brain Trauma Foundation guidance encouraged. Clinically justified departures, including transfusion for important ongoing bleeding, were adjudicated separately from unjustified protocol violations.3
  • Blinding:
    • Treating teams were unblinded. Trained assessors evaluated the primary and secondary outcomes centrally without knowledge of allocation; the statistical team was also blinded.
    • Locally collected tertiary outcomes and adverse events were assessed by personnel aware of allocation.
  • Statistics:
    • Power calculation: 712 patients were required to detect a 10-percentage-point absolute reduction in an unfavourable outcome, from 40% to 30%, with 80% power (β=0.20) at a two-sided α=0.05; the target was increased to 742 to accommodate anticipated losses.
    • The primary outcome used a prespecified sliding dichotomy of the eight-level Glasgow Outcome Scale–Extended (GOS-E), based on baseline prognosis calculated with the TBI-IMPACT model.
    • Patients were divided into prognosis tertiles: an unfavourable outcome was GOS-E ≤3 for the worst prognosis, ≤4 for the intermediate prognosis and ≤5 for the best prognosis. Thus, the primary outcome was not a uniform measure of death or severe disability.
    • Analysis followed the intention-to-treat principle within the retained cohort, with available primary outcomes. Robust hierarchical Poisson regression adjusted for sex and centre; missing prognostic covariates were imputed. Prespecified sensitivity analyses included per-protocol, conventional dichotomous and ordinal analyses.
    • The final statistical analysis plan was completed in February 2024 before analysis and unblinding. Secondary outcomes were adjusted for centre, sex and TBI-IMPACT prognostic covariates, without adjustment for multiplicity.5
  • Follow-Up Period:
    • Six months for neurological outcome, functional independence, quality of life, depression and survival; ICU and hospital outcomes were also collected.

Key Results

This trial was not stopped early. Recruitment reached the revised target of 742 patients. One prespecified formal interim analysis at 50% recruitment, using a Haybittle–Peto superiority boundary of P<0.001, recommended continuation.

