Article Text

Download PDFPDF

Cot-side electroencephalography for outcome prediction in preterm infants: observational study
  1. Claire R West1,
  2. Jane E Harding2,
  3. Chris E Williams3,
  4. Melinda Nolan4,
  5. Malcolm R Battin5
  1. 1Neonatal Intensive Care Unit, Middlemore Hospital, Auckland, New Zealand
  2. 2Liggins Institute, University of Auckland, Auckland, New Zealand
  3. 3Department of Surgery, University of Melbourne, Victoria, Australia
  4. 4Neurology Department, Starship Children's Health, Auckland, New Zealand
  5. 5Newborn Service, National Women's Health, Auckland, New Zealand
  1. Correspondence to Dr Claire R West, Neonatal Intensive Care Unit, Middlemore Hospital, Private Bag 93311, Otahuhu, Manukau 1640, Auckland, New Zealand; westc{at}


Objective To assess the use of two-channel electroencephalographical (EEG) recordings for predicting adverse neurodevelopmental outcome (death or Bayley II mental developmental index/psychomotor developmental index < 70) in extremely preterm infants and to determine the relationship between quantitative continuity measures and a specialist neurophysiologist assessment of the same EEG segment for predicting outcome.

Design Observational study.

Setting The study was conducted in a neonatal intensive care unit.

Patients Preterm infants born <29 weeks' gestation.

Interventions Two-channel EEGs using the reBRM2 monitor (BrainZ Instruments, Auckland, New Zealand) within 48 h of delivery. One-hour segments were analysed, blinded to the clinical outcome, by off-line quantitative analysis of continuity and a review of the raw EEG by a neurophysiologist.

Main outcome measures Developmental assessment at a median of 15 months' corrected age.

Results 76 infants had an EEG within 48 h of delivery and a developmental assessment. The analysed segment of the EEG was obtained at 24 (3–48) h of age (median (range)). The neurophysiologist's assessment was a better predictor of adverse outcome than the continuity measures (positive predictive value 95% CI 75 (54% to 96%) vs 41 (22% to 60) at 25-µV threshold, negative predictive value 88 (80% to 96%) vs 84 (74% to 94%) and positive likelihood ratio 9.0 (3.2 to 24.6) vs 2.0 (1.2 to 3.6)). All the infants with definite seizures identified by the neurophysiologist had poor outcomes.

Conclusions Modified cot-side EEG has potential to assist with identification of extremely preterm infants at risk for adverse neurodevelopmental outcomes. However, analysis by a neurophysiologist performed better than the currently available continuity analyses.

View Full Text

Statistics from


Amplitude-integrated electroencephalography (aEEG) has been widely incorporated into routine practice for term infants with neonatal encephalopathy and has been used for outcome prediction,1 seizure management and entry criteria for trials of neuroprotective strategies.2 3 However, interpretation of the preterm aEEG is more complex because the normal background pattern is discontinuous. Most reports on outcome prediction are related to variation in the number of bursts in a specified period,4 5 for which there is no readily available method of analysis. Early work using the Oxford Medilog recorder included an analysis of discontinuity and amplitude.6 7 However, this system has not been incorporated into clinical practice for preterm infants. Nevertheless, information from early EEGs may help identify preterm infants at highest risk for adverse outcome and allow targeted interventions to be assessed.

What is already known on this topic

  • Conventional EEG has some use for predicting outcome in preterm infants. However, neonatal neurophysiology services are not universally available in neonatal intensive care units.

  • Cot-side limited-channel EEG may assist outcome prediction for term infants with encephalopathy and is available in many neonatal units.

What this study adds

  • Quantitative analyses of early two-channel cot-side EEG may assist identification of extremely preterm infants at risk for adverse neurodevelopmental outcome.

  • Neurophysiologist interpretation of raw two-channel EEG from cot-side recordings had similar sensitivity but greater specificity and positive and negative predictive values than quantitative analyses.

Cot-side EEG recordings offer some potential in this role. Some devices allow generation of quantitative neurophysiological parameters, including amplitude and measures of continuity,8 9 which may assist clinicians' interpretation of recordings. These EEG parameters must be critically examined to determine which, if any, may assist with outcome prediction. However, excessively discontinuous EEG activity is linked with adverse outcome in preterm infants.10

The aims were, first, to assess the utility of two-channel EEG recordings in prediction of adverse neurodevelopmental outcome in extremely preterm infants and, second, to determine the relationship between specialist neurophysiologist and quantitative continuity assessments of the same segments of EEG in the prediction of adverse neurodevelopmental outcome.



