Development and external validation of an artificial intelligence-based tool for PROGression Risk assessment in Non-muscle invasive Bladder Cancer (PROGRxN-BCa)

Introduction & Objectives

Current tools to predict risk of progression in non-muscle invasive bladder cancer (NMIBC) perform poorly and do not completely reflect current practice. We aimed to develop and validate PROGRxN-BCa (PROGression Risk assessment in NMIBC) – an artificial intelligence tool to better predict tumour progression.

Materials & Methods

PROGRxN-BCa, based on a gradient-boosted survival forest, was trained on NMIBC patients treated from Jan-2005 to Dec-2015 at one of three academic or community-based hospital networks: University Health Network, Sinai Health System, and Trillium Health Partners (n=2,002). Internal validation was then performed on patients treated from Jan-2016 to Jun-2022 at the same institutions (n=1,321). External validation was performed on patients treated from May-2012 to Jun-2023 across 13 academic institutions affiliated with the Canadian Bladder Cancer Information System (n=3,708). Primary outcome was time to progression, defined as first relapse of ≥T2 (at TURBT or cystectomy), nodal, or metastatic disease. PROGRxN-BCa was compared to the European Association of Urology (EAU) risk calculator, the most widely used clinical prediction model for NMIBC progression.

Results

During a median follow-up of 36 months (IQR 17-65), 1,006 out of 7,031 (14%) patients developed progression. PROGRxN-BCa outperformed the EAU risk calculator, achieving a c-index between 0.75 to 0.81 (Table). This performance benefit was consistent across clinically relevant subgroups, including age, sex, and tumour history. PROGRxN-BCa was well-calibrated for risks between 0-40%. At 5 and 10 years, PROGRxN-BCa demonstrated a higher net benefit (i.e. avoid unnecessary treatment escalation) compared to the EAU risk calculator for clinically relevant decision thresholds between 15-35%. When applied to intermediate risk patients (n=1,555), PROGRxN-BCa identified 19% of patients with an actual average 5-year progression risk of 32% – revealing a subset of patients who may benefit from treatment intensification. Similarly, the model identified 32% of patients with an actual average 5-year progression risk of 2.5%. This approach outperformed sub-stratification based on intermediate risk factors. Table. C-index (95% CI) of the prediction models.

Conclusions

PROGRxN-BCa outperformed the current gold standard in NMIBC prognostication in both academic and community settings. A particular strength is its ability to further sub-stratify the heterogeneous intermediate risk group. PROGRxN-BCa has the potential to further improve NMIBC risk stratification, inform clinical decision-making, and determine eligibility for clinical trials.