Objective To determine the sensitivity and specificity of amplitude integrated electroencephalogram (aEEG) compared to simultaneous standard electroencephalogram (sEEG) for seizure detection and background discontinuity.
Design Prospective paired cohort.
Setting Tertiary academic neonatal intensive care unit.
Patients Infants were recruited from 2005 to 2008. Neonates requiring sEEG were recruited for simultaneous aEEG.
Interventions Following sEEG and aEEG, seizures were recorded as present or absent, and background was recorded as normal or discontinuous in each format.
Main outcome measures Presence of at least one seizure during recording. The background activity was reported as normal or discontinuous. Discontinuity of brain activity was further ranked as mild, moderate or severe.
Results 51 sEEG and aEEG studies were completed. 44 studies were analysed for presence of seizures and 46 were analysed for background discontinuity. Sensitivity for presence of seizures by aEEG was 80% and specificity was 50%. The proportion of infants with seizures were overdiagnosed by aEEG (63.6% vs 45.5% for sEEG p=0.045). Discontinuity of background activity had higher sensitivity (88.6%) and specificity (54.5%) when compared with seizure detection.
When stratified by indication for EEG, hypoxic episode (n=14) or suspected seizures (n=33), similar sensitivity for presence of seizure (80%) was noted by aEEG and sEEG. However the specificity of aEEG for seizure detection was higher in neonates undergoing EEG for suspected seizures (66.7% vs 22.2%).
Conclusions Background abnormalities were detected with fair accuracy by aEEG but aEEG criteria alone would result in the overdiagnosis of neonatal seizures. Therefore seizures noted on aEEG require sEEG confirmation prior to implementing anticonvulsant therapy for neonatal seizures.
Statistics from Altmetric.com
Neurophysiological studies in the neonatal intensive care unit (NICU) are used as critical evidence for selecting neuroprotective interventions.1,–,4 Electroencephalogram (EEG) recordings aid in the diagnosis of encephalopathy in neurologically depressed infants as well as serve to confirm suspected seizures or alert the clinician to sub-clinical seizures.5,–,8 Neurological injury is frequent with hypoxic ischemic encephalopathy (HIE), metabolic disorders, infections, stroke and drug exposure resulting in neurological dysfunction in 2–6 neonates per 1000 live births.9 10
What is already known on this topic
Neonatal seizures are often missed by clinical examination alone and may be associated with increased risk of death and severe cerebral palsy, irrespective of aetiology.
The use of amplitude integrated EEG (aEEG) is becoming increasing popular in NICU, but the accuracy and reliability of interpretation of aEEG is not established.
What this study adds
Detection of discontinuous background activity by aEEG was highly sensitive (88.6%) and specific (54.5%).
Presence of seizures was overdiagnosed by aEEG (63.6 vs 45.5% for sEEG p=0.045) signifying the need for sEEG confirmation prior to implementing anticonvulsant therapy.
Neonatal seizures are very difficult to diagnose as clinical signs may be very subtle2 or absent in as many as 80% of electrographic seizures.11 12 Diagnosis of seizures and neonatal encephalopathy is more challenging in critically ill neonates that are intubated, sedated or paralysed due to respiratory failure. Electrographic seizures in neonates with HIE are associated with increased likelihood of mortality and poorer long-term neurological outcome, when compared with HIE infants without seizures.13 14 Recurrent or prolonged seizures in the neonatal rat model result in reduced neurogenesis.15,–,19 This data supports the treatment of frequent seizures; however specificity in seizure diagnosis is also important because treatment of all seizure-like movements may be detrimental to the developing brain. Anticonvulsant medications commonly used in neonates have been shown to increase neuronal apoptosis in the immature rodent brain.20 The accurate and timely detection of neonatal seizures and appropriate institution of anticonvulsant therapy will likely play a role in minimising brain injury attributable to either seizure or inappropriate therapy. This may improve long term neurological outcomes.
A standard EEG (sEEG) routinely used to detect encephalopathic processes or an epileptic seizure requires the placement of 16 leads by an expert technician. Interpretation of sEEG requires a clinical neurophysiologist with training and experience in reading neonatal EEGs. These requirements present logistic challenges for many NICUs. These limitations have driven the emergence of a simpler methodology for monitoring cerebral function.
