From prediction to practice: Barriers to implementing artificial intelligence in blood inventory management and transfusion support

Abstract

BACKGROUND: Artificial intelligence (AI) is increasingly applied to blood-demand forecasting, donor management, inventory optimisation, wastage reduction and transfusion-related decision support; however, routine implementation remains limited. OBJECTIVES: To review current applications of AI in blood inventory management and transfusion support and identify the principal barriers to safe and sustainable implementation. METHODS: A focused narrative review informed by structured searches of PubMed, Scopus, EBSCO and Google Scholar, supplemented by targeted searches of transfusion-specific and healthcare AI implementation literature. Evidence was synthesised thematically. RESULTS: Published studies demonstrate technical feasibility across forecasting, donor-return prediction, inventory management and clinical decision support. Major implementation barriers include limited external and prospective validation, incomplete reporting, data-access and privacy constraints, poor interoperability, limited explainability, workforce training gaps and unclear governance. CONCLUSIONS: The principal challenge is no longer developing predictive models but integrating them safely and sustainably into routine transfusion services. Future progress requires multicentre evaluation, privacy-preserving data collaboration, workflow-integrated design, role-specific training and explicit governance with continued human oversight.