Development and Validation of Unplanned Extubation Prediction Models Using Intensive Care Unit Data: Retrospective,...

Area of Science:

• Critical Care Medicine
• Health Informatics
• Machine Learning in Healthcare

Background:

• Patient safety in the intensive care unit (ICU) is paramount, with unplanned extubation (UE) being a significant adverse event.
• Preventing and detecting UE is crucial for quality care but remains challenging.

Purpose of the Study:

• To develop and validate machine learning models for predicting UE in ICU patients.
• To identify key predictors and assess the clinical utility of these models.

Main Methods:

• Retrospective study of 6914 extubation cases in an academic tertiary hospital (2010-2018).
• Development of UE prediction models using Random Forest (RF), Logistic Regression (LR), Artificial Neural Network (ANN), and Support Vector Machine (SVM).
• Performance evaluation using Area Under the Receiver Operating Characteristic Curve (AUROC), sensitivity, specificity, calibration curves, Brier score, and Integrated Calibration Index (ICI).

Main Results:

• Out of 6914 cases, 248 (3.6%) experienced UE.
• The Random Forest (RF) model achieved the highest AUROC of 0.787.
• UE was associated with male gender, physical restraints, fewer surgeries, and night shifts, with higher reintubation and mortality rates.

Conclusions:

• Machine learning models, particularly RF, can effectively predict UE in ICU patients using electronic health record data.
• The developed RF model demonstrates good calibration and potential clinical usefulness with widely available variables.
• This predictive tool may help reduce the incidence of UE and improve patient safety in ICUs.