Incidence of nonconvulsive seizures in different populations of critically ill children and adults. The confidence intervals were not reported by the studies, but were calculated based on the number of subjects in the study and the proportion of patients in whom nonconvulsive seizures were detected. Data is derived from (a) Abend NS, et al. Electrographic seizures in pediatric ICU patients: cohort study of risk factors and mortality. Neurology 2013; 81:383–391. (b) Abend NS, et al. Electroencephalographic monitoring during hypothermia after pediatric cardiac arrest. Neurology 2009; 72:1931–1940. (c) Arndt DH, et al. Subclinical early posttraumatic seizures detected by continuous EEG monitoring in a consecutive pediatric cohort. Epilepsia 2013; 54(10):1780–1788. (d) Carrera E, et al. Continuous electroencephalographic monitoring in critically ill patients with CNS infections. Arch Neurol 2008; 65 (12):1612–1618. (e) Claassen J, et al. Detection of electrographic seizures with continuous EEG monitoring in critically ill patients: Neurology 2004; 62:1743–1748. (f) Claassen J, et al. Electrographic seizures and periodic discharges after intracerebral hemorrhage. Neurology 2007; 69:1356–1365. (g) Crepeau AZ, et al. Value analysis of continuous EEG in patients during therapeutic hypothermia after cardiac arrest. Resuscitation 2014 (85):785–789. (h) Gilmore EJ, et al. Acute brain failure in severe sepsis: A prospective study in the medical intensive care unit utilizing continuous EEG monitoring. Intensive Care Med 2015; 41(4):686–694. (i) Mani R, et al. The frequency and timing of epileptiform activity on continuous electroencephalography in comatose post-cardiac- arrest syndrome patients treated with therapeutic hypothermia. Resuscitation 2012 (83):840–847. (j) O’connor KL, et al. High risk for seizures following subarachnoid hemorrhage regardless of referral bias. Neurocrit Care 2014; 21:476–482. (k) O’neill BR, et al. Incidence of seizures on continuous EEG monitoring following traumatic brain injury in children. J Neurosurg Pediatr 2015; 16: 167–176. (l) Oddo M, et al. Continuous Electroencephalography in the medical intensive care unit. Crit Care Med 2009; 37 (6): 2051–2056. (m) Payne ET, et al. Seizure Burden is independently associated with short-term outcome in critically ill children. Brain 2014; 137: 1429–1438. (n) Ronne-Engstrom E , Winkler T. Continuous EEG monitoring in patients with traumatic brain injury reveals high incidence of epileptiform activity. Arch Neurol Scand 2006; 114: 47–53. (o) Schreiber JM, et al. Continuous video EEG monitoring for patients with acute encephalopathy in a pediatric intensive care unit. Neurocrit Care 2012; 17:31–38. (p) Topjian AA, et al. Electrographic status epilepticus is associated with mortality and worse short-term outcome in critically ill children. Crit Care Med 2013; 41 (1):210–213. (q) Vespa PM, et al. Nonconvulsive seizures after traumatic brain injury are associated with hippocampal atrophy. Neurology 2010; 75 (9):792–798. (r) Vespa PM, et al. Acute seizures after intracerebral hemorrhage: A factor in progressive midline shift and outcome. Neurology 2003; 60:1441–1446. (s) Westover B, et al. The probability of seizures during EEG monitoring in critically ill adults. Clinical Neurophysiology 2015; 126:463–471 (Published with kind permission from © Lawrence J. Hirsch, MD 2016. All Rights Reserved)
Nonconvulsive status epilepticus. This is the EEG from a 29-year-old man with a history of liver transplantation and chronic immunosuppression who presented with convulsive status epilepticus due to encephalitis. He was treated with intravenous anticonvulsants and movements ceased, but he remained comatose. His EEG demonstrated electrographic seizure activity without clinical correlate. Low-frequency filter (LFF) = 1 Hz, high-frequency filter (HFF) = 70 Hz, notch off (Published with kind permission from © Lawrence J. Hirsch, MD 2013. All Rights Reserved)
