Skip to main content

Foundational Trials

TRAIN

Image: Shutterstock / ChaNaWiT

Publication

  • Title: Restrictive vs Liberal Transfusion Strategy in Patients With Acute Brain Injury: The TRAIN Randomized Clinical Trial
  • Acronym: TRAIN
  • Year: 2024
  • Journal published in: JAMA
  • Citation: Taccone FS, Rynkowski Bittencourt C, Møller K, et al; TRAIN Study Group. Restrictive vs Liberal Transfusion Strategy in Patients With Acute Brain Injury: The TRAIN Randomized Clinical Trial. JAMA. 2024;332(19):1623-1633.

Context & Rationale

  • Background
    • Anaemia is common during ICU care for acute brain injury, yet optimal transfusion thresholds are uncertain because cerebral oxygen delivery is uniquely vulnerable to reductions in haemoglobin concentration.
    • General ICU evidence and many transfusion policies support restrictive haemoglobin triggers (~7–8 g/dL), but neurocritical care patients were under-represented in many foundational ICU transfusion trials, and clinicians worry about secondary cerebral ischaemia.
    • Red blood cell transfusion can increase arterial oxygen content but also carries recognised risks (eg, transfusion reactions, lung injury, circulatory overload, thrombosis, infection) and consumes a scarce resource; therefore, equipoise existed for neuro-specific thresholds.
  • Research Question/Hypothesis
    • Among adults with acute brain injury and anaemia (haemoglobin ≤9 g/dL) in ICU, does a liberal transfusion threshold (<9 g/dL) improve 180-day neurological outcome compared with a restrictive threshold (<7 g/dL)?
  • Why This Matters
    • Even modest changes in transfusion thresholds could materially affect neurological outcomes, blood utilisation, and harm profiles in a large global population of neuro-ICU patients.
    • Practice variation in neurocritical care is substantial; high-quality randomised evidence is needed to support consistent, defensible transfusion targets.

Design & Methods

  • Research Question: In ICU patients with acute brain injury and haemoglobin ≤9 g/dL, does a liberal red blood cell transfusion strategy (transfuse if haemoglobin <9 g/dL) reduce the proportion with an unfavourable 180-day neurological outcome compared with a restrictive strategy (transfuse if haemoglobin <7 g/dL)?
  • Study Type: Pragmatic, international, multicentre, randomised, open-label, parallel-group superiority trial with blinded assessment of 180-day functional outcome; conducted in 72 ICUs across 22 countries; 1:1 allocation; stratified by centre, acute brain injury type, and Glasgow Coma Scale (3–5, 6–9, 10–13) with variable block sizes (4, 6, 8).
  • Population:
    • Setting: adult ICUs (academic 57/72 [79.2%]; non-academic 15/72 [20.8%]); participants from high-income and low-/middle-income countries (low-/middle-income 229/827 [27.7%]).
    • Inclusion: age 18–80 years; acute brain injury (traumatic brain injury, subarachnoid haemorrhage, intracerebral haemorrhage); Glasgow Coma Scale ≤13; haemoglobin ≤9 g/dL; randomisation within 10 days of the inciting brain injury/haemorrhage; ICU physician anticipated ICU stay ≥72 hours and survival ≥24 hours.
    • Key exclusions: other acute neurological diagnoses (eg, post-anoxic coma; ischaemic stroke; status epilepticus without underlying brain injury; CNS infections); major pre-existing neurological disability; intracerebral haemorrhage from arteriovenous malformation or tumour; inability/refusal to receive blood products; active/uncontrolled bleeding or imminent surgery requiring transfusion; GCS 3 with bilaterally fixed dilated pupils; brain death/imminent death; pregnancy; clinical requirement to correct anaemia with a target haemoglobin >9 g/dL; limitation of life-sustaining treatment at enrolment; history of alloimmunisation limiting red blood cell availability.
  • Intervention:
    • Liberal transfusion strategy: transfuse red blood cells whenever haemoglobin <9 g/dL, aiming to maintain haemoglobin ≥9 g/dL.
    • Strategy applied for up to 28 days after randomisation, or until ICU discharge/death (whichever came first).
    • Red blood cell product type and transfusion delivery followed local standards; co-interventions otherwise at clinician discretion.
  • Comparison:
    • Restrictive transfusion strategy: transfuse red blood cells whenever haemoglobin <7 g/dL, aiming to maintain haemoglobin ≥7 g/dL.
    • Same exposure window (up to 28 days post-randomisation or ICU discharge/death).
    • Permitted transfusion outside thresholds for clear clinical indications (eg, uncontrolled bleeding), at clinician discretion.
  • Blinding: Treating teams were not blinded (strategy-based transfusion cannot be practically masked); 180-day neurological outcome assessment was performed by trained assessors blinded to group assignment, reducing detection bias for the primary endpoint but not for ICU management or clinician-driven diagnoses (eg, imaging-confirmed cerebral ischaemia).
  • Statistics: A total of 794 patients were required to detect an 11% absolute reduction in unfavourable neurological outcome (from 50% to 39%) with 85% power at a two-sided 5% significance level; the trial ultimately randomised 850 patients. Primary analysis used a modified intention-to-treat approach (excluding patients withdrawing consent after randomisation) with adjusted risk ratios for the primary endpoint.
  • Follow-Up Period: 180 days for the primary neurological outcome; key secondary outcomes and adverse events assessed to day 28 (or ICU/hospital discharge for some endpoints).

