Linking Trials to Publications: Enhancing Recall by Identifying Trial Registry Mentions in Full-Text

Area of Science:

• Biomedical Informatics
• Clinical Trial Research
• Meta-science

Background:

• Linking clinical trials to their publications is crucial for evidence synthesis.
• Existing tools struggle to identify publications when registry numbers are not in metadata.
• Textual clues in titles or abstracts identify only a fraction of linked publications.

Purpose of the Study:

• To enhance a machine learning tool for linking clinical trials to publications.
• To improve the identification of publications reporting clinical outcome results.
• To increase the recall of linked publications by searching full-text content.

Main Methods:

• Developed a web-based tool using machine learning to predict publication links.
• Expanded the tool to search for registry numbers (NCT numbers) within full-text articles.
• Included data from PubMed Central, EuroPMC, and OpenAlex for full-text analysis.
• Retrieved biomedical citations from the ClinicalTrials.gov registry.

Main Results:

• The initial tool showed high precision but limited recall for publications without explicit metadata links.
• Searching full-text articles significantly increased the identification of linked publications.
• Retrieving citations from the registry itself further enhanced connectivity.

Conclusions:

• The expanded tool greatly improves the ability to identify publications reporting clinical trial results.
• This enhanced recall assists researchers in evidence synthesis and meta-science studies.
• The tool provides a valuable resource for understanding the dissemination of clinical trial findings.