JMIR Publications

ABSTRACT

Background:

Delirium in intensive care units (ICUs) poses a significant challenge and affects not only global patient outcomes but also healthcare efficiency. The development of an accurate, real-time prediction model for delirium represents a crucial advancement in critical care and addresses the need for timely intervention and resource optimization in ICUs worldwide.

Objective:

This study aimed to create a novel machine-learning model for real-time delirium prediction in ICUs using the random forest method.

Methods:

Distinct from existing approaches, the model integrated routinely available clinical data such as age, sex, and patient monitoring device outputs to ensure its practicality and adaptability in diverse clinical settings. Using these data, we trained a random forest model to predict the occurrence of delirium in patients. Retrospective data were used for training and internal validation. Retrospective data were used for training and internal validation. Prospective data were used to confirm the reliability of the delirium determination. CAM-ICU records assessed by ICU nurses were collected and validated by qualified investigators performing CAM-ICU measurements prospectively on the same patients and then determining Cohen's kappa coefficient. In addition, we additionally verified the performance of the model using a temporal validation cohort and performed external validation using data from an external hospital.

Results:

The Kappa coefficient between labels generated by ICU nurses and prospectively verified by qualified researchers was 0.81. This indicates that the recorded CAM-ICU results were reliable. The model showed robust performance in internal validation (area under the receiver operating characteristic curve [AUROC]: 0.82, area under the precision–recall curve [AUPRC]: 0.62) and maintained its accuracy in temporal validation (AUROC: 0.73, AUPRC: 0.85), confirming its reliability over time. External validation across various patient populations and time frames further confirmed its effectiveness (AUROC: 0.84, AUPRC: 0.77).

Conclusions:

Our model represents a significant breakthrough in the management of delirium in ICUs and offers a real-time, data-driven approach for improving patient care. The proven accuracy and adaptability of this model in various clinical scenarios underscore its potential to substantially improve patient outcomes and operational efficiency in ICUs. The integration of this model into current healthcare practices may lead to major advancements in early delirium detection and treatment, thereby reducing the ICU stay and improving the recovery rate.