Abstract

Patients needing emergency department (ED) services are sorted into urgency categories using triage, often through severity indexes like the Emergency Severity Index (ESI), traditionally done manually by nurses. This manual triage process, while effective, can be time-consuming and prone to human error due to the subjective nature of the assessment. In this talk, it will be introduced a network-based patient modelling approach using Graph Neural Networks (GNNs) to automate triage by leveraging inter-patient similarities and inter-feature relationships. This approach aims to streamline the triage process, enhancing both accuracy and efficiency. The proposal presented in this session considers two models: one that views patients as nodes in a similarity graph (Patient-Level Modelling), and another that forms a graph for each patient with nodes as features that connect based on mutual information (Feature-Level Modelling). The preliminary findings from applying these models confirm the effectiveness of these methods. The automated triage system shows promise in accurately categorizing patients according to urgency levels, thereby potentially reducing the workload on medical staff and minimizing the chances of human error. Moreover, there are exciting future possibilities for enhancing transparency and clinical applicability through explainability techniques and the integration of the two models.

Google Meet joining info: https://meet.google.com/tmt-xpeg-xhm