Unstructured Electronic Health Records of Dysphagic Patients Analyzed by Large Language Models.

Area of Science:

• Medical Informatics
• Natural Language Processing
• Clinical Research

Background:

• Dysphagia is a complex disorder with challenging diagnoses and treatments.
• Electronic health records (EHRs) for dysphagia are often unstructured, hindering systematic analysis.

Purpose of the Study:

• To apply natural language processing (NLP) and large language models (LLMs) for analyzing unstructured clinical narratives.
• To extract diagnostic information from diverse EHRs and cluster patients based on swallowing dysfunctions.

Main Methods:

• Utilized NLP techniques and LLMs to process unstructured diagnostic information from 486 patients' EHRs.
• Employed clustering algorithms on extracted features to identify patient groups with similar pathophysiological swallowing dysfunctions.

Main Results:

• Basic NLP methods yielded limited insights due to data variability.
• LLMs effectively bridged the gap in understanding nuanced dysphagia information.
• Closed-source LLMs successfully clustered different categories of dysphagia.

Conclusions:

• LLMs show significant promise for future dysphagia research and clinical applications.
• This study demonstrates LLMs' capability in preprocessing unstructured EHRs for improved diagnosis and patient clustering.