Agentic AI System for No-Code Network Study in Healthcare

Problem Background: Network studies enable researchers to run the same analysis across multiple institutions using a shared data model like OMOP CDM. This ensures consistency while preserving local data privacy. However, setting up such studies typically requires writing complex R code, which poses a barrier for clinical researchers without programming experience.

Proposed Solution: The Strategus framework from OHDSI offers a modular, portable approach to observational research by allowing study designs to be represented in a standardized JSON format. While this improves reproducibility and reusability, creating the JSON specification still relies on R programming and understanding complex study components.

This project introduces an AI-powered, natural language interface to generate Strategus-compatible study specifications without requiring R coding. Following are the key features:

• A multi-agent system using the CrewAI framework to decompose the study design process into modular, conversational tasks handled by specialized AI agents.
• The system uses Model Context Protocol (MCP) to manage memory, maintain shared understanding between agents, and track study context across user interactions.
• Built in Python, the system transforms user inputs into a complete, validated JSON specification, ready to be executed using Strategus pipelines.