pediatrics
February 2017, VOLUME139 /ISSUE 2

Prognostic Accuracy of Electroencephalograms in Preterm Infants: A Systematic Review

  1. Emilie Pi Fogtmanna,
  2. Anne Mette Plomgaard, MD, PhDa,
  3. Gorm Greisen, MD, DMSca, and
  4. Christian Gluud, MD, DMScb
  1. aDepartment of Neonatology, Copenhagen University Hospital, Rigshospitalet, Copenhagen, Denmark
  2. bThe Cochrane Hepato-Biliary Group, Copenhagen Trial Unit, Centre for Clinical Intervention Research, Rigshospitalet, Copenhagen, Denmark
  1. Ms Fogtmann contributed to the conception and design of the study, participated in study selection, data extraction, quality assessment, data analysis, and interpretation of data, and drafted the initial manuscript; Dr Plomgaard contributed to the conception and design of the study, participated in study selection, data extraction, and quality assessment, and critically reviewed and revised the manuscript; Drs Greisen and Gluud contributed to the conception and design of the study, participated in the interpretation of data, and critically reviewed and revised the manuscript; and all authors approved the final manuscript as submitted.

Abstract

CONTEXT: Brain injury is common in preterm infants, and predictors of neurodevelopmental outcome are relevant.

OBJECTIVE: To assess the prognostic test accuracy of the background activity of the EEG recorded as amplitude-integrated EEG (aEEG) or conventional EEG early in life in preterm infants for predicting neurodevelopmental outcome.

DATA SOURCES: The Cochrane Library, PubMed, Embase, and the Cumulative Index to Nursing and Allied Health Literature.

STUDY SELECTION: We included observational studies that had obtained an aEEG or EEG within 7 days of life in preterm infants and reported neurodevelopmental outcomes 1 to 10 years later.

DATA EXTRACTION: Two reviewers independently performed data extraction with regard to participants, prognostic testing, and outcomes.

RESULTS: Thirteen observational studies with a total of 1181 infants were included. A meta-analysis was performed based on 3 studies (267 infants). Any aEEG background abnormality was a predictor of abnormal outcome. For prediction of a developmental quotient <70 points, cerebral palsy, or death, the pooled sensitivity was 0.83 (95% confidence interval, 0.69–0.92) and specificity 0.83 (95% confidence interval, 0.77–0.87).

LIMITATIONS: All studies were at high risk of bias. Heterogeneity was evident among the studies with regard to the investigated aEEG and EEG variables, neurodevelopmental outcomes, and cutoff values.

CONCLUSIONS: aEEG or EEG recorded within the first 7 days of life in preterm infants may have potential as a predictor for later neurodevelopmental outcome. We need high-quality studies to confirm these findings. Meanwhile, the prognostic value of aEEG and EEG should be used only as a scientific tool.

  • Abbreviations:
    aEEG
    amplitude-integrated electroencephalogram
    BSID-II
    Bayley Scales of Infant Development–II
    CI
    confidence interval
    GRADE
    Grades of Recommendation, Assessment, Development, and Evaluation
    LR+
    positive likelihood ratio
    LR−
    negative likelihood ratio
    MeSH
    Medical Subject Headings
    QUADAS-2
    Quality Assessment of Diagnostic Accuracy Studies-2
    QUIPS
    Quality in Prognosis Studies
    SROC
    summary receiver operating characteristic
  • Preterm infants (gestational age <37 weeks) are at greater risk of acquiring neurodevelopmental impairments compared with term infants,1,2 and the lower the gestational age at birth, the higher the risk. The number of infants being born preterm is increasing, and although mortality has been declining over the past decades, morbidity has not been reduced to the same degree.3 Therefore, the number of preterm infants at risk for long-term health problems is increasing.4,5 Neurologic damage may lead to neurodevelopmental impairment spanning from mildly delayed psychomotor development to severe cognitive problems and cerebral palsy.