Key Results

This trial was not stopped early. A single interim analysis was planned (after enrolment of 300 patients) and did not lead to early stopping.

Outcome Liberal (Hb <9 g/dL) Restrictive (Hb <7 g/dL) Effect p value / 95% CI Notes
Unfavourable neurological outcome at 180 days (GOS-E 1–5) 246/393 (62.6%) 300/413 (72.6%) aRR 0.86 95% CI 0.79 to 0.94; P=0.002 Primary outcome; modified intention-to-treat; adjusted analysis
GOS-E (ordinal shift analysis) at 180 days n=393 n=413 aOR 1.37 95% CI 1.07 to 1.75; P=0.01 Higher odds of better functional outcome with liberal strategy
All-cause mortality at 28 days 82/397 (20.7%) 94/418 (22.5%) RR 0.95 95% CI 0.74 to 1.22 No clear mortality difference at 28 days
Composite: death or any organ failure at 28 days 318/397 (80.1%) 328/419 (78.3%) RR 1.03 95% CI 0.97 to 1.10 Composite endpoint; organ failure definitions prespecified
Any organ failure at 28 days 236/397 (59.4%) 241/423 (57.0%) RR 1.03 95% CI 0.96 to 1.10 No clear difference
ICU length of stay (days) Mean 14.9 Mean 16.1 Mean difference −1.19 95% CI −3.34 to 0.97 Direction favours liberal strategy; CI crosses 0
Hospital length of stay (days) Mean 26.4 Mean 29.9 Mean difference −3.50 95% CI −8.62 to 1.62 Direction favours liberal strategy; CI crosses 0
Any red blood cell transfusion after randomisation 345/397 (86.9%) 170/423 (40.2%) Not reported Not reported Large separation in exposure; total transfusion episodes 910 vs 373
Red blood cell units per participant (post-randomisation) Median 2 (IQR 1–3) Median 0 (IQR 0–1) Mean difference 1.0 unit 95% CI 0.87 to 1.12 Protocol successfully increased transfusion intensity
Cerebral ischaemia (serious adverse event) 35/397 (8.8%) 57/423 (13.5%) RR 0.65 95% CI 0.44 to 0.97 Lower frequency with liberal strategy
Venous thromboembolism (serious adverse event) 19/397 (4.8%) 17/423 (4.0%) RR 1.19 95% CI 0.63 to 2.25 No clear increase with liberal strategy
Acute respiratory distress syndrome (serious adverse event) 29/397 (7.3%) 36/423 (8.5%) RR 0.86 95% CI 0.54 to 1.36 No clear increase with liberal strategy
Transfusion-related acute lung injury 0/397 (0.0%) 2/423 (0.5%) RR 0.21 95% CI 0.01 to 4.31 Rare events; imprecise estimate
Transfusion-associated circulatory overload 2/397 (0.5%) 2/423 (0.5%) RR 1.02 95% CI 0.15 to 7.53 Rare events; no clear difference
  • Primary endpoint favoured the liberal strategy: unfavourable 180-day neurological outcome 62.6% vs 72.6% (aRR 0.86; 95% CI 0.79 to 0.94; P=0.002).
  • Protocol separation was substantial: any post-randomisation transfusion 86.9% vs 40.2%; median post-randomisation red blood cell units 2 (IQR 1–3) vs 0 (IQR 0–1).
  • Major adverse events were broadly similar; cerebral ischaemia was less frequent with the liberal strategy (8.8% vs 13.5%; RR 0.65; 95% CI 0.44 to 0.97).
  • Subgroups (primary outcome, relative risks): treatment effects were directionally consistent across prespecified subgroups, with no statistically significant interactions reported (eg, traumatic brain injury RR 0.87 [0.76–1.00] vs subarachnoid haemorrhage RR 0.87 [0.72–1.04] vs intracerebral haemorrhage RR 0.85 [0.71–1.01]; P for interaction 0.98).
  • Sensitivity/post hoc (primary outcome definition): using GOS-E 1–4 as unfavourable, outcomes remained directionally consistent (50.4% vs 60.5%; adjusted RR 0.85; 95% CI 0.75 to 0.95).

