Toward personalized prediction: a multicenter machine learning model for omalizumab response duration in moderate-to-severe perennial allergic rhinitis

Background: Omalizumab effectively improves quality of life in patients with moderate-to-severe perennial allergic rhinitis (PAR) uncontrolled by conventional medications. However, the duration of its efficacy remains unclear, and there is a lack of effective tools for individualized prediction. Objective: This study aimed to identify predictors of Omalizumab duration of efficacy in moderate-to-severe PAR patients, then develop and validate an interpretable, machine learning-based predictive model to forecast the duration of efficacy following treatment. Methods: (D2), comorbid conditions (asthma, urticaria, conjunctivitis, atopic dermatitis), and injection frequency. Univariate and multivariate Cox regression were employed to investigate independent predictors of Omalizumab efficacy, with restricted cubic splines for dose-response analysis. Five survival machine learning models (CPH, XGBoost, RSF, SSVM, CoxBoost) were constructed and compared by comprehensive metrics. The optimal model was interpreted using the SHapley Additive exPlanations (SHAP) and deployed as an online web application. Results: Univariate Cox regression identified age, D1, D2, asthma and injection frequency as independent factors affecting Omalizumab efficacy duration. Multivariate analysis did not confirm D2 as significant. Dose-response analysis demonstrated enhanced protective effects beyond four injections. Among the five models, RSF demonstrated robust predictive performance. SHAP analysis identified injection frequency, age, D1, D2, and coexisting asthma as the most critical factors. Conclusion: This study developed and validated a machine learning-based model capable of forecasting the duration of Omalizumab efficacy in moderate-to-severe PAR based on readily available clinical variables.