Article Text

PDF

Monitoring of seizures in the newborn
  1. Divyen K Shah1,
  2. Geraldine B Boylan2,
  3. Janet M Rennie3
  1. 1Department of Neonatology, Barts and The London Children's Hospital, London, UK
  2. 2Neonatal Brain Research Group, University College Cork, Cork, Ireland
  3. 3Department of Neonatology, University College Hospitals, London, UK
  1. Correspondence to Dr Janet M Rennie, Department of Neonatology, Elizabeth Garrett Anderson Obstetric Wing, University College Hospital, 2 North 250 Euston Road, London NW1 2PQ, UK; jmr{at}janetrennie.com

Abstract

Neonatal seizures are a distinct and not uncommon sign of neurological disease in the newborn, most often occurring in association with hypoxic-ischaemic encephalopathy at term. The diagnosis and monitoring of seizures in the newborn is a considerable challenge, with many suspected clinical seizures having no electrographic correlates, while many electrographic seizures have no clinical correlate. Continuous video-EEG is the gold standard for seizure monitoring, but few centres have the resources or expertise required. Amplitude-integrated EEG can be a helpful monitoring tool in experienced hands, but has potential for error when used by inexperienced staff. Automated seizure detection algorithms show much promise and some cotside systems are already available. The efficiency and accuracy of these systems is likely to improve.

Statistics from Altmetric.com

Introduction

Seizures in the newborn often occur in association with serious underlying cerebral pathology. It is important to reach a diagnosis and treat the underlying condition rapidly and establish a treatment plan for the seizures. Seizures present clinicians as well as researchers with many challenges and unanswered questions. What do seizures represent? How much do seizures reflect cerebral pathology? Do they exacerbate any pre-existing brain injury and to what extent do they impact on the longer term neurodevelopmental outcome of the affected infant? Given that seizures are difficult to recognise in the newborn, how can they be detected and monitored effectively?

Background

Population based studies suggest the incidence of clinical seizures in the newborn is between 1 and 3/1000 live births.1 2 Seizures often accompany underlying brain pathology in the newborn, and are associated most commonly with hypoxic–ischaemic encephalopathy (HIE), periventricular haemorrhage, neonatal stroke and central nervous system (CNS) infection. However, several other conditions may predispose a newborn to seizures, such as neonatal abstinence syndrome, hypoglycaemia and inborn errors of metabolism (see table 1).3 Of 92 newborns with electrographic seizures recorded over a 4-year period at a single centre by Scher et al,4 60% had seizures in the first 48 h of life. Two-thirds of the infants were preterm and 45% (28/62) of these had experienced severe intraventricular haemorrhage (ie, with accompanying ventricular dilatation or intraparenchymal extension). In contrast, the majority of term infants with seizures (23/30) had cerebral infarction on imaging or at postmortem; 13 of the babies had HIE. Seizures are thought to be more common in preterm than in term babies, but few reliable prospective surveys exist.1

Table 1

Causes of seizures in the newborn

Why monitor seizures in the newborn?

Many clinicians believe that seizures are harmful to the developing newborn brain, and there is some support for this belief. In a 10-day-old rat pup model, Wirrell et al5 showed significantly increased brain injury after seizures induced by kainic acid superimposed on prolonged hypoxia–ischaemia in comparison with prolonged hypoxia–ischaemia alone. Using magnetic resonance spectroscopy in a group of babies with HIE, Miller et al6 showed that increased seizure duration and frequency were independently associated with raised lactate/choline ratios in the watershed zones of cerebral white matter as well as the basal ganglia. The raised cerebral lactate/choline ratio which can be seen in HIE is probably related to secondary energy failure. There is also some evidence that neurodevelopmental outcome in babies is related to the duration and frequency of electrographic seizures in the newborn period.7

Treating seizures in the neonate

Most neonatologists choose to treat persistent clinical seizures in the newborn, and many would use cotside monitoring to detect electrographic seizure activity if it was reliable and accessible, not least in order to monitor the effects of treatment. There is emerging evidence that the rapidly developing newborn brain may not respond to anticonvulsants in the same way as the brain of an older child or adult. In the perinatal period, γ amino butyric acid (GABA) activation is associated with chloride efflux and excitation, rather than influx and inhibition as occurs after GABA activation of the major GABA-A receptor in the mature neuron.8 Many anticonvulsants are thought to act by facilitating GABA activation. Clearly much research is required to develop anticonvulsants targeted at the developmental stage of the newborn brain. The forthcoming EU FP7 NEonatal seizures with Medication Off-patent study (www.nemo-europe.com) is one such study which will soon begin recruitment and which aims to assess the efficacy of bumetanide as an adjunct to phenobarbitone for seizure treatment in newborns with seizures.