Internal Validity

  • Randomisation and allocation concealment: centralised computer randomisation with stratification (centre, diagnosis category, GCS strata) and variable block sizes supports allocation concealment and minimises selection bias.
  • Post-randomisation exclusions and missingness: 850 randomised; 30 withdrew consent post-randomisation (11 liberal; 19 restrictive) and were excluded from the modified intention-to-treat analysis; primary outcome available for 806/820 analysed (393/397 liberal; 413/423 restrictive), with 14 missing primary outcomes.
  • Performance bias: open-label transfusion strategy permits clinician behaviour changes (eg, ICU interventions, imaging, withdrawal decisions) that could influence outcomes; the primary endpoint is patient-centred but not purely objective.
  • Detection bias: primary outcome assessment at 180 days was blinded, which strengthens validity for the primary endpoint; adverse events such as cerebral ischaemia were dependent on clinical detection and diagnostic work-up (risk of differential ascertainment not fully eliminable).
  • Protocol adherence: protocol violations were low and similar across groups (approximately 5% in each arm in supplementary reporting), supporting fidelity to the assigned transfusion strategy.
  • Separation of the variable of interest: post-randomisation red blood cell transfusion occurred in 345/397 (86.9%) vs 170/423 (40.2%), with median 2 (IQR 1–3) vs 0 (IQR 0–1) units per participant (mean difference 1.0 unit; 95% CI 0.87 to 1.12).
  • Baseline comparability: groups were broadly similar at baseline (eg, median haemoglobin at randomisation 8.5 g/dL in both arms; median GCS at randomisation 6 in both; acute brain injury mix with traumatic brain injury ~58–61%, subarachnoid haemorrhage ~22–25%, intracerebral haemorrhage ~17–18%).
  • Timing: randomisation occurred a median of 3 days after admission (IQR 2–5/6), which may reduce confounding from immediate resuscitation but could miss very early secondary injury windows; however, eligibility required active ICU-level brain injury physiology and haemoglobin ≤9 g/dL.
  • Heterogeneity: inclusion of multiple acute brain injury syndromes increases clinical applicability but can dilute syndrome-specific effects; subgroup analyses did not identify significant effect modification, although interaction testing is typically underpowered.
  • Outcome measurement choice: the dichotomised GOS-E (1–5 vs 6–8) can discard ordinal information; the reported ordinal (shift) analysis remained supportive (aOR 1.37; 95% CI 1.07 to 1.75).
  • Statistical rigour: power assumptions were revised during the trial (with a single interim analysis planned); primary analysis remained prespecified (modified intention-to-treat) and effect estimates were presented with confidence intervals.

Conclusion on Internal Validity: Overall, internal validity is moderate-to-strong given central randomisation, strong protocol separation, and blinded primary outcome assessment; the main threats are open-label care (performance/ascertainment bias) and the exclusion of post-randomisation consent withdrawals.

