A Decision Support System for Rapid Deployment of Emergency Healthcare Facilities for Infectious Disease Using Augmented Large Language Models

The emergency healthcare facilities for infectious disease deployment (EHFIDD) is a crucial measure to alleviate the pressure placed on healthcare systems by public health emergencies caused by infectious outbreaks. However, the rapid EHFIDD involves short timelines, stringent quality requirements, and numerous decision points. Therefore, it is urgent to develop a decision support system (DSS) capable of guiding a scientific and efficient deployment process from site selection to final delivery within a limited timeframe. Previous studies have primarily developed rule-based and pretrained DSS for the rapid EHFIDD, which heavily relied on predefined knowledge and manual configuration, resulting in limited efficiency and scalability. To address this limitation, this study used an augmented large language model (LLM) to develop and deploy a DSS prototype for rapid EHFIDD, simultaneously achieving high development efficiency and reliable decision-making performance. Specifically, this study established an LLM-enabled knowledge graph (KG) development method tailored for EHFIDD-related standards; developed a retrieval-augmented generation (RAG)-based approach to augment six recent LLMs using EHFIDD-KG; and proposed a comprehensive evaluation system for DSS performance, including accuracy, traceability, and relevance. Results show that (1) medium-scale LLMs achieve more augmentation from EHFIDD-KG than ultralarge LLMs; (2) the augmentation is more pronounced in answering open-ended questions than closed-ended ones; (3) the augmented LLMs achieve average improvements of 0.20, 0.40, and 0.21 in accuracy, traceability, and relevance over the original model; and (4) KG containing image information outperforms text-only KG with average gains of 0.08, 0.05, and 0.12 in accuracy, traceability, and relevance. Finally, the DSS’ robustness and generalizability are discussed, alongside its theoretical contributions, managerial implications, and human-centric governance for deployment safety. This study provides a systematic and reliable methodology for DSS development in rapid EHFIDD contexts and offers a novel technical pathway for enabling explainable LLM applications in emergency engineering domains.