Amplitude Integrated Electroencephalogram as a Prognostic Tool in Neonates with Hypoxic-Ischemic Encephalopathy: A Systematic Review

Introduction Perinatal management and prognostic value of clinical evaluation and diagnostic tools have changed with the generalization of therapeutic hypothermia (TH) in infants with hypoxic-ischemic encephalopathy (HIE) Aim to ascertain the prognostic value of amplitude integrated electroencephalogram (aEEG) in neonates with HIE considering hours of life and treatment with TH. Methods A systematic review was performed. Inclusion criteria were studies including data of neonates with HIE, treated or not with TH, monitored with aEEG and with neurodevelopmental follow-up of at least 12 months. The period of bibliographic search was until February 2016. No language restrictions were initially applied. Consulted databases were MEDLINE, Scopus, CINHAL and the Spanish language databases GuiaSalud and Bravo. Article selection was performed by two independent reviewers. Quality for each individual paper selected was evaluated using QUADAS-2. Review Manager (RevMan) version 5.3 software was used. Forest plots were constructed to graphically show sensitivity and specificity for all included studies, separating patients treated or not with hypothermia. Summary statistics were estimated using bivariate models and random effects approaches with the R package MADA from summary ROC curves. Meta-regression was used to estimate heterogeneity and trends. Results from the 403 articles initially identified, 17 were finally included and critically reviewed. In infants not treated with hypothermia the maximum reliability of an abnormal aEEG background to predict death or moderate/severe disability was at 36 hours of life, when a positive post-test probability of 97.90% was achieved (95%CI 88.40 to 99.40%). Positive likelihood ratio (+LR) at these hours of life was 26.60 (95%CI 4.40 to 94.90) and negative likelihood ratio (-LR) was 0.23 (95%CI 0.10 to 0.44). A high predictive value was already present at 6 hours of life in this group of patients, with a positive post-test probability of 88.20% (95%CI 79.80 to 93%) and a +LR of 4.34 (95%CI 2.31 to 7.73). In patients treated with TH the maximum predictive reliability was achieved at 72 hours of life (post-test probability of 95.70%, 95%CI 84.40 to 98.50%). +LR at this age was 24.30 (95%CI 5.89 to 71.30) and–LR was 0.40 (95%CI 0.25 to 0.57). Predictive value of aEEG at 6 hours of life was low in these patients (59.10%, 95%CI 55.70 to 63%). Conclusion This study confirms that aEEG´s background activity, as recorded during the first 72 hours after birth, has a strong predictive value in infants with HIE treated or not with TH. Predictive values of traces throughout the following 72 hours are a helpful guide when considering and counselling parents about the foreseeable long-term neurological outcome


Introduction
Perinatal hypoxic ischemic encephalopathy (HIE) is an important cause of early mortality and permanent major disability in full term infants [1,2]. Therapeutic hypothermia (TH) remains the only neuroprotective treatment available for these patients today [3,4]. Perinatal management and prognostic value of clinical evaluation and diagnostic tools have changed with the generalization of TH [5][6][7].
Amplitude-integrated electroencephalography (aEEG) is a simplified bedside neurophysiology tool that is nowadays widely used in neonatal units today [8] In term newborns with HIE, different authors have described a good correlation of aEEG with conventional EEG findings [9][10][11]. aEEG traces are assesed using two methods: voltage and background pattern [12,13] Both methods seem to be comparable for rapid clasification of electrocortical activity in newborns [14].
Prior to the use of TH for neonates with HIE, abnormal aEEG background pattern during the first 3-6 postnatal hours was reported to be highly predictive of adverse outcome, with the combination of aEEG pattern and clinical examination being more predictive than either of these parameters independently [12,[15][16]. With the use of TH some authors described that normal aEEG background pattern continued to relate to good neurological outcome whilst persistance of an abnormal aEEG tracing beyond 48 hours of life related to later altered neurodevelopmental outcome [17]. However, the temporal profile of the predictive ability of the aEEG in infants treated with TH during the first 72 hours of life has not been established [18].
A systematic review of the medical literature was performed to ascertain the prognostic value of aEEG in newborns with HIE considering hours of life and treatment with hypothermia. This review was part of the Spanish National Health System Clinical Practice Guideline (CPG) on HIE [19].

Methods
We conducted a systematic review and performed a meta-analysis (MA). Inclusion criteria were studies including newborns with HIE, treated or not with hypothermia, monitored with aEEG and with neurodevelopmental follow-up of at least 12 months.

