Skip to main content

Foundational Trials

HEMOTION

Image: Shutterstock / ChaNaWiT

Publication

  • Title: Liberal or Restrictive Transfusion Strategy in Aneurysmal Subarachnoid Hemorrhage
  • Acronym: SAHARA
  • Year: 2025
  • Journal published in: New England Journal of Medicine
  • Citation: English SW, Delaney A, Fergusson DA, Chassé M, Turgeon AF, Lauzier F, et al.; SAHARA Trial Investigators on behalf of the Canadian Critical Care Trials Group. Liberal or Restrictive Transfusion Strategy in Aneurysmal Subarachnoid Hemorrhage. N Engl J Med. 2025;392(11):1079-1088.

Context & Rationale

  • Background
    Anaemia is common during the acute phase of aneurysmal subarachnoid haemorrhage (aSAH), driven by phlebotomy, inflammation, haemodilution, operative blood loss, and critical illness physiology.
  • Background
    Red-cell transfusion is frequently used to augment arterial oxygen content, but it carries biological and logistical downsides (e.g., transfusion-associated pulmonary and immunological complications, thrombosis, resource strain, and exposure to multiple donor units).
  • Background
    Prior evidence in aSAH was dominated by observational associations and small mechanistic studies, leaving equipoise on whether aiming for “higher haemoglobin” meaningfully improves patient-centred neurological outcomes.
  • Research Question/Hypothesis
    Among adults with aSAH who develop anaemia (haemoglobin ≤10.0 g/dL) in the first 10 days of admission, does a liberal transfusion strategy (threshold 10.0 g/dL) improve 12‑month neurological outcome compared with a restrictive strategy (threshold 8.0 g/dL)?
  • Why This Matters
    Transfusion thresholds in neurocritical care are highly variable in practice, are costly, and have potential downstream harms; aSAH is a disease where delayed cerebral ischaemia and secondary injury are plausible targets for oxygen-delivery interventions, making definitive randomised evidence clinically and methodologically important.

Design & Methods

  • Research Question: In adults with aSAH and haemoglobin ≤10.0 g/dL within 10 days of admission, does a liberal red-cell transfusion threshold (≤10.0 g/dL) reduce unfavourable neurological outcome at 12 months compared with a restrictive threshold (≤8.0 g/dL)?
  • Study Type: Pragmatic, investigator-initiated, open-label, randomised controlled trial with blinded outcome and imaging adjudication; conducted at 23 centres (Canada, Australia, USA); pilot phase (Oct 2015–Nov 2016) followed by full trial phase (Mar 2018–Jul 2023).
  • Population:
    • Adults (≥18 years) with first-ever aneurysmal subarachnoid haemorrhage and haemoglobin ≤10.0 g/dL within the first 10 days after hospital admission.
    • Key exclusions (high-level): non-aneurysmal SAH; active bleeding causing haemodynamic instability; contraindications to transfusion or known objection to receiving blood products.
  • Intervention:
    • Liberal strategy: leukoreduced red-cell transfusion was mandatory when haemoglobin was ≤10.0 g/dL; transfused 1 unit at a time; protocol applied from randomisation until 21 days after hospital presentation, death, or discharge (whichever came first).
    • Daily haemoglobin monitoring to at least day 10 or ICU discharge (whichever occurred first), with continued monitoring within protocol-defined intervention period as clinically indicated.
  • Comparison:
    • Restrictive strategy: leukoreduced red-cell transfusion was permitted only when haemoglobin was ≤8.0 g/dL; 1 unit at a time; same intervention window and monitoring framework as the liberal arm.
  • Blinding: Open-label treatment allocation (clinical teams and patients); blinded assessment of the primary 12‑month modified Rankin Scale (mRS) outcome; blinded adjudication of radiographic/clinical secondary brain injury outcomes.
  • Statistics: Power calculation: 740 patients to detect a 25% relative reduction in unfavourable neurological outcome (mRS ≥4) from 40% to 30% with 80% power at a two-sided 5% significance level (allowing ~3% loss to follow-up); primary analysis by intention-to-treat using a binomial generalised linear mixed model (log link), adjusted for centre and sex.
  • Follow-Up Period: Primary endpoint at 12 months post-randomisation; in-hospital secondary outcomes assessed during the intervention window (up to day 21 after presentation) with follow-up for key in-hospital events up to 28 days where specified.

