Predictive Modeling of Surgically Recalcitrant Chronic Rhinosinusitis Using Integrated Feature Sets

Area of Science:

• Immunology
• Otorhinolaryngology
• Computational Biology

Background:

• Chronic rhinosinusitis (CRS) is a complex inflammatory condition with varied treatment responses.
• Accurate patient stratification for revision surgery is challenging due to disease heterogeneity.

Purpose of the Study:

• To apply machine learning (ML) to clinical and inflammatory data for classifying CRS patients undergoing revision surgery.
• To identify which data types are most effective for predicting recalcitrant CRS.

Main Methods:

• Utilized data from 634 CRS patients, categorizing features into histopathology, cytokines/chemokines, demographics, comorbidities, and medications.
• Employed random forest (RF) classifiers with leave-one-out cross-validation, training on individual and combined feature sets.

Main Results:

• For CRS without nasal polyps (CRSsNP), cytokine/chemokine markers showed the highest predictive accuracy.
• For CRS with nasal polyps (CRSwNP), combining cytokine/chemokine markers with eosinophil/neutrophil markers yielded the best performance.
• Integrating multiple feature sets significantly improved the overall classification model performance.

Conclusions:

• Multimodal data integration substantially enhances predictive modeling for surgical history in CRS patients.
• These findings underscore the value of combining diverse data for improved ML model accuracy in CRS management.
• The study provides a foundation for developing more precise clinical ML tools for CRS.