Introduction

Neonatal neurocritical care is dedicated to providing multidisciplinary brain-focused care in the neonatal intensive care unit (NICU).1,2,3,4 Neuromonitoring tools commonly used in the NICU include electroencephalography (EEG) and near-infrared spectroscopy (NIRS).

EEG records cerebral electrical activity through electrodes placed on the scalp. Continuous EEG (cEEG) monitoring uses at least eight electrodes to record multichannel EEG over a prolonged period, typically 24 h or more. In the NICU, EEG is often displayed at the bedside as a simplified, time-compressed trend called amplitude integrated EEG (aEEG). Stand-alone aEEG systems use fewer electrodes (typically two to four) to facilitate easy application and use. With aEEG, visual inspection of the raw or source EEG recording is essential for confirming aEEG findings. For clarity in this paper, we will use the term “aEEG” when referring to aEEG trends and their source EEG signals, and “cEEG” when referring to multichannel recordings. Notably, both aEEG and cEEG can be supplemented with synchronized video recording and extracerebral channels, such as electrocardiogram, abdominal respiration belt, and electromyography.

cEEG has been used for decades in neonates to assess brain maturation, functional state, cerebral injury, seizure identification, and to predict outcome.5,6,7,8,9,10,11 Initial work relied on brief (30–60 min) EEG recordings; however, longer term (hours to days) cEEG monitoring in the NICU has increasingly become standard. Longer recording is essential for accurate identification of seizures and can provide real-time information regarding changes in brain function. The American Clinical Neurophysiology Society has published guidelines for the use of continuous EEG in neonates, and guidelines for terminology to categorize neonatal EEG findings.12,13

aEEG is complimentary to cEEG and has the advantage of being easier to apply and to interpret by non-neurophysiologists. aEEG gives an overview of trends in cerebral activity (including recovery from injury), easily identifies sleep wake cycling (SWC), and can screen for seizures.14,15 While multiple approaches to aEEG interpretation have been described, the most commonly used system relies on pattern recognition to classify aEEG into five distinct patterns: continuous (C), discontinuous (DC), burst suppression (BS), low voltage (LV), and inactive or flat trace (FT)16 (Fig. 1). This classification is most useful for term infants. A more detailed scoring system that accounts for maturation changes of prematurity has been developed and could have clinical utility.17 aEEG has limitations, including lower sensitivity for detecting seizures and increased susceptibility to artifacts as compared to cEEG. Monitoring of neonates with aEEG requires proper training of providers and nursing staff. NICU providers should be competent in aEEG interpretation, incorporation of aEEG into workflow, and documentation of aEEG findings in the medical record. Bedside nurses need training to be competent placing leads, monitoring electrode impedance, using seizure detection algorithms and documenting aEEG findings in regular NICU assessments. Ideally, ongoing education and case reviews improve skills continuously for all.

Fig. 1: Common classification used to describe aEEG tracing, especially in term infants: Continuous (C), Discontinuous (DC), Burst Suppression (BS), Low Voltage (LV) and Inactive, flat trace (FT).
figure 1

These figures are derived from 2 different brands of aEEG devices which is reflected in the apparent difference in format. The Y axis is the standard semi-logarithmic aEEG axis (linear from 0–10 µv and logarithmic from 10–100 µv). X-axis represents time with the label representing one hour of recording.

NIRS is a non-invasive, bedside monitoring system that provides continuous data on cerebral tissue oxygenation. NIRS relies on the relative transparency of near-infrared light and the wavelength-dependent absorption of oxyhemoglobin (HbO2) and deoxyhemoglobin (HHb). Most commercial NIRS systems use two to four wavelengths of near-infrared light to estimate the regional tissue oxygen saturation of hemoglobin (rStO2), expressed as a percentage. Changes in oxygenation reflect changes in cerebral perfusion, oxygen delivery, and/or changes in tissue oxygen extraction due to metabolic demand. There are no precise normative values for rStO2, as different monitors and different sensors provide different normal values. Until there is more definitive information on the utility of absolute cutoffs, rStO2 is best utilized as a trend to allow recognition of relative changes for a given patient over time.

Cerebral rStO2, in combination with pulse oximetry (SaO2), allows real-time estimation of the balance between oxygen supply and demand, calculated as the cerebral fractional tissue oxygen extraction (FTOE) ratio: (SaO2-rStO2)/SaO2. NIRS used in combination with invasive blood pressure monitoring allows clinicians to derive information regarding autoregulation of cerebral blood flow.18,19,20 In addition to EEG and NIRS, multimodal monitoring may also include the use of evoked potentials and / or other dynamic markers of neonatal brain health.

