Comparative Analysis of NLP Models for Automatic LOINC Document Ontology Named Entity Recognition in Clinical Note Titles.

In order to utilize clinical notes for research studies, it is necessary to identify the most relevant notes. Mapping to the LOINC Document Ontology makes this process easier by reducing the variability of note types. We experimented with three models to automatically identify LOINC DO entities in VA note titles. The supervised BERT model performed best, but the open-source large language models (LLMs) performed well despite a lack of fine-tuning. Future work will aim to improve note classification by including additional note metadata and contents, hybridizing with rule-based approaches, testing fine-tuned LLMs, and mapping to exact LOINC codes.