A new AI assisted approach aligns data standards and accelerates interoperability in biomedical research

Area of Science:

• Biomedical Informatics
• Artificial Intelligence in Healthcare
• Data Science

Background:

• Biomedical data harmonization is complex and time-consuming.
• Manual data integration presents significant barriers to research collaboration.
• Standardization of data elements is crucial for interoperability.

Purpose of the Study:

• To demonstrate how Large Language Models (LLMs) can automate the generation of Common Data Elements (CDEs).
• To assess the efficiency and accuracy of LLM-driven metadata generation for biomedical datasets.
• To develop a system for identifying semantic equivalences and building a standardized data repository.

Main Methods:

• Utilized OpenAI's GPT-4 (API Model gpt-4-0613) to process 31 diverse biomedical datasets.
• Employed a template-based system for comprehensive metadata generation.
• Implemented ElasticSearch with weighted field matching for semantic equivalence identification.
• Validated outputs with subject-matter experts and tested with Alzheimer's Disease Neuroimaging Initiative (ADNI) and Global Parkinson's Genetic Program (GP2) datasets.

Main Results:

• Achieved 94% of generated metadata fields requiring no revision by experts, with 83.8% unweighted accuracy for semi-structured sources.
• Demonstrated significantly faster processing compared to manual methods.
• Successfully mapped 32.4% of unseen headers to CDEs, achieving an average interoperability score of 53.8/100.
• Reduced duplicate CDEs and built a standardized repository.

Conclusions:

• LLMs substantially accelerate biomedical data harmonization through automated CDE generation.
• The developed system effectively reduces barriers to cross-study collaboration by automating data integration.
• This approach offers a scalable solution for standardizing diverse biomedical data, improving research efficiency.