Federated Insights Across Distributed Health Datasets in Canada
Feasibility study to explore federated machine learning.
Project Overview
Updated October 2, 2024
The Problem
In order to inform decision making and improve health outcomes for Canadians, government stakeholders and organizations across the country need to be able to make better use of health data. Health data is distributed across many different institutional, provincial, and national systems. A major problem is a lack of connectedness and interoperability between distributed health datasets, making it impossible to leverage their collective power to make insights.
The traditional approach of centralized “data sharing” – where datasets are combined by uploading and analyzing them in one place – is not suitable for health data because it is inefficient, insecure, not scalable, not sovereign, and does not empower data stewards to retain physical and administrative control over data. The many technical, cultural, political, logistical and regulatory issues with the centralized data sharing model are significant barriers to enabling better health outcomes for Canadians.
How We Are Solving It
This feasibility study will explore the use of DNAstack’s federated data analysis technology which could serve in new use cases of interest to stakeholders in the Canadian health ecosystem. It will seek to confirm the ability of DNAstack’s technology to provide aggregative insights using data from individual health data stewards; demonstrate its ability to derive insights across a federated network of multiple health data stewards; and validate an ongoing consortium of organizations to participate in federated networks to help overcome long-standing barriers to health data insights.
The Result
The primary technological output of this project was a modular system composed of multiple software products to enable federated analysis. Its technical architecture was successfully deployed within The Hospital for Sick Children (SickKids) and Holland Bloorview Kids Rehabilitation Hospital (Holland Bloorview), with data scientists from both institutions having successfully conducted analyses and validated machine learning algorithms on connected networks of federated health data.
Autism Speaks, SickKids, Holland Bloorview, and King’s College London, all partners from a previously-funded DIGITAL project, the Autism Sharing Initiative, are utilizing the developed framework to create a proof-of-concept federated analysis initiative involving additional datasets.
