Related Experiment Video
Updated: Jan 15, 2026
Introduction of an Integrated Pathology Image Management, Artificial Intelligence, and Reporting System
Published on: July 11, 2025
Development and International Evaluation of an Artificial Intelligence-based Model (PROGRxN-BCa) Using the World
Jethro C C Kwong1, Zizo Al-Daqqaq2, Yashan Chelliahpillai2
1Division of Urology, Department of Surgery, University of Toronto, Toronto, Canada; Division of Urology, Department of Surgery, University Health Network, Toronto, Canada; Temerty Centre for AI Research and Education in Medicine, University of Toronto, Toronto, Canada.
Background And Objective:
Current non-muscle-invasive bladder cancer (NMIBC) tools perform suboptimally in predicting progression risk to potentially lethal muscle-invasive disease. We aimed to improve risk assessment using artificial intelligence approaches (PROGression Risk assessment in NMIBC [PROGRxN-BCa]).
Methods:
PROGRxN-BCa was trained using 14 clinicopathological features on 3324 NMIBC patients treated from 2005 to 2022 at four Canadian institutions. External testing was performed on 9335 patients treated from 2005 to 2023 across 30 North American and European institutions. The primary outcome was time to progression (muscle-invasive or metastatic disease). PROGRxN-BCa was compared with the European Association of Urology (EAU) risk calculator. Performance was characterised using concordance index (c-index), calibration plots, instability assessments, decision curve analysis, and an algorithmic audit.
Key Findings And Limitations:
During a median follow-up of 3.3 yr (interquartile range 1.6-5.8), 1405 of 12659 patients progressed. In the external testing cohort, PROGRxN-BCa had significantly higher c-index (0.79, 95% confidence interval [CI] 0.77-0.80) and net benefit overall and across different subgroups compared with the EAU risk calculator (0.71, 95% CI 0.70-0.73, p < 0.001). This improvement was consistent regardless of treatment with bacillus Calmette-Guérin, adherence to guideline-concordant care, and World Health Organization 1973 or 2004/2022 grading system, and was consistent among patients with at least 5 yr of follow-up. It also outperformed other guideline-endorsed tools and a previously published artificial intelligence model. Compared with guideline recommendations, PROGRxN-BCa improved substratification of intermediate-risk patients into distinct risk tertiles, with estimated 5-yr progression risks of 2%, 7%, and 15%-the latter in line with high-risk NMIBC.
Conclusions And Clinical Implications:
PROGRxN-BCa outperformed current tools in the largest NMIBC cohort of its kind. Its integration into guidelines could improve risk stratification and patient management.