How to Monitor
Obtaining high-quality cEEG recordings in the ICU is a challenge. Adequate technologist coverage is necessary to connect patients promptly, including off hours, and maintain those connections 24 h/day. Critically ill patients are frequently repositioned and transported to tests, which makes maintaining electrode integrity difficult. In both of our centers, we often employ collodion to secure disk electrodes and check the electrodes twice daily, usually supplemented by keeping the live recordings visible remotely to see which patients require electrode maintenance. Newer electrodes, such as subdermal wires, which may be more secure and lead to less skin breakdown, may be appropriate for comatose patients who are expected to undergo cEEG for many days to weeks . While these electrodes may take more time to apply, they require less maintenance and are MRI and CT compatible (both safe and not affecting image interpretation), thereby saving substantial technologist time. Concerns for image artifacts and patient safety make it necessary to remove and then reapply standard disk electrodes when patients undergo brain MRIs, but there has been some progress in creating practical MRI- and CT-compatible electrodes , including conductive plastic electrodes. Figure 3.3 shows CT and MRI images taken with these electrodes in place displaying minimal image artifact and CT images in a different patient with considerable image artifact caused by the electrodes. MRI-compatible disposable plastic cup electrodes are now commercially available and can be used in the ICU setting to minimize risk of transmitting infections .
CT- and MRI-compatible EEG electrodes. (a) A CT “scout” image demonstrating the placement of a full montage of conductive plastic electrodes (white arrows) on the scalp of a patient undergoing cEEG monitoring. (b) An axial image from the CT scan from the same patient demonstrates minimal artifact due to the electrodes (white arrows). Note there is no evidence of “streaking” common to head CT with conventional electrodes. (c) Axial FLAIR MRI performed 3 days later with electrodes in place. Note only minimal image artifact near the scalp (white arrows). (d) Axial CT in a different patient with multicompartmental intracranial hemorrhage showing beam hardening CT artifact caused by intact standard metal electrodes (blue arrows) (Modified with kind permission from © Lawrence J. Hirsch, MD 2013. All Rights Reserved)
There are numerous sources of artifact in the ICU environment that make cEEG challenging. Some are easily identified and filtered out such as 60 Hz (or 50 Hz in Europe) line noise from nearby electrical equipment. Others, however, such as pacemaker artifact, chest percussion, vibrating beds, ventilator activity, and intravenous drips, may be difficult to distinguish from seizures or other rhythmic or periodic cerebral activities [18–20] (Fig. 3.4). Simultaneous video recording is useful for distinguishing brain signals from artifact, especially rhythmic patterns such as those seen with chest percussion. In addition, video recording helps correlate EEG patterns with patient behaviors. In some cases, periodic EEG patterns can be determined to be ictal if they are time-locked to subtle patient movements . In addition, some significant EEG patterns in the critically ill appear after the patient is stimulated, which is easily determined by reviewing the video [22, 23] (see below).