Infants born before 29 weeks' gestation and admitted to the National Women's Hospital, Auckland, New Zealand, between March 2002 and February 2004 were recruited. Infants with serious congenital head malformations or serious scalp sepsis were excluded. Written informed parental consent was obtained before EEG recordings began. This study was approved by the Auckland ethics committee.

EEG measurements

Recordings were obtained using hydrogel electrodes (Hydrospot neonatal electrodes; Physiometrix Inc, North Billerica, Massachusetts, USA) placed on the C3, P3, C4 and P4 regions as defined by the modified International 10–20 System after appropriate skin preparation (Nuprep; D O Weaver & Co, Aurora, Colorado, USA). The electrodes were stabilised with hydrogel tape (Klear-Tape; Cas Medical Systems, Branford, Connecticut, USA). The EEGs were recorded on the ReBRM2 (research version of the BRM2; BrainZ Instruments, Auckland, New Zealand) in 2- to 12-h periods. Conventional ventilation or continuous positive airway pressure was provided according to the infant's clinical status, and changes were made as required according to standard protocols. Routine nursing procedures were performed as required and documented on the recording. Episodes of physiological instability, drug administration and blood sampling were also marked on the recording. No infants received anticonvulsant medication during their neonatal intensive care unit (NICU) admission.

Right and left EEG signals were amplified 5000 times and band-pass filtered with a first-order high-pass filter at −3 dB with a frequency at 1 Hz and a fourth-order low-pass Butterworth filter at −3 dB with a frequency at 50 Hz. The signal was digitised by the computer at a sampling rate of 256 Hz. The mean signals were analysed off-line using customised software (Chart Analyser; Liggins Institute, Auckland, New Zealand). EEG data were defined as valid and included in the analyses if the electrode impedance was <15 kΩ per pair and the continuous data epoch was ≥120 min. Continuity measures were determined as the percentage of each minute during which the amplitude of the raw EEG (assessed at 2-s intervals) was above the determined threshold amplitude (25 or 50 μV).9 The maximum positive amplitude in a 2-s section of raw EEG was added to the maximum negative amplitude in the same section of EEG. If the sum of these amplitudes was above the threshold amplitude, then that section of EEG was considered continuous for the analysis; otherwise, it was considered not continuous (figure 1). The EEG continuity measurements were determined and stored to a disk at 1-min intervals to give the percentage of each minute that the EEG amplitude was above the threshold.

Figure 1

Examples of low- and high-continuity EEG traces. Each example shows a 10-s segment of a raw EEG trace divided into 2-s segments. The top trace is from the left side of the infant's head, and the bottom one is from the right side. Note the difference of scales on the vertical axes with (A) being ±50 µV and (B) being ±100 µV. The vertical black bar indicates 50 µV in both traces. The horizontal dashed lines in some segments illustrate the maximum positive and negative amplitudes for that segment. If this maximum amplitude is above the threshold value (25 or 50 µV), then that segment is considered to be continuous. If the maximum amplitude for that segment is below the threshold value, it is considered to be discontinuous.

EEG analysis

Sixty-minute segments were analysed, and median values of each quantitative variable were recorded. To minimise potential bias, all the segments for the analysis were chosen using a screen showing electrode impedance but no quantitative analyses. Segments were chosen to avoid marked events whenever possible. If a prolonged segment of EEG without event marks was available, a 60-min portion in the middle of this was chosen. If there was no 60-min segment of EEG without events, then a 60-min segment in the middle of the recording was chosen. The same segment of raw EEG was analysed for each of the quantitative measures. The means of the left and right quantitative values measured over the same segment are presented, as the patterns were the same for both sides.

The receiver operating characteristic curves for prediction of adverse outcome at 18 months for the first EEG recorded after delivery have been constructed previously in a larger cohort, of which the infants in this study comprised 63%.11 These produced optimal cutoff values for the prediction of adverse outcome of ≤60.3%/min for the 25-μV continuity threshold and ≤30.6%/min for the 50-μV continuity threshold. These cutoff values were used in the current study.

