Standardizing Multi-site Clinical Note Titles to LOINC Document Ontology: A Transformer-based Approach.

Area of Science:

• Health Informatics
• Medical Informatics
• Clinical Data Standardization

Background:

• Clinical notes in electronic health records (EHRs) are diverse, hindering data retrieval and exchange.
• Standardization of clinical document types is essential for unified data management.
• The LOINC Document Ontology (DO) offers a framework for naming and describing clinical documents, but its real-world deployment needs exploration.

Purpose of the Study:

• To evaluate the utility of LOINC Document Ontology (DO) for standardizing diverse clinical note titles.
• To develop and assess an automated pipeline for mapping clinical note titles to LOINC DO codes across multiple institutions.
• To analyze the coverage and potential extensions of LOINC DO based on real-world clinical note titles.

Main Methods:

• Mapped clinical note titles from five institutions to the LOINC Document Ontology (DO).
• Classified mappings based on semantic similarity between note titles and LOINC DO codes.
• Developed an automated pipeline using large language models to map note titles without accessing clinical note content.

Main Results:

• The automated standardization pipeline achieved an accuracy of 0.90 in mapping clinical note titles to LOINC DO codes.
• Manual and automated mapping results were compared to analyze LOINC DO coverage.
• Performance variations between different large language models used in the pipeline were assessed.

Conclusions:

• The developed automated pipeline effectively standardizes clinical note titles using LOINC DO across multiple institutions.
• The study provides insights into the coverage of LOINC DO for diverse, multi-site clinical notes.
• Findings suggest areas for LOINC DO extension to improve its applicability in real-world clinical settings.