Personalised artificial intelligence–based approach to assessing objective response to immunotherapy in patients with solid tumors.

e13640 Background: Predicting response to immune checkpoint inhibitors is a crucial need in real-world clinical practice that has not yet been met. Various approaches exist to predict objective response or the development of immune-mediated adverse events mostly based on clinical or molecular biomarkers. Machine learning approaches are a novel way to incorporate clinical data sets into predictive model development. Aim: to develop a predictive model to assess immune checkpoint inhibitors efficacy based on machine-learning approach in patients with solid tumors. Methods: a retrospective study included 1001 patients who were treated between 2018 and 2022. Patient cohort included 585 (58.5%) men and 415 (41.5%) women aged 21 to 91 years (mean age 62.37±12.0 years) with various malignancies: cutaneous melanoma (n = 316), renal cancer (n = 247), gastrointestinal cancer (n = 96), non-small cell lung cancer (n = 247), small cell lung cancer (n = 33), cervical cancer (n = 62)] receiving immune checkpoint inhibitor therapy. Overall response rate (ORR) to treatment was defined as partial or complete tumor regression using iRECIST criteria. Python 3 packages, including scikit-learn and Catboost, were employed to develop machine-learning models, seaborn and matplotlib for data visualization, SMOTE for dataset balancing and SHapley Additive exPlanations (SHAP) for interpreting black-box results. Results: the most significant predictors of achieving ORR were body mass index (F score 472.0), age (F score 365.0) and creatinine (F score 185.0), platelet (F score 178.0) and haemoglobin (F score 163.0) levels. Less significant predictors of objective response to immunotherapy were pre-treatment ALT (F score 132.0), AST (F score 126.0) and WBC (F score 116.0) levels, as well as immunotherapy line (F score 126.0) and T stage of disease (F score 120.0). A model based on these factors reflects the achievement of an objective response by the patient in 80% of cases. Conclusions: In patients with solid malignancies undergoing immune checkpoint inhibitor therapy, body mass index, age, and levels of creatinine, platelets, and hemoglobin are the most significant predictors of achieving an ORR. These findings definitively highlight the potential of integrating routinely available clinical data into predictive models to enhance the personalisation of immunotherapy strategies.