The neurophysiologist, blinded to the clinical outcome, reviewed the chosen EEG segment analysed, as described previously, for each infant using the EEG viewer developed for the cot-side monitor (EEG Viewer V.9.26; BrainZ Instruments). In addition, the neurophysiologist received a list of the marked events for the period analysed. The EEG traces were assessed for interburst interval, continuity, seizure activity, amplitude (qualitative assessment) and developmental features (eg, the presence of delta brushes) to categorise the EEGs as normal, within normal limits, mildly abnormal, moderately abnormal and severely abnormal (table 1, figure 2) based on published criteria.12 To improve the power of the analyses, by increasing the numbers in each group, these categories were then grouped as normal/mildly abnormal (the normal, within normal limits and mildly abnormal groups) or abnormal (the moderately abnormal and severely abnormal groups).

Figure 2

Example of subclinical EEG seizure activity. The illustration shows a 20-s segment of raw EEG trace with repetitive rhythmic discharge consistent with seizure activity, more evident on the upper (left) trace. Throughout the hour of analysed trace, there were multiple subclinical seizure events such as that lasting between 10 and 90 s. (The upper trace reflects activity recorded over the left head, and lower trace reflects that over the right.)

Table 1

Qualitative assessment of EEG by a neurophysiologist

Developmental outcome

The infants were assessed by one of five psychologists at 18 months' chronological age using the Bayley II assessment to generate a mental developmental index (MDI) and a psychomotor developmental index (PDI). The infants who died or had an MDI or a PDI < 70 (>2 SD below the mean) were considered to have an adverse outcome, and those with MDI and PDI ≥ 70 (within 2 SD of the mean) were considered to have a satisfactory outcome.

Data analysis

Using two-by-two tables, we calculated the sensitivity, the specificity, the positive and negative predictive values, and the positive likelihood ratios with their 95% confidence intervals (CIs) for the prediction of adverse outcome by the neurophysiologist, and continuity measured at the 25- and 50-µV thresholds.

From our hospital data,13 we anticipated that the incidence of adverse outcome would be 25%. Our previous analysis had suggested that the continuity measures would have a specificity of approximately 80%.11 A sample size of 80 infants would, therefore, provide a 95% probability of estimating the specificity of the neurophysiologist or the continuity measurements for predicting adverse outcome with a lower CI of 60%.14


We recruited 84 infants with gestations < 29 weeks in the 2-year period (figure 3). Two infants did not have their first EEG within 48 h of delivery. The EEG trace for one infant was unsatisfactory. No outcome data were available for five infants (5.9% of the cohort), including two who emigrated. Therefore, we report the data from 76 infants (90% of the complete cohort) who had an EEG within 48 h after delivery and had outcome data available. They had a median (range) gestation at delivery of 26 (24 to 28) weeks and birth weight of 910 (420 to 1620) g. Their median (range) Apgar score at 1 min was 6 (1 to 9) and at 5 min was 9 (3 to –10). Their median (range) clinical risk index for babies score15 at 12 h was 4 (0 to –12). No infants were receiving sedation at the time of EEG recording.

Figure 3

Flow diagram for inclusion of preterm infants in the study.

The EEG recording was started at 23 (2 to –48) h. The 60-min segment of EEG reviewed was obtained at a median (range) of 24 (3 to –48) h of age. Valid EEG recordings were obtained for a median (range) of 247 (106 to –549) min. Fourteen infants had a variety of events during the analysed portion of the EEG. Some events had the potential to influence the raw EEG, including the administration of intravenous caffeine and the commencement of morphine infusion, one infant for each medication. However, exclusion of these infants did not alter the analysis; thus, they are included in this report. On early cranial ultrasound scanning (routinely at postnatal day 5 in our NICU), there were five infants with grade 1 intraventricular haemorrhage (IVH),16 four unilateral and one bilateral; two with grade 2 IVH, one unilateral and one bilateral; one with bilateral grade 3 IVH and four with grade 4 IVH, three unilateral and one bilateral. In addition, one infant showed significant white matter echogenicity on early cranial ultrasound scanning. Nine infants died before discharge. The 67 remaining infants had outcome assessments performed at 15 (13 to 41) months' corrected age. Two infants did not have a Bayley II assessment but were allocated a satisfactory outcome: one after an alternative formal developmental assessment at 18 months' chronological age and one after an assessment by a paediatrician, including an assessment of development, at 24 months' chronological age. Overall, 19 (25%) had adverse outcomes (death or an MDI/PDI < 70), and the remaining 57 (75%) had a satisfactory outcome. Nine infants died, and 10 had an MDI/PDI < 70.