Anticonvulsant treatments may also increase electro-clinical ‘uncoupling’, the situation when clinical manifestations of seizure activity are largely eradicated but the abnormal rhythmic electric discharges seen on EEG continue or, in some cases, worsen.9 10

Clinical versus electrographic seizures

Babies are often treated with anticonvulsants on the suspicion of clinical seizures alone. Many suspected clinical seizures or abnormal movements do not have electrographic correlates; conversely, many electrographic seizures have no clinical correlate.11 Clinical seizures may be subtle and can go unrecognised in a busy clinical setting. Using video-EEG, a study by Murray et al12 reported a total of 526 electrographic seizures in nine of a cohort of 51 term at-risk neonates. Of these 526 seizures, 34% had simultaneous clinical manifestations evident on video surveillance. Only 9% (48/526) of the electrographic seizures had accompanying clinical manifestations which were actually documented at the time by attending staff. Conversely, of 177 episodes of ‘suspected’ clinical seizures, only 27% had accompanying electrographic seizure activity.

Clinical seizures in the newborn

Despite the limitations noted by Murray et al,12 clinical observation by nursing and medical staff in the neonatal intensive care unit (NICU) is extremely important. Increased vigilance and further investigation is the appropriate response when a baby is noted to have abnormal movement, posturing or other behavioural manifestation which could possibly represent a clinical seizure.

The common manifestations of clinical seizures in the newborn differ from those in older children, and this is probably related to the stage of neuroanatomical and neurophysiological brain development.8 Unlike in older children and adults, well organised tonic–clonic seizures are rare in the newborn. Seizures in the newborn are usually focal in origin and commonly display multifocal characteristics as the seizure develops. Subtle seizures are the commonest type of clinical seizure in the newborn. They typically involve the eyes, lips and mouth. Eye deviation, fixed open starring, blinking and nystagmus may occur. Mouthing, chewing, lipsmacking and smiling may also be seen in these seizures. Babies may have associated apnoeas, abnormal respiratory patterns and hiccups as well as autonomic features such as tachycardia, change in respiratory rate or rise in blood pressure. Clonic seizures typically present with rhythmic jerking at 1–4 Hz that is not dampened by putting the examiner's hand against the limb. Clonic seizures may start on one side of the face or in one limb and spread.3 They may be focal or multifocal. If focal, an underlying brain lesion should be sought. They are frequently associated with neonatal stroke.13 They may also occur in association with a metabolic cause and occasionally herpes encephalitis. Myoclonicseizuresconsist of rapid jerks with a predilection for flexor muscle groups, and are more rapid than the movements in clonic seizures which are most frequently noted in preterm infants with major cerebral pathology. The EEG background is usually abnormal in babies with myoclonic seizures. Tonicseizures are much less common than either subtle or clonic seizures in the newborn and are associated with focal or generalised sustained posturing of limbs and/or the trunk, or deviation of the head or eyes. Clinical manifestations of seizures in preterm infants may be more subtle.

Electrographic seizures

A single channel EEG waveform represents the voltage difference between two electrodes on the scalp which record summated postsynaptic potentials from the vicinity of the underlying cerebral cortex. Electrographic seizures are typically defined as repetitive, rhythmic, stereotypic activity lasting at least 10 s that evolve over time and on multichannel recordings can be seen to evolve spatially (figure 1). A seizure results from excessive synchronous electrical discharge of neurons within the CNS.

Figure 1

Multichannel EEG recording from a term baby with hypoxic–ischaemic encephalopathy and seizures. Note how the frequency and morphology of the seizure discharge evolves over the left hemisphere (red channels). In addition, a slower seizure discharge over the left fronto-central region (top two blue channels) ceases as the seizure discharge changes over the right hemisphere.

How should seizures be monitored in the newborn?

Multichannel video-EEG is the gold standard for monitoring seizures in the newborn. The video component not only allows correlation of electrographic seizure activity with clinical seizure activity but (more importantly) allows for the identification of artefacts that may mimic electrical seizures. Common artefacts during EEG monitoring can be caused by patting, stroking, respiratory effort and ECG pick up.

