Identifying children at higher risk for electrographic seizures is complex since electrographic seizures have been reported in both large heterogeneous cohorts [13] and smaller more homogeneous cohorts of children with single brain insult etiologies [7, 11, 12, 16]. Several risk factors have been reported including younger age (infants as compared to older children) [8, 11, 13, 16, 18], the occurrence of convulsive seizures [9, 13, 14] or convulsive status epilepticus (CSE) [8] prior to initiation of monitoring, the presence of acute structural brain injury [7–9, 11, 12, 14, 16], and the presence of interictal epileptiform discharges [8, 12–14] or periodic epileptiform discharges [3]. Although the reported risk factors are statistically significant, the absolute difference in the proportion of children with and without electrographic seizures based on the presence or absence of a risk factor is often only 10–20 %, so these risk factors may have limited clinical utility in selecting patients to undergo monitoring.

EEG monitoring is resource intense, and seemingly small changes in utilization may have substantial impacts on equipment and personnel needs [20, 21]. Seizure prediction models combining multiple known seizure risk factors could allow targeting of EEG monitoring to children with the highest risk for experiencing electrographic seizures within the resource limitations of an individual medical center. A recent study derived a seizure prediction model from a retrospectively acquired multicenter dataset and validated it on a separate single-center dataset. Both datasets were derived from clinically indicated EEG monitoring performed for critically ill children with heterogeneous etiologies for their acute encephalopathy. The model had fair to good discrimination including the validation dataset, indicating that most (but not all) patients were appropriately classified as having or not having electrographic seizures. The model could be applied clinically in three steps. First, the clinician would obtain two clinical variables (age and whether there were clinically evident seizures) and two EEG variables (background category and interictal epileptiform discharge presence). Second, using these variables, the clinician could determine a model score. Third, patients with model scores above an institutional cutoff score would be selected to undergo cEEG monitoring. Individual institutions could select different model cutoff scores based on center-specific criteria. A center with substantial EEG monitoring resources might perform EEG monitoring for any patient with a model score >0.10. At this lower cutoff, 14 % of patients with electrographic seizures would not undergo EEG monitoring, so the seizures would not be identified and managed. However, 58 % of patients without electrographic seizures would be identified as not needing EEG monitoring, so limited resources would not be expended. Given a seizure prevalence of 30 %, this cutoff would have a positive predictive value of 47 % and negative predictive value of 91 % [22]. Further development might yield improved predictive models by incorporating additional variables or focusing on more homogeneous cohorts.

Decisions regarding the duration of EEG monitoring must balance the goal of identifying electrographic seizures with practical concerns regarding the substantial and limited resources required to perform EEG monitoring. Observational studies of critically ill children undergoing clinically indicated EEG monitoring have reported that about 50 % and 90 % of patients with electrographic seizures are identified with 1 h and 24–48 h of EEG monitoring, respectively (Fig. 2) [3, 6, 8, 9, 12, 14, 15, 18]. Thus, 1 h of EEG will fail to identify many children who will subsequently experience electrographic seizures, while 48 h of monitoring identifies most children with electrographic seizures.

There are two important limitations regarding the electrographic seizure timing data described above. First, most of the studies providing the data above calculated timing at the onset of EEG monitoring and not the onset of the acute brain insult. However, in clinical practice patients may present at varying durations after the onset of acute brain insult. Furthermore, patients may experience clinical changes potentially producing additional brain injury while in the ICU, and it is unclear if the timing considerations should restart with each of these clinical occurrences. Second, most of the studies providing the data above were observational studies in which patients underwent 1–3 days of clinically indicated EEG monitoring. Thus, some patients may have experienced electrographic seizures after EEG monitoring was discontinued. In specific circumstances electrographic seizures are known to occur later in time, such as following cardiac arrest resuscitation [7].

Based on the data described above, the Neurocritical Care Society’s Guideline for the Evaluation and Management of Status Epilepticus strongly recommends performing 48 h of EEG monitoring to identify electrographic SE in comatose children following an acute brain insult [23]. The American Clinical Neurophysiology’s Consensus Statement on Continuous EEG Monitoring in Critically Ill Children and Adults recommends performing EEG monitoring for at least 24 h in children at risk for seizures [24].

