The Clotbust Review

A Restrictive Versus a Liberal Transfusion Strategy in Patients With Spontaneous Intracerebral Hemorrhage: A Secondary Analysis of TRAIN Randomized Clinical Trial

· DOI: 10.1161/STROKEAHA.125.050729 · PMC12372721 · stroke deep-dive intracerebral-hemorrhage transfusion critical-care TRAIN

Stylized illustration of anemic ICH transfusion threshold debate
AI-generated editorial illustration. Generated for editorial use.

Clinical Question (PICO)

P — 144 anemic (Hb ≤9 g/dL within 10 days of admission) adult ICU patients with spontaneous intracerebral hemorrhage (ICH), drawn from the TRAIN trial (72 ICUs, 22 countries). I — Liberal red blood cell transfusion strategy: trigger at Hb <9 g/dL. C — Restrictive strategy: trigger at Hb <7 g/dL. O — Unfavorable neurological outcome at 180 days, defined as GOS-E 1–5. Prespecified subgroups — none for the primary outcome in this secondary analysis; the parent TRAIN trial stratified by brain-injury type, GCS band (3–5 / 6–9 / 10–13), and center.

Bottom Line

In anemic ICU patients with spontaneous ICH, a liberal transfusion strategy (Hb <9 g/dL) reduced the rate of unfavorable 180-day GOS-E from 84.7% to 71.8% versus the restrictive strategy (Hb <7 g/dL), a 12.9-point absolute drop that did not reach statistical significance (RR 0.85, 95% CI 0.71–1.01; P=0.06). The adjusted odds ratio (2.31, 95% CI 0.89–5.99; P=0.082) likewise trended toward harm from restriction but with a confidence interval that crosses 1. The composite of 28-day mortality/organ failure was similar between arms.

Design

  • Trial type: Pre-specified secondary analysis of a prospective, multicenter, phase 3 randomized clinical trial (TRAIN). Open-label for clinicians (Hb-driven, impossible to blind transfusion triggers); patients/families blinded; outcome assessors blinded.
  • N: 144 patients with spontaneous ICH (subgroup of the parent 850+ TRAIN cohort; 7.5% of the enrolled brain-injury population).
    • Restrictive (Hb <7 g/dL): 73
    • Liberal (Hb <9 g/dL): 71
  • Randomization: 1:1, stratified by brain-injury type, GCS band (3–5 vs 6–9 vs 10–13), and center (in the parent trial).
  • Setting: 72 ICUs across 22 countries, 2017–2022.
  • Enrollment: 2017–2022.
  • Mean follow-up: 180 days (primary endpoint).
  • Analysis: Intention-to-treat. Multivariable logistic regression adjusted for hematoma volume >30 mL, hydrocephalus on admission, and admission GCS.
  • Primary outcome: Unfavorable neurological outcome at 180 days, defined as GOS-E 1–5 (death, vegetative, or severe disability).

Population

Inclusion Criteria

  • Spontaneous (non-traumatic) intracerebral hemorrhage.
  • Acute brain injury with Hb ≤9 g/dL within first 10 days of ICU admission.
  • Expected ICU stay ≥72 hours.
  • Enrolled in parent TRAIN trial at one of 72 participating ICUs.

Exclusion Criteria

  • Traumatic ICH, subarachnoid hemorrhage, or ischemic stroke (analyzed separately in TRAIN).
  • Hb never fell to ≤9 g/dL during the first 10 days.
  • Anticipated ICU stay <72 hours or imminent death.

Baseline Characteristics (overall)

  • Mean age: 58.4 years (SD 13.4).
  • Sex: 45.8% male.
  • Admission GCS: mean 7.3 (SD 3.3) — heavily comatose cohort.
  • Hematoma volume >30 mL: 75.7%.
  • Baseline characteristics were comparable between groups.
  • HIV prevalence was higher in the restrictive arm (mentioned as a potential driver of the sepsis signal).

Interventions

  • Restrictive: Transfusion of 1 unit packed RBCs when Hb <7 g/dL. Maintained for up to 28 days post-randomization, hospital discharge, or death.
  • Liberal: Transfusion of 1 unit packed RBCs when Hb <9 g/dL. Same 28-day / discharge / death window.

Daily Hb was measured from routine labs or blood-gas analyses. Protocol violation = transfusion outside assigned threshold or cross-matching error (5 patients in liberal [7.0%], 1 in restrictive [1.4%]).

Outcomes

Primary Outcome (unfavorable GOS-E 1–5 at 180 days):

  • Liberal: 51/71 (71.8%)
  • Restrictive: 61/72 (84.7%)
  • Risk ratio 0.85 (95% CI 0.71–1.01); P = 0.06
  • Absolute risk difference: −12.9 percentage points (95% CI −26.3 to +0.47).
  • Adjusted OR for unfavorable outcome, liberal vs restrictive: 2.31 (95% CI 0.89–5.99); P = 0.082, adjusted for hematoma volume >30 mL, hydrocephalus on admission, and admission GCS.

