Related Experiment Video
Updated: Aug 6, 2026
A Machine Learning Approach to Design an Efficient Selective Screening of Mild Cognitive Impairment
Published on: January 11, 2020
A research algorithm to improve detection of delirium in the intensive care unit
Margaret A Pisani1, Katy L B Araujo, Peter H Van Ness
1Department of Internal Medicine, Pulmonary & Critical Care Section, and the Program on Aging, Yale University School of Medicine, New Haven, CT 06520-8057, USA. Margaret.Pisani@yale.edu
Introduction:
Delirium is a serious and prevalent problem in intensive care units (ICU). The purpose of this study was to develop a research algorithm to enhance detection of delirium in critically ill ICU patients using chart review to complement a validated clinical delirium instrument.
Methods:
Prospective cohort study of 178 patients 60 years and older admitted to the Medical ICU. The Confusion Assessment Method for the Intensive Care Unit (CAM-ICU) and a validated chart review method for delirium were performed daily. We assessed the diagnostic accuracy of the chart-based delirium method using the CAM-ICU as the gold standard. We then used an algorithm to detect delirium first using the CAM-ICU ratings, then chart review when the CAM-ICU was unavailable.
Results:
When using both the CAM-ICU and the chart-based review the prevalence of delirium was 143/178 (80%) patients or 929/1457 (64%) of patient-days. Of these, 292 patient-days were classified as delirium by the CAM-ICU, and the remainder (n=637 patient-days) were classified as delirium by the validated chart review method when the CAM-ICU was missing due to weekends or holidays (404 patient-days), when CAM-ICU was not performed due to stupor or coma (205 patient-days), and when the CAM-ICU was negative (28 patient-days). Sensitivity of the chart-based method was 64% and specificity was 85%. Overall agreement between chart and the CAM-ICU was 72%.
Conclusions:
Eight of 10 patients in this cohort study developed delirium in the ICU. Although use of a validated delirium instrument with frequent direct observations is recommended for clinical care, this approach may not always be feasible, especially in a research setting. The algorithm proposed here comprises a more comprehensive method for delirium detection in a research setting taking into account the fluctuation that occurs with delirium, a key component to accurately determining delirium status. Improving delirium detection is of paramount importance first to advance delirium research and, subsequently to enhance clinical care and patient safety.
