Summary

Optimization of the blood supply chain is vital for ensuring the availability of blood products and patient safety. This dissertation introduces data-driven approaches to address challenges such as the perishability of blood, stochastic supply and demand, and donor responsibilities. It focuses on improving logistics between blood supply organizations and hospitals through matching of blood products, optimizing facility locations, predicting demand, and optimizing inventory levels.

Alloimmunization, caused by antigen mismatches during transfusions, is a significant complication. The first part of this dissertation proposes the MINRAR model, a Mixed-Integer Linear Programming approach, that leverages genotyping advances to minimize alloimmunization by up to 93.7%. Incorporating patient-specific priorities further refines the model, reducing mismatches by 30% and shortages by 92%. For patients requiring frequent transfusions, such as those with sickle cell disease, Deep Reinforcement Learning and Neural Networks are applied to optimize long-term immune responses.

The second part explores the use of Deep Reinforcement Learning for platelet inventory management, minimizing waste while ensuring compatibility. A study performed both in the Netherlands and Finland identifies optimal locations for distribution centers, reducing costs without compromising the reliability of delivery. A novel demand forecasting system adapts to stochastic demand, enhancing supply chain efficiency.

This dissertation highlights the potential of data-driven methods to improve safety, efficiency, and cost-effectiveness in blood supply chains, paving the way for better patient outcomes and more resilient healthcare operations.