Using artificial intelligence to predict prolonged mechanical ventilation and tracheostomy placement.

Area of Science:

• Critical care medicine
• Medical informatics
• Artificial intelligence in healthcare

Background:

• Early identification of patients requiring prolonged mechanical ventilation (PMV) is challenging.
• Predicting the need for tracheostomy placement in intensive care unit (ICU) patients is also difficult.

Purpose of the Study:

• To develop and evaluate machine learning classifiers for early identification of patients at risk for PMV.
• To identify patients requiring tracheostomy placement using machine learning.

Main Methods:

• Utilized the Multiparameter Intelligent Monitoring in Intensive Care III database.
• Developed gradient-boosted decision tree classifiers using ICU admission data, including severity-of-illness scores and comorbidities.
• Defined PMV as mechanical ventilation exceeding 7 days.

Main Results:

• Identified 20,262 ICU stays; 13.6% required PMV and 6.6% underwent tracheostomy.
• Achieved high predictive accuracy for PMV (AUC 0.820) and tracheostomy (AUC 0.830).
• Models showed enhanced performance in surgical ICUs (AUCs 0.852 for PMV, 0.869 for tracheostomy).

Conclusions:

• Demonstrated the efficacy of AI-driven machine learning for early risk identification of PMV and tracheostomy.
• Early identification facilitates timely interventions, potentially improving patient outcomes.
• Logistic organ dysfunction score pulmonary component and cardiac arrhythmia were key predictors.