Artificial Intelligence In Blood Banking: Current Advances, Clinical Applications, Challenges, and Future Directions

Abstract

Background-Aim: Blood banking and transfusion medicine remain a costly element of healthcare, with major problems being irregular donor availability, component wastage, and supply-demand mismatches. Artificial intelligence (AI) and machine learning (ML) have been suggested to minimize these gaps, but few studies have compared model performance to real-world blood banking data. This research analyzed blood-component utilization patterns in a hospital blood bank and the accuracy of ML models in forecasting daily blood-component demand.

Materials and Methods: A retrospective observational study reviewed routinely maintained hospital blood bank records, integrating an AI-based predictive modelling component. Data encompassed utilization of RBCs, platelets, and FFP; units requested, issued, transfused, returned, expired, and discarded; hospital and ICU admissions; surgeries and trauma cases; and opening and closing blood inventory. Five ML algorithms (Linear Regression, Random Forest, Gradient Boosting, XGBoost, LightGBM) were trained and tested on MAE, RMSE, and R² using a held-out test set.

Results: Mean total daily blood-component utilization was 111.0 ± 27.4 units, highest for RBC and platelets, lowest for FFP. RBC utilization was highest for the O⁺ and A⁺ groups. XGBoost's performance (MAE 7.84, RMSE 10.31, R² 0.90) exceeded that of the other four models, with R² reaching 0.92 for RBC.

Conclusion: For the first time, routine information from blood banks is used to predict blood-component usage per day with high accuracy based on AI, specifically ML algorithms like XGBoost, allowing the use of the tool in the clinical context for demand forecasting and blood banking inventory management.