Amplitude-integrated EEG (aEEG) was developed as a cerebral function monitor for adults undergoing bypass surgery and later adapted for neonatal use.21,–,24 aEEG uses two to four leads placed over the parietal area of the brain that can be applied after minimal training. aEEG interpretation has been portrayed as relatively easy for clinicians without detailed training in neonatal neurophysiology, though evidence supports a significant learning curve.25,–,27 aEEG is used to monitor neurologic status, seizures, effects of therapies and to predict neurological outcome following HIE.3 27,–,29 aEEGs use has been increasing in the NICUs across the country,29,–,31 especially to determine the eligibility for therapeutic hypothermia following HIE.31
The goal of the present study was to determine the sensitivity and specificity of aEEG for seizures and for background discontinuity when compared with sEEG. A secondary goal was to evaluate use of aEEG in the neonates with and without HIE.
Study design and methods
This study was approved by the University of California Los Angeles Institution Review Board (UCLA Office of the Human Research Protection Program). Infants were recruited from the UCLA Medical Center NICU if they qualified, and an investigator was available to consent parents at the time of the study. Infants were eligible for study if they required continuous EEG monitoring as determined by the treating physician in the NICU who was not a member of the research team.
EEG application and interpretation
EEG was performed following protocols standard for bedside EEG. Electrode placement was completed by an EEG technologist with experience in working with newborns. Placement used the international 10–20 system modified for neonates (figure 1A). Gold disc electrodes were attached using paste. Respiratory, chin EMG channels and eye leads were also placed per neonatal protocol. The EEG was recorded in a standard neonatal montage and used the proper impedance checks and calibration common to all neurophysiology laboratories.
The duration of continuous EEG monitoring was determined by the clinical team with the pediatric neurology consultant independent of the research team. A digital continuous ICU EEG monitoring system from either the Stellate VITA (Montreal, Quebec) or a Cardinal Health Nervus (Dublin, Ohio) was used to record simultaneous scalp EEG and performed amplitude integration from the C3, C4, O1 and O2 leads (figure 1B). The aEEG was produced from the central and occipital leads on both the right and left sides. Both the sEEG and aEEG recorded from each infant were interpreted by paediatric neurologists with special training in neonatal EEGs and certified in clinical neurophysiology. The aEEG and sEEG were evaluated independently. Raw sEEG was used to confirm the seizures observed on aEEG. When evaluating the aEEG, seizures were recorded if there was at least one raised notch in both the lower and upper aspect of the band of cerebral activity.
Seizures were reported as present if the electrical activity in sEEG became rhythmic with fast synchronous potentials which evolved and were present for at least 10 s at least once during the complete EEG recording.32 33 Discontinuity on sEEG was dependent on the postmenstrual age and the sleep state of the patient. The interburst interval duration, synchrony in the waveforms during quiet and active sleep and the amplitude of the waveforms during a burst were reviewed to determine discontinuity for age.33 Background activity was recorded as normal, mild, moderate or severely discontinuous on both aEEG and sEEG.32,–,34
To control for any bias introduced by the use of a single neurologist, the last 20 complete aEEG and sEEG recordings were reviewed by a second neurologist (JTL). The second neurologist was blinded to both medical history and patient identity when reviewing either aEEG or sEEG.
Overall good agreement was found between the first and second neurologists' interpretations. The agreement for seizure detection between the two readers was 95% by sEEG and 75% by aEEG. The agreement for background discontinuity was 90% for both sEEG and aEEG. Of the five studies discrepant for seizure diagnosis by aEEG, two studies were interpreted as having no seizure and three studies were found to have seizure by the second neurologist, where the first had come to the opposite conclusion.
Sample size and statistical analysis
Inclusion of 45 subjects was determined to yield a power of 80% to detect the sensitivity within 15% in paired testing. Data analysis was completed using SAS V.9.1. Sensitivities and specificities were calculated using frequency tables. Continuous variables were compared using Wilcoxon rank sum test. Proportions were compared using McNemar's test for paired analyses. Fisher's exact test was used to compare categorical data. An α-value of 0.05 was considered statistically significance. Multivariate analysis was performed on the complete cohort using the generalised estimating equation with outcome of seizure agreement between aEEG and sEEG. Independent variables included in the model were sex and postmenstrual age. Logistic regression was also performed using the same variables on the first EEG from each patient.