Outcome Liberal strategy
≤10 g/dL		 Restrictive strategy
≤7 g/dL		 Effect p value / 95% CI Notes
Randomised / retained cohort 371 / 369 371 / 367 — — Primary outcome available in 364 and 358 patients.
Unfavourable neurological outcome at six months: primary sliding dichotomy 249/364 (68.4%) 263/358 (73.5%) Adjusted RR 0.93 95% CI 0.83 to 1.04 Adjusted absolute difference, restrictive minus liberal: 5.4 percentage points; 95% CI −2.9 to 13.7.
Unfavourable outcome using a fixed GOS-E ≤4 225/364 (61.8%) 240/358 (67.0%) RR 0.92 95% CI 0.83 to 1.03 Prespecified unadjusted sensitivity analysis; death, vegetative state or severe disability.
Mortality at six months 99/369 (26.8%) 96/365 (26.3%) Adjusted HR 1.01 95% CI 0.76 to 1.35 Hazard ratio, not risk ratio.
Overall Functional Independence Measure at six months 119 (95–125) 115 (76–124) Adjusted median difference 4.34 95% CI 0.22 to 8.45 Survivors; higher scores indicate greater independence.
EuroQol visual analogue score at six months 70 (50–80) 60 (40–75) Adjusted median difference 5.19 95% CI 0.52 to 9.86 Survivors; higher scores indicate better health.
EQ-5D-5L utility index at six months 0.74 (0.45–0.87) 0.64 (0.33–0.82) Adjusted median difference 0.06 95% CI 0.01 to 0.10 Survivors; higher scores indicate better health.
TBI-specific quality of life: QOLIBRI at six months 64 (45–80) 56 (39–77) Adjusted median difference 3.72 95% CI −1.13 to 8.56 Survivors; higher scores indicate better quality of life.
Depressive symptoms: PHQ-9 ≥10 at six months 82/227 (36.1%) 95/222 (42.8%) Adjusted RR 0.85 95% CI 0.63 to 1.17 Survivors with assessment available.
Haemoglobin during ICU stay, g/dL 10.8 (10.3–11.5) 8.8 (8.1–9.6) Median difference 2.00 g/dL 95% CI 1.97 to 2.03 Achieved concentrations, distinct from transfusion triggers.
Patients receiving red cells after randomisation 365/369 (98.9%) 141/367 (38.4%) Not reported Not reported Exposure during the ICU treatment period.
Red-cell units per patient / total units 3 (2–5) / 1516 0 (0–1) / 307 — — Per-patient values are medians (IQR).
Any infection 204/369 (55.3%) 192/367 (52.3%) RR 1.06 95% CI 0.92 to 1.21 No clear difference.
Venous thromboembolism 31/369 (8.4%) 31/367 (8.4%) Not reported Not reported Deep venous thrombosis or pulmonary embolism.
Acute respiratory distress syndrome 12/369 (3.3%) 3/367 (0.8%) Not reported Not reported Numerical safety signal; few events.
Renal replacement therapy 12/369 (3.3%) 3/367 (0.8%) Not reported Not reported Additional numerical imbalance in the supplement.4
Transfusion reactions among transfused patients 6/365 (1.6%) 1/141 (0.7%) RR 2.33 95% CI 0.35 to 58.32 None was severe; denominators include only transfused patients.
Mechanical ventilation, days 12 (8–17) 11 (7–17) Median difference 1.00 day 95% CI −0.52 to 2.52 Median (IQR).
ICU length of stay, days 15 (10–22) 15 (10–22) Median difference 0.00 days 95% CI −1.85 to 1.85 Hospital stay was also similar: 33 (18–50) versus 33 (19–55) days.
  • The primary estimate favoured liberal transfusion, but its confidence interval crossed no effect. The absolute-difference interval permits both a small disadvantage and a substantial neurological benefit from liberal transfusion; equivalence was not demonstrated.
  • Continuous values are medians (IQR). Functional and quality-of-life findings apply to assessed survivors, and their confidence intervals were not adjusted for multiple comparisons; they support a possible benefit without establishing it.
  • No convincing subgroup effect emerged. For moderate TBI, the RR was 0.94 (95% CI 0.75 to 1.18); for severe TBI, 0.93 (95% CI 0.82 to 1.05). In patients aged ≤55 years the RR was 0.88 (95% CI 0.76 to 1.03), compared with 0.98 (95% CI 0.84 to 1.14) above 55 years; these estimates do not establish an age-specific treatment rule.