    Several prognostic methods are in general use or have been proposed for predicting neurodevelopmental outcomes in preterm infants, including the early use of molecular biomarkers, electrophysiological tests within the first days of life, neuroimaging including early cerebral ultrasound and MRI at term equivalent age, and neurologic examinations later in the newborn’s life.68 The electrophysiological tests include the conventional EEG and the more easily applicable and widely used amplitude-integrated EEG (aEEG). In general, there appears to be a good correlation between findings in the aEEG and EEG,9,10 and both methods have been suggested as potential predictors of neurodevelopmental outcome if measured early in life. In term infants with hypoxic ischemic encephalopathy, background activity of the aEEG or EEG during the first 72 hours of life has proven to be one of the best predictors of neurodevelopmental outcome.11 Several studies have investigated variables of the aEEG or EEG as potential predictors of neurodevelopmental outcome in preterm infants.7,1214 In addition to using different variables of the aEEG or EEG, the studies apply a variety of methods to assess neurodevelopmental outcome.7,15

    Because of the various approaches, the prognostic values of an early aEEG or EEG in preterm infants remain unclear. To our knowledge, no systematic review on the topic has previously been conducted. Therefore, we have conducted a systematic evaluation of the literature in the field.

    The primary objective was to assess the prognostic test accuracy of aEEG or EEG obtained within the first 7 days of life in preterm infants for predicting neurodevelopmental outcome 1 to 10 years later.

    Methods

    Our methods have been described in detail in our protocol for this systematic review.16

    Criteria for Considering Studies for This Review

    Types of Studies

    Observational studies and randomized clinical trials published in any language were eligible for inclusion.

    Participants

    We included preterm infants (gestational age <37 weeks) of both sexes, both singletons and infants from multiple births. If both preterm and term infants were included in a study, data of the preterm infants were extracted if possible.

    Prognostic Tests

    Studies with an aEEG or EEG recording within the first 7 days of life and a description of the background activity were eligible for inclusion.

    Outcome Measures

    We included studies where a psychomotor developmental test or a neurologic examination had been performed at the age of 1 to 10 years. If a psychomotor developmental test was not available in a study, but an age-appropriate test for measuring IQ was conducted, that was sufficient for inclusion.

    One or more of the following target conditions had to be identified at follow-up: psychomotor impairment, cerebral palsy, or death.

    Search Methods for Identification of Studies

    We performed a comprehensive literature search to identify relevant studies in the following databases: The Cochrane Library, PubMed (National Library of Medicine), Embase (OvidSP), and the Cumulative Index to Nursing and Allied Health Literature. The searches were carried out as specified in our protocol.16

    Free text words and Medical Subject Headings (MeSH) were applied as follows:

    • Search 1: EEG OR aEEG OR electroencephalog*

    • Search 2: “Infant, Premature” [MeSH] OR “Infant, Low Birth Weight” [MeSH]

    • Search 3: Bayley OR neurodevelopment* OR cerebral palsy OR motor OR psychomotor OR cognitive OR mental OR intelligence OR IQ OR DQ OR death OR mortality

    • Search 4: #1 AND #2 AND #3

    No time limitations were applied. To assess ongoing and future studies, we searched www.clinicaltrials.gov and www.who.int/trialsearch. The last search was conducted on October 1, 2015. Reference lists from relevant reviews and included studies identified in our initial electronic searches were searched for additional studies eligible for inclusion.

    Data Collection and Analysis

    Selection of Studies

    Two authors (E.P.F. and A.M.P.) independently assessed all studies for inclusion based on title and abstract. Studies relevant for inclusion were retrieved in full text and assessed by the same 2 authors. Disagreements were resolved by consensus or by the decision of a third author (G.G.).

    Data Extraction and Management

    Two authors (E.P.F. and A.M.P.) independently extracted data from the included studies by using a data extraction sheet. The following data were extracted: title, authors, journal name, year of publication, study design, number of participants meeting the criteria, gestational age, aEEG or EEG variables used to classify the background activity including the classification, age of the participants when the prognostic testing was conducted, clinical reference standard tests, target conditions being diagnosed including the classification of the outcomes, and age of the participants at follow-up.

