Machine Learning-Based Prediction of Subsyndromal Delirium in Intensive Care Patients

Area of Science:

• Critical Care Medicine
• Artificial Intelligence in Healthcare
• Predictive Analytics

Background:

• Subsyndromal delirium is a frequent complication in Intensive Care Unit (ICU) patients.
• It is associated with prolonged hospital stays, increased healthcare costs, and poorer patient prognosis.
• Developing methods for early intervention is crucial for improving patient outcomes.

Purpose of the Study:

• To develop and validate a machine learning-based model for predicting the risk of subsyndromal delirium in ICU patients.
• To identify key predictors of subsyndromal delirium in this patient population.

Main Methods:

• A prospective cohort study involving 447 ICU patients.
• Identification of eight independent predictors using least absolute shrinkage and selection operator (LASSO) and multivariate logistic regression.
• Construction and validation of four machine learning models and one logistic regression model.

Main Results:

• Subsyndromal delirium was present in 20.1% of ICU patients.
• Key predictors included sleep questionnaire scores, sedative use, restraint use, renal replacement therapy/ECMO, intra-abdominal pressure, prealbumin levels, alcohol history, and stroke.
• The random forest model demonstrated the best performance with an AUC of 0.885.

Conclusions:

• Machine learning models, particularly random forest, show promise in predicting subsyndromal delirium risk in ICU patients.
• Early screening using validated models can facilitate individualized intervention strategies.
• Implementing these models may lead to improved patient prognoses and reduced healthcare expenditures.