While various guidelines and consensus statements exist for individual modalities, there is no comprehensive guidance synthesizing their application in the NICU. In this two-part review, we describe neuromonitoring tools in four common NICU scenarios, Part I: neonatal encephalopathy and neonatal seizures, and Part II: extreme prematurity and the critically ill neonate. Finally, in part II we describe the importance of multimodal and sleep monitoring in the NICU and summarize the application of neuromonitoring across common scenarios. By this approach, we aim to inform best practices for using these tools in these vulnerable populations to promote neuroprotective care.

Neonatal encephalopathy

Neonatal encephalopathy (NE) is a clinically defined syndrome of disturbed neurologic function in the earliest days of life manifesting as abnormal level of consciousness or seizures, and is often accompanied by difficulty with respiration, abnormalities of tone and impaired developmental reflexes. NE may reflect underlying brain injury, brain malformations, inborn errors of metabolism, or other causes. Hypoxic-ischemic encephalopathy (HIE) is a specific diagnosis that applies only when a neonate has encephalopathy that is known or highly suspected to be due to hypoxic ischemic brain injury.21,22,23 The incidence of HIE is about 1.5 per 1000 in developed countries and much higher in middle to low-income countries.24,25 Despite recent advances, HIE remains associated with considerable mortality and morbidity. Neuromonitoring has multiple benefits in the first few days of life for those with NE, and specifically for those receiving therapeutic hypothermia (TH) for presumed HIE.

aEEG in neonatal encephalopathy

aEEG can be used to classify the severity of NE, identify evolution of encephalopathy over time, and inform prognostication. aEEG is also used to screen for neonatal seizures, as detailed separately below.

The predictive value of aEEG in infants with NE varies depending on whether or not a neonate has received hypothermia. In neonates who are not receiving hypothermia, an abnormal aEEG background pattern between 3–12 h after birth correlates with subsequent adverse neurodevelopmental outcome.26,27,28,29,30 In a study from the pre-hypothermia era, 20% of neonates with perinatal asphyxia had deterioration in their aEEG background between three and six hours after birth, illustrating the importance of ongoing aEEG monitoring until six hours after birth or beyond.27 At six hours after birth, an abnormal aEEG pattern has a sensitivity of 84% and a specificity of 79% for predicting death and moderate/severe disability in normothermia.31 Predictive ability improves further when aEEG is combined with clinical examination.30 Ongoing continuous recording is also important; recovery of aEEG background within 24 h was shown to be associated with favorable outcome.32 Likewise, the onset of SWC before 36 h of life was associated with good neurological outcome.33

In the era of TH, aEEG has been used to screen for TH eligibility as well as for prognostication. Although an abnormal aEEG was used as an inclusion criteria in some hypothermia trials as evidence of encephalopathy,34,35 its use as an entry criterion has been challenged given concerns for falsely reassuring patterns.36 At the same time, it is possible for abnormal aEEG background to identify moderate or severe encephalopathy when exam findings are unclear. At six hours after birth, an abnormal aEEG pattern has a high sensitivity of 95-96% and modest specificity of 39–61% for predicting death and moderate/severe disability in the setting of hypothermia.31,37 The lower specificity of an early abnormal aEEG trace in the setting of TH as compared to normothermia reflects the modifying effects of hypothermia on the risk of poor outcome. Furthermore, after starting TH, brain electric activity may also be altered by the TH itself and/or the associated sedation. Although the high sensitivity of aEEG supports its use as a screening tool to identify those at risk for poor outcomes who might benefit from hypothermia treatment, the negative predictive value (NPV) of a normal aEEG at six hours is still just 77%.37 Furthermore, artifact is common in aEEG and may result in a falsely normal appearing trace, so careful inspection of the source signal is important. For these reasons, aEEG findings should not preclude initiation of TH for an otherwise eligible neonate.

