Optimal ordering policy for platelets: Data-driven method vs model-driven method

Platelets, one of the most significant materials in treating leukemia, have a limited shelf life of approximately five days. Because platelets cannot be manufactured and can only be centrifuged from whole or donated blood directly, an accurate ordering policy is necessary for the efficient use of this limited blood resource. Given this motivation, the present study examines an ordering policy for platelets to minimize the expected shortage and overage. Rather than using the two-step model-driven method that first fits a demand distribution and then optimizes the order quantity, we solve the issue using an integrated data-driven method. Specifically, the data-driven method works directly with demand data and does not rely on the assumption of demand distribution. Consequently, we derive theoretical insights into the optimal solutions. Through a comparative analysis, we find that the data-driven method has a mean anchoring effect, and the amounts of shortage and overage reduced by this method are greater than those reduced by the model-driven method. Finally, we present an extended model with the service level requirement and conclude that the order decided by the data-driven method can precisely satisfy the service level requirement; however, the order decided by the model-driven method may be either higher or lower than the service level requirement and can lead to a higher cost.