DHDP: A health data sharing platform putting federated learning into action for the benefit of Canadians

The Digital Health & Discovery Platform is developing a made-in-Canada solution that enables researchers to securely access and analyze health data without moving sensitive patient information between institutions, while working with partners across the country to test and strengthen its privacy protections.

Across Canada, hospitals and healthcare institutions generate vast amounts of health data that, when analyzed together, could help answer important research questions, advance personalized treatments, and improve outcomes for patients. But much of this information remains locked in institutional silos, limiting its potential to drive discoveries and improve care.

To address this challenge, the Terry Fox Research Institute-led Digital Health & Discovery Platform (DHDP) is developing technology that enables researchers to securely access and analyze health data across institutions while protecting patient privacy.

The solution? An approach known as federated learning.

The inherent risks of data sharing

Advancing precision medicine requires access to large, diverse collections of health data — more than any single hospital or research centre can generate on its own. Combining data from multiple institutions can help researchers and AI tools identify patterns, improve diagnoses, and develop more personalized treatments.

But sharing sensitive patient information comes with significant risks and responsibilities. Transferring data outside an institution can increase the risk of privacy breaches, unauthorized access, and misuse. Hospitals must also navigate complex regulatory requirements, protect patient confidentiality, and maintain control over who accesses their data and how they are used.

These challenges can make institutions reluctant to share valuable information, even when doing so could accelerate research and improve care.

The solution isn't to stop sharing data, but to change how researchers access and analyze them. By keeping sensitive patient records within the institutions that hold them, federated approaches can reduce risks while enabling collaboration across sites.

How federated learning enables secure data sharing

Imagine a researcher in Vancouver wants to study breast cancer treatments using patient data held at hospitals in Montreal and Toronto. Traditionally, this might require transferring sensitive records between institutions, raising privacy and security concerns. But what if the researcher could analyze those data without moving them?

That's the approach behind the DHDP, which uses a federated architecture to enable secure data discovery and analysis while keeping patient information under local control.

Here's how it works: A central hub connects participating hospitals and research institutions, each operating an edge node within its own infrastructure. Rather than collecting patient records, the hub receives only aggregate metadata — such as table names, record counts, and date ranges — allowing authenticated researchers to discover what data are available across the network.

In our example, the Vancouver researcher could discover that hospitals in Montreal and Toronto hold relevant breast cancer data without accessing individual patient records. To analyze those data, the researcher must first obtain approval through each hospital's governance process. Once authorized, analytical tools and AI models can be brought to the data at each institution rather than transferring sensitive information between sites.

The result is a system designed to enable researchers to collaborate across Canada while institutions retain control over their patients' data.

Privacy and security built into the Platform

Keeping patient data within institutions is only one layer of protection. The DHDP's edge nodes connect only to approved data sources and use automated tools to identify potentially sensitive information, including names, health numbers, and dates of birth. Meanwhile, the central hub manages user authentication and authorization, helping ensure that only approved users can access Platform services.

Importantly, the technology does not replace institutional governance. Participating institutions retain control over their data access policies and approval decisions, while the Platform provides tools to support those processes.

But federated architecture alone cannot guarantee privacy. Even aggregate metadata or AI models can potentially reveal sensitive information. That's why privacy protections must be continually evaluated and strengthened through technical safeguards, institutional oversight, and real-world testing.

Putting the Platform to the test through collaboration

Through the Digital Health Innovation Fund (DHIF), the DHDP is supporting collaborations among researchers, healthcare institutions, and Canadian technology companies to develop innovative solutions to healthcare challenges while putting the emerging Platform to the test.

These projects provide opportunities to evaluate how the Platform performs in real-world research settings, including how data are discovered, accessed, and analyzed across institutions. Technical assessments, security reviews, and ongoing feedback from users will help identify vulnerabilities, strengthen safeguards, and guide Platform development.

This collaborative approach is helping build technology that responds to the needs of researchers and institutions while maintaining strong protections for sensitive patient information.

Unlocking the potential of Canada's health data

Protecting patient privacy doesn't mean keeping valuable health data out of reach. It means creating secure, trusted ways for researchers to use those data responsibly.

By combining federated technology, institutional oversight, and real-world testing, the DHDP is working to make that possible — helping researchers collaborate across Canada, accelerate discoveries, and bring the benefits of more personalized care to patients, without compromising the privacy of their health information.