PD784 - SYNERGISTIC INTEGRATION OF LIPIDOMIC PROFILING AND BODY COMPOSITION PATTERNS VIA AI FOR PRECISION RISK STRATIFICATION IN ADVANCED NSCLC TREATED WITH ICIS

PD784

SYNERGISTIC INTEGRATION OF LIPIDOMIC PROFILING AND BODY COMPOSITION PATTERNS VIA AI FOR PRECISION RISK STRATIFICATION IN ADVANCED NSCLC TREATED WITH ICIS

G. Mentrasti^1,*, M. Ferrara², N. Chiodi³, A. Sbrollini⁴, S. Villani¹, S. Lunetti², B. Marozzi², L. Angeli Temperoni², E. N. Serritelli³, T. Galassi¹, A. Lancianese¹, R. Marchitelli¹, V. Agostinelli¹, E. Ambrosini³, F. Pecci⁵, F. Bianchi³, A. Parisi¹, M. B. L. Rocchi⁶, A. Vignini⁷, P. A. Corsetto⁸, L. Burattini⁴, M. Taus², R. Berardi^1,3 on behalf of Funded by The Italian Ministry of University and Research through the Programmes of Relevant National Interest (PRIN 2022) (PNRR–NextGenerationEU) funds. Project Code 2022J343TP_001

¹Clinical Oncology, ²Dietology and Clinical Nutrition, Azienda Ospedaliero-Universitaria delle Marche, ³Department of Clinical and Molecular Sciences, ⁴Department of Information Engineering, Università Politecnica delle Marche, Ancona, ⁵Department of Medicine and Surgery, Università degli Studi di Parma, Parma, ⁶Department of Biomolecular Sciences, Università degli Studi di Urbino Carlo Bo, Urbino, ⁷Department of Clinical Sciences, Section of Biochemistry, Biology and Physics, Università Politecnica delle Marche, Ancona, ⁸Department of Pharmacological and Biomolecular Sciences, Università di Milano, Milano, Italy

Rationale: Immune checkpoint inhibitors (ICIs) have led to a clear survival gain in NSCLC, but 50% of patients (pts) still fail to respond. Lipid metabolism and body composition (BC) contribute to immune modulation. With a comprehensive nutritional assessment, we aim to identify a lipidomic and BC signature using an artificial intelligence(AI)-driven approach.

Methods: From March 2025, we prospectively enrolled advanced NSCLC treated with ICIs +/- chemotherapy. Circulating lipids, adipokines and BIA-assessed BC were evaluated at baseline (V0) and at 9 ± 3 weeks (V1) after ICIs start. Data were correlated to ORR. AI investigated the association between patterns and ICIs response. Categorical variables were compared using Chi-square test, non-parametric with Wilcoxon test. Due to the high dimensionality and small sample, a LOSO cross-validation strategy was adopted. More than 20 machine learning models were implemented and re-evaluated through PCA.

Results: 23 pts were enrolled, median age was 70, 39,1% (N=9) were overweight/obese. BMI was higher in non-responders (NRs) vs responders (Rs) (p=0.03). At V0 Rs showed a significant association with hypercholesterolemia and statin use (p=0.01), at V1 Rs had higher HDL, but lower LDL cholesterol and triglycerides (p<0.001) vs NRs. Among saturated fatty acids, at V0 Rs had significantly lower blood concentration of stearic acid (C18:0) (p=0.03). For adipokines, after ICIs Rs tended to have higher plasma adiponectin levels compared with NRs (p=0.06). Regarding BC at V0, Rs were significantly associated with higher fat-mass/height (kg/m)(p<0.001), NRs were characterized by lower free-fat mass/height (kg/m)(p<0.001).

Conclusion: The AI modeling supports the clinical utility of a multidimensional integration of circulating lipidomic and BC patterns into a patient-tailored nutritional intervention to impact ICIs response.

Disclosure of Interest: None declared