Transfusion Indexes and Machine Learning Prediction of Red Cell Transfusion in Brain Tumor Surgery

Abstract

Effective perioperative blood management in brain tumor surgery has been a challenge, and there is a high tendency to use preoperative cross-matching. Machine learning (ML) models can be used to enhance the prediction of the need to transfuse. The objective of the study was to assess transfusion indexes based on surgery and externally test a web application based on ML to predict packed red cell (PRC) transfusion.

A retrospective cohort of 850 patients undergoing brain tumor surgery (2023–2025) was analyzed. Key results were transfusion indices, crossmatch-to-transfusion (C/T) ratio, transfusion probability (Tp), and transfusion index (Ti). Secondary outcomes included predictive performance of an ML model using the coefficient of determination (R 2), root mean square error (RMSE), and mean absolute error (MAE), calibration analysis, Bland–Altman agreement, and decision curve analysis (DCA).

The overall transfusion rate was 39.3%, with a C/T ratio of 4.15, Tp of 39%, and Ti of 0.90, indicating inefficient blood utilization. The ML model demonstrated good predictive performance (R 2 = 0.70, 95% CI 0.63–0.75), with RMSE 0.99 and MAE 0.62 units. The slope of calibration was 0.80, indicating slight overestimation in high-risk patients. There was greater variation in greater ranges of transfusions. The DCA showed that the model had a net clinical benefit that was beneficial overall and exceeded the threshold probabilities.

Blood utilization in brain tumor surgery remains suboptimal. The ML-based model showed acceptable predictive performance but reduced reliability in high-transfusion scenarios. Integration into clinical workflows should consider safety thresholds and risk-adjusted decision-making.