Fifty-one studies were examined on 47 neonates (table 1). Fifty-three per cent of the EEG recordings were 24 h long with a range from 12 to 360 h. In the comparison of aEEG to sEEG only the first study was included from the four subjects who underwent two EEGs. The amplitude integration software failed during one recording. Two studies were of such low amplitude that seizure detection on aEEG was impossible and so were excluded from the comparison of seizures but not the comparison of the background. Forty-six neonates were therefore included in the analysis of background activity and 44 infants were included in the analysis of seizure detection.
Thirty-three subjects with suspected seizures (table 1) were either exhibiting abnormal movements or were experiencing prolonged or frequent episodes of apnoea and bradycardia. Ten of 14 infants in the hypoxia group (table 1) had low Apgar scores and abnormal neurological examination, while the other four infants required cardio-pulmonary resuscitation due to an acute decompensation in their clinical status.
When subjects were stratified by their indication for EEG, (hypoxia n=14 vs suspected seizures n=33), they did not differ by median gestational ages, (37 vs 37), or postmenstrual age at time of EEG (39 vs 40). However, the hypoxia group had lower median Apgar scores at 5 min, (4.5 vs 8, p=0.003) and tended to be of lower median age at EEG (2.5 days vs 20 days, NS)
In the 44 studies examined for presence of seizures 20 (45.5%) were found to have seizures on sEEG. Seizures were significantly overdiagnosed by aEEG with seizures detected in 63.6% of infants (p=0.045) (table 2A). The most common reason for false positive seizure diagnosis by aEEG was movement artifact. Sensitivity of seizure detection by aEEG was 80.0±11.8% (95% CI, figure 2A). Specificity for seizure detection was only 50.0±14.8% (figure 2A). In the hypoxia group (n=14) seizures were overdiagnosed by aEEG when compared with sEEG (78.6% vs 35.7%) (p=0.039, Table 2A). No difference in detection of seizures was noted in the suspected seizures group (56.7% vs 50.0%; p=0.48, table 2A).
Movements consistent with clinical seizures were noted in 81.8% of infants, however not all infants with seizure on sEEG had clinical manifestations. Two infants were identified as having electrographic seizures without any clinical indications.
Seizure detection by aEEG had the same sensitivity in both infants undergoing EEG for hypoxia and those undergoing EEG for suspected seizures (80±21.0% vs 80±8.6). However the specificity for seizure detection was higher in those infants undergoing EEG for suspected seizures (66.7±16.9% vs 22.2±21.7%). Of the four infants who had repeat EEG testing, two had seizures which were correctly identified by aEEG during both initial and repeat testing. Two infants were seizure free on sEEG; however aEEG misdiagnosed a seizure in one of these infants undergoing a repeat study. This was due to significant artifact during the sEEG recording which was picked up as a seizure by aEEG.
In the 46 studies examined for discontinuous background activity sensitivity and specificity for aEEG were higher (88.6±9.2% and 54.5±14.4% respectively, figure 2B). Discontinuous background activity was observed in 78.3% of the 46 aEEG studies compared with 76.3% on sEEG, which was not statistically different (p=0.739, table 2B). Similarly, no difference was detectable between the sEEG and aEEG in identification of discontinuous background activity in the hypoxia group or the suspected seizures group (table 2B).
Agreement between sEEG and aEEG, defined as both studies producing the same severity of discontinuity, was 55.0% for normal studies, 50.0% for mildly discontinuous studies, 45.8% for moderately discontinuous studies and 0% for severely discontinuous studies (table 3).
The sensitivity of aEEG to find discontinuous background activity in the suspected seizures group was similar to the hypoxia group (86.6±11.6% vs 91.7±14.5%). Specificity for discontinuous background in the seizure group was 66.7±24.7%. Specificity for discontinuous background in the hypoxia group was not calculated as only 2/14 backgrounds were continuous in the hypoxia group; however both were misidentified by aEEG. Of the four infants who had repeat EEG testing all four had discontinuous background activity which was correctly identified by aEEG during both initial and repeat testing.
