Agent-Based dynamic optimization for managing the workflow of the patient's pathway

Abstract The emergency department (ED) is an important health care service representing the bottleneck of the hospital. The ED often faces overcrowding problems in many countries around the world. One of the causes of the ED overcrowding is the permanent interference between three types of arriving patients: already programmed patients, non-programmed patients and urgent non-programmed patients. This interference, which is subjected to stochastic behaviors of consultation time and to patients’ random arrival flows, makes the ED more complex to manage, which prevents any prior planning. Overcrowding situations are not easy to control and cause problems to patients and medical staff, implying the deterioration of several performance indicators such us waiting times, quality of care, treatment time, length of stay, etc. This paper aims to study and develop an orchestration architecture that drives the patient's pathway workflow in real time, thus improving the Performance Indicators step by step during the execution. This architecture uses a multi-agent system in order to ensure the coordination between the medical staff so as to provide the best patient care. Thanks to the agents’ behavior and communication protocols, the proposed system establishes a direct real-time link between the required performances and the effective actions in order to reduce the overcrowding impact. In this study, we discuss the relevance of each metric in order to select those that best match with a given study context (environment, type of patients, objective, etc.). We focus then on three performance indicators: the Remaining Patient Care Load (RPCL) which gives an estimation of what remains in care within the complete patient care process, the Cumulative Waiting Time (CWT) and the Length of Stay (LOS). Experimental results indicate that improvements of the performance indicators are achievable thanks to the agents’ driven patient pathway workflows during their execution through the dynamic re-orchestration.