Neurology · Point-of-care EEG delirium screening and monitoring
Ceribell Delirium Monitor System
Ceribell, Inc.
Described by its manufacturer as the first device the FDA has cleared for delirium screening and monitoring — a December 2025 clearance that turns Ceribell's widely deployed point-of-care EEG headband into a continuous bedside check on the brain. The 510(k) summary reports 81 per cent sensitivity and 81 per cent specificity against clinician DSM-5 diagnosis in 225 critical-care adults; the peer-reviewed record for the cleared algorithm itself is still to come.
Performance Metrics
Clinical Evidence
The evidence base divides into the regulatory record and an independent literature that predates the clearance. The K251936 510(k) summary (section 7.2) reports the pivotal validation: EEG recorded from 225 adults aged 22 or older in critical-care environments (ages 24 to 92, mean 62.4; 60 per cent male; 49 per cent aged 65 or over) on previously cleared Ceribell hardware, with clinician DSM-5 delirium assessment up to three times a day as the reference standard, the data collected prospectively and analysed retrospectively (a sponsor data-collection study is registered as NCT04962815). Against 857 clinician assessments the algorithm reached [[sensitivity]] of 81 per cent (95 per cent CI 76 to 87), [[specificity]] of 81 per cent (78 to 84), AUROC 0.87 (0.83 to 0.89), positive predictive value 0.57 and negative predictive value 0.93; the pre-set performance goals were a lower confidence bound of at least 70 per cent for both sensitivity and specificity. Test-retest repeatability across 614 paired 30-minute segments gave a Cohen's kappa of 0.76 (0.71 to 0.80). These figures sit in the regulatory summary; no peer-reviewed publication of the cleared algorithm's performance has been located. The independent literature establishes the principle the device rests on: delirium has a measurable EEG signature, dominated by diffuse slowing. Williams Roberson and colleagues at Vanderbilt (Clinical Neurophysiology, 2023) enrolled 28 mechanically ventilated ICU patients for 24-hour continuous EEG and found delta and theta activity and reduced alpha power independently identified delirium or coma — explicitly framing limited- montage devices as a pragmatic route to delirium monitoring. A separate line of pilot work by Mulkey and colleagues used Ceribell's own limited-lead hardware in critically ill older adults: a gamma-band machine-learning pilot (Research in Nursing & Health, 2022; n=13) reached around 70 per cent predictive accuracy, and a vision-transformer analysis of the same 13-patient rapid-response EEG dataset (Scientific Reports, 2023) reported roughly 97 per cent testing accuracy. These are proofs of concept on related hardware, not validations of the cleared algorithm; the honest reading is a plausible mechanism with regulatory validation reported at 81 per cent sensitivity and specificity and a peer-reviewed record that has not yet caught up.
| Study | Design | n | Sensitivity | Specificity | AUC | Published |
|---|---|---|---|---|---|---|
| Williams Roberson S, et al. (independent, Vanderbilt CIBS Center) | ProspectiveProspective | 28 | — | — | — | Clin Neurophysiol 2023;146:40–48 (PMID 36529066). 24-hour continuous EEG in mechanically ventilated ICU patients (25 completed); delta variability, high-beta variability, relative alpha and theta power independently identified delirium or coma. Foundational qEEG evidence, not a study of the cleared algorithm. |
| Mulkey MA, Albanese T, Kim S, Huang H, Yang B (independent) | ProspectiveProspective | 13 | — | — | — | Res Nurs Health 2022;45(6):652–663 (PMID 36321335). Pilot machine-learning study of gamma-band features from a limited-lead rapid-response EEG device in ventilated adults ≥50 y; predictive accuracy ≥70% across three models. |
| Mulkey MA, Huang H, Albanese T, Kim S, Yang B (independent) | ProspectiveProspective | 13 | — | — | — | Sci Rep 2023;13:7890 (PMID 37193736). Vision-transformer deep-learning analysis of limited-lead rapid-response EEG for delirium prediction, using the same 13 ICU patients (7 delirious) as the Res Nurs Health pilot; ~97% testing accuracy in that small research cohort. |
| Ceribell, Inc. (K251936 pivotal validation; manufacturer-sponsored) | RetrospectiveRetrospective | 225 | 81% (95% CI 76–87) | 81% (95% CI 78–84) | 0.87 (95% CI 0.83–0.89) | FDA 510(k) summary K251936 §7.2 (December 2025), https://www.accessdata.fda.gov/cdrh_docs/pdf25/K251936.pdf; prospectively collected EEG from 225 critical-care adults (ages 24–92, mean 62.4; 60% male) analysed retrospectively against clinician DSM-5 delirium diagnosis up to 3×/day (857 assessments); PPV 0.57, NPV 0.93, Cohen's κ 0.76 across 614 paired segments; performance goals lower-bound CI ≥70%. Not peer-reviewed |
Inside the algorithm
Editorial featureHow the Delirium Monitor reaches a notification.
Five stages — from raw input to verdict — drawn from manufacturer documentation and the public regulatory record.
- INGEST
- NORMALISE
- DETECT
- LOCALISE
- VERDICT
Stage 01 · INGEST
A headband EEG, applied at the bedside.
The input is electroencephalogram signal from Ceribell's point-of-care system: a limited-montage headband and portable recorder designed to be applied by any trained clinician in minutes, without an EEG technologist. The delirium module takes the same continuous signal the platform already records for seizure monitoring in emergency departments and intensive care units.
Input
Continuous point-of-care EEG
Inference
Continuous, per EEG segment
Inside the Auris+ Listing
Five more sections complete this device’s Auris+ Listing.
Decision Ledger
ProPro unlocks a private, cross-vendor log of every case you read on this device — what the AI called, what you concluded, and a one-line reason if you overrode it. Ready for the EU AI Act's deployer logging obligations when they land in 2028.
Clinical Evidence Deep Dive
ProPro unlocks the structured clinical-evidence summary — study count, target patient population, and a tabular accuracy-metrics view drawn from peer-reviewed sources.
Peer-Reviewed Publications
ProPro unlocks the curated peer-reviewed publication list with PubMed cross-links — the citation backbone of every editorial verdict.
Post-Market & Regulatory Conditions
ProPro unlocks the post-market surveillance summary, recall record, and the conditions of approval that bound real-world use.
AI Algorithm Version History
ProPro unlocks the chronological record of algorithm version changes — what changed when, drawn from manufacturer changelogs and regulatory filings.
Regulatory Approvals
Safety Record
The post-market record is eight months old and correspondingly thin. No recall, FDA safety communication or adverse-event report specific to the Delirium Monitor System was located in indexed public sources as of August 2026; MAUDE could not be queried directly from this environment, so that is "none found in public reporting" rather than an exhaustive audit. The platform's cleared role bounds the risk: the device notifies clinicians of EEG patterns associated with delirium and does not diagnose, and the labelled indication keeps the confirmatory clinical assessment with the care team.
Intended Use & Indications
Per the FDA 510(k) record (K251936), the Ceribell Delirium Monitor System is intended to analyse features associated with diffuse slowing electroencephalogram (EEG) patterns that may be indicative of delirium, and is intended to aid in the screening and monitoring of delirium with clinical assessments in adult patients in critical care settings within hospitals. The system analyses discrete segments of EEG and notifies clinicians when EEG patterns associated with delirium are detected while the patient is monitored. It is an adjunct to clinical assessment, not a stand-alone diagnostic: delirium remains a clinical diagnosis, and the device's role is to surface an objective neurophysiological signal that bedside screening tools are known to miss.