Abstract 6836: Analysis of 81 plasma cytokine levels in a phase II trial of gastrointestinal cancer patients treated with metronomic chemotherapy: Incorporation of artificial intelligence and mathematical modeling

Abstract Metronomic, or continuous low-dose, chemotherapy can inhibit tumor growth via inhibition angiogenesis and immune system activation. At AACR 2024, we reported multiplex analysis of 75 cytokines in 34 out of 38 patients with gastrointestinal (GI) tumors, who were evaluated in a phase II clinical trial of metronomic UFT (a 5-fluorouracil prodrug; 100 mg/twice per day p.o.) and cyclophosphamide (an i.v. bolus on day 1 of 500mg/mq2 and then 50 mg/day p.o.) plus celecoxib (200 mg/twice a day p.o.); the COMET trial (CLINICAL ID NCT02926911; Allegrini et al. Angiogenesis 2012: 15(2):275-86). We now present analysis, including Artificial Intelligence (AI) input (Claude AI built by Anthropic, and Gemini AI built by Google), and mathematical modeling, of plasma levels of 81 cytokines in 34 of the 38 patients. Our analysis included baseline (day 0), days 28 and 56, and days 84 and 112 of the trial, and correlated specific cytokines levels with Overall Survival (OS) and Progression Free Survival (PFS). We now report previosly not evaluated potential biomarkers for this trial, including soluble Her-3 (sHER3). We used AI as confirmatory analysis of data elaboration, firstly performed by R and Stata packages. Building on our previous report of plasma levels of TRAIL and MCP-4 cytokines showing positive correlations with OS, we now report that sHER3 showed an inverse relationship with OS. Patients with sHER3 values ≤185.4 pg/ml presented better prognosis (median OS: 12.1 months) compared to the group with higher levels (median OS: 5.8 months). With regards to PFS, patients with sHER3 levels below the cut-off of 185pg/ml had a better PFS (5.8 months), compared to those above the cut-off (2.3 months; p=0.005). These results add to our previous reporting that TRAIL and MCP-4 levels had a direct association with PFS. As a proof of concept multifaceted analytical approach, we also built mathematical modeling assisted by AI. This approach developed a mathematical model for PFS prediction based on three interacting components: tumor dynamics (by using RECIST measurements), relative pro-angiogenic activity (using baseline VEGF as proxy), and immune response (using baseline IL2 as proxy). AI helped us to optimize the parameters of Ordinary Differential Equations and to predict PFS. Our approach could allow the stratification of GI patients for future metronomic chemotherapy trials, and the potential identification of patients that should instead be offered alternative therapies. Limitations of the present study are that all methods are calibrated on retrospective single-center data. Additional validation will be needed to confirm the robustness of our model. Citation Format: Maria L. Manca, Briana E. Pinales, Juan C. Silva Espinoza, Ryley D. Stewart, Diana L. Prospero, Mariana N. Frias, Mia E. Montelongo, Hector M. Padilla, Claire E. Perucho, Brianna E. Ortega, Valeria V. Lopez, Paola Orlandi, Arianna Bandini, Mariana J. Silva, Marta Banchi, Vladimir Georgiev, German Rosas-Acosta, Kazumasa Fujiwara, Alessandro Michelangeli, Guido Bocci, Giacomo Allegrini, Giulio Francia. Analysis of 81 plasma cytokine levels in a phase II trial of gastrointestinal cancer patients treated with metronomic chemotherapy: Incorporation of artificial intelligence and mathematical modeling [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 6836.