Biomedical Text Normalization through Generative Modeling.

Area of Science:

• Biomedical Informatics
• Natural Language Processing
• Artificial Intelligence

Background:

• Electronic health records (EHRs) contain vast amounts of unstructured text, posing challenges for data analysis and utilization.
• The inconsistent formatting of medical text hinders predictive modeling, clinical decision support, and data mining.
• Large language models (LLMs) show promise in understanding and standardizing complex medical language.

Purpose of the Study:

• To develop and evaluate clinical text normalization pipelines using LLMs.
• To assess the effectiveness of different LLM-based normalization strategies against a baseline method.
• To improve the standardization of unstructured medical text for enhanced data usability.

Main Methods:

• Implemented four LLM-based normalization strategies: Zero-Shot Recall, Prompt Recall, Semantic Search, and Retrieval-Augmented Generation (RAGnorm).
• Included a baseline approach using TF-IDF based String Matching.
• Evaluated performance on three SNOMED-mapped condition term datasets (oncology, institutional sample, common codes) and TAC 2017 drug annotations (MedDRA).
• Measured performance using mean shortest path length and micro F1 score.

Main Results:

• RAGnorm demonstrated superior performance across all evaluated datasets.
• RAGnorm achieved the lowest mean shortest path lengths: 0.21 (domain-specific), 0.58 (sampled), and 0.90 (top terms).
• RAGnorm attained a micro F1 score of 88.01 on TAC 2017 task 4, outperforming other models without prior training data.

Conclusions:

• Retrieval-focused LLM approaches effectively address limitations in clinical text normalization.
• RAGnorm and similar retrieval techniques show significant potential for normalizing biomedical free text.
• Further exploration of these methods is recommended for advancing biomedical natural language processing.