Machine learning-based delirium prediction in surgical in-patients: a prospective validation study

Area of Science:

• Clinical Informatics
• Artificial Intelligence in Medicine
• Surgical Patient Care

Background:

• Delirium is a common, severe complication in hospitalized patients, often preventable.
• Identifying at-risk patients is challenging due to clinical workload and limitations of current screening tools.

Purpose of the Study:

• To validate a machine learning (ML)-based delirium prediction tool for surgical in-patients.
• To assess the tool's performance in real-time using existing electronic health record (EHR) data.

Main Methods:

• Prospective validation study involving 738 surgical in-patients across vascular, trauma, and orthopedic departments.
• Delirium screening using the DOS scale twice daily.
• Real-time delirium risk prediction by an ML algorithm using EHR data at admission and evening of admission.

Main Results:

• 103 patients (14.0%) screened positive for delirium.
• The ML algorithm correctly identified 85 (82.5%) of patients with delirium.
• The algorithm achieved an Area Under the Receiver Operating Characteristic curve (AUROC) of 0.883, indicating high discriminative performance.

Conclusions:

• The ML-based delirium prediction tool demonstrated high discriminative performance in surgical patients.
• This technology has the potential to replace time-intensive screening methods.
• Future implementation could lead to more efficient delirium prevention strategies.