Sixteen (21%) infants had abnormal EEGs as assessed by the neurophysiologist. The remainder of the EEGs were normal/mildly abnormal. Twenty-seven infants had EEGs with continuity calculated for the 25-μV threshold below the cutoff, and 22 infants had EEGs with continuity calculated for the 50-μV threshold below the cutoff.

The neurophysiologist's assessment of the EEG had similar sensitivity but higher specificity than continuity measured at the 25-µV threshold and a trend towards better specificity than continuity measured at the 50-µV threshold for prediction of poor outcome. There was also a trend towards a better positive likelihood ratio for the neurologist compared with both continuity thresholds (tables 2 and 3).

Table 2

Prediction of outcome at 18 months by qualitative neurophysiologist assessment, continuity at 25-µV threshold and continuity at 50-µV threshold

Table 3

Comparison of prediction of adverse outcome at 18 months by the neurophysiologist and the continuity measured at the 25- and 50-µV thresholds

Seizures and outcome

None of the infants were clinically suspected of having seizures, and none received anticonvulsant medication. Five infants were found to have definite seizures by the neurophysiologist, two had probable seizures and there were possible seizures found in four other infants. Three infants had periods of rhythmic activity on the EEG without evolution that was considered to represent artefact. There were no other abnormalities in background activity in these infants.

Four of the five infants with definite seizures died before discharge, and the other developed severe long-term neurodevelopmental disabilities. Both of the infants with probable seizures had MDIs and PDIs within the normal ranges.

In all of the infants with possible seizures, the duration of the observed episodes was ≤10 s. Of these infants, two had normal neurodevelopmental assessment (MDI/PDI of 88/86 and 95/101), and the other two had an MDI within the normal range but a PDI >1 SD below the mean (94/70 and 95/81).

Exclusion from the analyses of the infants with definite seizures or all infants with any suspicion of seizure activity did not substantively change the comparisons presented in table 3.


Formal 12-lead EEGs have been used for prediction of neurodevelopmental outcome in preterm infants10 but are labour intensive and technically challenging to perform. Older limited-channel cot-side EEGs, such as the Oxford Medilog system, have some predictive value for preterm infants,7 but the systems have not been widely used in NICUs. More recently, devices such as the BrainZ two-channel EEG have become prevalent in the management of term infants with encephalopathy and/or seizures. However, there have been few predictive data for preterm infants. Recently, Wikström et al reported the automated analysis of interburst interval from single-channel aEEG over the first 72 h of life and showed that increased interburst interval was an indicator of acute brain damage and associated with adverse neurodevelopmental outcome at 2 years of age.17 Seizure activity on review of raw EEG obtained using the BrainZ monitor in preterm infants was associated with poor outcome as defined by cranial ultrasound or magnetic resonance imaging,18 but neurodevelopmental outcomes are not yet published.

In our study, the expert neurophysiologist was better at identifying preterm infants at risk for adverse neurodevelopmental outcome than the quantitative neurophysiological measures of continuity when the same periods of two-channel EEG were analysed. Although only the differences in specificity were statistically significant, the positive predictive value and positive likelihood ratios were also better for the neurophysiologist, with overlapping CIs likely to reflect limited power because of the size of our cohort. The neurophysiologist's assessment relied on two main factors for assignment of EEG category: interburst interval and seizures. Interburst interval can be assessed at a number of different amplitude thresholds, as can continuity. However, these two parameters are different. Interburst intervals were manually calculated as the time between two segments of EEG activity above threshold amplitude. Quantitative continuity measurements reflect the maximum amplitude in 2-s segments of raw EEG from which the percentage of each minute above a set threshold is calculated. At a set threshold, a given continuity can result from several different EEG patterns. For example, a 50%/min continuity could be produced by alternating segments of 2 s of low-amplitude trace followed by 2 s of burst (higher-amplitude) trace or by 30 s of normal (higher-amplitude) trace followed by 30 s of low-amplitude trace. Between these extremes are numerous other possibilities. These two patterns would be interpreted very differently by a neurophysiologist. The first would be within normal limits, whereas the second is pathological at any gestation. This suggests that modification of current continuity algorithms to better reflect interburst interval may improve outcome prediction using quantitative EEG.