Common ICU EEG artifacts. (a) Rhythmic bitemporal artifact due to chest compression in a medical ICU patient (arrow). (b) Respirator artifact due to fluid collecting in the tubing (arrow). These patterns are easily recognized on simultaneous video recording as they are synchronized with respirations. (c) Left temporal rhythmic waveforms (arrow) due to patting in an infant. This pattern is sometimes easy to confuse with seizures without video as it often shows a physiological field with evolution in frequency and amplitude. (d) Semirhythmic right temporal artifact (black arrow) due to chest percussion mimicking right LPDs or potentially ictal activity in a patient with true left hemisphere LPDs (white arrows). (e) Right occipital 6 Hz rhythmic artifact (arrow) due to automatic bed oscillation. (f) Rhythmic 1–1.5 Hz artifact (arrow) due to chewing in an edentulous patient. LFF = 1 Hz, HFF = 70 Hz, notch off (Published with kind permission from © Lawrence J. Hirsch, MD 2013. All Rights Reserved)
The number of electrodes used in cEEG studies varies considerably. In both our centers, we typically perform “full electrode” recordings using 16 or more active electrodes in addition to one or two reference electrodes and cardiac leads. Other authors have used reduced electrode configurations . The advantage of a reduced electrode system is that it is faster to apply and easier to maintain. It is also easier to work around other neuro-monitoring devices, surgical wounds, or ventricular drains common in neuro-ICU patients. However, a full electrode configuration improves the ability to distinguish brain signals from artifact, aids in spatial localization of pathological activity, and provides a safety factor in case one or more leads fail, including allowing qEEG calculations and alarms to continue to function adequately . In addition, reduced electrode methods, especially when coupled to qEEG tools, may miss clinically significant events. For instance, Shellhaas et al.  found that neonatologists evaluating amplitude-integrated EEG (aEEG) using only two electrodes for seizure detection, a technique employed in purpose-built devices common in neonatal ICUs, detected only 12–38% of seizures identified using conventional electrode arrangements. Although emergent below-the-hairline EEG recordings have only moderate sensitivities and specificities , they are almost certainly better than no EEG at all; a full EEG should be done when possible to confirm or refute the results. Several disposable headpieces including pre-gelled electrodes are now available and can be utilized, allowing fast application by ICU nurses, house staff, and other staff not fully trained in EEG electrode application .
Several days of cEEG generates gigabytes of data that, in its raw form, is time consuming for a neurophysiologist to review, especially if many patients are being monitored simultaneously. Furthermore, the raw EEG may be difficult for non-experts, such as ICU physicians and nurses, to interpret at the bedside. Therefore, concerning electrographic events may not be noticed until several hours later, when the neurophysiologist reviews the file, unless real-time remote monitoring is performed continuously (currently available only in a minority of academic centers). Computing advances have enabled the use of qEEG algorithms to reduce the data and provide graphical representation of significant patterns and trends to speed review. Some of the commonly employed qEEG methods are discussed below (see  for a detailed review).
Many qEEG data reduction and trending tools are based on transforming the raw cEEG into a time-frequency series using algorithms such as short-time Fourier transform or continuous wavelet transform. Several hours of cEEG recordings can be reduced to a single screen of time-frequency values using a compressed spectral array or density spectral array. The time-frequency data can be averaged over scalp regions or hemispheres to further reduce the data. Using these techniques, the abrupt changes in cEEG spectral power in a relatively narrow frequency range during seizures are highlighted, allowing quick assessment of seizure frequency and duration (Fig. 3.5). Time-frequency transformation of the cEEG can be further manipulated to provide a single scalar value for each epoch of time. For instance, Claassen et al.  showed that the ratio of total hemispheric power in the alpha-frequency band (8–13 Hz) to the total power in the delta-frequency band (1–4 Hz) after maximal alerting, or poststimulation alpha-delta ratio (ADR), was the most useful qEEG parameter for detecting delayed cerebral ischemia in patients with high-grade subarachnoid hemorrhage (SAH) (see later). Hemispheric asymmetries in spectral power, computed as ratio of left and right total power for all EEG frequencies or as relative differences at each frequency, can be used to quickly identify focal seizures (e.g., Fig. 3.5). The greatest utility of reducing the cEEG to single scalar values is that these values can easily be displayed and interpreted on bedside monitors like heart rate and blood pressure. This could allow for early identification of neurophysiological events by the ICU staff and alarms to trigger patient examination and could lead to more responsive treatment. Another measure used as part of quantitative EEG trending analysis is the rhythmicity spectrogram. This measure highlights the rhythmic or periodic component of different frequencies, thus facilitating seizure identification. A diagonal pattern in particular is characteristic of seizures as it shows rhythmicity rapidly and consistently changing in frequency, a typical pattern of ictal evolution .
