Predicting ICU Delirium in Critically Ill COVID-19 Patients Using Demographic, Clinical, and Laboratory Admission...

Area of Science:

• Critical Care Medicine
• Data Science in Healthcare
• Infectious Diseases

Background:

• Delirium is a frequent, underdiagnosed issue in intensive care units (ICUs), linked to worse patient outcomes.
• Predicting delirium early is challenging due to its complex causes and variable presentation.
• Identifying high-risk patients at ICU admission can enable timely interventions.

Purpose of the Study:

• To evaluate machine learning models for predicting delirium in critically ill patients with SARS-CoV-2 infection.
• To assess the performance of interpretable models for early delirium risk stratification.

Main Methods:

• A prospective cohort of 426 critically ill SARS-CoV-2 patients was analyzed.
• Five machine learning models (Logistic Regression, SVM, Decision Tree, Random Forest, Naïve Bayes) were trained using 112 features.
• Model performance was evaluated using 10-fold cross-validation and feature selection via Information Gain.

Main Results:

• The Naïve Bayes model demonstrated moderate predictive ability (AUC 0.717, accuracy 65.3%).
• Key predictors identified included mechanical ventilation, benzodiazepine sedation, SARS-CoV-2 diagnosis, ECMO, constipation, and male sex.
• The model provided high interpretability for risk factors.

Conclusions:

• Interpretable machine learning models can effectively stratify delirium risk in critically ill COVID-19 patients.
• Routinely available ICU admission data is sufficient for developing predictive models.
• Early identification of high-risk patients can guide targeted delirium management strategies.