Design: pooled patient-level analysis of 11 randomised heart-failure trials, Cox models and Poisson regression with restricted cubic splines
Population: 25,003 patients with HFrEF and 17,210 with HFpEF or HFmrEF
Primary outcome: first heart-failure hospitalisation or cardiovascular death, its components, and all-cause death by baseline haemoglobin
Effect: lowest observed risk at about 14 g/dL in women and 15 g/dL in men; excess risk persisted at 0 to <1 g/dL above WHO anaemia thresholds versus at least 2 g/dL above

Patient-level data from 11 heart-failure trials — 25,003 patients with reduced ejection fraction and 17,210 with preserved or mildly reduced — were pooled to ask where on the haemoglobin range risk is actually lowest. The WHO anaemia thresholds, under 12 g/dL in women and under 13 g/dL in men, were derived from distributions in healthy populations rather than from outcomes.

Risk was lowest at roughly 14 g/dL in women and 15 g/dL in men. Patients whose haemoglobin sat between 0 and 1 g/dL above the WHO threshold — comfortably 'not anaemic' on a report — still carried excess risk of heart-failure hospitalisation or cardiovascular death compared with those at least 2 g/dL above it.

What this challenges is a binary reading of the full blood count. A haemoglobin of 13.4 g/dL in a man with heart failure is not reassurance; on these data it sits in a band associated with worse outcomes, and it should prompt the same question a frankly low value would — is there iron deficiency, is there occult bleeding, is there renal disease.

The authors are careful about what follows. This is prognostic information that should prompt a search for reversible contributors, not a target to treat towards.

Read haemoglobin in heart failure as a continuum, not as anaemic or not anaemic
Check ferritin and transferrin saturation in a heart-failure patient whose haemoglobin is low-normal, not only when it is frankly low
Look for occult gastrointestinal loss where haemoglobin is drifting down across visits, whatever the absolute value
Record the trend, not just the latest value — the direction carries information the threshold hides
Do not transfuse towards 14 or 15 g/dL; this is an observed association, not a treatment target

Don't overread it

This is a pooled observational association from trial baselines — it does not establish that correcting haemoglobin to these values improves outcomes.

The statistics, in plain English

The optimal values come from restricted cubic spline models, which fit a smooth curve to risk across the whole haemoglobin range instead of splitting patients into groups — that is how a minimum can be identified at a point no threshold would pick out. Because this pools baseline measurements from trial participants and observes what happened afterwards, the relationship is an association. Low haemoglobin may be marking worse underlying disease rather than causing worse outcomes, and there is no randomised evidence here that raising haemoglobin improves anything.