Secondary Outcomes:

  • Composite (28-day mortality + organ failure): similar between groups.
  • Transfusion exposure: 151 units (liberal) vs 54 units (restrictive); P<0.001. 88.7% of liberal patients received transfusion vs 42.4% of restrictive; P<0.001. Median units per patient: 2 (1–3) liberal vs 0 (0–1) restrictive.
  • Median lowest daily Hb differed significantly between groups (P<0.001; Figure 1).
  • Median GOS-E at 180 days: 3 (1–6) liberal vs 4 (1–5) restrictive.

Adverse Events / Safety:

  • Sepsis: higher in restrictive group (attributed to higher HIV prevalence in that arm and post-ICH infection biology).
  • Transfusion-associated circulatory overload (TACO) and transfusion-related acute lung injury (TRALI): similar between groups, possibly because both arms used a relatively conservative fluid background.
  • Most other adverse events were comparable.

Figures

Median daily lowest hemoglobin by arm
Figure 1. Median daily lowest hemoglobin concentration at baseline and after randomization in the two groups. Baseline values were the last blood hemoglobin level measured before randomization. Day 1 was defined as the day after randomization. Bars indicate the 25th and 75th percentiles.

Source: PMC PMC12372721 — str-56-2617-g002.jpg. Click image to expand.

GOS-E distribution at 180 days by arm
Figure 2. Distribution of Glasgow Outcome Scale Extended (GOS-E) scores at 180 days after randomization in the restrictive and liberal group. Each cell corresponds to a GOS-E category; bars show the percentage of patients in each category within each arm.

Source: PMC PMC12372721 — str-56-2617-g004.jpg. Click image to expand.

Criticisms

  • Secondary analysis of a parent RCT not powered for ICH. N=144 is too small for a definitive answer; the parent TRAIN trial was designed for mixed acute-brain-injury enrollment and was negative on its primary endpoint. The 95% CI for the relative risk (0.71–1.01) brushes the null — the trial is underpowered, not negative.
  • P=0.06 is not “significant”, but the point estimate (a 13-point absolute reduction in unfavorable outcome with liberal transfusion) is clinically meaningful. This is the textbook zone for “trending toward benefit” — neither practice-changing nor dismissable.
  • Open-label for clinicians (impossible to blind Hb-trigger transfusion strategies), though outcome assessors were blinded — risk of differential co-intervention.
  • Confounding by indication: patients who required transfusion had worse baseline comorbidity; the authors explicitly note this as a confounder in prior literature.
  • HIV imbalance between arms drove a sepsis signal in the restrictive arm — a chance imbalance in a small subgroup, not a clean causal finding.
  • Generalizability: 22-country ICU cohort with mean admission GCS 7.3 and 75.7% with hematoma >30 mL is a sick ICH population. Results may not extrapolate to milder ICH or to ward-level (non-ICU) care.
  • No information on withdrawal-of-care decisions — a critical confounder in severe ICH trials where early limitation can drive 180-day GOS-E independent of transfusion strategy.

Funding

The parent TRAIN trial was an investigator-initiated study coordinated by the Université Libre de Bruxelles (Fabio Silvio Taccone, principal investigator). The secondary analysis itself does not list a separate dedicated funding line in the abstract; the article’s funding statement should be consulted on PMC for sponsor and grant identifiers. ClinicalTrials.gov registration identifier for the parent trial: NCT02970154 (TRAIN — Transfusion Strategies for Patients in Intensive Care Units, ISRCTN registry number ISRCTN88413182). No NCT id is provided specifically for this secondary analysis.

The paper

  • Authors. Faso C, Gouvea Bogossian E, Bittencourt Rynkowski C, Moller K, Lormans P, Quintana Diaz M, Caricato A, Dabrowski W, Gonzalez Perez I, Steblaj S, Quintard H, Justo P, Righy C, Roman-Pognuz E, Huet O, Mahmoodpoor A, Blandino-Ortiz A, Junttila E, Funes N, Izzo G, Zattera L, Giacomucci A, Dibu J, Rodrigues A, Bouzat P, Vincent JL, Taccone FS, et al. (TRAIN Study Trial Group).
  • Title. A Restrictive Versus a Liberal Transfusion Strategy in Patients With Spontaneous Intracerebral Hemorrhage: A Secondary Analysis of TRAIN Randomized Clinical Trial.
  • Journal. Stroke.
  • Year. 2025.
  • DOI. 10.1161/STROKEAHA.125.050729
  • PMCID. PMC12372721
Deep Dive — click to expand

What this is

A secondary analysis of the multicenter TRAIN RCT asking whether ICU patients with spontaneous ICH and anemia should be transfused at Hb <7 g/dL (restrictive) or Hb <9 g/dL (liberal). The headline is a 13-point absolute drop in unfavorable 180-day GOS-E favoring the liberal arm that missed significance by a hair (P=0.06). In a sick, comatose ICH cohort (mean GCS 7, three-quarters with hematoma >30 mL), the trend toward benefit is large enough to deserve a real Phase 3 trial dedicated to ICH.

