Introduction:

Per national guidelines, low-grade non-muscle invasive bladder cancer (LG NMIBC) can be managed with active surveillance, intravesical chemotherapy, or BCG. Understanding the risk of grade or stage progression can impact decisions on treatment selection. Herein we leveraged the Computational Histology AI (CHAI) platform, which has previously been used to identify prognostic and predictive biomarkers in high-grade NMIBC, to develop and validate a biomarker that prognosticates high grade or stage progression risk in LG NMIBC.

Methods:

We conducted a retrospective study of 425 patients with LG NMIBC with available index bladder tumor pathology specimens from 2007 to 2024, splitting 123 (30%) into a discovery cohort and 302 (70%) into a external held-out validation cohort. A board-certified pathologist selected representative diagnostic samples from transurethral resection specimens for each patient. We used the previously described CHAI platform to extract quantitative histologic features from digital whole slide images (WSIs) of hematoxylin and eosin (H&E)-stained tumor sections. Features most significantly associated with progression risk in the development cohort were used to construct the LG-prog biomarker, which scores each tumor on a continuous scale and assigns a binary risk category using a threshold fit to the development cohort. Progression was defined as any grade or stage progression. In the validation cohort, the degree of association between the LG-prog biomarker and the primary study endpoint of progression free survival (PFS) was assessed using a Cox proportional hazards (CPH) model. 

Results:

Of the 302 patients with LG NMIBC included in the validation cohort, the LG-prog biomarker classified 41 (13.6%) as high-risk and 261 (86.4%) as low-risk. Cases identified as LG-prog high-risk had significantly increased progression risk compared to the low-risk group (HR=4.75 [2.48, 9.09], p<0.0001), even after controlling for known IBCG prognostic clinical risk factors of size, multifocality, failure of prior intravesical therapy, frequent prior recurrence, recurrence within 1 year and first line induction treatment after index TURBT (HR=5.66 [2.81, 11.44], p<0.0001). Progression risks over time showed a marked difference between biomarker risk groups at 12 months (17% in the high-risk group vs 4.6% in the low-risk group, p=0.11), 36 months (50% vs 12%, p<0.05), and 60 months (73% vs 15%, p<0.001). 

Conclusion:

Using pre-treatment H&E-stained tumor specimens, the CHAI-based LG-prog biomarker can identify LG NMIBC patients with significantly increased progression risk who may be candidates for treatment escalation. With the growing availability of effective therapies, this risk stratification tool enables precision medicine for patients with low grade non muscle invasive bladder cancer. 

Funding: Valar Labs