Internal Validity

  • Randomisation and Allocation:
    • Central concealed allocation with variable block sizes protected against foreknowledge of treatment assignment. Stratification by centre addressed important variation in local practice.
  • Post-Randomisation Exclusions and Follow-Up:
    • Six patients were excluded from the retained cohort: one incorrectly enrolled patient with aneurysmal subarachnoid haemorrhage in each group, plus one consent withdrawal in the liberal group and three in the restrictive group.
    • A further five liberal-group and nine restrictive-group patients lacked the primary outcome, leaving 722 patients: 98.1% of the 736-person retained cohort. This was an available-outcome analysis within that cohort, rather than a complete analysis of every randomised patient.
    • Per-protocol and complete-covariate analyses both yielded RR 0.93 (95% CI 0.83 to 1.04), supporting stability of the main result.4
  • Performance and Detection Bias:
    • Unblinded treatment could influence cointerventions and clinical event ascertainment. Central blinded assessment of six-month outcomes substantially reduced detection bias for the principal endpoint.
    • Decisions to withdraw life-sustaining treatment were numerically identical: 50 patients per group in ICU and 63 per group during hospitalisation. This is reassuring for a major potential source of bias in TBI trials, although it cannot exclude differences in timing or decision-making.
  • Baseline Characteristics:
    • Patients had substantial injury burden: mean age 48.7±18.9 years, 73.2% severe TBI and 72.7% men. Haemoglobin at randomisation was 9.1±0.8 g/dL in both groups.
    • Some prognostic characteristics favoured the liberal group: both pupils were reactive in 78.7% versus 71.8%; admission hypotension occurred in 22.7% versus 28.8%; and a GCS motor score of 1 occurred in 26.0% versus 32.7%.
    • Marshall V/VI lesions were more frequent in the liberal group: 37.1% versus 33.2%. Overall predicted unfavourable-outcome probabilities were similar, at 0.54±0.23 versus 0.55±0.22; covariate-adjusted sensitivity analyses remained consistent with the primary result.
  • Timing and Delivery:
    • Median time from injury to randomisation was 55 hours (IQR 38–90) versus 56 hours (37–83). HEMOTION therefore primarily tested treatment of anaemia developing during ICU care after the initial resuscitation.
    • Median time from the qualifying haemoglobin measurement to transfusion was 134 minutes (57–190) versus 104 minutes (75–215), within the intended three-hour window at the median but with appreciable delays in some patients.
  • Dose, Adherence and Separation:
    • The intervention produced an early, sustained haemoglobin separation: 10.8 versus 8.8 g/dL, with a median difference of 2.00 g/dL (95% CI 1.97 to 2.03). Total red-cell use was 1516 versus 307 units.
    • Adjudicated major violations affected 14/369 (3.8%) versus 7/367 (1.9%). These percentages do not describe all departures from the intended delivery window: lesser deviations were more frequent, particularly in the liberal group.
    • Transfusion above threshold without a valid rationale occurred in one liberal-group patient and six restrictive-group patients. Receipt of transfusion in the restrictive arm was otherwise part of the assigned strategy, not automatically crossover.4
  • Cointerventions and Heterogeneity:
    • After randomisation, hyperosmolar treatment was used in 35.8% versus 33.2%, neuromuscular blockade in 40.1% versus 38.4%, and barbiturates in 5.7% versus 2.7%. There was no consistent pattern of greater rescue treatment in the restrictive group.
    • Intracranial hypertension occurred in 52.1% versus 52.7% of monitored patients. Invasive brain oxygen monitoring after randomisation was uncommon: 3.8% versus 3.5%.
    • Variation in injury patterns, neurosurgical treatment and centre practice was clinically relevant, but centre-adjusted analyses and broadly consistent subgroup estimates support the overall comparison.4
  • Outcome Assessment and Statistical Rigour:
    • GOS-E is an established functional measure, assessed using a defined, prognosis-dependent classification. Mortality was included in the primary endpoint, while survivor-only scales addressed complementary aspects of recovery.
    • The sliding dichotomy was planned in the original protocol. The final analysis plan clarified sex adjustment and quantile regression before the allocation code was opened; the principal outcome was not selected after seeing the treatment effect.5
    • The principal limitation to inference is precision: the prespecified superiority comparison did not establish benefit, while secondary and subgroup analyses cannot overcome uncertainty in the primary result.

Conclusion on Internal Validity: Internal validity is strong for the randomised comparison, supported by concealed allocation, blinded central outcome assessment, sustained treatment separation and high follow-up. Unblinded care, baseline imbalances and limited precision qualify the conclusions, particularly for secondary outcomes and uncommon harms.

External Validity

  • Population Representativeness:
    • The age, sex distribution and predominance of severe injury resemble many contemporary neurocritical care TBI populations. Inclusion of extracranial injuries in 64.5% versus 70.8% improves relevance to patients with multiple trauma.
    • Selection required anaemia and continued ICU treatment. Patients needing transfusion immediately after ICU admission, those with uncontrolled life-threatening bleeding, penetrating injury or treatment limitation, and people without a fixed address were not represented.
    • The screening flow chart identified 716 otherwise eligible patients who were not enrolled, with overlapping reasons including declined consent, physician decisions and research or blood-bank logistics. This limits representativeness more than it threatens the concealed comparison among those recruited.4
  • Applicability:
    • The intervention is operationally straightforward and does not require advanced cerebral monitoring. It is most applicable to adults with moderate or severe blunt TBI who become anaemic after initial resuscitation.
    • Most recruitment occurred in well-resourced trauma systems; only 5.2% of participants were recruited outside very high-income countries. Reliable blood availability, transfusion safety and access to rehabilitation may change the balance elsewhere.4
    • These findings do not directly determine thresholds in children, isolated mild TBI, aneurysmal subarachnoid haemorrhage, spontaneous intracerebral haemorrhage, post-cardiac-arrest brain injury or active haemorrhagic shock.

