Skip to main content
CLP 2024 Annual Meeting Main banner
Final Program


CLP 2024 Program
Back

Brief Oral Papers

Neuropsychiatry

Brief Oral Papers Session 09 - Delirium and Neuropsychiatry

The Stanford-Ceribell Study: Feasibility and Accuracy of a Rapid Response Electroencephalogram (rrEEG) in the detection of Delirium Among Critically Ill Patients

Friday, November 8, 2024
1:35 PM – 2:50 PM EST
Location: Jamsine
Abstract:

Background: The electroencephalogram (EEG) is the gold-standard for clinical monitoring of cerebral electrophysiological activity. Previous studies have reported a relationship between delirium development and EEG slowing and reduced functional connectivity. This study aims to determine the diagnostic accuracy of an EEG-based diagnosis of delirium against a neuropsychiatric assessment based on DSM-5 criteria, as well as any differences in EEG signatures based on delirium severity or clinical phenotypes.

Methods:

This is a prospective, non-randomized, observational clinical proof-of-concept study. A portable EEG (i.e., Ceribell Rapid Response EEG System (rrEEG)), was utilized to measure the brainwaves of patients admitted to the cardiovascular intensive care unit to monitor the development of delirium. All subjects underwent multiple daily assessments of delirium, including rrEEG, CAM-ICU, Stanford-Proxy Test for Delirium, and a blinded diagnostic neuropsychiatric assessment based on DSM-5 criteria (determining both the presence, severity, and phenotype of delirium) by one of the study’s Critical Care Psychiatrists. The data collected will confirm the study’s hypothesis, while at the same time being utilized to retrain the EEG’s artificial intelligence (AI) algorithm. The original AI-model was trained on datasets using CAM-ICU results as ground truth. We are building a new algorithm, based on DSM-5 based assessment as ground truth in order to enhance the diagnostic accuracy and scope.

Results: As of this submission, we have enrolled 142 out of 200 subjects. By the time of the ACLP conference, we anticipate having completed both recruitment and data analysis. The analysis of the current partial sample is as follows:

From each EEG recording, we extracted the average variability in alpha power, alpha/delta power ratio, alpha/theta power ratio and normalized alpha power in the 30 minutes periods surrounding each DSM-5 diagnostic assessment. We then compared these measures against the results of the various delirium assessments for both positive and negative episodes based on DSM-5 criteria using Mann-Whitney U tests. Positive cases of delirium were correlated with lower variability in alpha power (p < 0.001, effect size r=-0.20), and higher alpha/delta power ratio (p=0.004, effect size r=0.122) when compared to non-delirious subjects.

Discussion:

Despite the availability of delirium screening tools, delirium is missed in 66-84% of patients (Maldonado, 2019). Our data suggests that rrEEG methods can accurately detect delirium in a population of critically ill patients, when compared to the gold-standard of a neuropsychiatric assessment. Given the abundance of our data collection, we are retraining the AI-algorithm for the detection of subsyndromal states, as well as the various delirium phenotypes, especially hypoactive states, often missed by conventional delirium assessment tools.

Conclusion:

Clinical EEG measurements have the potential to provide vital, objective diagnostic information regarding the development of delirium without the inherit biases associated with current bedside assessments (e.g., operator error, patient’s lack of cooperation). Portable EEG techniques have the promise of further detecting the presence of subsyndromal or pre-delirium states, as well as assessing the patient’s response to the introduction of various treatment modalities.

Sponsored by the Critical Care Psychiatry SIG.