For thousands of liver transplant recipients worldwide, routine follow-up can quickly turn uncertain when liver enzyme levels rise—an early signal of potential graft injury. The diagnostic gold standard, a liver biopsy, is invasive and often delayed. Additionally, physicians may adjust management and immunosuppression based on intuition, sometimes before receiving biopsy results, which can lead to complications. A new AI-driven tool called GraftIQ provides a safer approach to the early management of patients with elevated liver enzymes.
Developed at the University Health Network (UHN), GraftIQ is a first-of-its-kind hybrid model that blends clinician expertise with machine learning to provide a multi-class prediction, meaning it can distinguish between several causes of liver graft injury. This “human-in-the-loop” approach enables the system to learn from both data and domain knowledge, thereby enhancing its ability to detect six major causes of graft injury.
Trained and validated on more than 8,000 biopsies across three continents, GraftIQ consistently outperformed traditional models, showing strong generalizability and clinical relevance. In trials, it helped identify conditions like acute cellular rejection, biliary obstruction, and recurrent hepatitis C with high accuracy—without the need for invasive testing.
Importantly, international partners collaborated on the external validation, including Drs. Joseph Ahn (Mayo Clinic, USA), Richard Taubert (Hannover Medical School, Germany), and Eunice Tan (National University Health System, Singapore). GraftIQ performed strongly in these validations, demonstrating a level of generalizability across three continents that is rare for health care AI tools.
“This kind of technology does not replace clinical judgment—it enhances it,” says Dr. Mamatha Bhat, Hepatologist and Co-Lead of the Transplant AI initiative and Scientist at UHN, who envisioned and led this project. “GraftIQ is a step toward faster diagnosis and more timely treatment decisions.”
By integrating seamlessly into clinical workflows, GraftIQ represents a scalable, multi-class decision-support tool for transplant programs worldwide. Its success underscores the value of global collaboration in evaluating health care AI tools. UHN stands at the forefront of applying responsible AI in medicine.
Dr. Divya Sharma, Senior Biostatistician at Princess Margaret Cancer Centre, and Assistant Professor in the Department of Mathematics and Statistics at York University, is co-first author of the study.
Dr. Neta Gotlieb, from the Department of Medicine at the University of Ottawa and Dr. Daljeet Chahal, from the Vancouver General Hospital, are co-first authors of the study.
Dr. Wei Xu, Clinician Scientist at the Princess Margaret Cancer Centre, and Professor at the Dalla Lana School of Public Health at the University of Toronto, is co-senior author of the study.
Dr. Mamatha Bhat, Clinician-Scientist and Hepatologist at Ajmera Transplant Centre, Scientist at the Toronto General Hospital Research Institute (TGHRI), and Associate Professor in the Department of Medicine at the University of Toronto, is co-senior author of the study.
This work was supported by grants to UHN investigators from the Canadian Society of Transplantation, the American Society of Transplantation (AST), the Canadian Institutes of Health Research (CIHR), and UHN Foundation.
Sharma D, Gotlieb N, Chahal D, Ahn JC, Engel B, Taubert R, Tan E, Yun LK, Naimimohasses S, Ray A, Han Y, Gehlaut S, Shojaee M, Sivanendran S, Naghibzadeh M, Azhie A, Keshavarzi S, Duan K, Lilly L, Selzner N, Tsien C, Jaeckel E, Xu W, Bhat M. GraftIQ: Hybrid multi-class neural network integrating clinical insight for multi-outcome prediction in liver transplant recipients. Nat Commun. 2025 May 28;16(1):4943. doi: 10.1038/s41467-025-59610-8. PMID: 40436838.