What question did this study set out to answer?

This research examines the optimal post-transfusion hemoglobin levels to minimize in-hospital mortality in critically ill patients with severe anemia.

April 28, 2026Open Access

Impact of post-transfusion hemoglobin levels on survival in critically Ill patients: a machine learning–based causal inference analysis

Puntos clave

This research examines the optimal post-transfusion hemoglobin levels to minimize in-hospital mortality in critically ill patients with severe anemia.
Analyzed 3,710 ICU admissions from MIMIC-IV and 4,719 from eICU for transfusion outcomes.
Used a double machine learning causal forest to estimate treatment effects on post-transfusion hemoglobin targets.
Identified optimal hemoglobin levels and evaluated subgroup characteristics and predictors of lower targets.
Optimal post-transfusion hemoglobin levels were 11.0 g/dL in training, 10.9 g/dL in testing, and 11.0 g/dL in external validation.
Lower baseline hemoglobin and estimated glomerular filtration rate linked to lower targeted optimal levels.
Indicators of inflammation correlated with optimal levels below 10.0 g/dL.

Resumen

Restrictive red blood cell transfusion strategies (hemoglobin trigger 7.0–8.0 g/dL) are well established, yet existing evidence addresses primarily when to initiate transfusion. The optimal achieved post-transfusion hemoglobin level in critically ill patients with severe anemia (pre-transfusion hemoglobin ≤ 8.0 g/dL) remains unknown. We analyzed 3,710 intensive care unit admissions in the Medical Information Mart for Intensive Care IV (MIMIC-IV; 3,339 training, 371 test) and externally validated findings in 4,719 admissions from the eICU Collaborative Research Database. Eligible subjects received red blood cells within 48 h of ICU admission, had a pre-transfusion hemoglobin ≤ 8.0 g/dL, and had no additional transfusions between 48 h and seven days. Using a double machine learning causal forest, we estimated treatment effects for post-transfusion hemoglobin targets from 8.0 to 14.0 g/dL relative to 8.0 g/dL, identified the target with the lowest predicted in-hospital mortality, evaluated subgroups, and modeled predictors of targets below 10.0 g/dL. Dose–response curves were U-shaped in all datasets. The estimated optimal hemoglobin was 11.0 g/dL in the training set, 10.9 g/dL in the test set, and 11.0 g/dL in external validation. Estimated optimal levels near 11.0 g/dL were consistent across chronic kidney disease, end-stage kidney disease, myocardial infarction, congestive heart failure, hypertension, and diabetes. Lower baseline hemoglobin and lower estimated glomerular filtration rate were associated with lower individualized optimal levels. Indicators of inflammation or hemodynamic instability, including hypotension, fever, leukocytosis, hyperlactatemia, acidosis, and vasopressor requirements, were associated with estimated optimal achieved post-transfusion hemoglobin levels below 10.0 g/dL. Among ICU subjects transfused under a restrictive trigger (hemoglobin ≤ 8.0 g/dL), an achieved post-transfusion hemoglobin around 11.0 g/dL was associated with the lowest predicted in-hospital mortality.

Me gusta

Guardar

Ver artículo completo