Patient-level AI Collaboration for Precision Medicine in Canada: A Scoping Review

Precision medicine runs on patient-level data. In Canada, most patient-level AI collaborations are international, and many draw on datasets housed outside the country. We mapped the landscape of Canadian healthcare-AI collaborations using patient-level data and found that privacy-preserving data-sharing practices are extremely rare. Using a custom large language model pipeline, we screened 245,886 articles (2018–Feb 2025) and identified 3,100 relevant studies. The main dataset drawn from PubMed after a pilot assessment against IEEE Xplore, ACM Digital Library, Scopus, and Web of Science indicated equivalent coverage for our eligibility criteria. We identified trends in domestic and international collaborations and examined implications for data sharing, privacy, and interoperability. Among 3,100 studies, only 35.8% were conducted solely by Canadian institutions and nearly two thirds involved international partners. Imaging data were the most common modality. Alarmingly, out of 160 multi-site collaborations, just 5% (8 studies) used any privacy-preserving method, and only one of those was a fully within-Canada collaboration using Canadian patients' data. Canada's slow adoption of privacy-preserving collaboration technologies is limiting the country's ability to fully leverage AI in healthcare. Without immediate investment in secure, decentralized data sharing infrastructure, Canada will fall behind in health AI. Prioritizing such infrastructure is critical to enable innovation and to ensure Canadian patient data benefit Canadian patients.