The positive predictive value (PPV) of aEEG changes over the course of hypothermia; aEEG background may gradually improve over the first 48–72 h of cooling.38,39,40 The positive predictive value of an abnormal aEEG in cooled neonates increases from 66% at 24 h to 85% at 48 h and 89% at 72 h.37 The predictive value of an abnormal aEEG is highest at 72 h of life in cooled neonates, as compared to 36 h in non-cooled neonates.31 These findings highlight the importance of monitoring neonates throughout the duration of cooling and rewarming to inform the trajectory of recovery, or absence of recovery over time, which have significant prognostic value.

cEEG in neonatal encephalopathy

Review of the full array cEEG is important for neuromonitoring in NE. Compared to aEEG trends, full cEEG provides more detailed information for encephalopathy grading and prognosis; it may also reveal focal abnormalities such as focal injuries and seizures.

Depending on insult severity, perinatal acquired injury is characterized by temporal evolution of EEG abnormalities. This most often consists of initial EEG suppression, seizures in the first 24 h, and gradual improvement of background activity after three days.41 When present, these EEG features can help determine the timing of brain injury within broad time windows42 as well as the severity of injury.43 When other patterns of cEEG evolution are observed, reconsideration of the underlying etiology is appropriate. Early cEEG is a reliable predictor of outcome in NE,44 with the best predictive ability for infants who are not cooled at around 24 h of life45 As expected, the greatest prognostic value of EEG background in those receiving TH is later than for neonates who do not receive TH, with the greatest prognostic value of EEG recording around 48 h of life.46 Similar to aEEG, the trajectory of the cEEG background is even more informative. Thus, cEEG evolution over time needs to be followed to best inform prognostication.

While interpretation of background categories has good interrater agreement for neonates with HIE,47 there is notable variability in some discrete features in the visual interpretation of neonatal cEEG.48 This highlights the need for specialized training in neonatal EEG and improved classification systems. Development of automated analysis algorithms for clinical use are eagerly awaited. A systematic review has shown that burst suppression, low voltage, and inactive trace on cEEG have the highest predictive value for adverse outcomes in neonates with HIE.49 As in aEEG, SWC on EEG is a valuable marker for prognosis: healthy newborns should have well-developed SWC at 6–12 h of age, whereas absence of SWC in encephalopathic, normothermic infants at 48 h is a poor prognostic sign.45,50 While the presence of seizures alone does not necessarily predict outcome, a high seizure burden is associated with adverse outcome.51,52,53 An example of cEEG over time for a neonate with NE is presented in Fig. 2.

Fig. 2: Progression of cEEG in a term infant with HIE.
figure 2

cEEG at age 2 days with (a) markedly discontinuous background activity (interburst intervals up to 40 s) and absent sleep wake cycling, consistent with severe NE. In addition, frequent seizures occurred; here (b) starting from the mid to left central region. (c) EEG at age 10 days has improved with nearly continuous background activity in wakefulness (interburst intervals up to 5 sec) but in sleep (d) interburst intervals are still very long (up to 20 s). At day 10, there are still no normal graphoelements seen, and there remains abnormal sleep wake cycling, which indicate poor neurodevelopmental prognosis.

Neonates at high risk for seizures, including those with NE, should undergo cEEG for at least 24 h to monitor for seizures and inform prognostication.12 Continued monitoring during both TH and rewarming should be considered, as there is a small but recognizable risk of seizure in the rewarming phase.54,55,56 It has been suggested that early background abnormalities may predict which neonates will go on to have seizures: seizures are most likely to occur following significantly abnormal initial EEG background, while a normal/mildly abnormal cEEG during the first 24 h of TH indicates a very low risk of subsequent seizures.54,57 Those neonates with normal/mildly abnormal cEEG, especially in presence of limited resources, may be considered to transition to aEEG monitoring instead after first 24 h. Work is ongoing to develop algorithms to guide the efficient use of cEEG in neonates with NE.

NIRS in neonatal encephalopathy

NIRS monitoring can provide information regarding individual pathophysiology in NE, particularly in those receiving TH for presumed HIE, as significant cerebral hemodynamic and metabolic disturbances evolve over time. The first four hours of TH are characterized by decreased cerebral rStO2,58 which is hypothesized to reflect decreased oxygen delivery coinciding with the abrupt decrease in body temperature.59 By 12 h and onwards, there is an increase in rStO2,58,60 as well as a sustained decrease in FTOE.60 These changes may be explained by the effect of mild hypothermia on cerebral metabolism, lowering oxygen utilization. This may also reflect the effect of cumulative analgesia/sedatives during TH, resulting in decreased oxygen extraction, although further study is needed.61