Few neonatal centres have the resources required for continuous video-EEG surveillance or its interpretation. Conventional multichannel EEG studies are often performed for infants thought to be having abnormal movements. Typically these are usually recorded for 30–120 min. One important drawback is that although this ‘snapshot’ may provide much useful information regarding the EEG background, it may not capture sporadic seizure activity. In a study using continuous multichannel EEG surveillance over a median period of 18.6 h, seven of 21 term infants had 41 non-status seizures.14 The seizures were observed from 5 to 21 h after recording began.

The use of amplitude-integrated EEG to monitor seizures

Amplitude-integrated EEG (aEEG) was first used in the late 1960s to monitor electrocortical ‘cerebral function’ in adult patients undergoing cardiac by-pass surgery or in intensive care.15 It was devised at a time when heart rate, blood pressure and oxygen saturations could be monitored in patients requiring intensive care but cerebral function could not be easily checked.16 The aEEG represents the processed EEG signal from one or two channels which have been filtered and rectified and then displayed on a semi-logarithmic scale. Low (<2 Hz) and high frequencies (>15 Hz) of the EEG signal are attenuated, and a time-compressed trace is obtained so that typically 6 cm of recording represents 1 h on the x-axis.17 The y-axis displays the voltage potential difference between the electrodes, using a semi-logarithmic scale so that low voltage EEG activity is enhanced, while allowing higher voltage activity, such as that seen during seizures, to be recognised. In general terms, the lower margin of the aEEG trace reflects EEG ‘continuity’ and the upper margin reflects EEG wave amplitude. Typically, an electrographic seizure will bring about a rise in the lower and upper margins of the aEEG (figure 2).18

Figure 2

Electrographic seizure activity as seen on a digital two-channel amplitude-integrated EEG (aEEG) monitor. The lower two panels represent the aEEG traces from the left and right hemispheres, showing frequent rises in the lower aEEG margins corresponding to electrographic seizure activity on the raw EEG trace as shown by repetitive rhythmic spike-wave activity in the upper panels. Although the appearance (particularly in the right hemisphere) may be similar to an ECG trace, assessment of the complete tracing shows evolution.

Modern digital aEEG monitors provide not only one or two channels of the aEEG trace but also the raw unprocessed EEG trace from those channels. Using digital two-channel aEEG with the accompanying raw EEG trace monitoring, 76% (31/41) of non-status neonatal seizures in 21 at-risk term infants were correctly identified off-line by experienced raters.14 Of the 10 seizures missed, seven had low frequency with a focus distant to the aEEG recording electrodes. Nine false positives were obtained in this study with over 351 h of recording. These were thought to be related to muscle, electrode and infant patting artefacts. When just one or two channels of EEG are used, focal seizures originating at a site distant from the electrodes may not be detected. Of 851 neonatal seizures captured on conventional EEG, Shellhaas and Clancy19 showed that 78% originated in the ‘cross-cerebral’ C3–C4 channel and 81% from the centro-temporal or midline vertex electrodes.

Reliance on the aEEG, sometimes known as the cerebral function monitor (CFM) trace alone, may not be sufficiently sensitive or specific for seizure detection.14 20 Our initial study using raters who have received 3–5 h of training and three different CFM trace speeds produced low sensitivity and poor interobserver agreement.20 Other studies also show poor sensitivity for seizure detection using the aEEG trace alone.14 21 An increase in the upper and lower aEEG voltages can be produced by handling of the baby causing a high aEEG artefact, whereas a high frequency EEG artefact such as that from an oscillator or electricity mains, will raise the lower aEEG margin. Also, a time-compressed trace such as that of the aEEG will miss brief seizures.22 There are also pitfalls to be aware of when using the raw EEG trace of the digital aEEG monitors. Electrode contact artefacts, associated with sweating and patting the infant, may bring about a rhythmic EEG appearance that mimics a seizure (figure 3).14 In addition, using just two channels does not allow visualisation of the temporal and spatial evolution of seizure activity.

Figure 3

EEG electrode artefact. Repetitive rhythmic spike-wave activity seen on the raw EEG trace on the digital amplitude-integrated EEG monitor (left). A simultaneous conventional EEG using shared electrodes reveals the abrupt onset of this artefact, probably related to electrode contact.

The aEEG can be a very useful tool for monitoring seizures in the newborn provided users are aware of the limitations. Few neonatologists have had formal training in aEEG interpretation for seizure detection.23 Formal training as well as understanding its limitations make for safer use of this technology. The use of the aEEG should be backed up with conventional multichannel EEG.

