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

Disconnected standards and adverse-event data: Medical electrical equipment risk management is hindered by a disconnection between unstructured standard documents and adverse event data.

Lack of annotated data: Medical electrical equipment risk management lacks high-quality annotated data.

Manual risk analysis burden: Risk analysis for medical electrical equipment relies on manual combing.

Disconnected standards and adverse-event data: Medical electrical equipment risk management is hindered by a disconnection between unstructured standard documents and adverse event data.

Lack of annotated data: Medical electrical equipment risk management lacks high-quality annotated data.

LLM prompt-engineered risk knowledge graph: The paper proposes constructing a risk knowledge graph by integrating large language models with prompt engineering for standards-based medical device risk management.

Three-layer risk knowledge system: The approach integrates multi-source standards to build a three-layer risk knowledge system using early childhood incubator adverse event data as a case study.

Multi-angle prompting for entity relations: The method designs multi-angle prompting strategies for extracting or modeling entity relationships.

Entity disambiguation and aggregation: The method uses entity disambiguation and aggregation to integrate and standardize knowledge.

Risk retrieval question-answering system: A question-answering system was developed for intelligent risk retrieval based on the constructed knowledge graph.

Best prompt F1 score:

Large constructed risk knowledge graph:

Complete fault-standard-measure link: