Abstract PR010: Exploratory analysis of tumor tertiary lymphoid structures using a novel artificial intelligence–based approach in patients with muscle-invasive urothelial carcinoma from the CheckMate 274 trial

Abstract Introduction: The phase 3 randomized CheckMate 274 trial in patients (pts) with high-risk muscle-invasive urothelial carcinoma (MIUC) demonstrated improved disease-free survival (DFS) with nivolumab (NIVO) vs placebo (PBO) in intent-to-treat (ITT) and tumor programmed death ligand 1 expression ≥1% pts (primary endpoints). With extended follow-up, continued efficacy was observed, and immature overall survival favored NIVO in both populations. Extending from previous exploratory tumor biomarker analyses (Necchi et al. ESMO 2022), we conducted a post hoc analysis of tertiary lymphoid structures (TLS), which are ectopic lymphoid structures often found within the solid tumor microenvironment. A subset of TLS has been associated with improved outcomes to neoadjuvant immunotherapy in UC. Using artificial intelligence (AI) –based algorithms, we explored the biomarker utility of TLS from CheckMate 274. Methods: We used a reproducible AI-based algorithm to identify and classify TLS in situ using hematoxylin and eosin (H 90. 1%), with a higher incidence of LA (237/272; 87. 1%) compared with mTLS (89/272; 32. 7%). There was heterogeneity of TLS within the tumors, with most tumors demonstrating more than 1 form of TLS (198/272; 72. 8%). In the PBO arm, median DFS was modest for most TLS categories (5. 7–8. 3 months) but was substantially improved for mTLS+ tumors (22. 1 months; 95% CI, 5. 4–not estimable NE) ; n=41), representing a 2. 7- to 3. 9-fold increase relative to any other category. In contrast, treatment with NIVO resulted in 2. 1- to 4. 0-fold improvements in median DFS relative to PBO across all but the mTLS+ category (22. 9 months; 95% CI, 11. 9–NE; n=48), which was similar to the PBO arm. The largest increases in median DFS (95% CI) for NIVO vs PBO were observed for LA+ (27. 6 16. 1–NE; n=114 vs 8. 3 5. 6–11. 9; n=123 months; HR, 0. 56; 95% CI, 0. 39–0. 80; P=0. 0015), iTLS+ (27. 0 16. 1–NE; n=91 vs 8. 2 5. 4–21. 2; n=105 months; HR, 0. 61; 95% CI, 0. 41–0. 91; P=0. 0143), and mTLS− tumors (24. 6 11. 4–NE; n=81 vs 6. 1 5. 2–8. 5; n=102 months; HR, 0. 53; 95% CI, 0. 36–0. 78; P=0. 0012]). Conclusion: These results suggest that a novel AI-based TLS maturity classification using H 2024 May 17-20; Charlotte, NC. Philadelphia (PA): AACR; Clin Cancer Res 2024;30 (10Suppl): Abstract nr PR010.