Among neonates undergoing TH, NIRS demonstrates evolving patterns of brain oxygenation that predict favorable and adverse outcomes.58,60 One study found that the increase of rStO2 from day one to day two was only significant among those with abnormal MRI outcome.62 Similarly, overall rStO2 is significantly higher in neonates with brain injury later documented by MRI58,62 or who had an adverse outcome60 (Fig. 3). Although one study suggested a better predictive value of a high rStO2 at ten hours of hypothermia,58 other studies suggest that the predictive value increases over time60 and that rStO2 is most predictive at the time of rewarming.62 It is notable that, in one study, those with moderate HIE (stage II) undergoing TH showed increasing rStO2 and decreasing FTOE trends, but these did not significantly differ between those with normal or abnormal MRI findings.63 The association of increased rStO2 with brain injury/poor outcome could be a reflection of decreased oxygen utilization associated with severe tissue injury64 or a reflection of luxury re-perfusion in those with brain injury,65,66 or a combination of both.

Fig. 3: Illustration of the value of NIRS in monitoring neonates receiving therapeutic hypothermia for neonatal encephalopathy.
figure 3

a aEEG and cerebral rStO2 progression and brain MRI brain in a term baby with severe encephalopathy (GA 40 1/7 weeks, maternal chorioamnionitis, non-reassuring fetal heart rate, Apgar scores 2,2, and cardiopulmonary resuscitation for 20 min, umbilical vein pH = 6.94). aEEG evolved from flat (FT) (upper panel) to burst suppression (BS) background (lower panel). rStO2 progressed to maximum measurable level of 95%. Post-rewarming MRI showed extensive brain injury (demonstrated on apparent diffusion coefficient (ADC) map sequence to the left and diffusion-weighted imaging (DWI) sequence to the right). b aEEG, rStO2 progression and brain MRI in a term baby with moderate encephalopathy. (GA 39 5/7, Floppy at birth, Apgar score 0,3 and 3, needed PPV for 10 min before intubation, UA pH 6.87) aEEG evolved from discontinuous (DC) (upper panel) to continuous (C) with cycling (lower panel). rStO2 was stable in mid 70 s during hypothermia and rewarming. Post-rewarming MRI was normal (demonstrated on ADC map sequence to the left and DWI sequence to the right).

Value of combining aEEG, and cEEG and NIRS in neonatal encephalopathy

Combined neuromonitoring with aEEG, cEEG and NIRS can be more informative than any modality alone. Many cEEG devices have the capability to display aEEG tracings to enable bedside identification of aEEG patterns without a separate aEEG device. Changes in electrical brain activity on aEEG tend to be associated with abnormalities in cerebral NIRS measures and correlate with hemodynamic disturbance.67 Initial studies suggest that multimodal neuromonitoring in NE may improve stratification of injury, assist with prognosis, and help in development of adjunctive neuroprotective strategies. In infants receiving TH for NE, elevated cerebral rStO2 > 90% and decreased variability <5% were associated with abnormal aEEG background (defined as LV, BS or FT) (OR 4.7) at 24–48 h of life and with electrographic seizures (OR 6.1) between 6–24 h of life.68 The combination of elevated rStO2 with abnormal aEEG background or presence of seizures improved specificity, PPV, and NPV for predicting MRI abnormalities63 and neurodevelopmental outcomes at 30 months of age, as compared to either modality alone.69 A combined aEEG pattern and cerebral rStO2 score at 12–36 h of life increased predictive accuracy for 18-month neurodevelopmental outcomes in asphyxiated neonates treated with TH, with a PPV of 91% compared to only 67% for rStO2 or 62% for aEEG background.60

When combined, aEEG, cEEG, and NIRS monitoring in NE provides real-time information regarding brain function and physiology, as well as information over time to guide prognostication.

Seizures

Seizures can be the primary reason for NICU admission, or they may be a complication of another illness. Clinical concern for seizures may arise due to sudden, stereotyped episodes of abnormal movements or posturing. However, abnormal movements due to seizures are difficult to distinguish from non-epileptic events through visual observation alone. A recent International League Against Epilepsy (ILAE) position paper states that a diagnosis of seizure cannot be definite, and is at best only probable, when made by clinical observation without EEG confirmation.70 Conditions like NE carry a high risk for seizures, and monitoring for seizures in these cases is therefore appropriate even without observed clinical events. In one observational study of 514 neonates with acute symptomatic seizures, neonates monitored with a screening cEEG strategy had greater odds of successful seizure treatment as compared to those neonates who received confirmatory cEEG only after there was clinical suspicion for seizures (OR 2.44).71 In all cases, EEG-based neuromonitoring is essential for accurate diagnosis of seizures and assessment of their response to anti-seizure medications (ASM). Neonatal neurocritical care programs incorporating consistent neuromonitoring have demonstrated improved detection and rapid treatment of electrographic seizures, reduced risk of status epilepticus, reduced cumulative doses and duration of ASM, reduced proportion of infants discharged on ASM, and a trend toward reduction in subsequent development of epilepsy.72,73,74,75