qEEG trends (comprehensive panel view) from Persyst 12™ (Persyst. Inc.; San Diego, California) in a 24-year-old woman with SE. (a) 2 h qEEG page showing long-term trends: Artifact intensity (first from top) displays the amount of muscle artifact, vertical and lateral eye movement present. The intensity of these artifacts may help determine the state of the patient. Seizure probability (2nd from top): Red bars display seizure probability on a scale from 0 to 1, as determined by Persyst seizure detection algorithm. Rhythmicity spectrogram for left and right hemispheres (3rd and 4th from top, respectively) illustrates rhythmic components of different frequencies, darker colors being more rhythmic. FFT spectrogram for left and right hemispheres (5th and 6th from top, respectively) demonstrates power of different frequencies at different time periods. Time is displayed on x-axis, frequencies on y-axis, and amplitude of power of different frequencies as different colors on z-axis (see color scale). Relative asymmetry spectrogram (7th from top): illustrates comparison of power of different frequencies at homologous electrodes in each hemisphere (blue if higher power on left, red if on right). Suppression percentage (8th from top) displays the percent of the EEG record that is below a determined threshold amplitude (e.g., 10 μV). No EEG suppression is seen in this panel. aEEG (9th from top; combined left and right hemispheres; left, blue; right, red; overlap, pink): displays mean filtered and smoothed EEG amplitude (y-axis) across time (x-axis). FFT power ratio (last from top) illustrates alpha/delta ratio across time in both left (blue) and right (red) hemisphere. 26 seizures were detected in this 2 h page (black arrow heads), evidenced by surges in FFT power and aEEG, as well as evolving rhythmicity on rhythmicity spectral analysis. All of these seizures were also detected by seizure probability index (red bars on seizure probability index; 2nd panel from top). (b) 6 h qEEG page for the same patient shows significant decline in the number of seizures in the second half of the page. Using longer time windows allows for greater appreciation of long-term trends and assists in monitoring response to therapy. (c) Raw EEG for one of the detected seizures. LFF = 1 Hz, HFF = 70 Hz, notch off (Published with kind permission from © Lawrence J. Hirsch, MD 2016. All Rights Reserved)
Other trending algorithms highlight amplitude measures, which can also be used to detect seizures. Amplitude-integrated EEG (aEEG) displays compressed, smoothed, and full-wave rectified EEG signal (Fig. 3.5). It is particularly useful for assessing background amplitude and burst suppression. aEEG is commonly used in commercial devices in neonatal ICUs  to assess the background EEG and occasionally as an initial screening tool for detecting seizures, although aEEG may be inadequately sensitive and is probably not specific enough for detecting seizures . Envelope trend displays median amplitude of raw EEG background activity within a specified frequency range in a chosen time period, thus minimizing the effect of transient change in EEG signal created by artifacts, which are common in the ICU environment . Multiple seizure detection algorithms are now available and can be utilized for automated seizure detection in the ICU setting . Nonetheless, reliance on qEEG tools without the ability to review the raw EEG for non-cerebral signals can lead to false-positive seizure detections; thus, qEEG should only be interpreted in conjunction with the raw EEG wave forms and in conjunction with skilled electroencephalographers with special training in ICU EEG whenever possible .
Quantitative EEG tools can also calculate the degree of burst suppression of the EEG background to allow for easy titration of medications to induce coma, a common treatment of status epilepticus or refractory elevated intracranial pressure . EEG-based monitors such as bispectral index , patient state index , Narcotrend  and entropy systems  have been in use in operating rooms and ICUs for nearly two decades to monitor depth of sedation. While these single-purpose devices use proprietary algorithms, evaluation of the raw cEEG or qEEG measures can also provide information about arousal in the paralyzed patient . Data on the utility of these algorithms in those with underlying neurological issues is limited, as is the use of these devices to detect seizures; this should not be done without confirmation via review with expert review of raw EEG .
