ICU EEG Monitoring
Continuous EEG, NCSz detection, and qEEG trends
Continuous EEG (cEEG) is the intensive-care instrument that revealed a hidden epidemic. Before its routine use, the working assumption was that seizures in critically ill patients would declare themselves with convulsions; cEEG demonstrated instead that a large fraction of ICU seizures are nonconvulsive - electrographic events without overt motor manifestations - and that they are common, recurrent, and associated with secondary neuronal injury. The clinical consequence is profound: in a comatose or pharmacologically paralyzed patient, the only window onto ongoing seizure activity is the EEG, and a single 20-to-30-minute routine study is grossly inadequate to exclude it. This module covers when to monitor, how electrographic seizures and the patterns adjacent to them are defined, what cEEG finds, and how the quantitative trends that make hours of data reviewable are read and, all too easily, misread.
Indications and the question of duration
The consensus indications for cEEG converge on a single principle: monitor when the probability of nonconvulsive seizures is meaningful and the clinical examination cannot reveal them. The highest-yield populations are patients with persistently altered consciousness after convulsive status epilepticus, a substantial minority of whom remain in nonconvulsive status despite the cessation of visible movements; patients with acute brain injury and unexplained depressed or fluctuating mental status, including subarachnoid and intraparenchymal hemorrhage, traumatic brain injury, ischemic stroke, central nervous system infection, and hypoxic-ischemic injury; patients with clinical paroxysmal events of uncertain nature such as posturing, ocular deviation, or autonomic spells; and patients undergoing titration of anesthetic therapy for refractory status epilepticus or for burst-suppression-targeted intracranial pressure control. Each indication shares the same logic: a meaningful pretest probability of seizures combined with an examination that cannot detect them.
Duration is dictated by detection yield, which rises with recording time but with diminishing returns. A practical and widely cited synthesis holds that roughly half of nonconvulsive seizures are captured within the first hour of recording, but reaching the 80-to-95 percent range requires approximately 24 hours in non-comatose patients and up to 48 hours in comatose patients, because seizures in deep coma cluster later and more sparsely. The single most useful early predictor of who will go on to seize is the presence of an interictal epileptiform abnormality, especially a periodic or rhythmic pattern, in the opening hour; its appearance is itself a risk marker that justifies committing to prolonged recording. The corollary is the central practical message of this module: a normal 30-minute routine EEG does not exclude nonconvulsive status in a high-risk patient; it merely fails to capture it, and acting as though absence of evidence were evidence of absence is a recurrent and consequential error.
Capturing nonconvulsive seizures is a joint function of whom you monitor and for how long. About 50 percent of patients who will seize do so within the first hour, but adequate sensitivity demands roughly 24 hours of recording in awake patients and up to 48 hours in comatose patients, in whom seizures cluster later. Practically, the appearance of periodic or rhythmic patterns on the ictal-interictal continuum within the first hour is the strongest single trigger to commit to prolonged recording rather than stopping after a routine study.
Why nonconvulsive seizures matter
The argument for aggressive detection is mechanistic, not merely diagnostic. Sustained seizure activity drives a mismatch between cerebral metabolic demand and supply: it sharply increases the cerebral metabolic rate of oxygen and glucose while frequently compromising delivery, it elevates extracellular glutamate and intracellular calcium toward excitotoxic ranges, and in the already-injured brain it raises intracranial pressure and worsens the ischemic penumbra. Invasive multimodal-monitoring and cerebral microdialysis studies have linked nonconvulsive seizures and intense ictal-interictal-continuum activity to elevated lactate-pyruvate ratios, a biochemical signature of metabolic crisis, and seizure burden - the cumulative time spent in seizure - has been associated in a graded fashion with hippocampal injury and worse functional outcomes. Nonconvulsive status epilepticus is therefore not a benign electrographic curiosity but an ongoing, potentially treatable secondary insult, which is exactly what justifies the cost and labor of continuous monitoring.
Recognition rests on the ACNS definitions of electrographic seizures and on the broader concept of the ictal-interictal continuum. An electrographic seizure is defined as epileptiform discharges averaging greater than 2.5 Hz for at least 10 seconds, or any rhythmic or periodic pattern with definite evolution in frequency, morphology, or location lasting at least 10 seconds. Electrographic status epilepticus is an electrographic seizure persisting for at least 10 continuous minutes, or recurring for a cumulative total of at least 20 percent of any 60-minute epoch of recording. Patterns that do not meet these thresholds but raise concern - generalized periodic discharges, lateralized periodic discharges, lateralized and generalized rhythmic delta activity, and the same patterns carrying plus modifiers - occupy the continuum. The empirical justification for worrying about the continuum is concrete: the probability of recorded seizures rises as periodic discharges exceed roughly 2 Hz and as plus-modifiers such as superimposed fast or rhythmic activity appear, so these features escalate the level of concern and may warrant a diagnostic anti-seizure-drug trial to determine whether treating the pattern improves the patient.
The Salzburg consensus criteria operationalize the diagnosis of nonconvulsive status epilepticus for the ambiguous patient. In broad terms, epileptiform discharges faster than 2.5 Hz, or discharges with typical spatiotemporal evolution, in a patient with impaired consciousness meet the criteria directly; slower or non-evolving periodic patterns require additional evidence such as subtle clinical correlates or unequivocal EEG and clinical improvement after a benzodiazepine to qualify. The criteria do not replace clinical judgment, but they discipline it by forcing an explicit accounting of frequency, evolution, and treatment response rather than a gestalt impression.
