The messages

For years, ICUs treated anemia in brain-injured patients like anemia in anyone else: transfuse only when hemoglobin falls to about 70 g/L. Two large 2024 randomized trials HEMOTION (NEJM) and TRAIN (JAMA) have overturned that habit, marking the biggest shift in neurocritical-care transfusion practice in a decade.

Why the injured brain is different

Injured brain tissue has almost no reserve to pull extra oxygen from blood or raise blood flow in the vulnerable “penumbra” around the injury. So anemia isn’t just a lab number here it becomes a direct driver of secondary ischemic injury at hemoglobin levels the rest of the body tolerates. Both trials tested whether keeping hemoglobin higher protects the brain.

HEMOTION: a strong signal, just short of significance

Adults with moderate-to-severe traumatic brain injury and anemia were randomized to a liberal strategy (transfuse at Hb <100 g/L) or restrictive (<70 g/L), with unfavorable Glasgow Outcome Scale “Extended (GOS-E) at six months as the primary outcome. Unfavorable outcomes: 68.4% liberal vs 73.5% restrictive  a ~5-point improvement favoring liberal, but not statistically significant. A “negative” trial that still leans one way.

TRAIN: larger, broader, clearly positive

Bigger and wider  850 patients, 72 ICUs, 22 countries, including traumatic brain injury plus subarachnoid and intracerebral hemorrhage. Liberal trigger <9 g/dL vs restrictive <7 g/dL, GOS-E at 180 days. Liberal was clearly better: 62.6% vs 72.6% unfavorable, an absolute risk difference of 10.0%, with no excess clotting or lung injury.

Do they really disagree? No.

The trials are concordant in direction, differing only in power and breadth. HEMOTION’s confidence interval includes TRAIN’s result, and both favor transfusing sooner. The likely mechanism is better cerebral oxygen delivery at the microvascular level supported by benefit appearing in functional neurologic recovery rather than survival.

What this means at the bedside

Shift toward a liberal threshold of ~90-100 g/L, individualized by evidence of brain-tissue compromise (brain-tissue oxygen tension, an evolving infarct, vasospasm risk) rather than a rigid number. A 70 g/L default is now hard to defend in structural brain injury. Two caveats: neither trial showed a mortality benefit (the case rests on function), and both were underpowered for disease subgroups. The open frontier is titrating transfusion to real-time cerebral oximetry or brain-tissue oxygen monitoring rather than a single systemic haemoglobin value.