Delirium screening in an acute care setting with a machine learning classifier based on routinely collected nursing...

Area of Science:

• Medical Informatics
• Clinical Prediction Models
• Geriatric Medicine

Background:

• Delirium screening in acute care is resource-intensive and prone to protocol deviations.
• A predictive model using daily nursing data could broaden screening coverage.

Purpose of the Study:

• To develop and validate machine learning models for delirium prediction using only nursing data.
• To assess the feasibility of integrating predictive models into existing data collection processes.

Main Methods:

• Utilized a dataset of 29,967 adult patients hospitalized for over 24 hours.
• Developed and tested machine learning models, including a gradient boosting machine.
• Assessed model performance using the area under the receiver operating characteristic curve (AUC).

Main Results:

• The gradient boosting machine model achieved the highest performance with an AUC of 0.933.
• Models demonstrated reliable prediction of patients at risk for delirium.
• The study included a large cohort with extensive nursing data.

Conclusions:

• Machine learning models based on structured nursing data can effectively predict delirium risk.
• These models offer a potential solution to reduce resource requirements for delirium screening.
• Integrating predictive models can enhance the efficiency of delirium detection in clinical practice.