Platelet inventory management in hospital networks: A reinforcement learning approach

This study proposes a reinforcement learning (RL)-based framework incorporating the Proximal Policy Optimization (PPO) algorithm to improve platelet inventory management. The proposed approach considers an inventory system with varying ordering intervals, incorporating ABO-Rh substitution decisions and hospital collaborations through transshipment. In this framework, transshipment is modeled as a fixed policy, reflecting real-world practices where blood units nearing expiration are proactively transferred from smaller local hospitals to larger hospitals, where they are more likely to be used in time. We extend our analysis by exploring several RL models, including Trust Region Policy Optimization (TRPO) and Soft Actor-Critic (SAC). The results show that PPO-Complete outperforms the other RL models, and all considered RL approaches outperform the base-stock strategy, which is commonly used in hospital platelet inventory management. The analyses indicate that lower transshipment costs, when coupled with effective substitution decisions, lead to a reduction in total cost and enable larger order sizes, thereby mitigating shortages.