e21549 Background: PD1, CTLA4 and LAG3 blockade have shown remarkable efficacy in melanoma patients. Predictive biomarkers are needed to predict response, prevent unneeded toxicity and reduce costs. Despite collective efforts, no biomarker is robust or accessible enough for routine use. Our objective was to evaluate mathematical modeling-based kinetic markers derived from routine biological parameters as early predictors of durable benefit for first-line immunotherapy in advanced melanoma. Our secondary objective was to assess if these modeling could predict progression-free survival (PFS). Methods: We retrospectively included all consecutive, treatment-naïve, metastatic or unresectable melanoma patients treated with PD1 blockade alone or in combination with CTLA4 or LAG3 blockade between April 2020 and 2022 in our department. Clinical, biological and radiological data were collected. The primary endpoint was durable clinical benefit (DCB) defined as partial or complete response at 6 months or stable disease of at least 6 months. 11 simple blood markers and their on-treatment kinetics were considered. 4 empirical mathematical models (constant, linear, double exponential and hyperbolic) were fitted to the data, allowing to reject the constant null hypothesis for all of them. The association of the best-models coefficients with DCB and PFS was assessed using logistic and Cox regression. To do so, the data was truncated at 1, 2 and 3 months and the model coefficients were retrieved using Bayesian estimation with a population prior identified on the full-kinetics dataset. This study was approved by our local IRB. Results: 61 patients were included with 24 months median follow-up. DCB was observed for 57% of them. A low model-based value at baseline using 1 month truncated data was significantly associated with DCB for neutrophils (OR = 4.54[1.69-12.5], p < 0.0001) and - neutrophils/lymphocytes ratio (NLR) (OR = 3.03 [1.47-6.67], p = 0.003). Model-based results increased AUC versus baseline raw data: 0.755 versus 0.699 for neutrophils and 0.76 versus 0.7 for NLR. An elevated model-based at baseline using 1 month truncated data was significantly associated with PFS for neutrophils (HR = 1.56 [1.25-1.96], p < 0.0001), AST (HR = 1.71 [1.24-2.35], p < 0.0001), LDH (HR = 2.0 [1.44-2.77], p < 0.0001), CPR (HR = 1.75 [1.14-2.67], p = 0.01) and ALT (HR = 1.37 [1.05-1.78], p = 0.02). At 1 month, the kinetics parameters on creatinine (HR = 1.38 [1.05-1.82], p = 0.02), albumin (HR = 0.68 [0.48-0.97], p = 0.03) and monocytes (HR = 1.49 [1.02-2.18], p = 0.04) were significantly associated with PFS. Conclusions: Modeling the kinetics of routine biological parameters shows potential ability to predict DCB. Perspectives include model improvement, kinetics machine learning integrative model development as well as a toxicity prediction.