Assessment of Neonatal Encephalopathy by Amplitude-integrated Electroencephalography

DISCUSSION

We have shown that the aEEG correlates closely with neurologic outcome in newborn infants with encephalopathy from different causes and also correlates with the results of other more specialized investigations.

Our aim was to study full or nearly full-term infants with encephalopathy. We deliberately chose an inclusive definition of encephalopathy and did not restrict the study to infants with a classical presentation of birth asphyxia. Previous work has demonstrated that aEEG is an accurate predictor soon after birth in infants preselected because of suspected asphyxia. However, the clinical diagnosis of birth asphyxia is difficult and frequently controversial and, for the clinician faced with an infant with encephalopathy, the cause of the problem frequently is of less importance than the prognosis. Consequently, we wished to know the value of aEEG in an unselected cohort of infants diagnosed as suffering from encephalopathy.

We did not investigate very preterm infants. Interpretation of the aEEG is more difficult in the preterm infant because of EEG changes related to gestational age: a burst suppression pattern observed in extremely preterm infants changes to a discontinuous pattern in the more mature infant.^{20 ,21} Only 2 of the study infants were born before 37 weeks' gestation (1 born at 35 and the other at 36 weeks' gestation). Therefore, it is very unlikely that immaturity influenced our results. Moreover, a suppressed pattern is probably ominous at any gestation.²²

We did not observe any complications related to the aEEG or the use of needle electrodes; these were tolerated well for prolonged periods and did not cause any injury. In every case, aEEG was begun without difficulty, usually by nursing or junior medical staff. Although we did not examine systematically the influence of drugs on the aEEG, others have reported transient suppression of the aEEG after administration of anticonvulsants.^{8 ,12} Because only 1 infant had a briefly abnormal aEEG, it is unlikely that medications influenced the results. By excluding from analysis the parts of the records that were associated with drug administration or handling of the infant or where the impedance signal was disturbed, we minimized possible confounding effects.

We were able to obtain normative data from healthy infants soon after birth. These infants exhibited regular changes in aEEG voltage that have been ascribed to sleep state^{9 ,15}; a wider band of aEEG, probably associated with quiet sleep and comparable with trace alternate pattern seen on EEG, alternated with a narrower band of activity, which is associated with active sleep or wakefulness. The duration of the wider band of aEEG activity was remarkably consistent at ∼20 minutes on most observations. These findings are similar to those reported from infants studied at 2 to 5 days after birth⁹ and suggest that the birth process does not have a prolonged influence on EEG activity. This normative dataset provides comparative data for those obtained from infants with encephalopathy very soon after birth.

We defined voltage criteria for classifying the aEEG because descriptions such as “continuous low voltage,” used in previous reports,¹³ may be ambiguous. The minimum lower voltage in the normal infants was 6.5 μV. We chose to set 5 μV as the lower limit of normal in our classification, for three reasons: 1) for simplicity of practical application because 5 μV is easily determined on the aEEG record; 2) to allow a small margin for error; and 3) because that value is very similar to the criteria set empirically by previous studies of the prognostic value of aEEG after birth asphyxia,^{8 ,12} thus allowing our results to be compared.

To assess the use of these criteria, we compared the classification of 50 aEEG records by 2 pediatric residents with an expert assessment. The value of the κ statistic was uniformly high, showing that there was very close agreement between the assessments, both for classification of amplitude and for recognition of seizures. The interpretation of the κ statistic is that values of 0.6 to 0.79 show substantial agreement, and values >0.8 show almost perfect agreement. This suggests that aEEG could be interpreted accurately by clinicians even in nonspecialist units after a modest period of training.

We defined the prognostic value of the test by calculating the conventional statistics of sensitivity, specificity, and positive and negative predictive value. We also applied a Bayesian approach and calculated the PP value of the test, which defines the probability that a single aEEG will predict correctly the prognosis of an infant to within 95% CIs, and is a useful summary statistic of the tests value. We also determined the likelihood ratios of positive and negative tests to facilitate further the application of the results.

In determining the prognostic value of the aEEG, we included death and neurodevelopmental impairment as a single outcome group. This is a practical classification, but might be incorrect if clinical decisions to discontinue care were made on the basis of the aEEG. In our unit, a decision that intensive care is not appropriate is never made on the basis of a single test, but rather after exhaustive investigation and discussion between professionals and parents have taken place. The close relation between aEEG, EEG, and MRI in both surviving and nonsurviving members of the poor outcome groups indicates that cessation of intensive care is unlikely to have introduced bias into our results.

We found a close association between the aEEG and the standard EEG, similar to findings from previous reports.^{10 ,11} A normal aEEG was associated very closely with a normal standard EEG and neurodevelopmental outcome. Only slight MRI abnormalities were identified in this group, apart from 1 infant with unusual patterns of abnormal signal in mesencephalon, brainstem, and cerebellum, but with normal cerebral hemispheres. However, infants with a unilateral cerebral infarct had normal-amplitude aEEG but focal abnormalities on the standard EEG. These data suggest that some deep-sited or focal lesions may be associated with a normal aEEG. Infants with moderately abnormal-amplitude aEEG had variable EEG and MRI findings, and the neurologic outcome correlated with findings on standard EEG and MRI. However, as has been reported previously,⁷ infants with suppressed amplitude or burst suppression pattern had a very poor prognosis, with all infants dying or surviving with multiple deficits. Although prognosis should be assessed after considering the results of all the investigations and will be dependent on the underlying diagnosis, our findings suggest that regardless of cause, encephalopathy associated with a suppressed aEEG is likely to have a very poor prognosis.

In infants with seizures, we found that the interictal aEEG activity correlated with subsequent outcome. Thus, 4 of the 6 infants with seizures and a normal-amplitude aEEG had a normal outcome, whereas 12 of 14 infants with seizures and a moderately abnormal or suppressed amplitude had a poor outcome. This was confirmed by the subsidiary analysis using only background activity to predict outcome. This showed that the PP value of the test was very similar to that when seizures were included, and well within the 95% CIs for the analysis. These results give confidence in the robustness of the technique.

Our findings confirm the high predictive value of aEEG soon after asphyxia. Eken and co-workers found the aEEG accurate for assessing prognosis within 6 hours of birth asphyxia.¹³ Our study was not designed to test the value of aEEG within 6 hours of delivery, and thus provides no direct data on the value of the technique in selecting infants for trials of neural rescue therapy, which need to be applied within this time. However, the predictive value of studies at a median age of 5 hours was greater than in later studies, which suggests that very early aEEG may be useful as a selection method for clinical trials.