Chia, a large annotated corpus of clinical trial eligibility criteria

Area of Science:

• Natural Language Processing
• Clinical Informatics
• Biomedical Data Science

Background:

• Clinical trial eligibility criteria are complex and often unstructured.
• Efficiently extracting this information is crucial for trial recruitment and analysis.
• Existing methods for processing eligibility criteria are limited.

Purpose of the Study:

• To present Chia, a novel, large annotated corpus of patient eligibility criteria.
• To provide a benchmark dataset for developing and testing information extraction methods.
• To facilitate the transformation of eligibility criteria into computable formats.

Main Methods:

• Extracted eligibility criteria from 1,000 interventional, Phase IV clinical trials registered in ClinicalTrials.gov.
• Annotated 12,409 eligibility criteria, identifying 41,487 entities and 25,017 relationships.
• Represented each criterion as a directed acyclic graph for potential Boolean logic transformation.

Main Results:

• Developed Chia, a comprehensive corpus of annotated clinical trial eligibility criteria.
• The corpus contains 15 entity types and 12 relationship types.
• Eligibility criteria are structured as directed acyclic graphs.

Conclusions:

• Chia serves as a valuable resource for advancing information extraction from clinical trial data.
• Enables the development of machine learning, rule-based, and hybrid approaches.
• Facilitates the creation of database queries from free-text eligibility criteria.