A human-centered hybrid AI framework for optimizing emergency triage in resource-constrained settings

The paper proposes a human-centred hybrid artificial intelligence system, capable of triaging emergencies in healthcare facilities that are resource-constrained. The system combines cognitive modelling, lightweight machine learning and resource-aware optimisation to provide interpretable and actionable triage recommendations. The framework provides adaptive decision support by integrating patient vital signs, accessible clinical resources, and clinician workload, allowing clinicians to override these inputs as needed. The architecture is a hybrid of a cognitive module that probabilistically learns clinical heuristics and cognitive biases, a convolutional neural network-based predictive model with explainability provided by SHAP, and a resource-conscious optimiser, based on either integer programming or heuristic reinforcement learning. Two-way communication is supported by a human-in-the-loop interface, allowing the clinician to provide feedback and override suggestions, with the system recording the reasons behind ongoing learning and adaptation. The 90% accuracy and over 0.86 receiver operating characteristic area under the curve were confirmed through simulated and pilot implementations. Compared to baseline methods, the hybrid system decreased the false negatives by 12 to 5%, thus significantly enhancing the detection of high-acuity cases. The AI-assisted decision-making lowered the time spent by clinicians in decision-making by 31% and resource optimisation strategies enhanced the critical resource use by 20%, lowering the override rates significantly. Significant results include better triage accuracy, faster and more confident clinician decisions, and more effective resource allocation. The flexibility of the framework to local constraints offers a viable avenue through which AI implementation in resource-constrained emergency care facilities can be done safely and effectively. The presented collaborative model suggests that a combination of human skills and AI capabilities can be used to improve the performance of emergency triage and support progress in the field of difficult healthcare settings without ethical and clinical supervision. • Hybrid AI triage system achieves 90% accuracy in resource-limited emergency care • Clinician decision time reduced by 31% with human-in-the-loop AI support • False negative rates decreased from 12% to 5% improving patient safety • Resource optimization increased critical care utilization by up to 20% • CNN-based risk prediction with SHAP explanations enhances interpretability