Abstract 3584: Artificial intelligence-derived body composition measurements reveal sex- and metformin-specific predictors of overall survival in non-small cell lung cancer

Abstract Background: Body composition (BC) is emerging as a prognostic factor in cancer, with metformin showing potential anti-tumor effects in overweight/obese patients with non-small cell lung cancer (NSCLC). However, the sex-specific impact of BC on overall survival (OS), its underlying mechanisms, and relationship to metformin's beneficial effects remain unclear. Methods: A total of 1, 275 patients with NSCLC were included. Six AI-derived BC measurements were extracted from chest CT scans: muscle density and volume, visceral (VAT) and subcutaneous adipose tissue (SAT), VAT/SAT ratio, and intra-muscular adipose tissue (IMAT). Clinical data included age, sex, histology, stage, smoking status, diabetes, metformin use, and BMI. Multivariable Cox proportional hazard model was used to assess the associations with OS. Correlations between BC and eight circulating inflammatory cytokines were evaluated in a subset of 93 patients and BMI-related mutation profiles were compared in an independent cohort of 5, 363 patients with NSCLC. Results: Sex-specific distributions showed significantly higher muscle volume, density, VAT, and VAT/SAT ratio in males, while females had higher SAT and IMAT. Higher muscle volume is associated with prolonged OS (HR = 0.71, T3 vs. T1 and HR = 0.78, T2 vs. T1, P < 0.01). In females, higher muscle density was associated with prolonged OS (HR = 0.76, T3 vs. T1, P = 0.04), while higher IMAT correlated with shortened OS (HR = 1.39, T3 vs. T1, P = 0.01). In males, increased VAT and SAT tended to associate with prolonged OS. Higher SAT (HR = 0.48, 0.42 for T3 and T2 vs. T1, P = 0.02 and 0.01) and IMAT (HR = 0.52, 0.42 for T3 and T2 vs. T1, P = 0.04 and 0.01) predicted prolonged OS specifically in metformin users, but not in diabetic non-metformin users. Cytokine analysis showed positive correlations between VAT, SAT, IMAT and leptin in both sexes, while IMAT positively correlated with IL-8 and MCP-1, and muscle density negatively correlated with leptin and IL-8 in females. EGFR mutation was significantly less frequent and KRAS mutations more frequent in patients with BMI > 25, particularly in females. Mutation analysis revealed lower EGFR and higher KRAS mutation frequencies in patients with BMI > 25, particularly in females. Mutations in metabolism and immune response genes (STK11, NF1, ATM, and SMARCA4) showed reversed sex-specific trends related to BMI, suggesting obesity may drive distinct mutation selection and immune responses between sexes. Conclusions: AI-derived BC measurements revealed sex-specific associations with OS in NSCLC, supported by distinct patterns in circulating cytokines and somatic mutations. Metformin's beneficial effects were specifically associated with adipose tissue measurements, suggesting its anti-tumor activity may be mediated through metabolic and immunomodulatory pathways. Citation Format: Xinan Wang, Akinori Hata, Noriaki Wada, Yi Li, Mark Schiebler, Hiroto Hatabu, David Christiani. Artificial intelligence-derived body composition measurements reveal sex- and metformin-specific predictors of overall survival in non-small cell lung cancer [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 3584.