National Diabetes Repository

Diabetes Action Canada has successfully launched the first National Diabetes Repository in Canada.  This novel, secure analytics platform designed and implemented by Dr. Michelle Greiver (University of Toronto) and colleagues now contains information from over 110,000 individuals with diabetes in Alberta, Manitoba, Ontario, Newfoundland and Quebec, along with the same number of age-matched non-diabetic controls.  The National Diabetes Repository was created through collaboration with the Canadian Primary Care Sentinel Surveillance Network https://cpcssn.ca , using de-identified and encrypted primary care electronic medical records (EMRs) data that can be accessed by approved Diabetes Action Canada investigators for population-based and observational studies.  The repository has since evolved to accept patient reported outcomes and experience measures directly from patients using digital tablets that are directly connected to EMRs.  Data from provincial administrative data sources can be linked to individuals in each province to provide social determinants of health and outcomes data.  To facilitate data integration, data sharing agreements are anticipated with the provincial organization members of the SPOR National Data Platform research program.

In collaboration with the Fields Institute Centre for Quantitative Analysis and Modelling (Fields-CQAM), and the Vector Institute for Artificial Intelligence and investigators from the Dalla Lana School of Public Health from the University of Toronto, we held a two-day data workshop June 17th and 18th 2019.  Trainees and established researchers applied advanced analytics to the de-identified dataset within our National Diabetes Repository. The exercise provided much needed insights into the feasibility of using advanced mathematic modelling and artificial intelligence learning models on Canadian EMR data in a secure high-performance computing environment.  Using two testing environments, the results of this workshop showed that our data can be used for: 1) machine learning to predict patient responses to SGLT2 Inhibitors with high accuracy based on their health records; and, 2) artificial intelligence to identify the characteristics of sub-group patients, including their medication history, that are associated with different HbA1c trajectories.  Taken together, we were able to demonstrate that artificial intelligence and advanced analytics could be applied to our dataset to provide useful information for both patients and physicians in selecting treatment options to manage their condition.