Key Results

This trial was not stopped early. A prespecified interim analysis was conducted after enrolment of the first 370 patients using a Haybittle–Peto stopping boundary, and the trial continued to completion.

Outcome Liberal strategy Restrictive strategy Effect p value / 95% CI Notes
Primary: unfavourable neurological outcome at 12 months (mRS ≥4) 122/364 (33.5%) 136/361 (37.7%) RR 0.88 95% CI 0.72 to 1.09; P=0.22 Primary analysis adjusted for centre and sex; outcome data available for 725/732 (99.0%) analysed participants.
Mortality at 12 months 40/364 (11.0%) 52/361 (14.4%) RR 0.77 95% CI 0.52 to 1.14; P not reported Not powered for mortality; confidence interval includes clinically important benefit and no effect.
Functional Independence Measure (FIM) at 12 months 99.7 ± 37.3 100.8 ± 36.4 Mean difference −1.1 95% CI −7.0 to 4.9; P not reported Higher score indicates better functional independence.
EQ‑5D‑5L health-utility index at 12 months 0.74 ± 0.26 0.72 ± 0.27 Mean difference 0.02 95% CI −0.04 to 0.07; P not reported Patient-reported health status; scale 0–1.
Quality of life: EQ‑5D visual analogue scale at 12 months 72.0 ± 23.9 71.7 ± 24.6 Mean difference 0.3 95% CI −5.4 to 6.0; P not reported Patient-reported; scale 0–100.
New radiographic vasospasm during intervention period 115/365 (31.5%) 149/366 (40.7%) RR 0.77 95% CI 0.63 to 0.95; P not reported Radiographic outcome adjudicated by a blinded assessor; fewer cases with liberal strategy.
New delayed cerebral ischaemia during intervention period 64/365 (17.5%) 73/366 (19.9%) RR 0.88 95% CI 0.64 to 1.21; P not reported No clear effect; confidence interval includes benefit and no effect.
New cerebral infarction during intervention period 58/365 (15.9%) 65/366 (17.8%) RR 0.89 95% CI 0.63 to 1.26; P not reported No clear effect on infarction despite lower radiographic vasospasm.
Tertiary (implementation): any red-cell transfusion during intervention period 365/366 (99.7%) 128/366 (35.0%) RR 2.86 95% CI 2.42 to 3.38; P not reported Marked separation in exposure to transfusion.
Tertiary (dose): red-cell units transfused during intervention period Median 2 (IQR 2–4) Median 0 (IQR 0–1) Median difference 2.0 95% CI 1.78 to 2.20; P not reported Substantial increase in cumulative donor exposure with liberal strategy.
Tertiary: new intubation 22/364 (6.0%) 40/363 (11.0%) RR 0.56 95% CI 0.34 to 0.93; P not reported Open-label outcome; no separation in ventilation duration or length of stay.
Safety: acute transfusion reaction 2/366 (0.5%) 1/366 (0.3%) OR 2.01 95% CI 0.18 to 22.3; P not reported Events were rare; wide confidence interval.
  • The liberal strategy did not significantly reduce the primary endpoint (mRS ≥4 at 12 months): RR 0.88; 95% CI 0.72 to 1.09; P=0.22.
  • Despite a reduction in radiographic vasospasm (RR 0.77; 95% CI 0.63 to 0.95), there was no clear signal for delayed cerebral ischaemia, infarction, or mortality.
  • Protocol separation was substantial: 99.7% of liberal-strategy patients received transfusion vs 35.0% in the restrictive group (RR 2.86; 95% CI 2.42 to 3.38), with a median of 2 vs 0 red-cell units.

