Daily Automated Prediction of Delirium Risk in Hospitalized Patients: Model Development and Validation

Area of Science:

• Medical informatics
• Artificial intelligence in healthcare
• Clinical prediction models

Background:

• Delirium is a frequent complication in hospitalized patients, associated with poor outcomes.
• Underdiagnosis of delirium is common due to insufficient screening resources.
• Effective and scalable delirium screening methods are needed.

Purpose of the Study:

• To develop an automated machine learning algorithm for daily delirium risk identification.
• To leverage electronic medical record data for high-risk patient detection.
• To reduce barriers to widespread delirium screening.

Main Methods:

• Developed and compared four machine learning models (logistic regression, multilayer perceptrons, random forests, boosted trees).
• Utilized a retrospective dataset of 23,006 adult inpatients with Confusion Assessment Method (CAM) screens.
• Input features included demographics, laboratory values, vital signs, prior CAM screens, and medications.

Main Results:

• The boosted tree model demonstrated the highest predictive power with an AUROC of 0.92.
• Random forest and multilayer perceptron models also showed strong performance.
• Performance varied when predicting delirium in patients with no prior history.

Conclusions:

• A boosted tree machine learning model effectively identifies hospitalized patients at high risk for delirium within 24 hours.
• This automated approach facilitates targeted delirium prevention strategies.
• Enables precise application of interventions to mitigate delirium risks.