External Validity

  • Population representativeness: broad international ICU enrolment (72 sites, 22 countries) supports generalisability to many neurocritical care systems; however, eligibility required haemoglobin ≤9 g/dL and anticipated ICU stay ≥72 hours, selecting a specific (anaemic, sicker) subset of acute brain injury patients.
  • Exclusions that limit generalisability: patients with other neurological diagnoses (eg, post-anoxic coma, ischaemic stroke), those unable to receive blood products, those with active bleeding/urgent surgical needs, and those with limitations of life-sustaining treatment at baseline were excluded.
  • Applicability across settings: the strategy is implementable wherever haemoglobin monitoring and blood bank support exist; feasibility and opportunity costs may differ in resource-limited environments given the substantially higher transfusion exposure in the liberal arm.
  • Clinical translation nuance: findings apply directly to traumatic brain injury, subarachnoid haemorrhage, and intracerebral haemorrhage in ICU; extrapolation to other neuro-ICU diagnoses (eg, large ischaemic stroke, hypoxic-ischaemic encephalopathy) is uncertain.

Conclusion on External Validity: External validity is good for adult ICU patients with moderate-to-severe acute brain injury and haemoglobin ≤9 g/dL, but is limited outside this anaemic neuro-ICU population and in settings where increased blood utilisation is not feasible.

Strengths & Limitations

  • Strengths:
    • Large, international, pragmatic neuro-ICU randomised trial addressing a high-impact, common decision with genuine clinical equipoise.
    • Substantial protocol separation in transfusion exposure and red blood cell units supports a valid test of the haemoglobin-threshold strategy.
    • Patient-centred primary endpoint at 180 days with blinded assessment mitigates detection bias for the primary outcome.
    • Consistent direction of effect across multiple analyses, including ordinal (shift) analysis and sensitivity analysis using a stricter “unfavourable” definition (GOS-E 1–4).
  • Limitations:
    • Open-label design permits co-intervention bias and differential diagnostic intensity for certain adverse events (eg, cerebral ischaemia) that are clinician-identified rather than protocol-mandated.
    • Modified intention-to-treat excluded 30 patients who withdrew consent after randomisation; if withdrawals were differential by prognosis, bias could occur.
    • Heterogeneous acute brain injury syndromes increase applicability but complicate syndrome-specific inference and can mask clinically important subgroup effects.
    • Eligibility anchored to haemoglobin ≤9 g/dL and ICU physician expectation of ≥72-hour ICU stay limits direct applicability to less severe or non-anaemic patients.

Interpretation & Why It Matters

  • Clinical signal
    The liberal strategy (transfuse if Hb <9 g/dL) reduced the proportion with an unfavourable 180-day neurological outcome compared with the restrictive strategy (Hb <7 g/dL), with supportive ordinal (shift) findings and a lower frequency of cerebral ischaemia.
  • Mechanistic plausibility
    In acute brain injury, maintaining higher haemoglobin may protect against secondary brain injury by supporting oxygen delivery in a setting of impaired autoregulation and microcirculatory dysfunction; the observed reduction in cerebral ischaemia is consistent with this mechanism.
  • Resource and safety trade-offs
    The liberal strategy substantially increased transfusion exposure (86.9% vs 40.2%) and red blood cell use (median 2 vs 0 units per participant) without a clear increase in major transfusion-related complications in the reported outcomes; implementation requires local assessment of blood availability and stewardship priorities.