Several studies have evaluated the utility of qEEG trends analysis for seizure identification in critically ill individuals. Stewart et al.  investigated the sensitivity of qEEG for seizure identification in critically ill children by qEEG-naïve neurophysiologists and found that the median sensitivity of compressed spectral array (CSA) analysis was 83.3%, while that of aEEG was 81.5%. Missed seizures were more likely to be strictly focal, of low amplitude, or short duration . In another larger study involving critically ill adults, 89% of seizures were identified utilizing CSA-guided review by qEEG-naïve neurology residents after receiving 2 h of qEEG training . Another recent study evaluated the use of multiple qEEG panels for seizure identification and found that the overall sensitivity for seizure identification using these panels across all reviewer types was 84%, while the overall specificity was 69%. Interestingly, there was no statistically significant difference in sensitivities between the different reviewer groups (neurophysiologists, technologists, and neuroscience ICU nurses). Among four qEEG trends used in this study, rhythmicity spectrogram seemed to help the most in seizure identification . These data further support the role for qEEG as a reliable tool for screening EEG and targeting raw EEG review by both experienced and non-experienced readers. qEEG-guided EEG review may save as much as 78% of EEG reading time without carrying a significant impact on sensitivity .
In our experience, no one qEEG tool is appropriate for all patients or even for the same patient at all times. Situations may occur where one tool is more susceptible to certain artifacts or is less sensitive to the seizures the individual patient may have. Instead, we employ multiple tools simultaneously to screen the initial cEEG record and focus particularly on reviewing the raw EEG data at times where there appear to be clear changes in the qEEG measures from baseline. Once the patient’s seizure pattern is identified, the parameters of the qEEG tools can be further refined to highlight this pattern and improve the recognition of subsequent seizures.
With Internet-based networking, it is now practical to monitor dozens of patients in multiple ICUs. If there is sufficient network capability in the hospital, cEEG can be streamed live over the network and can be interpreted in real-time if needed (and personnel are available). In addition, cEEG can be reviewed remotely from home or from a distant hospital site using virtual private networks and virtual network computing . However, in current practice, cEEG is not yet truly real-time “monitoring” at most centers. In both our centers, records are routinely reviewed by neurophysiologists or technologists two-three times daily; ACNS guidelines suggest a minimum of twice daily . All new records should be interpreted as soon as possible. Ongoing records should be reviewed more frequently than just a few times per day if there are suspicious clinical events or medications are being titrated. However, as most NCSzs have little or no detectable clinical correlate, they may go unrecognized for several hours with only intermittent review. It is clear that we need to move toward continuous real-time monitoring via use of quantitative EEG alarms and around-the-clock “neurotelemetrists” to respond to the alarms and review the long-term trends. Several academic centers are already doing this.
Who to Monitor
Recent studies using routine and continuous EEG monitoring have helped to identify which patients are at risk for NCSz and, therefore, may benefit from cEEG. The causes of NCSz and NCSE in ICU patients are similar to the causes of convulsive seizures in these patients. These include acute structural lesions, infections (including sepsis), metabolic derangements, toxins, withdrawal, and epilepsy, all common diagnoses in the critically ill patient . It is important to stress that the majority of seizures in these critically ill patients are nonconvulsive and can only be diagnosed with EEG [1, 6, 46–48]. NCSzs are even more common in the pediatric population, especially in infants [1, 25, 49, 50]. There are many studies using cEEG that have identified the incidence of NCSz and NCSE in various patient populations. These studies are summarized in Fig. 3.1.
While it may not be surprising that patients with acute brain injuries [1, 51] and recent convulsive status epilepticus  have a high risk of NCSz, NCSzs are not uncommon in medical or surgical ICU patients, including in those without known structural brain injury. Critically ill medical and surgical patients are susceptible to many toxic, electrolyte, and metabolic abnormalities that may cause both mental status changes and seizures [7, 47]. 17–21% of patients with toxic-metabolic encephalopathy and impaired mental status had electrographic seizures on cEEG monitoring in two retrospective studies [1, 2]. Moreover, in one study of 201 medical ICU patients without known brain injury that underwent cEEG monitoring, 22% of patients had periodic discharges (PDs) or seizures; sepsis and acute renal failure were significantly associated with both PDs and seizures . A more recent prospective study found that among 100 episodes of sepsis in 98 patients without diagnosed acute primary neurological illness, periodic discharges were identified in 25 episodes; 11 of whom had nonconvulsive seizures . Meanwhile, 16% of patients admitted to the surgical ICU undergoing cEEG monitoring had electrographic seizures in one recent study, while 29% had periodic discharges (PDs) .
