Rheum Project: Automated Analysis of Rheumatology Research Literature Using Large Language Models

The Rheum Project is a research initiative in partnership with Alberta Health Services. It aims to use fine-tuned large language models to automatically identify and classify clinical trials in rheumatology papers, build a knowledge graph of research evolution, and address the pain points of manual processing of massive medical literature—such as being time-consuming, labor-intensive, and prone to missing information.

Rheumatology is a complex discipline involving multiple autoimmune diseases, with a large number of clinical trials and research findings published each year. Researchers and clinicians need to track progress to guide practice, but manual reading and organizing of massive literature is time-consuming, labor-intensive, and easy to miss information. Thus, the Rheum Project was born, aiming to use large language models to automatically analyze and classify rheumatology papers, identify clinical trials in the literature, and track their inheritance and development relationships.

Alberta Health Services (AHS) is Canada's largest integrated healthcare provider, serving over 4 million residents in Alberta. It has rich experience in medical research, clinical practice, and patient care. Collaborating with AHS allows the project to access real clinical data and needs, ensuring the research direction aligns with actual medical scenarios and exploring new possibilities for the application of large language models in the medical field.

The technical core of the project lies in the text understanding and information extraction capabilities of large language models. For the specificity of biomedical literature, specialized fine-tuning and optimization are carried out: 1. Collect a large number of professional rheumatology literature (clinical trial reports, reviews, case studies, etc.) as training data and professionally annotate key information; 2. Use domain adaptation technology to fine-tune pre-trained large language models, improving their understanding of medical terminology and literature structure, as well as the accuracy of key information extraction.

Two core functions: 1. Automatic clinical trial identification: The system scans literature to intelligently identify information such as trial names, numbers, and research phases. It can understand context and accurately recognize trial names in abbreviated or variant forms; 2. Research evolution relationship mapping: Analyze citations and logical connections between different trials, construct a knowledge graph, and display inheritance, verification, or refutation relationships between trials to help grasp the development context of the field.

Multi-layered application value: 1. For researchers: Shorten literature review time, quickly locate relevant studies, and avoid duplicate work; 2. For clinicians: Trial correlation analysis helps them understand the evidence basis of treatment plans and make more informed decisions; 3. For research institutions: The knowledge graph identifies research gaps, guides future directions, and promotes disciplinary development through rational resource allocation.

Challenges and solutions during implementation: 1. Complex medical terminology: Introduce medical knowledge bases and terminology dictionaries to enhance the model's term recognition ability; 2. Diverse literature formats: Adopt flexible text parsing strategies, combining rule matching and deep learning to improve adaptability; 3. Relationship accuracy: Combine multi-round verification and manual review to ensure accurate relationship extraction.

Future plans: 1. Expand to more medical specialties and build a more comprehensive medical research knowledge graph; 2. Explore integration with clinical decision support systems to provide doctors with real-time research evidence references; 3. Integrate image analysis capabilities to automatically extract information from literature charts and graphs, enrich the content of the knowledge graph, and promote the intelligent transformation of medical research.