Development and Interpretation of a Machine Learning-Based Predictive Model Using Clinical Parameters for Eosinophilic Chronic Rhinosinusitis With Nasal Polyps

ObjectiveWe aim to predict eosinophilic chronic rhinosinusitis with nasal polyps (ECRSwNP) employing preoperative clinical parameters and machine learning algorithms, evaluating, and selecting the optimal model.MethodsRetrospective collection of preoperative clinical parameters from 331 individuals suffering from chronic rhinosinusitis with nasal polyps. Independent predictors for ECRSwNP were determined through the application of the least absolute shrinkage and selection operator in conjunction with multivariate logistic regression. Four ML models for classification were constructed and cross-validated through the training set (223 patients), with predictive performance further evaluated on the testing set (98 patients). Shapley additive explanations (SHAP) technology provides importance rankings for predictive variables and explains personalized predictions from optimal models. Net reclassification improvement and integrated discrimination improvement values assessed the improvement in predictive performance. The prognostic value of the model further validated through Kaplan-Meier analysis.ResultsPeripheral eosinophil percentage, visual analog scale, ethmoid/maxillary sinuses ratio, and nasal polyps score were identified as key variables for model construction. Extreme gradient boosting (XGBoost) proved to be the superior prediction model, attaining an area under the receiver operating characteristic curve of 0.981 in the training cohort and 0.928 in the testing cohort. The model demonstrated optimal net benefit across various risk thresholds and effectively predicted postoperative recurrence in patients with ECRSwNP.ConclusionWe constructed an XGBoost predictive model and visualized its interpretation using the SHAP method, providing a reference for preoperative noninvasive assessment of ECRSwNP patients and guiding individualized treatment.