Internal Validity

  • Randomisation and allocation: Central, concealed, computer-generated randomisation with variable block sizes (4 and 6) stratified by centre; minimises selection bias.
  • Drop out / exclusions: 742 patients were randomised (pilot + full trial); 10 were excluded post-randomisation because of withdrawn/declined consent or removal of data at the request of a research ethics board; primary outcome was available for 725/732 (99.0%) analysed patients (364 vs 361).
  • Performance/detection bias: Treatment was open-label (unavoidable given transfusion thresholds); primary endpoint (mRS) and imaging endpoints were assessed by blinded adjudicators, reducing detection bias for key outcomes.
  • Protocol adherence: Overall protocol violations were infrequent (reported as 12 violations affecting 11 patients, 1.5%); however, protocol deviations were common in implementation metrics (e.g., time-to-transfusion >6 hours occurred in 134 patients in the liberal arm).
  • Baseline characteristics: Groups were well balanced across severity and key prognostic factors; median haemoglobin at randomisation was 9.4 g/dL (IQR 9.0–9.7) in both groups.
  • Baseline “secondary injury already present”: Radiographic vasospasm before randomisation occurred in 36.6% (liberal) vs 35.0% (restrictive), and delayed cerebral ischaemia before randomisation in 13.7% vs 15.6%, implying that a meaningful proportion had already entered the vasospasm/DCI window before transfusion strategy initiation.
  • Timing: Randoms were initiated a median of 3 days after admission (IQR 2–5) and within 10 days of admission; the intervention window lasted a median of 17 days (IQR 11–21) vs 16 days (IQR 10–21).
  • Dose and separation of the variable of interest: Median pretransfusion haemoglobin was 9.6 g/dL (IQR 9.5–9.8) vs 7.6 g/dL (IQR 7.2–7.8); any transfusion occurred in 99.7% vs 35.0%; median red-cell units were 2 (IQR 2–4) vs 0 (IQR 0–1).
  • Crossover/contamination: Protocol deviations included early transfusion near or above the restrictive threshold (e.g., transfusion within 0.5 g/dL above threshold in 8 restrictive patients), and missed transfusion when haemoglobin was 9.5–10.0 g/dL in 33 liberal patients; separation remained large despite these deviations.
  • Adjunctive therapy use: Use of common aSAH co-interventions was broadly similar between groups in reported tables (e.g., endovascular therapy, CSF diversion, rescue vasospasm therapies), supporting limited confounding by co-interventions, though granular clinician-level decisions cannot be fully excluded.
  • Outcome assessment: Primary outcome and key imaging outcomes were prespecified with blinded adjudication; non-blinded tertiary outcomes (e.g., intubation decisions) remain more vulnerable to performance bias.
  • Statistical rigour: Prespecified intention-to-treat analysis with mixed models and adjustment for centre/sex; prespecified interim analysis used a conservative stopping boundary; no multiplicity correction for secondary outcomes (interpretation appropriately cautious).

Conclusion on Internal Validity: Overall, internal validity appears moderate-to-strong given concealed randomisation, high follow-up completeness, and blinded adjudication for key endpoints, tempered by unavoidable open-label delivery and frequent implementation deviations (particularly transfusion timing) that could dilute or complicate estimation of small treatment effects.

External Validity

  • Population representativeness: Adults with aSAH and clinically relevant anaemia (≤10.0 g/dL) cared for in contemporary, high-resource neurocritical care systems; median age 56 years; substantial burden of high-grade haemorrhage (Modified Fisher grade 3–4 comprised the majority) and severe clinical grade (WFNS IV–V ~ one-third).
  • Practice setting: Multinational (Canada, Australia, USA) tertiary centres with ready access to daily haemoglobin monitoring, leukoreduced blood products, aneurysm securing (coil/clip), CSF diversion, and vasospasm surveillance/treatment.
  • Exclusions and applicability boundaries: Findings may not generalise to non-aneurysmal SAH, patients with major active haemodynamic bleeding, those who refuse transfusion, or settings where blood supply/monitoring differs materially.
  • Intervention transportability: The strategy is operationally simple (threshold-based transfusion) and aligns with routine ICU workflows, but the observed delays to transfusion (common in practice) highlight that real-world effectiveness depends on implementation fidelity.

Conclusion on External Validity: Generalisability is moderate to similar high-income neurocritical care systems managing aSAH with frequent anaemia, but may be limited where blood product availability, monitoring cadence, or standard vasospasm/DCI management differs substantially.

Strengths & Limitations

  • Strengths:
    • Largest pragmatic randomised evaluation of transfusion thresholds in aSAH to date, addressing an important practice-variation question.
    • Multicentre, multinational design (23 centres) with concealed randomisation and high completeness of 12‑month follow-up.
    • Blinded adjudication of the primary functional endpoint and key imaging-based secondary outcomes.
    • Large achieved separation in haemoglobin at transfusion and transfusion exposure, supporting a biologically meaningful contrast.
  • Limitations:
    • Open-label delivery may bias clinician-driven outcomes and co-interventions (particularly tertiary outcomes such as intubation).
    • Notable proportion had vasospasm/DCI already present before randomisation, potentially limiting modifiability of downstream injury.
    • Implementation deviations (especially transfusion timing) were common and could attenuate differences attributable to the intended thresholds.
    • Secondary outcomes were not multiplicity-adjusted; signals outside the primary endpoint should be interpreted as hypothesis-generating.

