Waters Michael
Department of Radiation Oncology, Washington University in St Louis School of Medicine, St Louis, MO, United States.
JNCI Cancer Spectr. 2025 Mar 3;9(2). doi: 10.1093/jncics/pkaf028.
Clinical trials are fundamental to evidence-based medicine, providing patients with access to novel therapeutics and advancing scientific knowledge. However, patient comprehension of trial information remains a critical challenge, as registries like ClinicalTrials.gov often present complex medical jargon that is difficult for the general public to understand. While initiatives such as plain-language summaries and multimedia interventions have attempted to improve accessibility, scalable and personalized solutions remain elusive. This study explores the potential of Large Language Models (LLMs), specifically GPT-4, to enhance patient education regarding cancer clinical trials. By leveraging informed consent forms from ClinicalTrials.gov, the researchers evaluated 2 artificial intelligence (AI)-driven approaches-direct summarization and sequential summarization-to generate patient-friendly summaries. Additionally, the study assessed the capability of LLMs to create multiple-choice question-answer pairs (MCQAs) to gauge patient understanding. Findings demonstrate that AI-generated summaries significantly improved readability, with sequential summarization yielding higher accuracy and completeness. MCQAs showed high concordance with human-annotated responses, and over 80% of surveyed participants reported enhanced understanding of the author's in-house BROADBAND trial. While LLMs hold promise in transforming patient engagement through improved accessibility of clinical trial information, concerns regarding AI hallucinations, accuracy, and ethical considerations remain. Future research should focus on refining AI-driven workflows, integrating patient feedback, and ensuring regulatory oversight. Addressing these challenges could enable LLMs to play a pivotal role in bridging gaps in clinical trial communication, ultimately improving patient comprehension and participation.
临床试验是循证医学的基础，为患者提供获得新型治疗方法的途径，并推动科学知识的进步。然而，患者对试验信息的理解仍然是一个关键挑战，因为像ClinicalTrials.gov这样的登记处常常呈现复杂的医学术语，普通公众难以理解。虽然诸如通俗易懂的总结和多媒体干预等举措试图提高信息的可及性，但可扩展的个性化解决方案仍然难以实现。本研究探讨了大语言模型（LLMs），特别是GPT-4，在加强癌症临床试验患者教育方面的潜力。通过利用ClinicalTrials.gov的知情同意书，研究人员评估了两种人工智能（AI）驱动的方法——直接总结和顺序总结——以生成对患者友好的总结。此外，该研究还评估了大语言模型创建多项选择题对（MCQAs）以评估患者理解程度的能力。研究结果表明，人工智能生成的总结显著提高了可读性，顺序总结的准确性和完整性更高。多项选择题对与人工标注的答案高度一致，超过80%的受调查参与者表示对作者内部的BROADBAND试验的理解有所增强。虽然大语言模型有望通过提高临床试验信息的可及性来改变患者参与度，但对人工智能幻觉、准确性和伦理考量的担忧仍然存在。未来的研究应专注于完善人工智能驱动的工作流程、整合患者反馈并确保监管监督。应对这些挑战可以使大语言模型在弥合临床试验沟通差距方面发挥关键作用，最终提高患者的理解和参与度。
