Peng Cheng, Yu Zehao, Smith Kaleb E, Lo-Ciganic Wei-Hsuan, Bian Jiang, Wu Yonghui
Department of Health Outcomes and Biomedical Informatics, College of Medicine, University of Florida, Gainesville, Florida, USA.
NVIDIA, Santa Clara, California, USA.
AMIA Jt Summits Transl Sci Proc. 2025 Jun 10;2025:432-440. eCollection 2025.
The progress in natural language processing (NLP) using large language models (LLMs) has greatly improved patient information extraction from clinical narratives. However, most methods based on the fine-tuning strategy have limited transfer learning ability for cross-domain applications. This study proposed a novel approach that employs a soft prompt-based learning architecture, which introduces trainable prompts to guide LLMs toward desired outputs. We examined two types of LLM architectures, including encoder-only GatorTron and decoder-only GatorTronGPT, and evaluated their performance for the extraction of social determinants of health (SDoH) using a cross-institution dataset from the 2022 n2c2 challenge and a cross-disease dataset from the University of Florida (UF) Health. The results show that decoder-only LLMs with prompt tuning achieved better performance in cross-domain applications. GatorTronGPT achieved the best F1 scores for both datasets, outperforming traditional fine-tuned GatorTron by 8.9% and 21.8% in a cross-institution setting, and 5.5% and 14.5% in a cross-disease setting.
使用大语言模型（LLMs）进行自然语言处理（NLP）的进展极大地改善了从临床叙述中提取患者信息的能力。然而，大多数基于微调策略的方法在跨域应用中的迁移学习能力有限。本研究提出了一种新颖的方法，该方法采用基于软提示的学习架构，引入可训练的提示来引导大语言模型生成期望的输出。我们研究了两种类型的大语言模型架构，包括仅编码器的GatorTron和仅解码器的GatorTronGPT，并使用来自2022年n2c2挑战赛的跨机构数据集和佛罗里达大学（UF）健康中心的跨疾病数据集评估了它们在提取健康的社会决定因素（SDoH）方面的性能。结果表明，通过提示调整的仅解码器大语言模型在跨域应用中表现出更好的性能。GatorTronGPT在两个数据集上均取得了最佳的F1分数，在跨机构设置中比传统的微调GatorTron分别高出8.9%和21.8%，在跨疾病设置中高出5.5%和14.5%。