Controversies & Subsequent Evidence

  • Feasibility-driven protocol evolution: the published trial protocol described broader early ambitions (including larger sample size assumptions and interim monitoring plans) that were subsequently amended as recruitment realities became clearer; this raises legitimate concerns about optimistic effect-size assumptions in the final power calculation, despite prespecified statistical safeguards and transparent reporting. 12
  • Open-label care and the possibility of bias: transfusion strategy could influence clinician behaviour (eg, intensity of neuro-monitoring, imaging, or other supportive care), and it cannot be excluded that unblinded decisions around treatment limitation contributed to between-group differences; blinded 180-day outcome assessment reduces, but does not eliminate, this threat. 2
  • Outcome definition (dichotomised GOS-E): dichotomising the GOS-E at 1–5 vs 6–8 risks information loss and can amplify apparent effects depending on distribution; the trial’s ordinal analysis and sensitivity analysis using GOS-E 1–4 remained supportive, strengthening inferential robustness. 2
  • Context with other randomised evidence: in a large traumatic brain injury–specific trial (HEMOTION), a more liberal threshold (Hb <10 g/dL) did not show a statistically significant reduction in unfavourable 6-month outcome compared with a restrictive threshold (Hb <7 g/dL), highlighting residual uncertainty about effect magnitude and potential heterogeneity across neuro-injury syndromes and thresholds. 3
  • Earlier neurocritical trials were smaller and heterogeneous: prior trials tested different thresholds, populations, and co-interventions (including erythropoietin use), limiting direct comparability and underscoring why TRAIN was needed to resolve equipoise at scale. 4
  • Meta-analytic context: systematic reviews before and around TRAIN generally emphasised low certainty and heterogeneity in the brain-injury transfusion literature; updated evidence syntheses (including Cochrane and neurotrauma-focused meta-analyses) provide a framework for integrating TRAIN with newer large RCTs and for reassessing subgroup effects and net benefit. 56
  • Guidelines and translation: broad transfusion guidelines continue to recommend restrictive strategies for most critically ill adults and acknowledge condition-specific exceptions/uncertainty; TRAIN provides high-quality neuro-ICU evidence likely to influence future iterations, particularly for anaemic acute brain injury patients. 78
  • Neuro-focused guideline implications: major brain injury guidelines (eg, severe traumatic brain injury and aneurysmal subarachnoid haemorrhage) provide broader neurocritical care recommendations; transfusion thresholds have historically been underpinned by limited neuro-specific RCT evidence, and TRAIN strengthens the evidentiary base for this domain. 910

Summary

  • TRAIN randomised 850 ICU patients with acute brain injury and haemoglobin ≤9 g/dL to liberal (Hb <9) vs restrictive (Hb <7) transfusion strategies.
  • The liberal strategy reduced unfavourable 180-day neurological outcome (62.6% vs 72.6%; aRR 0.86; 95% CI 0.79 to 0.94; P=0.002) with supportive ordinal and sensitivity analyses.
  • Separation was large: post-randomisation transfusion 86.9% vs 40.2%; median red blood cell units 2 vs 0 (mean difference 1.0 unit; 95% CI 0.87 to 1.12).
  • Mortality at 28 days was similar (20.7% vs 22.5%; RR 0.95; 95% CI 0.74 to 1.22), and major adverse events were broadly similar; cerebral ischaemia occurred less frequently with liberal transfusion (8.8% vs 13.5%; RR 0.65; 95% CI 0.44 to 0.97).
  • Internal validity is strengthened by randomisation and blinded primary outcome assessment, but limited by open-label care and post-randomisation consent withdrawals.

Further Reading

Other Trials

Systematic Review & Meta Analysis

Observational Studies

Guidelines

Notes

  • TRAIN tested haemoglobin-trigger strategies (7 vs 9 g/dL) in anaemic ICU patients with traumatic brain injury, subarachnoid haemorrhage, and intracerebral haemorrhage; it does not directly address non-anaemic patients or other neuro-ICU diagnoses.

Overall Takeaway

TRAIN provides large, international randomised evidence that a liberal transfusion threshold (Hb <9 g/dL) can improve longer-term neurological outcomes in anaemic ICU patients with acute brain injury compared with a restrictive threshold (Hb <7 g/dL), without a clear signal of increased major transfusion-related harm in reported outcomes. The result challenges routine extrapolation of restrictive ICU transfusion thresholds to neurocritical care, while open-label care and the broader body of neuro-specific RCTs mean that implementation should remain context-aware and evidence-synthesised.

Overall Summary

  • Liberal transfusion (Hb <9) reduced unfavourable 180-day neurological outcomes vs restrictive (Hb <7) in anaemic acute brain injury ICU patients.
  • Separation was large (86.9% vs 40.2% transfused; median 2 vs 0 units), supporting a credible “strategy” test.
  • Key remaining uncertainties relate to open-label care effects and syndrome-specific applicability when integrating with other neurocritical transfusion trials.

Bibliography