Clinical Utility of Machine Learning Methods Using Regression Models for Diagnosing Eosinophilic Chronic...

Area of Science:

• Allergy and Immunology
• Computational Biology
• Otorhinolaryngology

Background:

• Eosinophilic chronic rhinosinusitis (ECRS) diagnosis relies on histological assessment.
• Blood eosinophil levels are potential biomarkers, but their predictive utility for ECRS requires further clarification.
• Machine learning (ML) models may enhance the prediction of histological eosinophil counts from peripheral blood eosinophil levels.

Purpose of the Study:

• To compare statistical approaches for diagnosing ECRS.
• To investigate the utility of ML methods in predicting histological eosinophil counts for ECRS diagnosis.
• To determine optimal blood eosinophil cutoffs for ECRS detection.

Main Methods:

• Retrospective analysis of data from 264 chronic rhinosinusitis patients.
• Utilized regression models to identify factors influencing histopathological eosinophil counts.
• Employed receiver operating characteristic (ROC) curves and ML-based regression models to establish blood eosinophil cutoffs.

Main Results:

• Blood eosinophil levels, nasal polyp presence, and asthma were significant predictors of histopathological eosinophil counts.
• ML methods provided distinct blood eosinophil cutoffs for ECRS diagnosis based on nasal polyp status.
• Identified cutoffs for ECRS with nasal polyps (≥1% or ≥100/μL) and without (≥6% or ≥400/μL).

Conclusions:

• ML-derived blood eosinophil cutoffs show promise for pre-biopsy ECRS suspicion.
• ML models offer advantages in covariate adjustment, overfitting mitigation, and predicting histological eosinophil counts.
• These findings support the use of ML in refining ECRS diagnostic strategies.