Optimal Hemoglobin Thresholds for ICU Transfusion: A Target Trial Emulation in a Large Electronic Medical Record-Based Real-World Database
Author(s)
Takahiro Kinoshita, MD, MPH1, Io Murayama, MD (candidate)2.
1MeDiCU, Inc., Osaka, Japan, 2The University of Tokyo, Tokyo, Japan.
1MeDiCU, Inc., Osaka, Japan, 2The University of Tokyo, Tokyo, Japan.
OBJECTIVES: To investigate the optimal hemoglobin threshold for initiating red‑blood‑cell transfusion in adult intensive care unit (ICU) patients.
METHODS: We emulated a three‑arm target trial within OneICU, an electronic medical record database with minute‑level physiologic data from Japanese ICUs (2013 - 2024). Adults whose hemoglobin ever fell below 9 g/dL were enrolled and followed in 6‑hour intervals from ICU admission to discharge. We compared three dynamic strategies: transfuse only when hemoglobin dropped below 7, 8, or 9 g/dL. The parametric g‑formula estimated causal effects, adjusting for static variables (age, sex, admission diagnosis, and hospital) and time-varying variables (body temperature, heart and respiratory rates, mean arterial pressure, oxygen saturation, pH, lactate, hemoglobin, and vasopressor use). Patients discharged from the ICU within 7 days were considered censored.
RESULTS: Among 16 233 admissions, 7 300 (45 %) received at least one transfusion. The mean age was 68.5 years, and 6855 (42.2%) were female. Estimated 7‑day ICU mortality was 7.9 % under the 7 g/dL strategy, 7.4 % under 8 g/dL (risk difference [RD] -0.5 %; 95 % CI -0.7 % to -0.4 %), and 6.7 % under 9 g/dL (RD -1.2 %; 95 % CI -1.6 % to -0.8 %). Subgroup analyses in patients ≥70 years, those with sepsis, and those with acute coronary syndrome showed a consistent benefit of a higher hemoglobin threshold.
CONCLUSIONS: A liberal transfusion strategy triggered at ≤9 g/dL resulted in clinically meaningful reductions in early ICU mortality compared with a restrictive 7 g/dL strategy. The use of a large real-world database with minute-level EMR records enabled estimation of intervention effects while accounting for treatment-confounder feedback.
METHODS: We emulated a three‑arm target trial within OneICU, an electronic medical record database with minute‑level physiologic data from Japanese ICUs (2013 - 2024). Adults whose hemoglobin ever fell below 9 g/dL were enrolled and followed in 6‑hour intervals from ICU admission to discharge. We compared three dynamic strategies: transfuse only when hemoglobin dropped below 7, 8, or 9 g/dL. The parametric g‑formula estimated causal effects, adjusting for static variables (age, sex, admission diagnosis, and hospital) and time-varying variables (body temperature, heart and respiratory rates, mean arterial pressure, oxygen saturation, pH, lactate, hemoglobin, and vasopressor use). Patients discharged from the ICU within 7 days were considered censored.
RESULTS: Among 16 233 admissions, 7 300 (45 %) received at least one transfusion. The mean age was 68.5 years, and 6855 (42.2%) were female. Estimated 7‑day ICU mortality was 7.9 % under the 7 g/dL strategy, 7.4 % under 8 g/dL (risk difference [RD] -0.5 %; 95 % CI -0.7 % to -0.4 %), and 6.7 % under 9 g/dL (RD -1.2 %; 95 % CI -1.6 % to -0.8 %). Subgroup analyses in patients ≥70 years, those with sepsis, and those with acute coronary syndrome showed a consistent benefit of a higher hemoglobin threshold.
CONCLUSIONS: A liberal transfusion strategy triggered at ≤9 g/dL resulted in clinically meaningful reductions in early ICU mortality compared with a restrictive 7 g/dL strategy. The use of a large real-world database with minute-level EMR records enabled estimation of intervention effects while accounting for treatment-confounder feedback.
Conference/Value in Health Info
2025-11, ISPOR Europe 2025, Glasgow, Scotland
Value in Health, Volume 28, Issue S2
Code
P23
Topic
Clinical Outcomes, Epidemiology & Public Health, Methodological & Statistical Research
Topic Subcategory
Public Health, Safety & Pharmacoepidemiology
Disease
No Additional Disease & Conditions/Specialized Treatment Areas, Systemic Disorders/Conditions (Anesthesia, Auto-Immune Disorders (n.e.c.), Hematological Disorders (non-oncologic), Pain)