Large Language Models Aid in Simplifying Clinical Trial Information for Patients
A recent study published in JNCI Cancer Spectrum explored the potential of large language models (LLMs), specifically GPT-4, to generate patient-friendly educational materials for cancer clinical trials. The research addresses the critical need for improved patient awareness and understanding to enhance trial recruitment, informed decision-making, and adherence to protocols.
The study employed GPT-4 to create concise clinical trial summaries and multiple-choice question-answer pairs derived from informed consent forms (ICFs) available on ClinicalTrials.gov. Two prompting strategies were utilized for summaries: direct summarization and a sequential extraction-summarization approach. For the development of question-answer pairs, a one-shot learning method was applied.
Evaluation of the generated content involved patient surveys assessing the effectiveness of the summaries and crowdsourced annotations to verify the accuracy of the question-answer pairs. The assessments focused on held-out cancer trial ICFs that were not part of the initial prompt development.
Findings revealed that both prompting approaches produced summaries with comparable readability and inclusion of core content. Patients reported that these summaries were understandable and enhanced their comprehension of clinical trials, increasing their interest in learning more. The multiple-choice questions generated by GPT-4 demonstrated high accuracy and strong agreement with the annotations from crowdsourced evaluators.
However, the study identified a tendency for GPT-4 to introduce inaccuracies when prompted to provide information not sufficiently detailed in the ICFs. This underscores the necessity of careful prompt design and human oversight in the deployment of LLMs for generating educational content.
The authors conclude that LLMs like GPT-4 show promise in producing patient-friendly educational materials for clinical trials with minimal need for trial-specific engineering. They emphasize that while LLMs can significantly aid in improving patient education and engagement, ongoing human oversight is essential to ensure the accuracy and reliability of the information provided.
This study serves as a proof-of-concept for integrating advanced artificial intelligence technologies into patient education strategies within oncology, aiming to facilitate better patient participation and adherence in clinical trials.
Reference
Gao M, Varshney A, Chen S, et al. The use of large language models to enhance cancer clinical trial educational materials. JNCI Cancer Spectr. 2025:pkaf021. doi: 10.1093/jncics/pkaf021