    To perform statistical analyses, the background activity of the aEEG or EEG and the neurodevelopmental outcome had to be dichotomized as either normal or abnormal, with specification of the number of participants in each group. Data for true-positive, false-positive, true-negative, and false-negative cases or sensitivity and specificity for all cutoffs available were extracted.

    If any relevant information was missing, the authors were contacted; 2 e-mails 14 days apart were sent to the corresponding author. If contact details on the corresponding author could not be obtained, we attempted to contact the coauthors.

    Assessment of Bias Risk

    Two authors (E.P.F. and A.M.P.) independently evaluated the risk of bias for each study included in the review. Disagreements between the authors were resolved by consensus or the vote of a third author (C.G.). The Quality in Prognosis Studies (QUIPS) tool17 and the Quality Assessment of Diagnostic Accuracy Studies-2 (QUADAS-2) tool,18 modified to evaluate risk of bias in prognosis studies,19 were applied.

    Statistical Analysis and Data Synthesis

    Statistical analyses were performed through the use of Chapter 10 of the Cochrane Handbook for Systematic Reviews of Diagnostic Test Accuracy20 and the Review Manager 5 software.21

    For each included study, 2×2 cross-classification tables showing the aEEG or EEG test results and neurodevelopmental outcomes were generated. Sensitivities, specificities, and positive and negative likelihood ratios (LR+ and LR−) were calculated for all available cutoff values.

    For descriptive analysis, coupled forest plots reporting the pairs of sensitivities and specificities with the corresponding 95% confidence intervals (CIs) for all relevant cutoff values were conducted. The studies were graphically visualized in summary receiver operating characteristic (SROC) plots (sensitivity versus 1 − specificity) and were presented as rectangles with sizes corresponding to the sample size of each study.

    We summarized data by estimating SROC curves by fitting a regression curve to the pairs of sensitivity and specificity by using Rutter and Gatsonis’ random-effects hierarchical SROC model.22

    Investigations of Heterogeneity

    Possible sources of heterogeneity include criteria to diagnose psychomotor impairment and cerebral palsy, ages of the participants at follow-up, techniques used to evaluate the background activity of the aEEG or EEG, gestational ages of the participants, and methodological sources of heterogeneity including the effect of study design and differences in bias risk. Supposing there is variation in threshold, the proximity of the plotted study results to the fitted SROC curve is important.20

    For visualization of the heterogeneity among the included studies, a plot depicting prognostic variables against outcomes was constructed.

    To address additional heterogeneity among studies, relevant study-level covariates were added to the hierarchical SROC regression model to identify factors associated with prognostic test accuracy.

    Differences Between Protocol and Review

    We applied the Grades of Recommendation, Assessment, Development, and Evaluation (GRADE) approach to determine confidence in our prognostic estimates.23

    Results

    Results of the Search

    We identified 823 records through electronic searches. No additional records were identified through other sources. We excluded 627 records based on title and abstract screening, and 72 duplicates. We assessed 124 full-text articles, of which 49 were possibly eligible for inclusion. Eight of the studies were included directly.7,14,15,2428 In the remaining 41 studies, an attempt to contact the authors was made to obtain missing data. In 9 studies, responses from the authors were received, and in 5 of these additional data were obtained leading to inclusion.2933 In 8 studies, no contact details could be retrieved, probably as a result of the year of publication (1964–2004). In 24 studies, the authors did not reply. Consequently, 13 studies were included in our review. Figure 1 outlines the study selection. The missing data for each of the 41 studies are listed in Supplemental Table 2.

    FIGURE 1

    Flowchart of study selection.

    Characteristics of Included Studies

    Table 1 summarizes the characteristics of the 13 included studies.

    TABLE 1

    Characteristics of Included Studies

    A total of 1181 preterm infants were included in our analyses, ranging from 7 to 295 participants per study. Twelve of the included studies included only preterm infants, and 1 study also included term controls.31 The gestational ages of the preterm infants ranged from 22 to 36 weeks.

