Review article: Machine Learning Algorithms Improve Blood Utilization in Surgical Transfusion Management

Claudia Seskin, Patricia Tille

Abstract

Red blood cell transfusions are essential in perioperative care but are frequently overutilized, increasing costs and exposing patients to unnecessary harm. Traditional transfusion risk scores lack the precision needed for personalized care, often not accounting for the complexity of patient-specific variables. Machine learning (ML) has emerged as a promising tool to improve the accuracy of transfusion risk prediction by analyzing large, complex datasets and identifying non-linear relationships among clinical factors. A comprehensive review of published ML-based transfusion prediction models was conducted, focusing on surgical applications in cardiac, orthopedic, and general procedures. Studies were analyzed based on algorithm type, performance metrics, input variables, and model transparency. Implementation challenges, including data quality, clinical acceptance, and infrastructure limitations, were also examined. ML enables more accurate, individualized prediction of perioperative transfusion needs. ML models outperformed traditional methods in predictive accuracy, particularly those built using large data sets and ensemble techniques such as gradient boosting. Simpler models like logistic regression performed well with smaller datasets. Barriers to implementation included fragmented electronic health records, variability in data standardization, and limited external validation. The “black-box” nature of some ML algorithms poses additional implementation challenges for providers including trust and adoption. For successful clinical integration, models must be transparent, validated across diverse populations, and supported by standardized, high-quality data. ML-based transfusion prediction models improve blood utilization and enhance surgical outcomes.

Keywords: Perioperative blood transfusion, machine learning, risk prediction, blood utilization

Int. J. Bio. Lab. Sci 2025(14)2:80-94 【PDF】