Predicting Tracheostomy Need on Admission to the Intensive Care Unit-A Multicenter Machine Learning Analysis

Area of Science:

• Critical Care Medicine
• Machine Learning Applications
• Respiratory Therapy

Background:

• Predicting tracheostomy need in mechanically ventilated patients is challenging, leading to potential complications like subglottic stenosis.
• Early identification of patients requiring tracheostomy can prevent unnecessary spontaneous breathing trials and reduce healthcare costs.

Purpose of the Study:

• To develop and evaluate a machine learning tool for predicting tracheostomy requirement at intensive care unit (ICU) admission.
• To identify key clinical features associated with tracheostomy need in critically ill patients.

Main Methods:

• A retrospective cohort study utilizing the eICU Collaborative Research Database (eICU-CRD).
• Inclusion of adult patients (age >18) requiring mechanical ventilation.
• Development and comparison of logistic regression, random forest, and Extreme Gradient Boosting (XGBoost) models for binary classification of tracheostomy.

Main Results:

• The XGBoost model demonstrated strong performance (AUROC 0.794) in predicting tracheostomy among 38,508 mechanically ventilated patients.
• A reduced feature set (20 features) for the XGBoost model minimally impacted performance (AUROC 0.778).
• Key predictors included pneumonia/sepsis diagnosis and chronic respiratory failure comorbidity.

Conclusions:

• The developed machine learning model accurately predicts the likelihood of tracheostomy in ICU patients upon admission.
• Prospective validation could enable earlier interventions, potentially reducing prolonged ventilation and its complications.