    Six studies used aEEG,14,15,24,25,29,33 and 7 studies used conventional EEG7,2628,3032 as the prognostic marker. The age at the time of recording was 6 hours to 7 days of life. Five of the 6 aEEG studies14,15,24,25,33 analyzed data according to a classification system based on burst suppression and continuity developed by Hellström-Westas et al.9 In addition to background activity, some studies investigated the prognostic value of sleep–wake cycling14,15,24,25,29,33 and seizure activity, the latter as an independent variable14,15,28,29,31,33 or as part of a composite variable also including background activity.7,15,27,28,30,33

    Age at follow-up was 1 to 7 years. The combination of a psychomotor developmental test and a neurologic examination was applied in 7 studies,7,14,15,24,25,31,33 and 5 of these used the Bayley Scales of Infant Development–II (BSID-II).14,15,24,25,33 The threshold for an abnormal outcome varied. Moreover, some studies included death as an abnormal outcome, and other studies excluded infants who died during follow-up. The variation in prognostic variables, outcomes, and thresholds is displayed in Fig 2.

    FIGURE 2

    Illustration of the prognostic electroencephalographic variables (x-axis) that were investigated in the included studies to predict various outcomes (y-axis). The scatter of points reveals the heterogeneity among the studies, with a maximum of 3 studies sharing the same coordinates. In meta-analyses, we investigated the encircled studies. aEvaluated by Tsumori–Inage Infant Developmental Scale/Questionnaire or Kyoto Scale of Psychological Development. bEvaluated by neurologic examination. cEvaluated by BSID-II. dEvaluated by BSID-II and Peabody Developmental Motor Scales–2. eEvaluated by Denver-II test. fEvaluated by various neurodevelopmental examinations.

    Methodological Quality of Included Studies

    The methodological quality was evaluated by QUIPS17 and a modified version of QUADAS-2.18,19 We found QUIPS most suitable for evaluation of bias risk in prognosis studies; however, no substantial discrepancy regarding the results was evident between the 2 tools. The reported results in this review are based on QUIPS and summarized in Fig 3. The methodological quality as evaluated by the modified QUADAS-2 is available in Supplemental Fig 7. All studies were at moderate or high risk of bias in multiple domains when either of the tools was applied.

    FIGURE 3

    Bias risk summary of included studies according to QUIPS. *Authors contributed essential information to accommodate our inclusion criteria. The supplement of nonpublished data was taken into account in assessing the risk of bias.

    Study Participation

    In the “study participation” domain, 5 studies were at low risk of bias,7,24,26,28,32 5 were at moderate risk,14,27,30,31,33 and 3 were at high risk of bias.15,25,29 Moderate and high bias risk was a result of inadequate participation in the study by eligible patients (inappropriate exclusions).

    Study Attrition

    In the “study attrition” domain, 5 studies had a low risk of bias,2527,29,30 2 studies had a moderate risk,28,32 and 6 studies had a high risk of bias.7,14,15,24,31,33 The high risk of bias resulted from a low response rate and no reasons for loss to follow-up or attempts to collect information on participants who dropped out.

    Prognostic Factor Measurement

    In the “prognostic factor measurement” domain, 7 studies had a low risk of bias15,2426,2830 and 6 studies had a moderate risk of bias.7,14,27,3133 Moderate bias risk was most often a result of no blinding and the application of data-dependent cutoff values.

    Outcome Measurement

    In the “outcome measurement” domain, 4 studies had a low risk of bias,14,15,32,33 6 were at moderate risk,7,24,26,28,30,31 and 3 studies had a high risk of bias.25,27,29 Studies with moderate or high risk of bias were mainly older studies and had no blinding, had no clear definition of outcomes, and used different methods to assess outcomes.

    Study Confounding

    Only 2 studies7,33 had a low risk of bias in the “study confounding” domain, as multivariate analyses were applied. The other studies did not control or only sparsely controlled for potential confounders and were classified as having a moderate15,24,28,30,31 or high14,2527,29,32 risk of bias. Methods to account for potential confounders involved exclusion of infants with congenital cerebral malformations and performance of separate analyses where no sedatives were given.

    Statistical Analysis and Reporting

    All but 1 study32 had a low risk of bias in the “statistical analysis and reporting” domain. The high risk of bias resulted from selective reporting of results and insufficient data presentation.

