Abstract:

Objective To develop a management decision-support system integrating a knowledge graph with a large language model for dysphagia in older adults，and to evaluate its usability. Methods A disease-specific knowledge graph was built in Protégé through a combined top-down and bottom-up approach. With a text corpus and the knowledge graph serving as dual knowledge sources，two parallel channels：vector-based semantic retrieval and subgraph retrieval，respectively retrieved supporting evidence from the literature and performed directed causal reasoning. DeepSeek-V3 was adopted as the generative foundation model，supplemented by a cross-model consistency check. Clinical nurses were positioned as the primary users，with family caregivers and patients as extension users in home settings；role-tailored prompt templates and an emotional safety control mechanism were predefined for these user groups. Role-specific prompt templates were predefined for nurses，family caregivers，and patients. The system was evaluated at three levels：model performance，decision output，and user experience. Results The system achieved a response accuracy of 88% in standardized tests for geriatric dysphagia，significantly higher than that of the general-purpose models（64%-67%，P<0.001） and the single-channel model（74%，P=0.012）. In the evaluation of the system-assisted decision-making scheme by 13 experts，the evidence trace ability dimension achieved the highest score of（4.42±0.17）. The system usability scores were 80.25±7.12 for 10 nurses and 74.50±8.80 for 10 caregivers. The analysis of interview data from 3 nurses and 3 caregivers revealed that the system strengthened the credibility and efficiency of decisions and improved caregivers’ practical skills，although the translation of technical terms and the depth of responses for severe cases warranted further refinement. Conclusion The system delivers high response accuracy and enables evidence traceability，showing favorable usability for assisted decision-making in the management of geriatric dysphagia in both clinical and home settings.

Key words: Dysphagia, Aged, Decision Support Systems, Clinical, Artificial Intelligence, Knowledge Graph