Development and Interpretation of a Machine Learning-Based Predictive Model Using Clinical Parameters for...

Area of Science:

• Otolaryngology
• Medical Informatics
• Computational Biology

Background:

• Chronic rhinosinusitis with nasal polyps (CRSwNP) presents diagnostic challenges.
• Differentiating eosinophilic CRSwNP (ECRSwNP) preoperatively is crucial for effective treatment.
• Current noninvasive methods for predicting ECRSwNP subtypes have limitations.

Purpose of the Study:

• To develop and validate a machine learning model for predicting ECRSwNP using preoperative clinical data.
• To identify key clinical predictors of ECRSwNP.
• To evaluate the performance of different machine learning algorithms for this prediction task.

Main Methods:

• Retrospective analysis of preoperative clinical data from 331 CRSwNP patients.
• Utilized least absolute shrinkage and selection operator (LASSO) and multivariate logistic regression to identify predictors.
• Developed and cross-validated four machine learning models, including XGBoost, with SHAP for interpretation.

Main Results:

• Peripheral eosinophil percentage, visual analog scale, ethmoid/maxillary sinuses ratio, and nasal polyps score were identified as significant predictors.
• The XGBoost model achieved high predictive accuracy (AUC 0.981 training, 0.928 testing).
• The model effectively predicted postoperative recurrence and demonstrated clinical utility across risk thresholds.

Conclusions:

• An interpretable XGBoost model can accurately predict ECRSwNP preoperatively.
• This model provides a valuable tool for noninvasive patient assessment and guiding individualized treatment plans.
• SHAP analysis enhances understanding of the model's predictions and identified key clinical drivers.