    Findings

    The results of the a priori defined prognostic test and outcome thresholds including all 13 studies are summarized in a forest plot and SROC plot (Fig 4 A and B). The sensitivity ranged from 0.35 to 0.94 and the specificity from 0.67 to 1.00 in 11 studies (estimates from Kidokoro et al and Nunes et al had to be excluded in the reporting of ranges because of no positive testing). The SROC curve indicates that the background activity of an aEEG or EEG recorded within 7 days of life in preterm infants is informative, but the spread of the studies along the curve indicates a substantial heterogeneity among the studies (see Fig 4B). The heterogeneity is also demonstrated by the scatter of points in the plot of prognostic variables and outcomes (Fig 2), where studies that investigated the same prognostic variable and outcome at similar cutoff values share the same coordinates. We found only studies that investigated the same prognostic variable and outcome suitable for meta-analysis.

    FIGURE 4

    A, Forest plot of a priori defined cutoff values. If binary divisions for normality and abnormality were made in a study with regard to aEEG, EEG, and outcome, these data were used in the analysis. Otherwise, the authors agreed on cutoff values before the examination of data. FN, false-negative; FP, false-positive; TN, true-negative; TP, true-positive. B, SROC plot of a priori defined cutoff values. The studies are presented as rectangles whose size corresponds to the sample size of the particular study. The height of the rectangle corresponds to the number of diseased, and the width of the rectangle corresponds to the number of nondiseased.

    Three studies with a total of 267 participants investigated the prognostic accuracy of a mildly, moderately, or severely abnormal aEEG according to Hellström-Westas et al9 for the prediction of a moderately or severely abnormal outcome (ie, developmental quotient <70 points according to BSID-II, cerebral palsy, or death).14,24,33 The pooled sensitivity was 0.83 (95% CI, 0.69–0.92), specificity 0.83 (95% CI, 0.77–0.87), LR+ 4.80 (95% CI, 3.50–6.58), and LR− 0.20 (95% CI, 0.11–0.38).

    Two studies with a total of 180 participants applied the same grading system9 and investigated the prognostic accuracy of a mildly, moderately, or severely abnormal aEEG to predict mildly, moderately, or severely abnormal outcome (ie, developmental quotient <85 points according to BSID-II, cerebral palsy, or death).15,33 The pooled sensitivity was 0.76 (95% CI, 0.66–0.84), specificity 0.87 (95% CI, 0.77–0.93), LR+ 5.77 (95% CI, 3.21–10.39), and LR− 0.28 (95% CI, 0.20–0.39).

    The forest plots and SROC plots for the 2 subgroups are shown in Figs 5 A and B and 6 A and B. The SROC curves indicate that any abnormality in the background activity of the aEEG according to a classification based on burst suppression and continuity9 is informative to predict developmental delay, cerebral palsy, or death in preterm infants. The plotted study points are not in close relation to the fitted SROC curve in either of the plots, suggesting there is heterogeneity among the studies within the subgroups. Because of sparse data it was not possible to address additional heterogeneity by adding covariates to the regression models.

    FIGURE 5

    A, Forest plot and B, SROC plot depicting the prognostic accuracy of a mildly, moderately, or severely abnormal aEEG for prediction of a developmental quotient <70 points, cerebral palsy, or death. FN, false-negative; FP, false-positive; TN, true-negative; TP, true-positive.

    FIGURE 6

    A, Forest plot and B, SROC plot depicting the prognostic accuracy of a mildly, moderately, or severely abnormal aEEG for prediction of a developmental quotient <85 points, cerebral palsy, or death. FN, false-negative; FP, false-positive; TN, true-negative; TP, true-positive.

    To determine confidence in our prognosis estimates, we conducted GRADE assessments.23 Our solitary inclusion of observational studies increased our confidence in the results, whereas the evident risk of bias across all studies decreased it. There was inconsistency in the sensitivities and to a lesser extent also the specificities in our a priori defined analysis. Subgroup analyses did not alter this inconsistency. Wide CIs indicated imprecision and decreased our confidence in the results. Our study population corresponded well to the population of interest, preterm infants. This generalizability strengthened the confidence in our results. Our subgroups were clearly defined, which minimized the concern of indirectness, and there were no clear indications of publication bias. According to our GRADE assessments, we concluded that the confidence in our results is low, because the true prognosis may be substantially different from our results.

