Federated Multi-Agent Architecture for Harmonizing Public Health Datasets into OMOP and FHIR Standards

Area of Science:

• Health Informatics
• Artificial Intelligence in Medicine
• Data Science

Background:

• Non-communicable diseases (NCDs) pose a significant global health challenge.
• Integrating diverse clinical data is crucial for NCD research and prevention.
• Existing data integration methods can be complex and time-consuming.

Purpose of the Study:

• To develop and validate a federated system for integrating retrospective and prospective clinical data.
• To leverage Large Language Model (LLM)-based multi-agent systems for automated data processing.
• To support scalable and interoperable data integration for clinical decision-making within the SHIELD project.

Main Methods:

• Development of a federated system using LLM-based multi-agent systems.
• Automated Extract, Transform, Load (ETL) processes for data harmonization.
• Natural language querying capabilities for data access.
• Harmonization of MIMIC-IV, ELSA, and synthetic data into OMOP Common Data Model (CDM) and Fast Healthcare Interoperability Resources (FHIR) standards.

Main Results:

• High accuracy in mapping diverse data sources to standardized formats (OMOP CDM and FHIR).
• Demonstrated feasibility of automated ETL processes using LLM-based agents.
• Successful integration of retrospective and prospective clinical data.
• Validation of the system's capability for natural language querying.

Conclusions:

• The federated system effectively integrates heterogeneous clinical data, enhancing interoperability.
• LLM-based multi-agent systems offer a powerful solution for automated data ETL and querying.
• The developed system is scalable and supports data-driven clinical decision-making for NCD reduction.