A Review of Algorithms and Models for Patient Triage in Outpatient Departments

Abstract

This paper is a comprehensive overview of algorithms and models used for patient triage in hospital outpatient departments. It starts by analyzing the systemic problems with conventional manual triage, such as low efficiency, great subjectivity and resource burden, underlining the need for intelligent systems. The nucleus of the work systematically identifies a variety of algorithmic paradigms starting from interpretable rule-based expert systems and traditional machine learning models (viz., Decision Trees, Logistic Regression, Support Vector Machines, Naïve Bayes) to state-of-the-art deep learning models like Recurrent Neural Networks, Long Short-Term Memory networks, Transformers, and Large Language Models. We also extend the frontier of multimodal fusion models that pool data from multiple modalities for comprehensive patient assessment. This survey combines performance results from many benchmark publications, providing not only a comparison between models, but also in comparison to the performance of human clinical experts, and summarize the most critical metric of evaluation. Finally, it also evaluates the practical and ethical challenges of real-world deployment of intelligent triage systems, including interoperability, data privacy/security, algorithmic bias, and the need for eXplainable Artificial Intelligence (XAI). The paper concludes that, although AI provides powerful means to improve triage efficiency and accuracy, its success in the future will depend on the creation of collaborative, transparent and ethically robust systems that complement, rather than replace, human clinical judgment.