Artificial Intelligence in Patient Blood Management: A Systematic Review of Predictive, Diagnostic, and Decision Support Applications.

Background: Patient blood management (PBM) is a patient-centered, evidence-based approach for optimizing anemia management, minimizing blood loss, and ensuring appropriate transfusion. Artificial intelligence (AI) provides powerful tools for prediction, diagnosis, and decision support across PBM, but current evidence remains emerging and not yet consolidated. Objectives: This review synthesizes AI applications in PBM, summarizing predictive, diagnostic, and decision support models; highlighting methodological trends; and discussing challenges for clinical translation. Methods: PubMed, Scopus, and Web of Science were searched from inception to 31 March 2025. Eligible studies reported AI models addressing the three established PBM pillars. Studies on transfusion safety and blood bank operations relevant to PBM were also included. Extracted data covered study characteristics, predictors, models, validation strategies, and performance. The findings were narratively synthesized given study heterogeneity. Results: A total of 338 studies were included, spanning anemia detection, bleeding risk stratification, transfusion prediction, transfusion safety, and inventory management. Deep learning (DL) predominated in image-based anemia detection, while ensemble and gradient boosting methods frequently outperformed baselines in bleeding and transfusion risk prediction. Recurrent and hybrid architectures proved effective for blood supply forecasting. Across domains, machine learning and DL models generally surpassed logistic regression, clinical scores, and expert judgment. Despite strong internal performance, external validation and clinical deployment remain limited. Conclusions: AI is advancing PBM by enabling earlier anemia detection, more accurate bleeding and transfusion prediction, and smarter resource allocation. Translation into practice requires standardized reporting, robust external validation, explainability, and workflow integration. Future work should emphasize multimodal learning, prospective evaluation, and cost-effectiveness.