Diabetica: Adapting Large Language Model to Enhance Multiple Medical Tasks in Diabetes Care and Management

2024-09-20

AI Summary

There is an urgent need for domain-specific large language models (LLMs) to support multi-stakeholder tasks—clinical care, medical education, and patient self-management—in diabetes management. Method: We propose the first end-to-end framework for customizing LLMs for diabetes. It features a high-quality, diabetes-specific data curation pipeline that collects, filters, augments, and refines heterogeneous clinical texts; introduces the first open-source diabetes training dataset and a comprehensive evaluation benchmark; and employs data-driven domain adaptation, instruction tuning, and multi-stage enhancement, validated through collaborative assessment by clinical experts. Contribution/Results: We release the first open-source diabetes-optimized LLM family. Evaluated across personalized clinical decision support, medical education assistance, and clinical workflow optimization, it achieves state-of-the-art performance and demonstrates empirically significant improvements in cross-role applicability and task effectiveness.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: (Large) Language Models
• Application Domains: Humanities & Computational Social Science

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Economics, Online Markets and Human Computation: Humans versus LLMs for data annotation and labeling
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Abstract

Diabetes is a chronic disease with a significant global health burden, requiring multi-stakeholder collaboration for optimal management. Large language models (LLMs) have shown promise in various healthcare scenarios, but their effectiveness across diverse diabetes tasks remains unproven. Our study introduced a framework to train and validate diabetes-specific LLMs. We first developed a comprehensive data processing pipeline that includes data collection, filtering, augmentation and refinement. This created a high-quality, diabetes-specific dataset and evaluation benchmarks from scratch. Fine-tuned on the collected training dataset, our diabetes-specific LLM family demonstrated state-of-the-art proficiency in processing various diabetes tasks compared to other LLMs. Furthermore, clinical studies revealed the potential applications of our models in diabetes care, including providing personalized healthcare, assisting medical education, and streamlining clinical tasks. Generally, our introduced framework helps develop diabetes-specific LLMs and highlights their potential to enhance clinical practice and provide personalized, data-driven support for diabetes management across different end users. Our codes, benchmarks and models are available at https://github.com/waltonfuture/Diabetica.

Problem

• Adapting LLMs for diabetes-specific medical tasks
• Developing a diabetes-focused dataset and benchmarks
• Enhancing diabetes care through personalized AI support

Innovation

• Developed diabetes-specific large language models
• Created comprehensive data processing pipeline
• Achieved state-of-the-art diabetes task proficiency