Predicting Agitation-Sedation Levels in Intensive Care Unit Patients: Development of an Ensemble Model.

Area of Science:

• Intensive care medicine
• Artificial intelligence in healthcare
• Machine learning applications

Background:

• Agitation and sedation management are crucial for patient safety in intensive care units (ICUs).
• Current nursing assessments for agitation and sedation are subjective and infrequent.
• Automating these assessments can significantly improve ICU efficiency and patient care.

Purpose of the Study:

• To develop a machine learning-based system for automated assessment of agitation and sedation levels.
• To compare different ensemble learning models for classifying agitation and sedation.
• To utilize interpretable AI (SHAP) for model explanation.

Main Methods:

• An ensemble learning model was developed using data from a hospital's ICU database (2020).
• The model classified agitation and sedation levels using 20 features and over 121,000 data points.
• SHAP (Shapley additive explanations) was used for interpretable AI analysis.

Main Results:

• The random forest model achieved high AUC values (sedation: 0.97, agitation: 0.88).
• The ensemble learning model improved agitation sensitivity to 0.82 while maintaining high AUC (>0.82).
• Model explanations derived from SHAP analysis were consistent with clinical expertise.

Conclusions:

• Machine learning-driven automation of ICU agitation-sedation assessment enhances efficiency and safety.
• Ensemble learning offers improved sensitivity for agitation detection without compromising accuracy.
• Future integration with real-time monitoring holds promise for advancing critical care.