Development of Machine Learning Prediction Models for Self-Extubation After Delirium Using Emergency Department Data

Area of Science:

• Emergency Medicine
• Artificial Intelligence in Healthcare
• Patient Safety

Background:

• Delirium is a frequent condition in emergency departments.
• Patients with delirium face risks such as self-extubation of medical devices.
• Predicting delirium and its complications is essential for timely intervention.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting delirium occurrence.
• To create a model for predicting self-extubation in patients who have delirium.
• To differentiate predictors for delirium versus self-extubation using interpretable AI.

Main Methods:

• Utilized machine learning algorithms to build two predictive models.
• Employed Shapley Additive Explanation (SHAP) for model interpretability and predictor visualization.
• Assessed the discriminative performance of the developed models.

Main Results:

• Both prediction models demonstrated high discriminative performance.
• Identified distinct predictors for delirium onset compared to self-extubation post-delirium.
• SHAP analysis provided insights into the factors influencing each prediction.

Conclusions:

• Machine learning models can effectively identify patients at high risk of delirium and self-extubation.
• Understanding distinct predictors supports tailored clinical decision-making.
• Data-driven approaches can improve patient care and reduce iatrogenic suffering.