Electrographic seizures may be as brief as ten seconds in duration or may be sustained over many minutes. Status epilepticus is defined as seizures occurring for ≥50% of any one-hour epoch.13 Paroxysmal clinical events must have an EEG correlate to be diagnosed as seizures. If no EEG is available, focal tonic and focal clonic seizures observed by experienced personnel can be diagnosed clinically with only “probable” diagnostic certainty, and other clinical events should be considered only “possible” seizures.76 According to the most recent ILAE classification, seizures in neonates can be electrical-only or can have motor (automatisms, clonic, epileptic spasms, myoclonic, tonic), non-motor (autonomic, behavior arrest), or sequential presentation.70 Seizure semiology may be indicative of certain etiologies. For example, focal clonic seizures are common in stroke, myoclonic seizures often occur in inborn errors of metabolism, and tonic or sequential seizures are common in genetic epilepsies.77,78 Recognition of specific non-epileptic clinical events that mimic seizures is also important; episodic tonic stiffening and apnea can be due to hyperekplexia, a genetic disorder that can cause fatal laryngospasm or apnea and is typically responsive to low dose clonazepam. Withdrawal from selective serotonin reuptake inhibitors may present with jitteriness and stimulus-induced non-epileptic myoclonus. Paroxysmal vital sign changes may also reflect systemic illness rather than seizures. Seizure semiology in neonates can be helpful, but by visual observation only it remains difficult to distinguish seizures from other non-epileptic paroxysmal events.

aEEG for the detection of neonatal seizures

While aEEG can detect many seizures, cEEG has higher sensitivity and specificity. Several studies have assessed the efficiency of a single- or two-channel aEEG trend to detect seizures, with and without display of the corresponding source EEG. Fewer than half of all seizures are detected when only the aEEG trend is inspected.79,80,81 However, when experienced users inspect the accompanying limited channel source EEG along with the aEEG trend, 76% of seizures82 and up to 80–90% of patients with seizures80,81 were identified. Additionally, aEEG showed a very distinctive pattern in infants with neonatal epilepsy associated with KCNQ mutations which could have significant clinical implications.83

Although useful in providing an overview of the temporal evolution of neonatal seizures, aEEG has limited ability to adequately characterize seizures due to numerous factors. First, aEEG is a very compressed trend: in standard displays, each millimeter of aEEG represents up to a minute of recording. Very brief seizures are thus not easily seen. In addition, aEEG reflects changes in EEG amplitude only, so it is more likely to miss low voltage seizures. In contrast to aEEG, cEEG can detect brief (<30 s) and low amplitude seizures although these seizures may be recognized if there is direct review of the the limited-channel source EEG of the aEEG.79,82,84,85 Also, aEEG typically has very few EEG electrodes, which may limit the ability to record focal seizures in all regions of the brain.86 Sensitivity is impacted by placement of the aEEG electrodes: most neonatal seizures originate from central and temporal regions and will not be detected by frontally placed leads.80,81 Therefore, frontal or forehead electrode placement should be avoided. Electrodes for aEEG should generally be biparietal (P3-P4) if one channel is used, and preferably parieto-central when two channels are used (C3-P3 and C4-P4). Reviewing raw EEG and understating the typical evolution of seizures on EEG is important since different artifacts and other rhythmic activities can look like seizures on aEEG (Fig. 4). Finally, the accuracy of aEEG interpretation is dependent upon the experience of the interpreting clinician. Statistical modeling shows that aEEG utility for neonatal seizure screening is best when used in those at highest risk for seizures, and when interpreted by an expert.87

Fig. 4: Examples of seizures and seizures-like artifacts on aEEG and the corresponding cEEG.
figure 4

a Seizures on a normal background; b Seizures on abnormal discontinuous background; c Ventilator artifact; d Patting artifact; e Chewing artifact; f Electrode artifact. (aEEG is displayed at 6 cm/h and cEEG is displayed at 15 mm/s. Arrows indicate where the CEEG is displayed) (Adopted from El-Dib et al. Amplitude-integrated electroencephalography in neonates, Pediatric Neurology 2019, with permission).