Choice of EEG portions for quantitative analysis should also be carefully considered, including the potential impact of movement, marked events, seizures and sleep state cycling. Whereas neurophysiologists can exclude portions of EEG traces that have movement artefact, most computer analyses cannot. Our data were consistent after the exclusion of infants with marked events or infants with definite or probable seizure activity. We limited the portion of EEG analysed to only 1 h, as reviewing the trace is a time-consuming process and 1 h reflects the period of conventional EEG usually assessed by a neurophysiologist. We also only examined EEG traces recorded in the first 48 h after birth. Many babies experience events such as IVH and white matter injury after this time, and it is possible that extending the analyses to include the whole recording and also recordings later in the neonatal admission may have enhanced outcome prediction by the neurophysiologist and possibly by the continuity analysis. However, events precipitating delivery or occurring around the time of birth are often the most significant and open to the potential for intervention and, therefore, are examined here.

The second EEG feature assessed by the neurophysiologist was seizure activity. None of the infants recruited were clinically diagnosed as having seizure activity, and none received anticonvulsant medication. However, the neurophysiologist was able to identify five infants with definite seizures. Four of these infants died, and the other had multiple disabilities. Another six infants had probable or possible seizures, and these infants had better outcomes, with none having a PDI or an MDI < 70. The poor outcome for infants with definite seizures may relate to the extent of seizure activity (being easily detected in a limited-channel EEG), the length of the seizures or the frequency of the seizures, as the seizures needed to occur within the 1 h of EEG analysed to be detected by the neurophysiologist.

Seizures may not be detected by two-channel monitoring. Without a complete montage of channels to obtain an overall picture of the EEG activity (ie, a conventional EEG), the discrimination between abnormal neuronal activity and artefact may be difficult and focal seizures may be missed. Clinician interpretation of cot-side EEG monitoring for seizure detection in term infants has been shown to be suboptimal.19

Seizures in preterm infants may be more difficult to detect by an algorithm because of the discontinuous background. Thus, without the expertise of a specialist neurophysiologist reading the raw EEG, seizures remain difficult to identify in preterm infants. Further research is required to examine whether these seizures can be detected by an automated system in real time. Although the numbers were small, the association between definite subclinical seizure activity and poor outcome was very strong, highlighting the need for further research into the identification of subclinical seizures and whether treatment of these seizures alters prognosis.

Conventional EEG reported by a neurophysiologist remains the criterion standard for infants in NICU. However, quantitative analyses of cot-side limited-channel EEG may add useful information when conventional EEG is not practicable. Neonatologists should be aware of the limitations of these quantitative analyses to avoid incorrect interpretation. Conventional EEG recordings might be expected to perform even better than our modified recordings for outcome prediction, as they provide further information in addition to interburst interval and improved seizure detection.

In summary, we consider the current study important for three reasons. First, it illustrates the potential for modified cot-side EEG to assist with identification of extremely preterm infants at risk for adverse neurodevelopmental outcome. This may, however, require further fine tuning, perhaps focusing on the factors used in the expert neurophysiologist analysis. Second, we are not aware of any literature comparing the use of quantitative EEG measures with expert interpretation of the raw two-channel EEGs from the same recording for outcome prediction beyond the neonatal period. Finally, the study reports five infants who had electrical but subclinical seizures and subsequently had poor outcome. Although further study is required to see if outcome can be moderated by seizure management, the first step is recognition of the subclinical seizures, and cot-side EEG offers potential to do this.


We are grateful to BrainZ Instruments for providing the use of the continuity analysis software developed by Dr Michael Navakatikyan.


View Abstract


  • Funding CRW received a University of Auckland Senior Health Research Scholarship during the period of data collection. The reBRM EEG machines were leased from BrainZ Instruments using a University of Auckland Staff Research Grant awarded to MRB.

  • Competing interests None.

  • Ethics approval This study was conducted with the approval of the Auckland Ethics Committee X and The University of Auckland ethics committee, and the Auckland District Health Board granted ethical approval.

  • Provenance and peer review Not 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.