Back to search

McMaster University

Platelet Ordering Strategies Using Reinforcement Learning

Abstract

dc:description.abstract

Blood is the ultimate “just-in-time” commodity. In the Canadian healthcare system, donated blood is separated in various components, such as red blood cells (RBCs) and plasma, which can be frozen for later use, but other components are characterized by extreme perishability, with platelets lasting a mere 5-7 days. This creates the unfortunate perishability paradox where maintaining a high “safety stock” is essential, which inevitably leads to high wastage, while aggressive waste reduction increases the risk of critical shortages during surges in demand. The vast and varied Canadian geography means that the logistics of maintaining a blood supply chain often leads to delayed care for countless communities. Traditional inventory models often fail to account for factors such as weather-related transport disruptions or sudden shifts in donor availability, which have reached a ten-year low in the post-pandemic landscape. This thesis explores a transition from static, rule-based ordering to a system using reinforcement learning (RL). RL treats blood replenishment as a sequential decision-making problem under uncertainty, integrating current viable inventory levels with demand forecasts. This thesis evaluates if RL can provide a more resilient logistics framework compared to the traditional approach and how this RL approach is structured, with a focus on platelet inventory. The performance of this model is presented with respect to average operational costs, as well as metrics such as the frequency of shortages or unit expiration. The system has demonstrated that this overall architecture is effective under the conditions tested, with its behaviour and limitations being explored as the conditions were altered. The results showed that this approach relies both on the characteristics of demand data, as well as the values of system parameters and training hyperparameters. This thesis should serve as a resource for researchers interested in implementing RL for perishable inventory management. The experiments performed were curated to highlight the use cases where RL shines and where future work would elevate its performance and viability.

Degree

thesis:*
Department dc:contributor.department
Computing and Software
Year dc:date.issued
2026

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Fabian, Anthony
Advisor dc:contributor.advisor
  • Down, Douglas

Rights

Language dc:language.iso
en

Identifiers

dc:identifier.*

Chain of custody

source
Harvested from
McMaster University
Base URL
macsphere.mcmaster.ca/server/oai/request
Last updated
2026-11-18
Source record
OAI-PMH GetRecord
related terms
citation

Fabian, Anthony. Platelet Ordering Strategies Using Reinforcement Learning. 2026. https://hdl.handle.net/11375/33156