A computational histology AI (CHAI) biomarker was developed using TCGA data and validated in an independent retrospective cohort of 134 patients with cT2N0M0 urothelial carcinoma from NCI-Designated Cancer Centers, using H&E-stained TURBT whole slide images to predict recurrence-free survival (RFS), cancer-specific survival (CSS), and overall survival (OS) after radical cystectomy ± neoadjuvant chemotherapy (NAC).
In the validation cohort, unfavorable CHAI risk classification was associated with significantly worse 3-year RFS (40% vs. 74%; HR 3.1 [1.7–5.7], P<0.001), CSS (HR 3.5 [1.5–7.8], P=0.003), and OS (HR 3.0 [1.5–5.7], P=0.001) vs. favorable risk; the biomarker remained independently prognostic after adjusting for NAC and other clinical variables (P<0.01), with an exploratory interaction signal between biomarker risk and NAC benefit (P=0.02).
- Small overall sample size (N=178; only 44 in development, 134 in validation) limits statistical power and generalizability. - Retrospective design with pooled real-world data introduces selection bias and confounding. - The NAC interaction finding is exploratory and hypothesis-generating only; prospective validation is needed.
This image-only AI biomarker applied to routine pre-treatment TURBT H&E slides may help stratify cT2 MIBC patients by prognosis and potentially by NAC benefit, but it is not yet ready for clinical use pending prospective validation.
Explore related topics