Quantitative EEG: making hours reviewable
No human can stare at days of raw EEG in real time, and quantitative EEG (qEEG) compresses long records into trends that a clinician at the workstation or a nurse at the bedside can scan at a glance. The foundational transform is the fast Fourier transform, which decomposes each short epoch of signal into its constituent frequency components. From it derive the two most widely used displays. The color density spectral array (CDSA), also called the compressed spectral array, plots time on the horizontal axis, frequency on the vertical axis, and spectral power as color, so that a seizure - a sudden increase in power that frequently climbs in frequency as it evolves - appears as a vertical flame or arrowhead rising out of the background. The amplitude-integrated EEG (aEEG) plots a smoothed, semilogarithmically compressed amplitude envelope over time; seizures appear as abrupt upward deflections in the lower and upper margins of the band, and this trend is especially entrenched in neonatal neurointensive care.
| qEEG trend | What it displays | Seizure signature | Main pitfall |
|---|---|---|---|
| Color density spectral array | Power across frequency over time, as color | Vertical flame rising in power and frequency | Rhythmic artifact mimics a flame |
| Amplitude-integrated EEG | Smoothed amplitude envelope over time | Abrupt rise in band margins | Smoothing hides brief or focal seizures |
| Rhythmicity / seizure-probability trend | Detector estimate of rhythmic-pattern likelihood | Sustained high-probability band | Periodic discharges trigger false positives |
| Suppression ratio | Percent of epoch that is suppressed | Falls when activity replaces suppression | Sedation and artifact shift the value |
| Asymmetry / relative alpha | Interhemispheric power difference over time | Focal change or declining alpha | Electrode problems mimic asymmetry |
Modern review systems add rhythmicity spectrograms and proprietary seizure-probability detectors that apply pattern-recognition algorithms to flag rhythmic, evolving activity, alongside asymmetry, relative alpha variability, and suppression-ratio displays useful for detecting ischemia and for titrating burst-suppression. Used well, these trends let an expert reviewer triage which segments of raw EEG to open first and let bedside staff alert the team to an evolving flame between formal reads, dramatically increasing the practical reach of a limited number of electroencephalographers. Their proper role, however, is screening and trend-tracking - drawing the eye to a suspicious epoch - and not standalone diagnosis. Studies of non-expert qEEG seizure detection report good but clearly imperfect sensitivity and specificity, which is exactly why the workflow must close the loop on the raw tracing.
Every qEEG abnormality must be confirmed against the raw EEG, because the failure modes are systematic and predictable. Rhythmic artifacts - chest physiotherapy, ventilator cycling, patting, bed percussion, and rhythmic movements - produce CDSA flames that are visually indistinguishable from seizures. Smoothing in aEEG hides brief, focal, or low-amplitude seizures and generates false-negative reassurance. Periodic discharges on the ictal-interictal continuum drive seizure-probability detectors into false positives because the detectors key on rhythmicity rather than on evolution. The trend points to a segment; the raw waveform, read for evolution, field, and physiologic plausibility, makes the diagnosis.
Beyond seizure detection, cEEG and qEEG serve as a continuous brain monitor in coma, exploiting the EEG's exquisite sensitivity to cerebral blood flow. In aneurysmal subarachnoid hemorrhage, a declining relative alpha variability and a worsening interhemispheric asymmetry can precede the clinical onset of delayed cerebral ischemia by hours, offering a monitoring window that intermittent imaging cannot provide and allowing intervention before infarction is established. In titrated anesthetic coma, the suppression ratio and burst-suppression metrics let the team target a defined depth - a chosen suppression ratio or interburst interval - rather than guessing from intermittent spot checks. In each application the principle is identical: qEEG turns the EEG into a continuous trend that tracks the brain's state over time, with raw confirmation always closing the loop, so that the trend triages attention and the waveform renders the verdict.
Two structural limitations deserve emphasis so that monitoring is deployed honestly. First, cEEG is resource-intensive, demanding electrodes that must be maintained for days, technologists, secure data handling, and expert review, so its indications should be driven by pretest probability rather than reflex. Second, the ictal-interictal continuum remains genuinely uncertain territory: for many continuum patterns it is still not established whether treating the EEG improves the patient, and the honest answer at the bedside is often a carefully monitored therapeutic trial with a prespecified endpoint rather than a confident pronouncement. Acknowledging this uncertainty is not a weakness of the method but a faithful description of where the evidence currently stands, and it protects patients from both undertreatment of true status and overtreatment of patterns whose causal role is unproven.
1. A comatose patient after convulsive status epilepticus has a normal 30-minute routine EEG. What is the most appropriate next step regarding nonconvulsive seizure detection?
2. A bedside nurse reports a vertical flame on the color density spectral array. The raw EEG over that epoch shows rhythmic muscle and movement artifact time-locked to chest physiotherapy. The correct conclusion is:
3. Which finding most increases the probability that a periodic pattern on the ictal-interictal continuum will be associated with recorded electrographic seizures?