When cEEG is not available, aEEG is strongly recommended and is clearly superior to clinical assessment alone in the diagnosis of neonatal seizures.88 aEEG is also useful as a screening tool when there are delays in cEEG availability. Additionally, bedside display of an aEEG trend concurrent with cEEG recording can assist in timely detection of seizures by NICU- based providers. While high seizure burden is a known risk factor for brain injury and adverse outcome; further evidence regarding how timely seizure detection impacts long-term outcomes is needed. Two small, randomized trials demonstrated that treatment of electrographic seizures in the context of HIE was associated with reduction of overall seizure burden.88,89 More recently, a study failed to demonstrate reduction of death or disability in a heterogenous group of neonates diagnosed with seizures when randomized to be treated for electrographic and clinical seizures as compared to clinical seizures only.90 However, the results of this study should be taken cautiously due to numerous limitations including population heterogeneity with variable seizure etiology, late start of monitoring, possible difference in seizure burden (though not statistically different), the use of aEEG with unclear use of raw EEG or seizure detection algorithms, and the equal use of ASM among groups.91,92

cEEG as the gold standard for diagnosing seizures in neonates

Diagnosis of seizures requires EEG for several reasons. First, as above, seizures may have clinical manifestations that are difficult to detect by observation, or may present as only electrographically.46,93,94,95 In addition, paroxysmal movements can mimic seizures and EEG is needed to confirm the diagnosis. Finally, ASM can cause electroclinical uncoupling, in which clinical signs of seizures are suppressed even as electrographic seizures continue.96,97 For all these reasons, cEEG is the gold standard for neonatal seizure diagnosis.

Prolonged cEEG monitoring, rather than a brief 30- or 60-min recording, is needed to capture seizures because an abnormal or normal EEG background cannot confirm or exclude a diagnosis of seizures. In neonates, seizures can be short, and the interval between seizures can be more than an hour.56,95,98 The duration of cEEG necessary to capture seizures varies by underlying etiology. In some studies, 95–99% of neonates with seizures are diagnosed within 24 h of cEEG,93,99 while up to 36 h may be needed for neonates after cardiac surgery involving bypass.100,101 One study compared time to seizure detection between neonates with HIE, neonates undergoing cEEG for differential diagnosis of stereotyped clinical events, and neonates with any other condition. The authors found that those with HIE and no epileptiform discharges on initial EEG segment took the longest- 73 h- to capture 99% of the seizures.55 Thus, the duration of cEEG required to effectively exclude seizures may vary depending on the clinical scenario and the initial EEG findings. Once seizures are detected, cEEG monitoring should continue for at least 24 h after the last seizure to ensure true seizure resolution.12

Early recognition of electroclinical phenotypes may allow for targeted treatment and can guide the diagnostic approach.78,83,102 While most seizures in the neonatal period are provoked by an acute insult, in a significant minority they may represent the onset of a neonatal epilepsy.103 Simultaneous video recording allows for the recognition of seizure semiology which may inform etiologic evaluation.70,77 Among neonates with acute symptomatic seizures, optimal seizure detection may increase ASM efficacy since seizures are easier to stop when they are treated early.104,105 Screening cEEG to identify seizures among neonates at risk is associated with better initial response to ASM, as compared to using cEEG only for confirmation of diagnosis after clinically suspected seizures begin.71 Moreover, minimizing seizure burden may improve outcomes.51,88,89

cEEG is essential for diagnosis and management of seizures in neonates. It is important to acknowledge that the use of cEEG requires expertise, both in applying electrodes and in the availability of rapid neurophysiologist interpretation. These resources are not universally available.106 Ongoing research is needed to guide best practices in identification of neonates at risk for seizures, protocols for cEEG monitoring, strategies to make cEEG more accessible, and to better delineate the related utility and risks of anti-seizure treatment.