Several studies in critically ill children have reported associations between high seizure exposures and worse outcomes. However, the extent to which electrographic seizures are actually producing secondary brain injury versus serving as biomarkers of more severe acute brain injury remains unknown. Further, the extent to which seizures produce secondary brain injury is likely dependent on a complex interplay between acute brain injury etiology, seizure exposure, seizure characteristics, and seizure management strategies. As summarized below, a number of recent studies have reported an association between electrographic seizures, particularly with high electrographic seizure exposures, and worse outcomes even after adjustment for potential confounders related to acute encephalopathy etiology and critical illness severity.

Several studies have described an association between electrographic seizures and unfavorable short-term outcome. A prospective observational study of 1–3 channel EEG in 204 critically ill neonates and children found that the occurrence of electrographic seizures was associated with a higher risk of unfavorable neurologic outcome (odds ratio 15.4) in a multivariate analysis that included age, etiology, pediatric index of mortality score, Adelaide coma score, and EEG background categories [10]. Several other studies aimed to evaluate the effect of seizure burden and classified children as having no seizures, electrographic seizures, or electrographic SE. A single-center study of 200 children in the pediatric ICU with outcome assessed at discharge identified an association between electrographic SE and higher mortality (odds ratio 5.1) and worsening pediatric cerebral performance category scores (odds ratio 17.3) in multivariate analyses including seizure category, age, acute neurologic disorder, prior neurodevelopmental status, and EEG background categories. Electrographic seizures not classified as electrographic SE were not associated with worse outcomes [25]. A larger multicenter study of 550 children in the pediatric ICU reported an association between electrographic SE and mortality (odds ratio 2.4) in a multivariate analysis that included seizure category, acute encephalopathy etiology, and EEG background categories. Electrographic seizures not classified as electrographic SE were not associated with worse outcomes [13]. A single-center prospective study evaluated 259 critically ill infants and children who underwent EEG monitoring described electrographic seizures in 36 % of subjects which constituted electrographic SE in 9 % of subjects. Seizure burden was calculated as the proportion of the hour containing seizures, and the maximum hourly seizure burden was identified for each subject. The mean maximum seizure burden per hour was 15.7 % in subjects with neurological decline versus 1.8 % in subjects without neurological decline. In a multivariate analysis that adjusted for diagnosis and illness severity, for every 1 % increase in the maximum hourly seizure burden, the odds of neurological decline increased by 1.13. Maximum hourly seizure burdens of 10, 20, and 30 % were associated with odds ratios for neurological decline of 3.3, 10.8, and 35.7. In contrast to some of the other studies described above, electrographic seizures were not associated with higher mortality [17].

A study addressing long-term outcome obtained follow-up data at a median of 2.7 years following pediatric ICU admission from 60 children who were neurodevelopmentally normal prior to admission and underwent clinically indicated EEG monitoring. Multivariate analysis including acute neurologic diagnosis category, EEG background category, age, and several other clinical variables identified an association between electrographic SE and unfavorable Glasgow Outcome Scale (Extended Pediatric Version) category (odds ratio 6.36), lower Pediatric Quality of Life Inventory scores (23.07 points lower), and an increased risk of subsequently diagnosed epilepsy (odds ratio 13.3). Children with electrographic seizures not classified as electrographic SE did not have worse outcomes [26].

Together, these studies suggest there may be a dose-dependent or threshold effect of seizures upon outcomes, with high seizure burdens having clinically relevant adverse impacts. This threshold may vary based on age, brain injury etiology, and seizure characteristics such as the extent of brain involved and electroencephalographic morphology. While further study is needed, these data suggest that at least in some patients and at high seizure exposures, electrographic seizures may be producing secondary brain injury, and identifying and managing those seizures might mitigate such injury.

A recent survey of EEG monitoring use in the pediatric ICUs of 61 large pediatric hospitals in the United States and Canada reported that the median number of patients who underwent cEEG monitoring per month increased about 30 % from 2010 to 2011 [27]. Indications for EEG monitoring included determining whether events of unclear etiology were seizures in 100 % of centers and identifying electrographic seizures in patients considered “at risk” in about 90 % of centers. Patients considered “at risk” included those with altered mental status following a convulsion, altered mental status in a patient with a known acute brain injury, and altered mental status of unknown etiology. About 30–50 % of centers reported using EEG monitoring as part of standard management for specific acute encephalopathy etiologies within a clinical pathway (i.e., following resuscitation from cardiac arrest or with severe traumatic brain injury) [27].

