[Development and validation of a machine learning-based prognostic model for chronic rhinosinusitis with nasal
Summary
Machine learning models can predict poor outcomes after surgery for chronic rhinosinusitis with nasal polyps (CRSwNP). The best model, glmBoost+GBM, identified key factors like polyps and age for risk stratification.
Area of Science:
- Otorhinolaryngology
- Medical Informatics
- Machine Learning
Background:
- Chronic rhinosinusitis with nasal polyps (CRSwNP) poses challenges for postoperative prognosis.
- Predictive models are needed to assess surgical outcomes in CRSwNP patients.
Purpose of the Study:
- To develop and compare machine learning (ML) models for predicting poor postoperative prognosis in CRSwNP patients.
- To evaluate the clinical utility of these predictive models.
Main Methods:
- Developed 113 ML models using 12 feature selection algorithms on data from 333 CRSwNP patients.
- Classified patients into favorable and poor prognosis groups based on 1-year follow-up data.
- Assessed model performance using Area Under the Curve (AUC) and validated in an independent test set.
Main Results:
- The incidence of poor postoperative prognosis was 36.34%.
- The optimal model, a generalized linear model boosting plus gradient boosting machine (glmBoost+GBM) ensemble, achieved an AUC of 0.894 in the training set and 0.813 in the test set.
- Key predictors included olfactory cleft polyps, age, eosinophil counts, neutrophil counts, Visual Analogue Scale (VAS) scores, and olfactory impairment.
Conclusions:
- The glmBoost+GBM ensemble model effectively predicts postoperative prognosis in CRSwNP patients.
- This interpretable model offers a reliable tool for clinical risk stratification.
- The findings support the clinical application value of ML in managing CRSwNP.