*To preserve the depth and detail of the original conversation, this article will be presented in two parts.
In 2024, Ruth Sapir-Pichhadze, BMedSc, MD, MSc, PhD, FRCPC, Associate Professor in the Department of Medicine at McGill University’s Faculty of Medicine and Health Sciences and clinician-scientist in the Division of Nephrology at the McGill University Health Centre (MUHC), approached the McGill Interdisciplinary Initiative in Infection and Immunity (MI4) with an ambitious idea: to use artificial intelligence to better predict kidney transplant outcomes.
With MI4’s leverage grant of $115,000, Dr. Sapir-Pichhadze successfully secured $921.6K over three years from Quebec’s Ministry of Economy and Innovation, for her project “Federated machine learning to create tools predicting kidney graft survival: a German and Canadian collaborative projects”. The work is now part of the NephroCAGE initiative, which aims to advance AI-driven predictive tools for kidney transplantation while strengthening collaboration between Canada and Germany.
We spoke with Dr. Sapir-Pichhadze about the project, the challenges of building international data collaborations, and what the future of transplant care could look like.
What was the moment you realized this project could truly change outcomes for kidney transplant patients?
As a clinician-scientist, I want to make sure the work I do ultimately improves patient outcomes and experiences. This project focuses on two key goals.
First, we are developing clinical prediction models that estimate the likelihood of complications after transplantation, such as organ rejection or infection, while also predicting how long a transplanted kidney is likely to function. These tools can help clinicians better match donors and recipients and refine organ allocation so that transplanted organs last longer.
Second, we are building a federated learning infrastructure that allows researchers to analyze data across multiple institutions without moving sensitive patient information. Instead of transferring patient-level data to a central database, algorithms are sent to local sites where the analysis is performed. This approach protects confidentiality while still allowing us to learn from large datasets.
Together, these advances can help improve clinical care while respecting the ethical and regulatory frameworks surrounding health data.
In this process, what was the biggest challenge in turning an international, AI-driven collaboration into something that actually works in practice?
This is still a work in progress, but we have already made significant progress.
One of the first challenges involved administrative agreements between institutions. Establishing international collaborations requires extensive contracts and approvals, and it took nearly two years to finalize our agreements.
Funding cycles were another hurdle. Initially, the project relied on resources from our German collaborators while we pursued funding opportunities in Canada.
Data harmonization was also important. Our collaborators operate within different healthcare systems, languages, and clinical practices than our own, so we needed to standardize data formatting, naming, and definitions into a single, consistent format.
Bringing scientists together, however, was the easiest part. There is a shared passion and a clear understanding of the importance of the opportunity at hand.
