Development and Validation of a Machine Learning-Based Risk Prediction Model for Delirium in Older Inpatients: A...

Area of Science:

• Geriatric Medicine
• Artificial Intelligence in Healthcare
• Clinical Informatics

Background:

• Delirium is a common complication in older inpatients, associated with adverse outcomes.
• Early identification and intervention are crucial for managing delirium.
• Existing prediction methods may lack accuracy and efficiency.

Purpose of the Study:

• To develop and validate a machine learning (ML)-based risk prediction model for delirium in older inpatients.
• To identify key clinical features predictive of delirium.
• To compare the performance of different ML algorithms for delirium risk prediction.

Main Methods:

• A prospective cohort study involving 973 older inpatients.
• Collection of 18 clinical features from electronic medical records.
• Development and validation of four ML models, including Random Forest.
• Performance evaluation using AUC, accuracy, sensitivity, and Brier score.

Main Results:

• The Random Forest model achieved an AUC of 0.908, accuracy of 0.935, and sensitivity of 0.992.
• ML models demonstrated high predictive performance in both training and test sets.
• The model effectively identified patients at risk of delirium.

Conclusions:

• The developed ML model shows excellent predictive performance for delirium in older inpatients.
• This tool can assist healthcare professionals in early diagnosis and informed clinical decision-making.
• Early identification facilitates preventive measures and personalized care strategies.