The American Clinical Neurophysiology Society (ACNS) recently published an official guideline entitled “Consensus Statement on Continuous EEG in Critically Ill Adults and Children.” The following are the proposed indications :
Persistent alteration of mental state following generalized convulsive status epilepticus (GCSE)
Altered mental state in association with acute supratentorial brain injury
Unexplained alteration of mental status without evidence of acute brain injury
Periodic discharges on routine or emergent EEG
Pharmacological paralysis in patients at high risk for seizures
Paroxysmal events suspected to be seizures to determine ictal vs. non-ictal nature of these events
Other indications recommended by ACNS include monitoring response to treatment of seizures and SE. In addition, the ACNS taskforce suggested the use of continuous EEG monitoring for detection of cerebral ischemia in high risk individuals, as an adjunct to other methods . Similar statements discussing indications of continuous EEG monitoring have been released by the European Society of Intensive Care Medicine (ESICM) , as well as the Neurocritical Care Society as part of recommendations on management of SE .
EEG Patterns Encountered During EEG Monitoring
The background, interictal, and ictal EEG patterns of the critically ill patient are significantly different from those encountered in ambulatory patients [58, 59]. Ictal patterns may include rhythmic epileptiform discharges or rhythmic waves at greater than 3 Hz (as with most seizures). However, in critically ill patients, rhythmic or periodic patterns occurring at a rate of less than three per second can be ictal as well. One set of criteria for defining NCSz are shown in Table 3.1. It should be noted that these criteria reflect expert consensus and there are periodic patterns common in critically ill patients where the relationship to seizures is unknown . In practice, it is often difficult to determine whether periodic or rhythmic activity at 1–3 Hz in a comatose patient reflects seizure activity or a brain at risk for seizures or is merely a marker of severe brain injury. These patterns have been considered to lie along the ictal-interictal continuum . Aiming to create common terminology for use by critical care electroencephalographers worldwide, the ACNS published standardized terminology for describing these patterns, initially proposed in 2005 and then revised and published as an official guideline in late 2012 [62, 63]. The current terminology is summarized in Table 3.2. In one recent study, the interrater reliability for ACNS terminology was near perfect for main terms (1) and (2), which describe the location and the nature of the pattern, respectively. However, the interrater reliability for evolution and some of the other modifiers was not as good .
Criteria for diagnosing nonconvulsive seizures
Patients without known epileptic encephalopathy:
• Epileptiform discharges (EDs) > 2.5 Hz
• EDs ≤ 2.5 Hz or rhythmic delta/theta activity >0.5 Hz and one of the following:
EEG and clinical improvement after IV AEDs
Subtle clinical ictal phenomena during the EEG pattern mentioned above
Typical spatiotemporal evolution
Patients with known epileptic encephalopathy:
• Increase in prominence or frequency of the features mentioned above when compared to baseline with observable change in clinical state
• Improvement of clinical and EEG features with IV AEDs
ACNS terminology for description of periodic and rhythmic patterns  (Published with kind permission from © Lawrence J. Hirsch, MD 2013. All Rights Reserved)
Main term A (for localization)
Main term B (pattern type)
Periodic discharge (PD)
Rhythmic delta activity (RDA)
Bilateral independent (BI)
Spike-wave or sharp-wave complex (SW)
There is accumulating evidence that certain periodic discharges may reflect injured tissue at high risk for seizures such as lateralized periodic discharges (LPDs; previously called periodic lateralized epileptiform discharges (PLEDs)) and generalized periodic discharges (GPDs) (Fig. 3.6) . There is convincing evidence to suggest that LPDs are sometimes ictal. For instance, LPDs can be time-locked to focal clonic movements in some patients with focal motor status epilepticus . This seems to be more common in cases in which LPDs primarily involve Rolandic cortex (not surprisingly) . Positron emission tomography in one patient with frequent LPDs demonstrated increased regional glucose metabolism similar to what is seen with focal seizures . Single-photon emission CT (SPECT) imaging in patients with LPDs demonstrated increased regional cerebral perfusion in some patients that normalized when the LPDs resolved [67, 68]. In addition, frequent LPDs in elderly patients have been associated with a confusional state that resolves spontaneously or with diazepam treatment . However, other studies have described cases where LPDs are clearly non-ictal such as in some epilepsy patients with chronic interictal LPDs . In addition, when some patients with LPDs and acute brain injury demonstrate seizures, the EEG pattern is often faster and with different morphology . Given the close association with seizures and the fact they are at times clearly associated with behavioral changes, some authors view LPDs as an unstable state in an “irritable” brain, lying along an ictal-interictal continuum [60, 72].
