The AI Wave in OHDSI: Accelerating Innovation Through Community Collaboration

Our community spent June diving deep into the intersection of generative AI and observational research, kicking off the month with a comprehensive high-level overview of how Large Language Models (LLMs) are transforming our ecosystem (see video). Throughout the subsequent presentations and community discussions, several core themes emerged, showcasing how these advanced models can streamline the entire data lifecycle. From automating complex cohort definition and accelerating phenotype development to simplifying data harmonization and ETL pipeline design, the presentations highlighted a powerful shift toward embedding intelligent automation directly into the OMOP Common Data Model architecture.

The incredible variety of individual breakthroughs presented this month shows just how energized our

Overview of LLM Research in OHDSI

network is by the potential of healthcare AI. To fully unlock the potential of these foundational models, our next great opportunity lies in bringing these exciting, innovative efforts together into unified, network-wide collaborations. By standardizing how LLMs interface with OHDSI vocabularies and ensuring these tools are reproducible across all data environments, we can scale these innovations globally through the same open, collective effort that has always driven our community forward.

It is important to remember that the insights and studies spotlighted this past month represent just a small fraction of the AI momentum building across our global network. The upcoming OHDSI Global Symposium in October will feature a massive expansion of this research, diving far deeper into validated real-world evidence generation and the future of healthcare AI. If you want to be part of shaping the official roadmap for AI integration in observational health, make sure your registration is secured for New Brunswick!

LLM Presentations

Title: Ariadne: Automated vocabulary mappings using AI
Presenter: Anna Ostropolets, Johnson & Johnson/Columbia University

Slides
Video

Title: LLM–Based Classification of ICD-10-CM to SNOMED Mappings for Improved Semantic Fidelity in OHDSI
Presenter: Dmytry Dymshyts, Johnson & Johnson

Slides
Video

Title: ACQIRE: Extending the OMOP Lifecycle
Presenter: Roger Carlson, Corewell Health

Slides
Video

Title: FastOMOP – multi agent cohort creation
Presenter: Niko Moeller-Grell, King’s College

Slides
Video

Title: Phenelope – tool for developing concept sets using LLM
Presenter: Joel Swerdel, Johnson & Johnson

Slides
Video

Title: LLM-Based Phenotype Refinement via CAPR
Presenter: Jared Houghtaling, Johnson & Johnson

Slides
Video

Title: Using Synthetic Data and Claude Code to Develop Transportable Analytic Code
Presenter: Adam Johnson, Duke University

Slides
Video

Title: Study Agent support for users of Hades
Presenter: Richard Boyce, University of Pittsburgh 

Slides
Video

Title: FastSSV, a semantic static validator for LLM Generated queries 
Presenter: Shihao Shenzhang, King’s College, London

Slides
Video

Title: Implementation of foundation models in the Trøndelag Health Study 
Presenter: Brooke Wolford, Norwegian University of Science and Technology / University of Oslo

Slides
Video

Title: FastOMOP: Agent Harness and Ecosystem for RWE 
Presenter: Vishnu V Chandrabalan, Lancashire Teaching Hospitals NHS Foundation Trust

Video