D1.460 - Diagnostic Opportunities in Type 2 Inflammations - A Machine-Learning Approach: EAACI Task Force Update

Therapeutic options for type 2 chronic rhinosinusitis with nasal polyps have improved significantly in recent years, particularly with the approval of several biologics. However, it is often difficult to identify the most appropriate targeted therapy for individual patients. A machine-learning program could help physicians decide on the best personalized management strategy. The aim of this EAACI Task Force is to establish a machine-learning prediction model to recommend targeted biologic therapy and support physicians in selecting the most appropriate biological treatment.

We designed a self-supervised machine-learning experimental model using published data from randomized controlled trials investigating dupilumab, mepolizumab, tezepelumab, and omalizumab in chronic rhinosinusitis with nasal polyps. Clinical and laboratory markers, including patient-reported outcome scores, nasal polyp scores, and blood eosinophil counts before and 24 weeks after initiation of biologic therapy, were included. The model was trained using data from the published randomized controlled trials.

Three machine-learning approaches (XGBoost, Random Forest, and a Generalized Linear Model model) were implemented to predict clinical markers at 3 and 6 months following initiation of biologic therapy. Missing data were addressed using imputation techniques, and feature selection analysis was performed to identify the most informative predictors. Based on these models, an open-source online tool was developed to enable individualized prediction of treatment response; model validation is currently ongoing.

We developed an open-source, research-use online tool capable of predicting SNOT-22 and nasal polyp scores at 3 and 6 months prior to initiation of biologic therapy. This approach represents a step toward precision medicine in chronic rhinosinusitis with nasal polyps and may support physicians in individualized treatment selection. By facilitating more informed decision-making, the tool has the potential to reduce unnecessary biologic switching and optimize patient outcomes.