Background: Deep neck infection (DNI) may rapidly lead to airway compromise. Although some patients require immediate intubation, others

initially appear stable but later deteriorate. This study developed and validated a CatBoost-based machine learning (ML) model to predict delayed intubation in initially stable DNI patients.

Methods: We retrospectively analyzed 413 DNI patients without immediate intubation and assessed late intubation within 24–72 hours. Clinical and radiologic variables were incorporated into a supervised CatBoost model with five-fold cross-validation. Two models were built: one using all features and another using the top 14 predictors selected by mean feature importance. Performance was evaluated using AUC, accuracy, F1 score, precision, recall, and specificity.

Results: Delayed intubation occurred in 75 patients (18.15%). The all-feature model achieved a mean AUC of 0.8839 and accuracy of 0.9056.

The 14-feature model showed improved classification balance, with a mean AUC of 0.8721, accuracy of 0.9250, F1 score of 0.7621, precision of 0.8785, recall of 0.6800, and specificity of 0.9793.

Conclusions: The CatBoost-based ML model reliably predicted delayed intubation in initially stable DNI patients. Feature reduction improved

performance balance and stability, and the 14- feature model may aid early identification of patients at risk of airway deterioration.

Shih-Lung Chen is a board-certified otolaryngologist specializing in head and neck surgery, committed to integrating clinical excellence with translational research. His clinical expertise includes sudden sensorineural hearing loss, tinnitus management, deep neck infections, and complex airway conditions. He has extensive experience in managing infection-related complications and acute airway deterioration in adult patients. Dr. Chen’s research centers on the application of machine learning in clinical medicine. He develops and validates predictive models that integrate clinical and radiological data to improve risk stratification and support timely intervention. His work on predicting delayed intubation in patients with deep neck infection aims to enhance patient safety and optimize clinical decision-making. Dedicated to evidence-based practice and interdisciplinary collaboration, Dr. Chen emphasizes data-driven approaches while maintaining a patient-centered philosophy. He is also involved in medical education and community outreach, promoting awareness of hearing health and otologic disorders.