Ensuring dynamic organisational performance of an agent-based organisation – Application to an emergency department

An Emergency Department (ED) is an essential department in the hospital and the healthcare system, which is known for its complexities and its operations in a dynamic environment. In order to ensure good-quality care for patients and manage patient flow well, the ED needs to make its decisions of resource scheduling effectively and regularly. However, making effective resource scheduling decisions in such a complex and dynamic environment is challenging. This thesis addresses this challenge by studying the impact of different scheduling strategies on the performance and dynamic behaviour of the ED. In order to perform this study, we explore and model different strategies of scheduling and evaluate their performance.

First, we explore different decision-making structures. We propose a hybrid structure as an agent-based simulation in which each agent runs a Mixed-Integer Linear Programming (MILP) solver. We compare it with the centralised structure that runs a single MILP solver to schedule all resources, and the decentralised structure relying on a triage-based first-come, first-served (FCFS) rule. The experiments show that the hybrid decision structure produces a good schedule in a short time.

Next, we study the hybrid decision-making structure in greater detail by modelling two different decision-making approaches: (1) an Agent-Based model with an Auction mechanism and Local Sequencing algorithms (AB-ALS) and (2) four variants of the Agent-Based model with Priority Queue policies (AB-PQ). The comparative study of these two models demonstrates the effectiveness of the AB-PQ models. Consequently, we focus on AB-PQ in order to consider ED settings closer to reality, including the decision to open a Fast Track. We compare the performance of the four PQ policies during normal and epidemic periods. The results show that no single PQ policy works the best under all circumstances.

Finally, to better account for dynamics in the evaluation of the four AB-PQ variants, we analyse and characterise the dynamic behaviour of the ED. We use a data-driven identification method to determine the differential equations of the ED and perform a stability analysis for each AB-PQ strategy. We are thus able to identify the stable points common to all strategies, as well as the different points leading to instability for each of them.

Keywords: Dynamic resource scheduling, emergency department, agent-based Simulation, decision structure, performance evaluation, organisational stability