Periodic discharges in critically ill patients. (a) Right frontal LPDs occurring at 1 Hz (arrow) in an 82-year-old man after resection of a bifrontal meningioma. The patient subsequently developed right frontal electrographic seizures. (b) Generalized periodic discharges at 1–2 Hz in a 79-year-old patient with dementia, renal disease, and altered mental status. Although these waveforms have a triphasic morphology at times, the pattern subsequently evolved to 2.5–3 Hz GPDs consistent with NCSz and was associated with modest elevations in neuron-specific enolase to 14 (reference range 3.7–8.9). Low-frequency filter (LFF) = 1 Hz, HFF = 70 Hz, notch off (Published with kind permission from © Lawrence J. Hirsch, MD 2013. All Rights Reserved)
A common practice used to distinguish ictal from non-ictal periodic EEG patterns in the critically ill is to see if they are abolished by a trial of short-acting benzodiazepines (Table 3.3). However, almost all periodic discharges are attenuated by benzodiazepines . Thus, unless there is clinical improvement accompanying the EEG change, the test is not helpful. Unfortunately, clinical improvement can take substantial time even if the activity represents NCSE and is aborted with benzodiazepines. However, a substantial portion of ICU patients with nonconvulsive seizures or NCSE will improve neurologically and usually within a day of treatment. For example, Hopp et al.  showed that 35% of patients with suspected NCSE receiving IV benzodiazepine (BZP) trial achieved positive clinical response. Moreover, positive clinical response correlated well with survival, recovery of consciousness, and achieving good functional outcome . In order to avoid the confounding effect of sedation, we often use loading doses of nonsedating IV antiepileptic drugs (AEDs) such as valproate, lacosamide, levetiracetam, and phenytoin, for these diagnostic trials (see Table 3.3). One recent retrospective study evaluated clinical response to antiepileptic drug trial in patients with unexplained encephalopathy and triphasic wave pattern on EEG and found that 42.2% of patients receiving nonsedating IV AED trial achieved positive clinical response, whereas only 18.9% of patients receiving IV BZP trial did . Our protocol for attempting to prove the presence of NCSE is shown in Table 3.3. It is important to recognize that lack of clinical improvement does not exclude NCSE—it simply does not help determine its presence or absence. This situation (EEG improvement without clinical improvement) has been referred to as “possible NCSE” .
Antiepileptic drug trial for the diagnosis of nonconvulsive status epilepticus (adapted from Hirsch and Gaspard  with permission from Wolters Kluwer Health, Inc.)