1. Shadow Audit

The article frames this as a “negative” secondary analysis but the effect size is in the same direction and roughly the same magnitude as HEMOTION (TBI), and consistent with the SAH anemia literature. The paper does not emphasize that a 13-point absolute difference in a 144-patient subgroup is large in absolute terms — the CI on the absolute scale runs from −26.3 to +0.5, meaning the data are compatible with anything from “liberal saves 1 in 4” to “essentially no difference.” The honest read: underpowered, not refuted.

The other shadow: transfusion burden. 151 units vs 54 units is a ~3× increase in red-cell exposure in the liberal arm. The paper does not cost this in dollars, donor exposure, or long-term immunomodulation risk. Even if liberal is neurologically better, the public-health frame is “more transfusion for a borderline signal.”

2. Inversion Engine

For the null to be true (no benefit from liberal transfusion), the 13-point absolute gap would have to be sampling noise around a true RR of 1.0. To flip the conclusion from “trend toward liberal benefit” to “trend toward liberal harm,” the underlying true RR would have to be above ~1.0 — which the data do not support (RR 0.85, upper CI 1.01). So the inversion is asymmetric: the data lean liberal, and only the upper bound brushes harm. A reasonable Bayesian posterior with a weakly informative prior favoring equivalence probably puts the probability of true RR<0.9 somewhere in the 60–75% range.

3. Second-Order Catalyst

If this result is real, the first institution to move would be large academic neuro-ICUs already running restrictive transfusion protocols by reflex (defaulting to <7 g/dL). The protocol change is trivial: nudge the transfusion trigger from 7 to 9 g/dL for anemic ICU patients with ICH in the first 10 days. That single number in the order set moves the needle. Smaller hospitals will wait for an ICH-specific RCT or a guideline update — likely 2–3 years out.

4. Asymmetric Leverage

The leverage is asymmetric: a small change in transfusion threshold (a number on a flowsheet) applied to a large denominator (every anemic ICU patient with ICH). Even a 5–10% relative reduction in unfavorable outcome at 180 days, applied across the ~30,000–40,000 severe ICH ICU admissions per year in the US, would translate to thousands of patients shifting from severe disability/vegetative/death to a better GOS-E band. The cost is incremental RBC units, not new drugs or new devices.

5. Paradigm Destroyer

The reflex this paper kills: “transfuse only when Hb <7 g/dL, regardless of brain injury subtype.” The two-sentence protocol change for tomorrow morning: in anemic ICU patients with spontaneous ICH and Hb between 7 and 9 g/dL, do not assume a <7 threshold is safe; consider transfusion at <9 g/dL pending a dedicated Phase 3 trial. The reflex to harden against this paper: clinging to <7 g/dL as universal in acute brain injury is now unsupported by either TRAIN-ICH or HEMOTION-TBI.

MVP — Minimum Viable Proof

The minimum numerical statement that, if true, would change bedside practice: a 10–15% absolute reduction in unfavorable 180-day GOS-E with liberal transfusion in anemic ICU ICH, sustained in a properly powered Phase 3 RCT dedicated to spontaneous ICH (n ≥ 600). TRAIN-ICH suggests the effect is real and in that range; it does not yet prove it.

Best Combination

Combine this finding with: (a) HEMOTION (TBI), which also trended liberal; (b) the SAH observational literature linking anemia to worse outcomes; (c) the broader ICU transfusion literature (TRICC, TRISS) showing restrictive is at least non-inferior in mixed ICU populations. The synthesis: for acute brain injury specifically — ICH, TBI, possibly SAH — the optimal transfusion threshold appears to sit above the generic ICU <7 g/dL default. A unified “brain-injury Hb <9 g/dL” policy is a defensible interim stance while awaiting a definitive ICH trial.

Overvalue Warning

(1) Don’t overread P=0.06 as “liberal is better.” The confidence interval brushes the null; this is a hypothesis-generating secondary analysis, not a Phase 3 verdict. (2) Don’t generalize the <9 g/dL threshold to non-ICU ward ICH patients, stable chronic anemia, or post-operative ICH — TRAIN-ICH enrolled only ICU patients with expected stay ≥72 hours and acute Hb drift, and the result may not transport.

Educational use notice. The Clotbust Review is dedicated to medical education. Figures from published articles are embedded for teaching purposes, with attribution to the original publication. We do not claim copyright over any embedded figure; all rights remain with the original publisher and authors under the article's published license. If you are a rights holder and would like a figure removed, please contact the editorial team. See our editorial notice for the full policy.