When Data Stays Local but Science Travels: A New Method for Cross-Border Health Research
A Canadian-led study shows that a novel statistical technique can analyse health data held in different regions without ever moving a patient record.
Some of the world’s most pressing health questions remain unanswered - not because scientists lack the tools, but because the data cannot cross borders. A new study from researchers at the University of Calgary, published in the International Journal of Population Data Science (IJPDS), offers a potential solution. The team – led by Megan Harmon and supervised by Drs. Tyler Williamson and Jessalyn Holodinsky – evaluated a novel statistical method known as the federated likelihood approach, which allows researchers in different regions to collaborate without transferring sensitive patient data across jurisdictions.
Around the world, laws and policies tightly restrict the transfer of personal health information to protect patient privacy. While these safeguards are essential, they can also make collaboration across institutions or regions difficult. As a result, many important public health questions – particularly those involving rare diseases, large populations, or cross-regional trends – remain challenging or even impossible to answer.
One common strategy used when data cannot be shared is meta-analysis, which pools estimates from separate models rather than combining the underlying data. However, because meta-analysis relies on these estimates, it can miss important nuances in the data.
In the study, the federated likelihood method produced more accurate results than meta-analysis, suggesting strong potential as an innovative approach for analysing health data in settings where privacy regulations prevent sharing individual-level information.
The method works by generating a mathematical matrix that acts as an encoded representation of each regional dataset. These matrices can be shared across sites without revealing the underlying data. Instead of centralising records, each team retains its data locally and shares only the “statistical fingerprint” - unique and complex enough to prevent reconstruction of the original data while preserving key information. When combined, these matrices allow researchers to identify broader patterns and trends across regions.
“This method allows us to learn from health data across different regions without compromising patient privacy. In an era of evolving data governance, that capability could transform how we conduct large-scale health research.”
— Megan Harmon, MSc
By enabling collaboration without moving sensitive data, the federated likelihood approach could help unlock answers to major global health questions that have long been out of reach. It may be particularly valuable for studying rare diseases and population-level health trends, where large and geographically diverse datasets are essential. If widely adopted, this approach could open new possibilities for global health research, allowing scientists to work together across borders while keeping patient data securely in place.
Click Here to read the full article
Megan Harmon, MSc Biostatistics and PhD student Epidemiology, University of Calgary, Canada