On bivariate and multivariate analysis none of the variables examined significantly improved the agreement between aEEG and sEEG. Most of these models failed due to small sample size.
Seizures occur frequently in the neonate but are under-recognised and may result in cerebral injury as in the case of status epilepticus35 or persistent seizures.8 12 35 aEEG is rapidly being adopted by NICUs across the country to identify these events, especially in the management and treatment of infants with HIE.3 24 31 However, the accuracy and reliability of interpretation of aEEG by clinicians is not established.30 In the present study, simultaneously recorded sEEG and aEEG were independently examined by two trained clinical neurophysiologists when EEG was indicated for confirming seizures or for the assessment of encephalopathy after a hypoxic episode. Our results in neonates with a suspected seizures or hypoxic episode indicate that aEEG is sensitive in detecting seizures as well as abnormal background. However, there is an overestimation of seizures by aEEG. Therefore our results support that seizures detected by aEEG assessment alone require further confirmation by sEEG prior to implementing anticonvulsant therapy for seizures.
Our observation of overestimation of seizures by aEEG contrasts with reports that have described difficulty in detecting brief and infrequent seizures that were readily detected by sEEG.25 However, similar to our results, false positives seizures and low sensitivity for seizure detection have also been previously reported from use of aEEG alone by other investigators.36 Few studies have cautioned against erroneous aEEG readings.30 37 This apparent discrepancy in results may be attributable to the different outcome measures in these studies.
While previous studies examined individual seizure episodes,25 this study examined the whole EEG recording for presence or absence of seizures. Counting individual seizures provides an accurate number of seizures by giving equal importance to each seizure regardless of seizure duration or seizure severity in addition to validating the accuracy of seizure detection by aEEG among many readers.25 In this study the evaluation of each complete recording for presence or absence of seizures was performed in order to provide a clinically relevant diagnosis of seizures. In addition, all studies were reported by expert neurophysiologists. Since aEEG may be used as a screening test, this analysis emphasised the diagnosis of multiple or a prolonged seizures rather than every seizure episode or testing the ability of reader. aEEG does have some clear limitations due to the loss of information from the full 12 lead montage. While most seizures do occur over the C3 and C4 areas, approximately 20% do not.25
All the recordings in this study were interpreted by neurologists with expertise in neurophysiology and aEEG with the institutional goal of training neonatologists to learn interpretation once aEEG was shown to be accurate. The intrareviewer agreement of two trained neurophysiologists in aEEG interpretation of seizures was lower despite a very good agreement in the assessment of background abnormality. Previous studies using neonatologists experienced in reading aEEG have pointed out that neonatal seizures are difficult to detect on an aEEG, especially when they are infrequent, brief, or of low amplitude.25 38 Our results show that this shortcoming in diagnosis of seizures by aEEG persisted even among expert neurophysiologists.
Background activity is routinely used in screening for therapeutic hypothermia for HIE.31 The background abnormalities were detected with fair accuracy in infants with suspected seizures as well as hypoxia episode. In addition very good intrareviewer agreement was observed for detection of abnormal background activity. Despite several limitations,25 30 aEEG complements the neurological exam and provides a reasonable alternative for newborn infants when continuous sEEG is not available. Current aEEG systems also display the simultaneously recorded EEG from channels the tracing based upon. This may improve the accuracy of aEEG diagnosis. Our observations and review of current literature indicates that sEEG should either be obtained simultaneously (and reviewed by experts) or after the initial assessment with aEEG in almost all circumstances.
EE was supported by the NICHD T32 training grant no. 07549. We thank the UCLA Pediatric Electrophysiology Laboratory for their technical support in conducting these studies.
Conflicts of interest None.
Ethics approval This study was conducted with the approval of the UCLA Institutional Review Board: Office for Protection of Research Subjects (OPRS), 11000 Kinross Avenue, Suite 102, Box 951694, Los Angeles, California 90095-1694.
Provenance and peer review Not commissioned; externally peer reviewed.
Patient consent Parental consent obtained.
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.