Enhanced NMIBC Progression Risk Prediction With a Novel AI Model

Jethro C.C. Kwong, a urology resident from the University of Toronto, discusses a novel artificial intelligence (AI)-based model, PROGRxN-BCa, that has been developed and validated to improve the prediction of progression risk in patients with non-muscle invasive bladder cancer (NMIBC), particularly within the challenging intermediate-risk group.

This study, encompassing the largest NMIBC cohort to date (n = 12,659), involved training the AI model on 3,324 patients using 14 readily available clinicopathological features. External validation on a separate cohort of 9,335 patients across North America and Europe demonstrated that PROGRxN-BCa significantly outperformed the current guideline-endorsed European Association of Urology (EAU) risk calculator in predicting progression to muscle-invasive or metastatic disease, with an approximate 10% improvement in overall performance.

“This was an AI model that was trained in the largest NMIBC cohort in the world, with over 12,000 patients, and it is actually over 100 times larger than the median cohort size of prior AI studies. We trained this model on over 3000 patients treated at 4 academic or community hospitals in Canada, and it is trained on 14 clinical pathological features, so no histopathological slides, no biomarkers, or anything like that. It is just things that are readily available in kind of routine day-to-day practice,” Kwong explains.

“We built this model, and then we externally validated this on over 9000 patients across Canada, US, and Europe, and there were over 30 institutions involved for this,” he continues.

Kwong emphasizes the critical need for accurate risk stratification tools in NMIBC, given the influx of new treatments and the limitations of existing calculators, which have shown poor performance in external validation and do not always reflect current clinical practice. He highlights that prior AI studies in this area were often of low quality and based on much smaller cohorts.

PROGRxN-BCa, trained on a substantially larger and more diverse patient population, offers a significant advancement. Notably, the model's superior performance was consistent regardless of adherence to guideline-concordant care.

Additionally, PROGRxN-BCa effectively sub-stratified the heterogeneous intermediate-risk group into distinct risk tertiles, allowing for better identification of patients with varying progression probabilities, a significant improvement over current methods that struggle to differentiate risk within this category.

“We were able to show that our AI model outperforms the current guideline-endorsed tool, generally by about 10%. It is also quite helpful. What we have shown is kind of a practical application of this tool in trying to sub-stratify intermediate-risk normal cell-based bladder cancer,” adds Kwong.