Enhancing Cross-Domain Generalizability in Social Determinants of Health Extraction with Prompt-Tuning Large Language Models.
Year: 2025
Type: article
Abstract: 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.
Source: PubMed
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
Funders: Patient-Centered Outcomes Research Institute, Nvidia, National Institutes of Health, University of Florida Health, Clinical and Translational Science Institute, University of Florida, National Institute on Aging, National Heart, Lung, and Blood Institute, National Institute on Drug Abuse, National Cancer Institute, National Institute of Allergy and Infectious Diseases
Awards: G1254130036, Advancing Drug Repositioning for Alzheimer’s Disease using Real-world Data, Post-Acute Sequelae of SARS-CoV-2 Infection and Subsequent Disease Progression in Individuals with AD/ADRD: Influence of the Social and Environmental Determinants of Health, G2654747551, G3398941559, G4156463097, G4691473461, G4696575845, The benefits and harms of lung cancer screening in Florida, G5717028053, G7257327873, G7884167642, G8156115585, G8743902842, An end-to-end informatics framework to study Multiple Chronic Conditions (MCC)'s impact on Alzheimer's disease using harmonized electronic health records, Developing and Evaluating a Machine-Learning Opioid Prediction & Risk-Stratification E-Platform (DEMONSTRATE), G982581970
OpenAlex ID: W4411256794