Rhythmic or periodic focal or generalized epileptiform discharges on EEG with neurologic impairment
Patients who are heavily sedated or paralyzed
EEG, pulse oximetry, blood pressure, electrocardiography, respiratory rate with dedicated nurse
Antiepileptic drug trial
Sequential small doses of rapidly acting short-duration benzodiazepine such as midazolam at 1 mg/dose or nonsedating IV antiepileptic drug such as levetiracetam, valproate, fosphenytoin, or lacosamide
Between doses, repeated clinical and EEG assessment
Trial is stopped after any of the following:
1. Persistent resolution of the EEG pattern (and exam repeated)
2. Definite clinical improvement
3. Respiratory depression, hypotension, or other adverse effect
4. A maximum dose is reached (such as 0.2 mg/kg midazolam, though higher doses may be needed if the patient is on chronic benzodiazepines)
The test is considered positive (“definite NCSE”)  if there is resolution of the potentially ictal EEG pattern and either an improvement in the clinical state or the appearance of previously absent normal EEG patterns (e.g., posterior-dominant “alpha” rhythm)
If EEG improves but patient does not, the result is equivocal (“possible NCSE”) 
Nonictal patterns may disappear after administration of benzodiazepines (always without clinical improvement)
Administration of too high a dose of benzodiazepine might improve the EEG but also leads to sedation, preventing the ability to detect clinical improvement
Negative or equivocal response does not rule out nonconvulsive status epilepticus
There is fairly consistent evidence that the presence of PDs and frequent nonconvulsive seizures are an independent risk factor for worse prognosis in ICH , SAH , and sepsis  and after GCSE [52, 80], even in patients without evidence of acute brain injury . In addition, there is accumulating evidence that increased electrographic seizure burden is correlated with worse outcome in both pediatric and adult populations [9, 82, 83]. Both NCSzs and LPDs have been shown to be independently associated with later epilepsy as well . Nonetheless, it is unclear whether these and other periodic discharges require treatment and how aggressive this treatment should be. Laboratory studies and computer modeling are beginning to probe the network mechanisms that mediate periodic discharges in the injured brain .
Another recently described pattern in critically ill patients is lateralized rhythmic delta activity (LRDA) . This pattern was identified as an independent predictor of increased risk of acute seizures in critically ill individuals. It has been suggested that it carries similar implication as LPDs as regards risk of acute seizures . One recent study suggested that LRDA is more likely to be associated with seizures if it occurs at a frequency of ≥1.5 Hz or is associated with a plus modifier .On the other hand, the presence of generalized rhythmic delta activity (GRDA), even when sharply contoured or superimposed by sharp waves or fast frequency activity (GRDA + S or GRDA + F, respectively), doesn’t seem to carry a significant increase in risk of electrographic seizures , at least based on the single retrospective study that has looked at this .
Epileptiform or rhythmic activity triggered by stimulation or arousal is also a common pattern in encephalopathic ICU patients. The evoked activity may be anywhere on the interictal-ictal spectrum and is collectively known as stimulus-induced rhythmic, periodic, or ictal discharges  (SIRPIDs) . There is usually no clinical correlate, as with most ICU seizures, but a small portion of patients will have focal motor seizures consistently elicited by alerting stimuli . On the other hand, there have been two case reports of SPECT-negative SIRPIDs, suggesting that some of these poststimulation discharges do not represent clear ictal phenomena or at least do not have the usual seizure-associated increased blood flow [88, 89]. SIRPIDs most likely occur as a result of hyperexcitable cortex that is activated by normal arousal pathways, which involve the upper brainstem, thalamus, and widespread thalamocortical projections. This epileptiform activity may become clinically apparent if it causes synchronous activation, propagates caudally in an organized fashion, and involves motor pathways. At both our centers, technologists stimulate patients twice daily to assess for state-dependent changes in the EEG including the appearance of SIRPIDs, but the relationship between ictal discharges and arousals raises the possibility that limiting unnecessary stimulation in patients with SIRPIDs may be beneficial. One recent study demonstrated that the presence of SIRPIDs was not independently associated with in-hospital mortality in critically ill patients . Another study reported that the presence of SIRPIDs is a poor prognostic marker in postanoxic patients, particularly when recorded during therapeutic hypothermia . The effect of SIRPIDs on long-term outcome in other settings as well as its therapeutic implication remains unclear. We treat stimulus-induced patterns the same as spontaneous patterns (other than potentially limiting stimulation) as there is no theoretical reason or evidence that they differ in their ability to cause neuronal injury.