Study selection and data extraction
Article selection was performed by two independent reviewers (DB and RR) and in case of discordance a third author intervened (AGA). Only studies including: 1) human infants with a gestational age of 35 weeks or more, 2) neonatal encephalopathy caused by perinatal asphyxia and 3) neurodevelopmental outcome data available at a postnatal age of at least 12 months were eligible. Review articles, opinion articles, and editorials were excluded as well as abstracts and conference proceedings. A standardized extraction form was used to record the characteristics of each study. For each paper aEEG background pattern at 6, 24, 26, 48 and 72 hours of life was analysed and neurodevelopmental outcome after follow up was recorded. Rates of true positive, true negative, false positive and false negative were recorded. aEEG background trace was considered abnormal according to the pattern classification if the trace was flat, low voltage or burst suppression [13]. Using the voltage classification, aEEG was considered abnormal if the trace's lower margin was below 5uV and its upper margin was below 10uV [12]. Adverse neurological outcome was defined as the combination of death or moderate/severe disability.

Quality assessment
Quality for each individual paper selected was evaluated using QUADAS-2 [20]. Quality was considered high if included population and outcomes were well defined. Quality was lowered if population was not well defined, follow-up was irregular or outcomes were not well specified. Influence of aEEG´s result on withdrawal of life support was not specifically described by any of the included articles except for the paper by Eken et al [21].
receving TH from patients not receiving this therapy. Summary statistics (sensitivity, specificity, likelihood ratios, diagnostic odds ratio) were estimated using bivariate models and random effects approaches with the R package MADA, from the summary receiver operative operating characteristic curve (SROC). MADA uses a linear mixed model with known variances of the random effects, similar to the computational approach by Reitsma et al [22]. Heterogeneity and trends were explored using meta-regression with the same package (S1 Table). Significant heterogeneity at 6 and 24 hours of life, depending on hypothermia treatment led us to separate summary estimates for patients receiving and not receiving hypothermia treatment. Post-test probabilities were computed for different hours of life, using our likelihood ratio estimations. For pre-test probabilities, the point estimations of incidences of death and moderate/severe disability reported by Tagin et al [3] were used.
ROC curves for different times of life for patients receiving or not TH are shown in Fig 4. Curves for patients receiving and not receiving TH are presented separately, as the prognostic value of aEEG differed for these two groups of patients at 6 and 24 hours of life. A statistical significant trend exists for sensitivity and specificity with increasing hours of life in patients receiving TH. This trend exists only for sensitivity in normothermia patients. After 36 hours of life, ROC curves for patients receiving and not receiving TH treatment overlapped. Pooled sensitivity and specificity, diagnostic odds ratios and an overall summary of this review are given in Table 3. In infants not treated with hypothermia the maximum reliability of an abnormal aEEG background to predict death or moderate/severe disability was at 36 hours of life, when a positive post-test probability of 97.90% (95%CI 88.40 to 99.40%) was achieved (Fig 5). Positive likelihood ratio at these hours of life was 26.60 (95%CI 4.40 to 94.90) and negative likelihood ratio 0.23 (95%CI 0.10 to 0.44). Positive post-test probability was already very high at 24 hours of life (97.60%, 95%CI 92.80% to 99%). In the following time epochs, aEEG´s predictive reliability remained high at around 95%. Notably, a high predictive value was already present at 6 hours of life in this group of patients, with a positive post-test probability of 88.20% (95%CI 79.80 to 93%) and a positive likelihood ratio of 4.34 (95%CI 2.31 to 7.73).
The maximum predictive reliability for death or moderate/severe disability of an abnormal aEEG in patients treated with hypothermia was achieved at 72 hours of life (post-test probability of 95.70%, 95%CI 84.40 to 98.50%). Positive likelihood ratio at this age was 24.30 (95%CI 5.89 to 71.30) and negative likelihood ratio 0.40 (95%CI 0.25 to 0.57). Compared with newborns not treated with TH, the predictive value of aEEG at 6 hours of life (Fig 6) was low in these patients (59.10%, 95%CI 55.70 to 63%).

