AI-Driven Medical Device Risk Management: A New Paradigm Integrating Large Language Models and Prompt Engineering for Standard-Risk Knowledge Graph Construction and Application.

Purpose: To address the problems in medical electrical equipment risk management caused by the disconnection between unstructured medical electrical equipment standard documents and adverse event data, the lack of high-quality annotated data, and the reliance on manual combing for risk analysis. Methods: This paper proposes a novel method for constructing a risk knowledge graph that integrates large language models and prompting engineering standards. Using adverse event data from early childhood incubators as a case study, it integrates multi-source standards to construct a three-layer risk knowledge system. It designs multi-angle prompting strategies involving entity relationships and employs a dual strategy of entity disambiguation and aggregation to achieve knowledge integration and standardization. Results: The thought chain reasoning suggestion has the best performance (mean F1 score of 0.871). The constructed knowledge graph contains 24,106 nodes and 18,053 relationships, achieving a complete "fault-standard-measure" link. Based on this, a question-answering system for intelligent risk retrieval was developed. Conclusion: This provides a low-cost, reusable knowledge graph construction path for the resource-constrained medical device field, promoting the transformation of risk management towards AI empowerment and assisting in intelligent supervision of adverse events related to medical devices.