Open Sharing Icons

CONNECT: Collaborative effort to optimize diabetes care

DHIF 1616

CONNECT: Collaborative effort to optimize diabetes care through LLM‑enabled clinical decision support

Privacy-preserving artificial intelligence that combines real-world clinical data, genomic risk scores and trusted guidelines to support safer, more personalized diabetes treatment decisions.

People living with diabetes often take multiple medications and need regular monitoring. Clinicians must balance guidelines, lab results, other health conditions, and each person's risk of complications. Today, the information needed to personalize decisions is often scattered across systems, recorded in different formats, and hard to reuse safely.

CONNECT will build a secure “diabetes prescriptome” — a structured, queryable view of medication history and diabetes care that links real-world electronic health record (EHR) data from Omnimed, genomic risk information (polygenic risk scores) from Optithera, and curated drug knowledge and guideline pathways from Wikimedica. The project will standardize data using widely used health-data standards (OMOP and HL7-FHIR) so results are interoperable and easier to validate.

Using privacy-preserving, the models will be trained and used where the data already live. In other words, patient-level data will not move between organizations. Instead, partners will exchange encrypted model updates and aggregated statistics. The decision-support tool will use a retrieval-first approach: it finds the most relevant guideline passages and evidence for a patient scenario, then an open-weight large language model will help draft a structured suggestion (for example, medication adjustments or recommended tests) that is linked back to the supporting evidence for transparency.

By the end of the project, the team aims to:

  1. deliver an interoperable diabetes prescriptome,
  2. validate and benchmark different LLMs (Large Language Models), and
  3. integrate a clinician-facing decision-support prototype into Omnimed’s EHR workflow.
The expected impact is safer, more consistent and more personalized diabetes care, with a clear pathway to responsible deployment in Canada.

 


“Canada needs more than great ideas, we need a trusted ecosystem where advanced AI algorithms can be tested, compared, and proven on real-world data under strong governance. CONNECT is designed for the real world. That shared infrastructure is what makes development, deployment and rigorous vetting of AI algorithms possible. CONNECT is about making diabetes care decisions safer and more consistent by bringing trusted clinical guidance together with real-world data.” Jean Noel Nikiema, Project Lead (Université de Montréal)

Project Images

↑