Clinical Trial Eligibility Criteria Decomposition and Parsing with Large Language Models.

Area of Science:

• Biomedical Informatics
• Natural Language Processing
• Clinical Trial Data Management

Background:

• Clinical trial eligibility criteria are often unstructured free text, hindering automated analysis.
• Efficient processing of these criteria is crucial for clinical trial recruitment and research.

Purpose of the Study:

• To develop and evaluate a Decomposition and Parsing (DP) workflow for systematically structuring clinical trial eligibility criteria.
• To leverage advanced large language models (LLMs) for automated extraction of study traits and their components.

Main Methods:

• A Decomposition and Parsing (DP) workflow was designed to break down criteria into 'study traits'.
• LLMs (GPT-4o, Llama3.3) with Chain-of-Thought prompting were employed for processing Alzheimer's disease trial data.
• Novel evaluation metrics were developed to assess extraction quality.

Main Results:

• The DP workflow demonstrated strong performance in logical relationship extraction and trait computability.
• Successful processing of Alzheimer's disease trial datasets was achieved.
• Proposed evaluation metrics showed superiority over traditional methods.

Conclusions:

• The developed framework offers a scalable and intuitive approach to representing clinical trial eligibility criteria.
• This advancement supports improved biomedical informatics applications.
• Further research is needed for domain-specific fine-tuning and broader dataset integration.