    Discussion

    Summary of Main Results

    In this review, we aimed to assess the value of aEEG or EEG obtained within the first 7 days of life in preterm infants for predicting neurodevelopmental outcome 1 to 10 years later. Thirteen studies with a total of 1181 participants were included. Meta-analyses of 2 subgroups indicated that any background abnormality in the aEEG according to the classification by Hellström-Westas et al9 was a predictor of abnormal outcome in preterm infants. For prediction of a developmental quotient <70 points, cerebral palsy, or death, pooled sensitivity was 0.83 and specificity 0.83. For a developmental quotient <85 points, cerebral palsy, or death, pooled sensitivity was 0.76 and specificity 0.87. However, the GRADE assessments suggested low confidence in the results.

    Strengths and Weaknesses of the Review

    This systematic review includes an exhaustive search strategy and is conducted according to a registered protocol.16 No language or time restrictions were applied. We included several studies published within the last few years, which provides an up-to-date assessment of the prognostic accuracy of aEEG or EEG. We attempted to contact authors when missing data prevented us from including studies. Five studies were included as a result of contact with authors and their contribution of nonpublished data.2933 More studies might have been included if all the contacted authors had replied. This limitation raises the risk of outcome reporting bias. To obtain an optimal assessment of methodological quality, we applied 2 different quality assessment tools. We rated the overall risk of bias as high in all studies, because clinically relevant information was missing.

    The results from pooling all 13 studies must be interpreted with caution, because heterogeneity was evident among the studies. One of the main sources of heterogeneity was the variation in cutoff values. This variation might be a result of the application of different scoring systems and subjectivity depending on the investigator.

    The illustration of prognostic variables and outcomes (Fig 2) revealed that no more than 3 studies could be grouped together. For our meta-analyses, we focused on the background activity consisting of measures of continuity and burst suppression, and an evaluation of the prognostic accuracy of sleep–wake cycling is yet to be conducted. Because of small sample sizes, it was not possible to perform subgroup analyses depending on gestational age, although differences in outcome are expected between infants born very prematurely compared with late preterm infants. In our review, 82% of the population had a gestational age <34 weeks, and therefore few infants were late preterm. We did not find it appropriate to insert summary points in the SROC plots because of the paucity of studies and between-study heterogeneity.

    Preterm infants are at elevated risk of developing intraventricular hemorrhage and periventricular leukomalacia, and cerebral ultrasound is used for the diagnosis of these conditions.34 Ultrasound-detected brain injury is related to an abnormal EEG and also to an adverse outcome35; this relationship may explain part of the association between aEEG or EEG findings and later cerebral outcome. Unfortunately, the present data do not allow us to analyze whether aEEG or EEG has prognostic value that is separate from what can be obtained by ultrasound.

    The development of brain pathologies is most common during the initial postnatal period, and most severe intraventricular hemorrhages occur within the first 72 hours of life.36 In contrast to term infants with hypoxic ischemic encephalopathy, the long-term neurodevelopmental outcome of preterm infants is also negatively affected by necrotizing enterocolitis, sepsis, and bronchopulmonary dysplasia, which often occur later than the first week of life. None of the studies in this systematic review adjusted for any of the 3 conditions, and this might explain why the prognostic accuracy of aEEG or EEG within the first week of life in preterm infants is low in comparison with term infants with hypoxic ischemic encephalopathy.

    Applicability of Findings to the Review Question

    We do not believe that our findings are subject to any significant applicability concerns regarding the review question, because all included studies were evaluated with the QUADAS-2 tool judging applicability concerns in 3 domains. Twelve out of 13 studies had a low risk of bias in all these domains (Supplemental Fig 7). When both preterm and term infants were included in a study, we extracted the data of the preterm infants. When the prognostic testing was carried out beyond the limit of 7 days, we extracted the relevant data. Although this approach strengthened the applicability of our findings to the review question, it led to smaller sample sizes in some of the studies,30,31,33 which resulted in poor estimates with wide CIs (see Fig 4A).

