Artificial Intelligence for EMS Triage: A Data-Driven Approach to Emergency Patient Prioritization in Kalasin Province, Thailand

Abstract

Emergency Medical Services (EMS) require accurate and timely triage to ensure efficient patient prioritization and optimal resource allocation. Traditional triage methods, like ESI, NEWS, and MEWS, often face issues of subjectivity, variability, and limited predictive accuracy, potentially delaying critical care. This study proposes an Artificial Intelligence (AI)-driven triage system designed for EMS in Kalasin Province, Thailand, leveraging deep learning for risk assessment and patient evaluation prioritization. A dataset of 1,683 EMS cases was utilized, incorporating patient demographics, vital signs, chronic conditions, mobility, self-care ability, pain/discomfort, anxiety/depression, and a self-reported health scale (from 0 to 100). A Deep Neural Network (DNN) was trained using the Adam optimizer and categorical cross-entropy loss, with hyperparameter tuning applied via grid search and Bayesian optimization. Model performance was evaluated using AUC-ROC, sensitivity, specificity, F1-score, and calibration analysis. The results show that the AI model achieved an AUC-ROC of 0.91, with 88.5% sensitivity and 87.3% specificity, outperforming conventional triage tools. This AI-powered system enables real-time risk assessment and provides hospital selection recommendations, enhancing EMS decision-making. Despite its effectiveness, continuous updates are required to mitigate model drift, and further validation is necessary for broader EMS applications. Future research will focus on expanding datasets, integrating real-time patient monitoring, and enhancing model adaptability. This study highlights the transformative potential of AI in EMS triage, paving the way for faster, more accurate, and data-driven emergency healthcare systems.