Reviewed by Lexie CornerJan 24 2025
According to a study published in Health Data Science, Emory University researchers created a new artificial intelligence (AI) model that can effectively predict the probability of blood transfusion in non-traumatic intensive care unit (ICU) patients. The study addresses challenges in predicting transfusion requirements for patients with diverse medical conditions.
Graphical Abstract. Image Credit: Alireza Rafiei, Emory University
Blood transfusions are critical for managing anemia and coagulopathy in ICUs, but existing clinical decision support systems often target specific patient groups or transfusion types, limiting their effectiveness in dynamic ICU environments. The AI model developed in this study uses clinical data, including laboratory results and vital signs, to predict transfusion needs within a 24-hour window.
Led by Alireza Rafiei from the Department of Computer Science and Rishikesan Kamaleswaran from the Department of Biomedical Informatics at Emory University, the research utilized a dataset of over 72,000 ICU patient records collected over five years. The AI system demonstrated high performance using a meta-model ensemble approach, achieving an AUROC of 0.97, an accuracy of 0.93, and an F1 score of 0.89.
Our model not only accurately predicts the need for a blood transfusion but also identifies critical biomarkers, such as hemoglobin and platelet levels, that influence transfusion decisions. This capability provides clinicians with a reliable decision-support tool, potentially improving patient outcomes and resource allocation in ICU settings.
Alireza Rafiei, Study Lead Author and Teaching Assistant, Emory University
The AI model was thoroughly tested across numerous scenarios to verify its resilience and dependability in real-world applications. The results showed consistent effectiveness across various ICU groups and medical conditions.
The team intends to incorporate this AI model into clinical workflows for real-time decision assistance, confirming its usefulness in real-world ICU scenarios.
Our ultimate goal is to personalize and optimize transfusion strategies, enhancing patient care and operational efficiency in hospitals.
Rishikesan Kamaleswaran, Associate Professor, Department of Biomedical Informatics, Emory University
This study represents a huge step forward in the use of AI in critical care medicine, demonstrating the potential of data-driven technology to change healthcare delivery.
Journal Reference:
Rafiei, A., et al. (2025) Robust Meta-Model for Predicting the Likelihood of Receiving Blood Transfusion in Non-traumatic Intensive Care Unit Patients. Health Data Science. doi.org/10.34133/hds.0197