Conclusion on External Validity: Generalisability is good for anaemic adults with moderate or severe blunt TBI receiving ongoing care in established trauma ICUs. It is substantially narrower for emergency haemorrhage resuscitation, other brain injuries and settings with constrained blood or rehabilitation resources.

Strengths & Limitations

  • Strengths:
    • A large international trial dedicated to TBI with anaemia, directly addressing an important gap in general ICU transfusion evidence.
    • Concealed randomisation, blinded central functional assessment and a prespecified analysis framework.
    • Clear haemoglobin and transfusion separation, low rates of major protocol violation and near-complete primary follow-up.
    • Patient-relevant assessment extending beyond mortality to independence, quality of life and depressive symptoms.
  • Limitations:
    • A superiority design with confidence intervals that do not exclude clinically important benefit; it cannot establish non-inferiority of restrictive transfusion.
    • Unblinded care, some prognostic imbalance and more frequent minor delivery deviations in the liberal arm.
    • A complex primary endpoint and survivor-only secondary outcomes with incomplete responses and no multiplicity correction.
    • Limited physiological monitoring, few safety events, and no comparison of intermediate thresholds or treatment tailored to documented cerebral hypoxia.

Interpretation & Why It Matters

  • What HEMOTION establishes
    • Liberal transfusion reliably increased haemoglobin and blood use, but did not establish superiority for the prespecified six-month neurological endpoint.
    • The result should not be translated into proof that waiting until haemoglobin reaches 7 g/dL is neurologically equivalent or universally safe in TBI.
  • Functional recovery matters
    • Similar mortality can coexist with clinically relevant differences in disability. HEMOTION’s principal contribution was to examine neurological recovery and patient experience rather than infer transfusion safety from survival alone.
    • The functional findings are compatible with benefit, but their size, survivor selection and multiple comparisons prevent a definitive claim of improved independence.
  • Clinical implication
    • HEMOTION alone does not mandate transfusion at 10 g/dL. Its findings support reassessment of an automatic 7 g/dL threshold in brain injury, particularly when considered alongside the subsequent randomised evidence.
    • The relevant decision concerns the balance between potential neurological benefit and additional transfusion exposure; the trial did not identify an optimal intermediate threshold or validate a physiological rescue algorithm.