    Implications for Research

    To determine the prognostic accuracy of aEEG or EEG in preterm infants more precisely, there is a need for prospective cohort studies with adequate sample sizes,37 predefined cutoff values, and high quality that investigate a standardized set of electroencephalographic variables in relation to neurodevelopmental outcomes and death. Our findings seem to contrast the more positive results of 2 nonsystematic reviews on the topic.38,39 Consensus regarding definitions and classifications of aEEG and EEG variables and neurodevelopmental outcomes is needed. To our knowledge, there is no gold standard for the classification of aEEG and EEG or the evaluation of neurodevelopmental outcome.

    The added value of using aEEG or EEG on top of other, more easily obtainable prognostic tests is a major question. The added value is the prognostic accuracy offered by aEEG and EEG over and above the prognosis provided by information that is already at hand at 7 days of life. At a minimum this information encompasses gestational age, birth weight, prolonged rupture of membranes, chorioamnionitis, Apgar scores, and severity of cardiorespiratory illness during the first week, all of which have independent prognostic values.4044 Also, cranial ultrasound is routinely used in most neonatal services. To demonstrate added prognostic value of aEEG and EEG, all these data must be available at the individual level, and appropriate adjustment done in the statistical analysis.

    To determine the clinical consequences for the implementation of aEEG and EEG, there is a need for randomized clinical trials that investigate the effect of coupling early electroencephalographic testing and developmental interventions.37 Moreover, systematic reviews of randomized clinical trials should be conducted to evaluate the actual effect of the test.37

    In future research, aEEG or EEG should be recorded as soon as possible after birth, preferably within the first week of life,9 and the time point of the recording should be available. At the same time, the more easily obtainable or already available prognostic markers should be registered. It should also be possible to separate data of preterm infants from those of term infants. Studies should not exclude infants who die but should consider death an adverse outcome or report it as a secondary outcome. Other inappropriate exclusions should also be avoided, and for infants lost to follow-up reasons and attempts to collect information should be described. The assessment of all data should be blinded, and selected cutoff values should not be data dependent. Data presentation should facilitate investigation of all possible cutoff values. Studies should control for potential confounders, and to limit publication bias, all protocols of prognosis studies must be easily accessible. This recommendation also applies to depersonalized individual patient-level data.45

    Implications for Practice

    Different intervention programs for preterm infants have been investigated in randomized clinical trials.46,47 A Cochrane meta-analysis48 found a positive effect of early developmental interventions for preterm infants on cognitive development at infancy and preschool age. For motor development, a small effect was found in infancy, but no effect was found on the proportion of children with cerebral palsy. The use of neonatal aEEG and EEG as an experimental investigation in preterm infants combined with a focus on early developmental care may help identify effective interventions.

    Conclusions

    Our systematic review suggests that aEEG or EEG recorded within the first 7 days of life in preterm infants may have potential as a surrogate marker for neurodevelopmental outcome 1 to 10 years later. However, we need studies of high quality to confirm these findings. Meanwhile, the prognostic value of aEEG and EEG should be used only as a scientific tool,eg, in trials of interventions to reduce neurodevelopmental impairment in preterm infants.

    Acknowledgments

    We thank the participants and their parents for their participation in the studies reviewed. We thank the authors of the studies who responded to our pleas for better understanding their results.

    Footnotes

      • Accepted November 8, 2016.
    • Address correspondence to Emilie Pi Fogtmann, Department of Neonatology, Copenhagen University Hospital, Rigshospitalet, Blegdamsvej 9, 2100 Copenhagen, Denmark. E-mail: empifo{at}gmail.com
    • This study has been registered at http://www.crd.york.ac.uk/PROSPERO/ (identifier CRD42014010514).

    • FINANCIAL DISCLOSURE: The authors have indicated they have no financial relationships relevant to this article to disclose.

    • FUNDING: No external funding.

    • POTENTIAL CONFLICT OF INTEREST: The authors have indicated they have no potential conflicts of interest to disclose.

    References