The Neurocritical Care Society’s Guidelines for the Evaluation and Management of Status Epilepticus recommends the use of EEG monitoring to identify electrographic seizures in at-risk patients including those with persisting altered mental status for more than 10 min after convulsive seizures or SE or encephalopathic children after resuscitation from cardiac arrest, with traumatic brain injury, with intracranial hemorrhage, or with unexplained encephalopathy. The guideline strongly recommends 48 h of EEG monitoring in comatose patients. If SE occurs (including electrographic SE), then the guideline recommends that management should continue until not only the clinical seizures are halted, but until all electrographic seizures are halted [23].

The American Clinical Neurophysiology Society’s (ACNS) Consensus Statement on Continuous EEG Monitoring in Critically Ill Children and Adults recommends EEG monitoring for 24–48 h in children at risk for seizures. Monitoring indications include recent convulsive seizures or CSE with altered mental status, cardiac arrest resuscitation or with other forms of hypoxic-ischemic encephalopathy, stroke (intracerebral hemorrhage, ischemic stroke, and subarachnoid hemorrhage), encephalitis, and altered mental status with related medical conditions. Detailed recommendations are provided regarding personnel, technical specifications, and overall workflow [24].

Increasing EEG monitoring use among critically ill children [27, 28] is resource intense and would benefit from improved seizure identification efficiency. Quantitative EEG (qEEG) techniques separate the complex EEG signal into components (such as amplitude and frequency) and compress time, thereby permitting display of several hours of EEG data on a single image that may be interpreted more easily than conventional EEG. QEEG techniques may facilitate more efficient EEG monitoring review by encephalographers and perhaps even earlier identification of seizures by non-encephalographer clinicians providing bedside care. These techniques are still being developed and their test characteristics are still being established.

Several studies have examined the utility of qEEG in critically ill children. In the first study, 27 color density spectral array (CDSA) and amplitude-integrated EEG (aEEG) tracings were reviewed by three encephalographers. The median sensitivity for seizure identification was 83 % using CDSA and 82 % using aEEG, but for individual tracings the sensitivity varied from 0 to 100 %. A false positive occurred about every 17–20 h [29]. In the second study, 84 CDSA images were reviewed by eight encephalographers. Sensitivity for seizure identification was 65 %, indicating that some electrographic seizures were not identified. Further, only about half of seizures were identified by six or more raters. Specificity was 95 %, indicating some non-ictal events were misdiagnosed as seizures [30]. A study of CDSA and envelope trend EEG found that seizure identification was impacted by both modifiable factors (interpreter experience, display size, and qEEG method) and non-modifiable factors inherent to the EEG pattern (maximum spike amplitude, seizure frequency, and seizure duration) [31].

Critical care providers have expertise at screening multiple monitoring modalities and are generally continually available within the ICU. Thus, if critical care clinicians are able to use qEEG, then electrographic seizures might be identified more rapidly. A study provided 20 critical care physicians (attendings and fellows) and 19 critical care nurses with a brief training session regarding CDSA and then asked them to determine whether each of 200 CDSA images created from conventional EEG derived from critically ill children contained electrographic seizures. The true seizure incidence was 30 % based on electroencephalographer review of the conventional EEG tracings. The CDSA seizure identification sensitivity was 70 %, indicating that some electrographic seizures were not identified. The specificity was 68 %, indicating that some images categorized as containing EEG seizures did not contain seizures. These errors may be problematic since they could lead to exposure of non-seizing children to antiseizure medications with potential adverse effects. Given the 30 % seizure incidence used in the study, the positive predictive value was 46 % and negative predictive value was 86 % [32].

These data indicate that commercially available qEEG techniques permit identification of many but not all seizures. Since seizures often occur early during EEG monitoring recordings and EEG technologists may not be readily available when EEG monitoring is needed [18], rapid bedside implementation may be an important advantage of these qEEG techniques. Seizure identification may improve with user training and experience, further development of qEEG trends, and implementation of qEEG panels with multiple trends. However, since qEEG leads to misclassification of some non-ictal events as seizures, potentially leading to unnecessary antiseizure mediation administration, confirmation by conventional EEG review may be indicated for when qEEG techniques suggest seizures are present. With further development these synergistic methods could make use of the efficiency and bedside availability of qEEG methods and the accuracy of conventional EEG tracings.