Development of blood demand prediction model using artificial intelligence based on national public big data

Area of Science:

• Healthcare Analytics
• Biomedical Informatics
• Data Science in Medicine

Background:

• Healthcare systems face persistent challenges in maintaining a stable and sufficient blood supply.
• Blood shortages can critically impact patient care and treatment outcomes.
• Accurate forecasting of blood demand is essential for efficient inventory management.

Purpose of the Study:

• To develop an artificial intelligence (AI) model for predicting monthly blood transfusion requirements.
• To leverage national open big data related to blood transfusions for accurate demand forecasting.
• To enhance the stability and sufficiency of blood supply in medical institutions.

Main Methods:

• Utilized Korean national open big data on blood types and components (2010-2021).
• Developed predictive models including eXtreme Gradient Boosting (XGBoost), Light Gradient Boosting Machine (LGBM), and CatBoost.
• Created an ensemble model combining the three individual predictive models.

Main Results:

• Individual models (XGBoost, LGBM, CatBoost) showed varying prediction accuracies for different blood products and types.
• Ensemble models significantly improved prediction performance, reducing mean absolute and root mean squared errors.
• The requesting department was the most influential factor; other key features included RBC antibody screens, crossmatch data, donations, and surgeries.

Conclusions:

• The developed AI-driven blood demand prediction algorithm effectively forecasts transfusion needs.
• This tool enables medical facilities to proactively manage blood inventory and ensure adequate supply.
• Utilizing open big data enhances the efficiency of blood resource allocation in healthcare.