Automated systems to assist seizure detection

Seizure surveillance with multiple channel EEG-video is resource and time intensive, limiting the feasibility of clinical monitoring of the newborn in most centres. Automated systems to assist this process have been developed.24 Automated systems use key features of the EEG signal to define the characteristics of a seizure as well as features that differentiate the seizure from background EEG activity. These systems typically calculate EEG features such as frequency, amplitude, rhythmicity, complexity and spectral edge frequency for the entire EEG recording; they then use a classifier to separate the EEG into seizure or non-seizure categories.25

The ideal automatic algorithm for neonatal seizure detection should have a high sensitivity (ie, should detect a high proportion of true electrographic seizures) and a low false positive rate so that infants with seizures are not missed and infants who do not have seizures do not receive unnecessary treatment. These goals are challenging because electrographic seizures in neonates have extremely varied morphology even within the same patient and because of the various artefacts that mimic electrographic seizure activity. As mentioned above, the NICU environment is a rich source of EEG artefacts that mimic seizures including ventilators, syringe pumps and other monitoring devices as well as comfort patting and stroking. Therefore, it is imperative that all potential neonatal seizure detection algorithms are validated with long duration multichannel EEG data containing multiple seizure and non-seizure epochs and a range of biological and non-biological artefacts.

A number of automated seizure detection algorithms are being developed26,,29 including one incorporated onto a digital aEEG monitor.30 The off-line use of this algorithm has been tested in the setting of a prospective aEEG monitoring study in term-born infants admitted to a NICU with encephalopathy and/or seizures.31 The algorithm was able to detect 615/1116 (55%) seizures in 25 of the 40 patients who had true seizure activity. The positive predictive value was 73% with a false positive rate of 1 per 11 h. Shorter seizures were more likely to be missed, with a detection rate of 19% (72/371) for seizures less than half a minute long, but increasing to 87% (343/395) for seizures lasting 1–10 min. Using the same automated system, van Rooij et al32 studied 214 seizures in 15 patients. The algorithm detected 140 (65%) of 214 seizures, with a false positive rate of 1 per 13.5 h of recording.

Although the field of computer-assisted seizure detection is exciting and has seen an explosion of research activity over the last 10 years, to date no single automated seizure detection system is reliable enough to substitute for an experienced electroencephalographer in the clinical environment; rather, these algorithms are used to complement review of the results. The different algorithms have been tested on different populations with non-uniform reporting of algorithm performance, so what may be impressive statistically may not be as useful clinically.33 If a seizure detection algorithm is to perform optimally in the NICU setting it must be ‘trained’ using a large amount of multichannel data from as many neonates as possible with and without seizures, with a variety of aetiologies and gestational ages.

In future, it seems likely that reliable automated seizure detection systems will be developed and incorporated into cotside EEG monitors. This will revolutionise seizure detection and management in the newborn.

Conclusion

So which babies should be monitored for seizures? Box 1 provides some suggestions. Seizures represent an important manifestation of cerebral pathology in the newborn and require urgent diagnosis and treatment. A large proportion of electrographic seizures have no clinical correlates and conversely a large number of infants with abnormal movements do not have electrographic seizures. Most neonatal units do not have access to continuous video-EEG surveillance. Digital aEEG used with the raw EEG trace by experienced users can probably detect three-quarters of all seizures. There is much exciting research being carried out on automated seizure detection systems which shows promise and is likely to deliver cotside equipment within the next decade.

Box 1

Which newborns should be monitored for electrographic seizure?

Babies suspected of clinical seizures or babies exhibiting abnormal movements

Babies known to have had a hypoxic–ischaemic insult

Babies with moderate or severe encephalopathy

Babies undergoing therapeutic hypothermia (particularly in the rewarming phase) or any other neuroprotective intervention

Consider monitoring very preterm infants in the first few days of life, and those with abnormal cranial ultrasound imaging

Acknowledgments

DKS would like to acknowledge his gratitude to Dr Paul Clarke and Dr Ajay Sinha for their suggestions for this article.

References

View Abstract

Footnotes

  • Funding JMR and GBB are collaborators on a Translational Award from the Wellcome Trust (85249/z/08/z) investigating automated seizure detection in the newborn. Some of this work was undertaken at UCLH/UCL who received a proportion of funding from the Department of Health's NIHR Biomedical Research Centres funding scheme. The views expressed in this publication are those of the authors and not necessarily those of the Department of Health.

  • Competing interests None.

  • Ethics approval This study was conducted with the approval of the hospitals concerned.

  • Provenance and peer review Commissioned; externally peer reviewed.

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.