Abstract
Background: Accurate identification of patients at high risk of perioperative blood transfusion is essential for optimizing patient blood management (PBM) strategies in oncological surgery. However, the performance of standard PBM eligibility criteria in real-world oncological settings remains incompletely characterized. Material and Methods: We conducted a retrospective, single-center analysis of 4228 consecutive patients undergoing elective oncological surgery of any complexity or liver transplantation over a 9-month period to assess transfusion need and estimate access to preoperative patient blood management (PBM) strategies to improve anemia management. Transfusion events were assessed within 24 h after surgery (PS24) and during the perioperative period (PO; 48 h before to 72 h after surgery). Two PBM eligibility strategies were applied to the same patient cohort and compared: (A) an observational approach, based on predefined PBM indicators (transfusion rate and transfusion index by surgical complexity), and (B) a multivariable modeling approach based on pre- and intraoperative anesthesiology assessment to estimate individual transfusion risk. Predictive performance of both strategies was evaluated using accuracy, Cramér's V, area under the receiver-operating characteristic curve (AUC-ROC), and Brier score. Results: Overall, 7.7% of patients received transfusion within PS24 and 9.2% during PO. According to the observational approach, 23.8% of patients were classified as PBM-eligible, accounting for 89.2% of PS24 transfusions and 87.1% of PO transfusions. In the multivariable modeling approach, independent predictors of transfusion included surgical type (e.g., sarcoma surgery: OR 22.8 for PS24; OR 6.3 for PO; vs. senology surgery OR 1 for PS24; OR 1 for PO, respectively), anemia severity (moderate anemia: OR 64.3 and OR 107.9, respectively and mild anemia OR 3.38 and OR 3.65, respectively), high surgical complexity, operative time >3 h (>3 h: OR 8.83 and OR 8.65, respectively vs. <3 h OR 1 and OR 1, respectively), and ICU admission risk. The observational approach demonstrated stronger alignment with actual transfusion events (Cramér's V = 0.44-0.47) and higher overall accuracy (90.8-92.3%); in contrast, a multivariable modeling approach showed superior discrimination (AUC = 0.94-0.95) and lower Brier scores, indicating better individual risk prediction. Conclusions: In a large real-world cohort of oncological surgical patients, standard PBM eligibility criteria effectively identified the majority of patients requiring perioperative transfusion. While multivariable modeling provided greater predictive precision, the observational PBM approach demonstrated strong clinical alignment and practical applicability. Integrating both strategies may support more effective transfusion risk stratification and PBM planning in oncological surgery.