Utilizing Large Language Models for Zero-Shot Medical Ontology Extension from Clinical Notes.

Area of Science:

• Biomedical Informatics
• Medical Ontology Engineering
• Natural Language Processing

Background:

• Ontologies are crucial for organizing biomedical knowledge.
• Clinical notes contain rich, unstructured data valuable for ontology extension.
• Current methods for leveraging clinical notes in ontology extension are limited.

Purpose of the Study:

• To develop a novel framework (CLOZE) for automated medical ontology extension using clinical notes.
• To leverage large language models (LLMs) for extracting medical entities and relationships.
• To ensure a scalable, accurate, and privacy-preserving solution.

Main Methods:

• Utilized pre-trained large language models (LLMs) for concept extraction from clinical notes.
• Developed a zero-shot framework requiring no additional training or labeled data.
• Implemented automated removal of protected health information (PHI) to ensure patient privacy.

Main Results:

• CLOZE accurately identifies disease-related concepts and hierarchical relationships.
• The framework demonstrates scalability for large datasets.
• Automated PHI removal ensures privacy-preserving ontology extension.

Conclusions:

• CLOZE offers an effective, cost-efficient method for extending medical ontologies from clinical notes.
• The framework has significant potential for biomedical research and clinical informatics applications.
• This approach enhances the utility of unstructured clinical data for knowledge representation.