Controversies & Other Evidence

  • Failure to Establish Superiority Is Not Equivalence:
    • The adjusted absolute effect was a 5.4-percentage-point reduction in unfavourable outcome with liberal transfusion, with an interval extending from 2.9 points worse to 13.7 points better. Dismissing that interval as showing “no effect” loses clinically important information.
    • Conversely, a favourable point estimate is not proof of benefit. The trial was neither designed nor analysed to establish non-inferiority of the restrictive strategy.
  • Power Assumptions and the Meaning of the Primary Endpoint:
    • The sample-size calculation assumed 40% unfavourable outcomes using a fixed GOS-E ≤4 definition. The comparable observed restrictive-group proportion was 67.0%; the 73.5% primary proportion used a different, sliding definition and should not be directly compared with the 40% assumption.
    • The anticipated 10-percentage-point effect was substantial. Failure to detect a smaller effect should be interpreted through the confidence interval, rather than a retrospective power calculation.
    • The sliding dichotomy recognises that attainable recovery differs with initial prognosis, but assigns different labels to identical outcomes: GOS-E 4 is favourable in the worst-prognosis tertile and unfavourable in the other two. The conventional cut-off and ordinal analyses are therefore valuable checks.
    • The prespecified ordinal analysis yielded OR 0.85 (95% CI 0.65 to 1.11), and the covariate-adjusted fixed-cut-off analysis yielded RR 0.94 (95% CI 0.79 to 1.11). Neither materially changes the interpretation.4
  • Prespecification, Timing and What Was Actually Compared:
    • The original protocol specified the sliding dichotomy. However, the detailed analysis plan was finalised after recruitment and six-month follow-up, although before analysis and unblinding. This provides less protection than finalisation before recruitment, without evidence that the endpoint was chosen in response to the treatment results.5
    • Randomisation around 55–56 hours after injury leaves uncertainty about transfusion during the earliest phase of secondary brain injury. It also reflects the intended question: treating anaemia arising during ICU care, rather than initial haemorrhage resuscitation.
    • The achieved restrictive-group haemoglobin was 8.8 g/dL; a 7 g/dL trigger did not mean maintaining every patient at 7 g/dL. The trial estimates the effect of two transfusion policies and cannot identify the effect of each additional gram of haemoglobin or unit of blood.
  • Secondary Benefits and Safety Require Restraint:
    • Functional and quality-of-life analyses excluded deaths. Randomisation does not guarantee comparability after conditioning on survival; similar mortality reduces, but does not remove, this concern.
    • Missing survivor assessments ranged from 14 versus 20 for functional independence to 41 versus 52 for QOLIBRI. The nominally favourable findings also arose among multiple correlated outcomes, without multiplicity adjustment.
    • Minimally important differences for these scales were not established in TBI. A 4.34-point adjusted FIM difference should not automatically be described as a clinically important improvement.
    • ARDS and renal replacement therapy each occurred in 12 versus three patients. These imbalances warrant attention but do not establish transfusion causality; venous thromboembolism was identical and severe transfusion reactions were absent.4
  • Earlier Dedicated TBI Trials:
    • In the 2014 factorial trial, favourable six-month outcome occurred in 33.0% with the 10 g/dL threshold versus 42.5% with 7 g/dL (P=0.28); thromboembolic events occurred in 21.8% versus 8.1% (P=0.009). Anaemia was not required, which limits comparison with HEMOTION’s anaemic population.1
    • The smaller TRAHT feasibility trial compared 9 versus 7 g/dL in 44 analysed patients. Hospital mortality was 1/21 versus 7/23 (P=0.048), but clinical outcomes were exploratory and the study was too small to establish efficacy.2
  • TRAIN: Subsequent Evidence Favouring Liberal Transfusion:
    • TRAIN randomised 850 adults with acute brain injury to thresholds of <9 versus <7 g/dL. Approximately 60% had TBI; the remainder had subarachnoid or intracerebral haemorrhage.
    • Unfavourable neurological outcome at 180 days occurred in 246/393 (62.6%) versus 300/413 (72.6%); adjusted RR 0.86; 95% CI 0.79 to 0.94; P=0.002. Its primary definition was a fixed GOS-E of 1–5.
    • The direction of effect is compatible with HEMOTION. Differences in statistical significance do not establish conflicting treatment effects, and a mixed-injury trial using 9 g/dL does not prove that 10 g/dL is the optimal TBI threshold.6
  • SAHARA: A Different Population and Comparator:
    • SAHARA randomised 742 adults with aneurysmal subarachnoid haemorrhage to mandatory transfusion at ≤10 g/dL or optional transfusion at ≤8 g/dL.
    • An unfavourable 12-month modified Rankin score of ≥4 occurred in 33.5% versus 37.7%; RR 0.88; 95% CI 0.72 to 1.09; P=0.22. The estimate again favoured liberal transfusion without establishing superiority.
    • The different disease, restrictive threshold, outcome scale and follow-up period make SAHARA complementary evidence rather than a direct replication of HEMOTION.7
  • Pooled Evidence:
    • The 2025 Cochrane review found more unfavourable neurological outcomes with restrictive transfusion across four brain-injury trials involving 2297 participants: RR 1.14; 95% CI 1.05 to 1.22; moderate-certainty evidence. Here the ratio compares restrictive with liberal transfusion.8
    • A December 2025 review restricted its principal analysis to patients with acute acquired brain injury and anaemia: five trials, 2364 participants; RR 0.89 for liberal versus restrictive transfusion; 95% CI 0.84 to 0.95. The authors rated the neurological evidence as high certainty and found no mortality difference.9
    • These analyses strengthen the case for a neurological benefit from more liberal transfusion, but draw heavily on overlapping trials. Differences in anaemia eligibility, outcome definitions and inclusion of mixed brain injuries affect interpretation; pooling does not establish the best threshold for every TBI patient.
  • 2026 HEMOTION Substudy: Multiple Trauma:
    • The principal analyses found no convincing interaction between transfusion strategy and multiple-trauma status across three definitions. In patients with extracranial injury and an Injury Severity Score >15, RR was 0.87 (95% CI 0.71 to 1.07; P for interaction=0.27).
    • Among patients requiring emergency extracranial surgery, RR was 0.79 (95% CI 0.59 to 1.05; P for interaction=0.13). An interaction appeared in the ordinal sensitivity analysis (P=0.02), but not in the principal analysis.
    • This is a hypothesis for further study, not a treatment-selection rule. Emergency surgery could occur before or after randomisation, adding a further limitation to causal subgroup interpretation.10
  • Haemoglobin Is an Incomplete Physiological Target:
    • HEMOTION compared transfusion policies triggered by haemoglobin; it did not select patients by cerebral hypoxia or demonstrate that raising haemoglobin improved cerebral oxygen delivery. The sparse brain oxygen monitoring limits mechanistic interpretation.
    • Red-cell storage duration has been proposed as an additional determinant of transfusion response. Published correspondence raised this as a possible explanation for differences between studies, but provided no new trial data establishing such an effect.11
    • That hypothesis does not invalidate HEMOTION’s randomised comparison or justify preferential use of fresher blood. Patient-specific physiology and blood-product characteristics remain questions for targeted trials.
  • Guidance After HEMOTION:
    • The 2025 Association of Anaesthetists guideline incorporates HEMOTION and TRAIN, recognises acute brain injury as an area of uncertainty and notes that higher thresholds of 80–100 g/L may be appropriate. It does not establish a single mandatory threshold for all patients with TBI.12
    • The 2025 CHEST guideline explicitly excludes neurological injury and trauma from the scope of its transfusion recommendations. Its general endorsement of restrictive transfusion should therefore not be presented as a recommendation to use 7 g/dL in HEMOTION-type patients.13
    • Clinical synthesis: subsequent randomised and pooled evidence makes routine tolerance of haemoglobin down to 7 g/dL less reassuring in anaemic acute brain injury. The exact threshold, duration and role of physiological individualisation remain unresolved.68

Summary

  • HEMOTION randomised 742 adults with moderate or severe TBI and anaemia to transfusion at haemoglobin ≤10 or ≤7 g/dL during ICU care.
  • Unfavourable six-month neurological outcome occurred in 68.4% versus 73.5%; adjusted RR 0.93; 95% CI 0.83 to 1.04. Liberal superiority was not established, and clinically important benefit remained possible.
  • Achieved haemoglobin was 10.8 versus 8.8 g/dL, with 1516 versus 307 red-cell units transfused. Similar mortality did not establish equivalent neurological recovery.
  • Some survivor functional outcomes favoured liberal transfusion, but multiplicity and survivor selection limit inference. ARDS occurred in 3.3% versus 0.8%, while venous thromboembolism was 8.4% in both groups.
  • Later evidence supports greater caution with a routine 7 g/dL threshold in acute brain injury, without establishing a universal requirement to transfuse at 10 g/dL.68

Overall Takeaway

HEMOTION provides a rigorous, clinically important comparison of transfusion strategies in anaemic patients with TBI: a 10 g/dL threshold did not establish superior neurological recovery, but the trial did not demonstrate that a 7 g/dL threshold was equivalent. Its lasting contribution is to place functional recovery at the centre of transfusion decisions and to show why general ICU mortality evidence cannot settle neurological safety.

Overall Summary

  • HEMOTION leaves open a clinically important neurological benefit from liberal transfusion, at the cost of substantially greater blood use. It should not be cited as proof that restrictive transfusion is neurologically equivalent in traumatic brain injury.

Bibliography

Last updated September 7th, 2026