Vessel branching patterns quantified by artificial intelligence as a novel prognostic biomarker in advanced renal cell carcinoma: A post-hoc analysis of the CM914 trial.

Abstract

4523 Background: Current adjuvant risk stratification in renal cell carcinoma (RCC) relies primarily on TNM-staging. However, disease progression and response to therapy are driven by biological subgroups reflected in tumor vessel architecture. We therefore analyzed vessel branching patterns in 1098 adjuvant RCC patients to evaluate whether an AI-derived biomarker quantifying vascular architecture improves prognostic discrimination beyond TNM. Methods: Digital images of representative tumor tissue and oncologic outcomes were obtained from the CM914 trial (NCT03138512). Tumor tissue was annotated by a pathologist to define vessel branching patterns as high-branching (HB) or low-branching (LB) patterns. HB and LB patterns were quantified using a deep learning model as the tumor-wide fraction of HB in digitized H&E whole-slide images from 1098 of 1641 evaluable patients (67%). Based on the tumor-wide HB fraction, patients were classified into risk groups using a data-driven cut-off optimized for Kaplan–Meier survival analysis, allowing comparison with TNM-based risk strata (PT2a, G3/G4, N0, M0 / PT2b, G ANY, N0, M0 / PT3, G ANY, N0, M0; intermediate-high risk, n=1016 vs PT4, G ANY, N0, M0 / PT ANY, G ANY, N1, M0; high risk, n=82). Risk stratification performance for overall survival (OS) and disease-free survival (DFS) was compared between branching pattern–based and TNM-based risk groups using Cox regression, Kaplan–Meier analysis, and concordance indices. Results: Higher tumor-wide HB fractions were associated with improved DFS and OS, showing a continuous effect on prognosis. Each 10% increase in HB fraction reduced risk of recurrence and death (DFS HR 0.92, 95% CI 0.89–0.95; OS HR 0.89, 95% CI 0.83–0.95; both p<0.005). Kaplan–Meier analyses showed longest survival in the 4 th quartile (OS HR 0.39, 95% CI 0.21–0.71, p=0.002; DFS HR 0.47, 95% CI 0.35–0.64, p<0.0001). This threshold defined the 75th percentile cutoff for risk stratification, separating patients into intermediate risk (≥74% HB; n=275) vs high risk (<74% HB; n=823) groups. High risk patients had an increased risk of recurrence and death (DFS HR 2.11, 95% CI 1.57–2.84, C-Index 0.57; OS HR 2.57, 95% CI 1.28–4.32, C-Index 0.56, both p<0.005). TNM-based risk showed lower prognostic discrimination (DFS: HR 1.95, 95% CI 1.39–2.74, C-Index 0.53; OS: HR 2.55, 95% CI 1.52–4.30, C-Index 0.55, both p<0.005). Combining branching-based risk with TNM improved DFS and OS discrimination compared with TNM alone (C-Index 0.59). Conclusions: Higher fractions of high-branching vessels are associated with improved DFS and OS, potentially enabling risk stratification beyond conventional TNM criteria in RCC. While prospective validation is required to confirm clinical applicability, these results suggest high-branching vessel architecture represents a promising prognostic feature.