Value of neonatal seizure detection algorithms

Absence of timely cEEG interpretation in presence of seizures may cause delays in treatment even when cEEG is used.107 A practical solution to this would be an automated seizure detection algorithm (SDA), providing real-time, continuous EEG interpretation in the bedside EEG system. Although SDAs show great promise, they are not widely used/ available. Machine-learning based algorithms for neonatal seizure detection are emerging.108,109,110,111,112 Their seizure detection accuracy can be equivalent to that of human experts.112 Recently, the first multicenter trial of a neonatal SDA was published.113 This study found a machine-learning algorithm was able to detect a higher percentage of neonatal seizures without a significantly increased false detection rate when compared to clinician performance. Implementation of automated seizure detection algorithms will undoubtedly become a priority in the coming years, though SDAs must be recognized as a decision support tool that signals a biomarker of neurological compromise, and not just a binary indicator to treat.

Seizures in extremely preterm infants

Acute seizures in preterm infants are commonly caused by high grade intraventricular hemorrhage, hypoxia-ischemia, transient metabolic disturbance, or infection. Seizures in preterm infants are associated with a poor prognosis.114,115 The true incidence of seizures in extremely preterm infants is unknown, and there is great variability in estimates depending on the method of inspection. The estimates of seizure incidence range from 0.34–5% in studies based on clinical signs alone,116,117,118 to 22–48% in studies using aEEG monitoring with limited-channel EEG,119,120,121 and 0.9–8.7% in studies using cEEG.122,123,124,125,126

This very large difference between estimates reflects the fact that seizure detection in preterm infants is particularly challenging due to their variable normal behavioral repertoire and their unique cEEG and aEEG features. Rhythmic EEG findings mistaken as seizures may commonly arise from artifacts, such as those resulting from care procedures.126,127,128,129 Rhythmic non-epileptic activity with higher amplitudes may mimic aEEG characteristics of a seizure, and should be assessed by cEEG. Caution is warranted to ensure extremely preterm infants are not treated with unnecessary ASM on the basis of clinical observation or aEEG alone; when possible, cEEG should be used to confirm seizures.130

While visual inspection of the cEEG signal and use of multichannel EEG is undoubtedly the most accurate way to detect all seizures, many challenges remain unresolved in the care of preterm infants with seizures. Even as diagnosis improves, optimal seizure management in extremely preterm infants is still unknown.

NIRS changes in combination with EEG during seizures

While NIRS is not used to diagnose seizures, it may provide complimentary information about physiology in neonates with seizures. Seizures may induce a local hemodynamic response in brain tissue, which could be detected with a concurrent NIRS recording if the seizure is of long enough duration.131 Continuous spikes correlating with significant fluctuations of rStO2 have been observed with multifocal seizures lasting several minutes, whether subclinical132 or clinical.133 Given the potential for NIRS to detect cerebral hypoxia, it can be used to monitor the effect of seizures as well as ASM on cerebral oxygenation. Indeed, clinical seizures associated with systemic hypoxia will lead to cerebral hypoxia. In addition, subclinical seizures associated with autonomic signs (mainly changes in blood pressure) have been associated with fluctuations in cerebral oxygenation, sometimes reaching the hypoxic threshold.134 One study found that the treatment of infants with phenobarbital bolus (>10 mg/kg) was associated with decreased FTOE and increased rStO2, persisting for at least an hour after dose administration.135 While further work is needed before widespread use, the combination of NIRS with EEG monitoring in neonatal seizures might prove useful in understanding and guiding individualized care in the newborn.

Conclusion

With the expansion of neonatal neurocritical care, skillful application of neuromonitoring techniques holds promise for improving care to high-risk infants. Improved monitoring and more precise prognostication may allow better selection of therapies and ultimately better outcomes.

Different monitoring modalities need to be selected according to the clinical condition and individual patient needs:

  • For term neonates with neonatal encephalopathy, aEEG has a significant role in screening neonates for hypothermia eligibility, though should not be used to exclude otherwise eligible neonates. During TH, cEEG monitoring for at least 24 h is helpful to monitor evolution of encephalopathy and to diagnose electrographic only seizures. While some evidence suggests cEEG is to be continued for the duration of TH to identify all neonates with seizures, those neonates with an initial 24 h of normal/mildly abnormal cEEG and no seizures are at lowest risk for seizures and may be considered to transition to aEEG monitoring instead. Continuous cEEG, aEEG and NIRS monitoring during cooling and rewarming can assist in prognostication.

  • For neonates with possible seizures, cEEG is the gold standard for detection and diagnosis of seizures. While not as sensitive as cEEG, aEEG is superior to clinical assessment alone for the identification of seizures. The use of seizure detection algorithms can help with timely seizures detection at the bedside