Sep 2026· Studies in Health Technology and Informatics· Vol 340, pp.
163-171
· 0 citations
Medicine
TL;DR
The findings indicate that LLM-based extraction is a promising approach for structuring complex eligibility criteria, particularly when combined with strategies to improve precision, in clinical applications.
Abstract
INTRODUCTION
Accurate interpretation of clinical trial eligibility criteria is essential for applications such as patient-trial matching and clinical decision support, particularly in precision oncology. However, relevant information, including genetic mutation requirements, is typically embedded in unstructured text within trial registries such as ClinicalTrials.gov, limiting accessibility for automated processing.
Methods
This paper reports on development, integration and evaluation of a system for extracting and structuring mutational eligibility criteria from trial records, focusing on identifying mutated genes, distinguishing inclusion and exclusion criteria, and assigning them to individual study arms. To address challenges like ambiguous abbreviations and context-dependent meaning, we combine large language models (LLMs) for context-aware extraction with rule-based validation against HUGO Gene Nomenclature. The system was implemented using local LLMs and integrated into the Community Annotated Trial Search (CATS) platform.
Results
Applied to 4,918 clinical trials, the system generated structured representations of genetic eligibility criteria for 1,010 studies. Expert review of 42 trials showed that 88.1% of studies were correctly annotated, with a precision of 80% at the level of individual eligibility criteria. Failures were mainly due to hallucinated genes and misinterpreted abbreviations, highlighting challenges in biomedical text processing.
Discussion
The findings indicate that LLM-based extraction is a promising approach for structuring complex eligibility criteria, particularly when combined with strategies to improve precision. The integration into an operational system demonstrates practical feasibility, while the observed limitations emphasize the need for careful dataset design, error mitigation strategies, and continued refinement to achieve reliable automation in clinical applications.
Background The rapid growth of biomedical publishing in radiation oncology has created a need for automated methods to extract structured information from unstructured text. Large language models (LLMs) show promise in biomedical natural language processing, but their use for clinical trial data curation remains to be...
F. Dennstaedt, Lena Kretzschmar, N. Cihoric et al.· Oncology· 0 citations
Abstract Objectives Large language models (LLMs) offer significant potential for automating clinical trial classification by eligibility criteria. However, the optimal input data remain unclear: while abstracts provide a condensed signal, full-text articles contain substantially more information. Whether this additiona...
J. Weyrich, F. Dennstädt, Robert Förster et al.· JAMIA Open· 0 citations
BACKGROUND
Precision medicine requires delivery and tracking of genome-informed risk assessments (GIRAs), often documented in unstructured electronic health record (EHR) notes, making large-scale evaluation reliant on labor-intensive manual chart review. Large language models (LLMs) offer a promising approach to automa...
Yi Xin, K. W. Davis, Wu-Chen Su et al.· JAMIA Journal of the America...· 0 citations
LLMs are suitable for information extraction of medications from clinical notes for use in research databases, however, for a clinical setting where the treatment of patients would be dependent on LLM performance, the current state-of-the-art open weight models are not accurate enough.
T. Spreuer, A. Günther, R. Majeed· Studies in Health Technology...· 0 citations
Spin in clinical trials includes reporting practices that distort the presentation of results. This is particularly critical in medicine, where spin is present in more than 50% of randomized controlled trials that fail to reach statistical significance. The comparison of primary and reported outcomes is crucial for det...
Tjaša Ajdovec, Marko Robnik-Šikonja, Simon Suster· 0 citations
An autonomous AI agent integrating Reasoning and Acting with Retrieval-Augmented Generation with Retrieval-Augmented Generation is developed to classify 10,211 Human Phenotype Ontology terms, enabling standardized panel design by providing reliable, automated classification supported by direct evidence.
Tohid Ghasemnejad, A. Argha, M. Grosser et al.· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.