Discussion
This systematic review provides a comprehensive summary of the prognostic value of aEEG in newborns with HIE. It is the first review that includes infants treated with hypothermia and, more importantly, stratifies the results according to postnatal age. This study confirms that aEEG background activity, as recorded during the first 72 hours after birth, has a strong predictive value in infants with HIE treated or not with TH.
In 2007 Spitzmiller et al [18] published a MA of the prognostic value of aEEG in newborns with HIE. Papers included were published between 1996 and 2005. In this MA, an abnormal aEEG had high positive and negative likelihood ratios for the prediction of death or moderate to severe disability (+LR 10.1, 95%CI 5.5 to 18; -LR 0.09, 95%CI 0.06 to 15). Limitations of this MA were heterogeneity between studies and different times of aEEG pattern assessment. Also no studies with hypothermia were included, as the analysis was performed in the pre-hypothermia era. Van Laerhoven et al recently reviewed the different prognostic tools for newborns with HIE [36]. These authors presented their results for aEEG as pooled sensitivities and specificities at 6, 24 and 72 hours of life. During the first 6 hours of life, pooled sensitivity for the Determining which infants with HIE will have significant brain damage is crucial. Among the diverse tools that can assist with prognosis are clinical examination, electrophysiological investigations including aEEG or imaging of the brain. Treatment with hypothermia has indeed changed the prognostic value of the initial aEEG trace and the time cutoffs for outcome prediction based on aEEG in newborn infants with HIE. The seminal work by Thoresen et al [17] showed that the positive predictive value of aEEG before 6 hours of life was significant in normothermia-treated neonates following perinatal HIE (84%), but lower among hypothermia-treated infants (59%). Our analysis showed similar predictive values, with clear differences existing between neonates treated or not with TH. As Infants with moderate or severe HIE are cooled within the first 6 hours of life, differences in aEEG´s predictive values at this early age deserves an explanation. Although profound hypothermia depresses EEG voltages, it has been shown that moderate cooling has no effect on aEEG background pattern both in animal models [37] and clinical studies [38]. Reduction in the predictive reliability of early aEEG (6 and 24 hours of life) for adverse outcome in hypothermia-treated infants compared with normothermic infants could be explained by the beneficial effect of TH on the ongoing brain damage [3]. Additional neuroprotection due to sedative drugs frequently administered in patients receiving hypothermia cannot be ruled out [39]. Another important finding in in the work by Thoresen  [17] was that whilst all infants cared for at normothermia who normalized their background pattern beyond 24 hours after birth had a poor outcome, some of the hypothermia-treated infants still developed normally after an abnormal aEEG at 24 hours as long as the aEEG recovered before 48 hours. Our study confirmed a similarly high predictive reliability for death or moderate/severe disability of an abnormal aEEG at 48 hours of life in patients treated with hypothermia (post-test probability of 93%, 95%CI 83.5 to 96.8%). The effect of TH on tools for neurological assessment of the newborn with HIE other than aEEG has been evaluated in individual studies. In a similar way to aEEG, the predictive value of the severity of clinical encephalopathy at less than 6 hours of age is lower for hypothermiatreated than for normothermia-treated infants [40,41]. The time course of clinical encephalopathy throughout TH is valuable in outcome prediction. While TH reduces brain injury on magnetic resonance imaging (MRI) in newborn infants with HIE, the predictive value of MRI for subsequent outcome is not affected by TH [42]. Combined with clinical and neuroimaging information, aEEG is of valuable help to the clinician when considering the need for neuroprotection and the likely outcome of infants with HIE.  The strength of the present work is the differentiation of prognostic values depending on postnatal age and use or not of TH. Traces at less than 6 hours and in normothermia conditions are useful for the selection of infants with significant HIE needing neuroprotection in order to improve outcome. Predictive values of traces throughout the following 72 hours are a helpful guide when considering and counselling parents about the foreseeable long-term neurological outcome. In order to provide strong evidence of the predictability of aEEG and its 72-hour time course in infants with perinatal HIE a thorough revision of available literature was needed.  We are conscious of several limitations of our work. The aim to include evaluation at different epochs of postnatal life lowered the number of patients included for each period. Also, heterogeneity was high when considering timing and duration of aEEG monitoring, duration of follow-up and neurodevelopmental tests used by different authors. In studies without TH the percentage of patients with mild HIE is higher than in studies using TH, a factor that might influence aEEG´s prognostic reliability. In addition, the ability of different recording equipment used from initial studies to the latest may be different. Our analysis did not include sleepwake cycling, seizures or administration of anticonvulsants or narcotics. One important limitation is the general lack of information in the original papers on the numbers of deaths that occurred after withdrawal of intensive care as a consequence of abnormal results of prognostic tools, including aEEG. Of the studies included, only one from the era preceding TH by Eken gave account of the causes of death [21]. In this study, 9 of 17 deaths followed withdrawal of intensive care occurring as a result of a flat aEEG trace beyond 24 hours of life. This limitation could introduce a bias in the prognostic reliability of aEEG.

Conclusion
This new MA of the prognostic value of aEEG in newborns with HIE incorporates for the first time hours of life and treatment with hypothermia. With TH the maximum capacity reliability of aEEG to predict adverse neurological outcome is delayed from 24-36 to 48-72 hours of life.