Using machine learning for the personalised prediction of revision endoscopic sinus surgery

Area of Science:

• Otolaryngology
• Medical Informatics
• Machine Learning in Medicine

Background:

• Revision endoscopic sinus surgery (ESS) is considered for chronic rhinosinusitis (CRS) when conservative treatments and initial ESS are insufficient.
• Predicting the need for revision ESS at an individual level remains challenging.
• This study investigates the accuracy of predicting revision ESS and identifies key individual risk factors.

Purpose of the Study:

• To evaluate the predictive accuracy of machine learning models for revision ESS.
• To identify and analyze the impact of individual-level risk factors for revision ESS.
• To enhance personalized treatment strategies for chronic rhinosinusitis patients.

Main Methods:

• Utilized electronic health records from 767 surgical CRS patients (≥16 years).
• Trained and validated machine learning models (logistic regression, gradient boosting, random forest) to predict revision ESS.
• Analyzed variable importance using Shapley values and partial dependence plots.

Main Results:

• Machine learning models demonstrated similar prediction accuracy (AUROC ~0.74) with data up to six months post-baseline ESS.
• Prediction performance improved with longer data collection periods (AUROC up to 0.784 at 12 months).
• Key predictors included: number of visits, time to baseline ESS, patient age, CRS with nasal polyps (CRSwNP), asthma, NSAID-exacerbated respiratory disease, and immunodeficiency.

Conclusions:

• Intelligent data analysis effectively predicts revision ESS at the individual level.
• Identified significant predictors: clinical visit frequency, patient age, Type 2 high diseases, and immunodeficiency.
• Findings support personalized risk assessment and management for CRS patients.