Interpretation & Why It Matters

  • Clinical practice signal
    A routine liberal transfusion threshold of 10 g/dL in anaemic aSAH patients did not improve 12‑month disability compared with a restrictive 8 g/dL approach, despite materially increasing transfusion exposure.
  • Mechanism vs outcome
    The reduction in radiographic vasospasm without concordant reductions in delayed cerebral ischaemia, infarction, or functional outcome suggests that radiographic vasospasm is an imperfect surrogate for the clinical consequences that matter most.
  • Resource and harm balance
    Given the large increase in red-cell use (and donor exposure) under a liberal strategy, the lack of functional benefit provides a strong argument against “default” higher haemoglobin targets in this population, reserving transfusion for lower thresholds or clear individual indications.

Controversies & Subsequent Evidence

  • Earlier aSAH randomised evidence was limited in size and scope (e.g., a single-centre trial targeting higher haemoglobin goals), leaving uncertainty about patient-centred neurological benefit and the plausibility of meaningful effect sizes in aSAH.1
  • The dissociation between improved radiographic vasospasm and unchanged infarction/DCI/12‑month function reinforces methodological caution about surrogate endpoints in aSAH and highlights the importance of long-term functional outcomes as primary endpoints; published commentary has emphasised this interpretive tension and the clinical costs of higher transfusion exposure.5
  • SAHARA’s results align with a broader acute brain injury transfusion literature: the TRAIN trial (acute brain injury population) similarly did not demonstrate clear functional benefit with liberal thresholds compared with restrictive thresholds.2
  • In traumatic brain injury, the HEMOTION trial found no improvement in neurological outcome with a liberal transfusion strategy (threshold 10 g/dL) compared with a restrictive strategy (threshold 7 g/dL), supporting the general direction of restrictive practice where specific indications are absent.3
  • Contemporary general transfusion guidelines (AABB) recommend restrictive transfusion thresholds for most haemodynamically stable adults, and SAHARA provides high-quality aSAH-specific randomised evidence that is directionally consistent with these recommendations (while still allowing for individualised decisions in exceptional circumstances).4

Summary

  • Pragmatic, multinational RCT (23 centres; Canada/Australia/USA) comparing liberal (≤10 g/dL) vs restrictive (≤8 g/dL) red-cell transfusion thresholds in anaemic aSAH.
  • No statistically significant improvement in 12‑month disability (mRS ≥4): RR 0.88; 95% CI 0.72 to 1.09; P=0.22.
  • Radiographic vasospasm was lower with liberal transfusion (RR 0.77; 95% CI 0.63 to 0.95), without corresponding reductions in DCI, infarction, or mortality.
  • Transfusion exposure was dramatically higher in the liberal arm (99.7% transfused; median 2 units) vs restrictive (35.0% transfused; median 0 units).
  • Overall, SAHARA argues against routine higher haemoglobin targets in aSAH, supporting restrictive thresholds unless individual clinical indications justify deviation.

Further Reading

Other Trials

Systematic Review & Meta Analysis

Observational Studies

Guidelines

Notes

  • Within SAHARA, transfusion thresholds produced strong separation in transfusion exposure and haemoglobin-at-transfusion, yet long-term disability was unchanged, supporting an interpretation that routinely targeting higher haemoglobin is unlikely to produce large functional benefits in aSAH.

Overall Takeaway

SAHARA provides high-quality, pragmatic randomised evidence that a liberal red-cell transfusion threshold (≤10 g/dL) in anaemic aSAH does not improve 12‑month disability compared with a restrictive threshold (≤8 g/dL), despite materially increasing transfusion exposure. The trial’s physiology-adjacent signal (less radiographic vasospasm) did not translate into fewer infarctions, DCI, or better function, reinforcing that patient-centred outcomes—not surrogates—should drive transfusion strategy selection in aSAH.

Overall Summary

  • In aSAH patients with haemoglobin ≤10 g/dL, a liberal transfusion strategy increased transfusion exposure substantially but did not improve 12‑month neurological outcomes.

Bibliography