Neonatal EEG

Neonatal EEG is a different language from adult EEG, and the first rule of that language is that there is no single normal. The neonatal brain is changing so rapidly that the same tracing is healthy at one age and pathological a few weeks later. Every neonatal EEG is therefore read against the postmenstrual age (PMA) - gestational age at birth plus chronological age, expressed in weeks since the last menstrual period - and the central interpretive act is to judge whether the background is appropriate for that age. A pattern of discontinuity that is entirely normal at 30 weeks is a serious abnormality at term. The reader who does not know the postmenstrual age cannot read the study at all; it is the indispensable denominator of the immature brain, the neonatal counterpart to the state-and-background principle that governs adult interpretation.

A terminology note matters here, because the literature is inconsistent and a careful reader must navigate it. The American Academy of Pediatrics recommends postmenstrual age for clinical use and explicitly discourages the older term conceptional age (gestational age minus about two weeks, dated from conception), on the grounds that 'conceptional' and its variants have been defined and applied inconsistently. Much of the classic neonatal EEG ontogeny literature nonetheless used conceptional age, and many authors still use the two terms interchangeably even though, taken literally, they differ by roughly two weeks. The practical discipline is to state the age and its basis explicitly and to read the established maturational milestones - which were largely described in those classic studies - against PMA, while recognizing that an older source quoting a 'conceptional age' is referring to a point about two weeks earlier than the same number in PMA.

Age-dependent patterns of the immature brain

The hallmark of the premature record is discontinuity. The most immature brains show trace discontinu - high-voltage bursts separated by prolonged, very-low-voltage interburst intervals - reflecting a cortex whose sustained activity is not yet established. As the brain matures, the interburst intervals shorten and their voltage rises, and by term the record becomes largely continuous in the waking and active-sleep states, reaching full continuity by around 44 to 46 weeks PMA. A caution for any quantitative claim: the absolute interburst-interval durations quoted by week disagree substantially across sources, in part because some report a maximum tolerable interval and others a typical one, so an interburst-interval table should be treated as approximate; only the trend - shortening and rising in voltage with maturation - and the continuity endpoint are robust.

The age-specific normal of quiet sleep near term is trace alternant: a continuous alternation between higher-voltage bursts (around 1 to 3 seconds) and lower-voltage - but not flat - interburst segments. Trace alternant appears around 34 to 36 weeks PMA and is the normal quiet-sleep pattern of the healthy term neonate, resolving as it is replaced by the more continuous slow-wave sleep of infancy by around 44 to 46 weeks. It must not be mistaken for the pathological discontinuity of an injured brain: the distinction lies in the interburst voltage, which is suppressed in pathological burst-suppression but preserved in trace alternant. This single discrimination - is the interburst segment merely lower in voltage, or is it flat? - is among the most consequential a neonatal reader makes, because it separates a healthy term infant in quiet sleep from a brain with severe injury.

The other signature graphoelement of the premature brain is the delta brush (also called the beta-delta complex or 'ripples of prematurity'): a slow delta wave with superimposed fast, beta-range oscillations riding on its crest. Delta brushes are a normal and expected feature of prematurity. They emerge around 26 to 28 weeks PMA, are most abundant - their peak - around 32 to 34 weeks, shift in their predominant location as the brain matures (central regions earlier, occipital and temporal later), and then become rare at term, largely gone by around 40 weeks and at most persisting to about 42. Their abundance, location, and timing are themselves maturational markers; their frank persistence well beyond term or gross asymmetry can indicate abnormality. A modern refinement worth noting is that delta brushes are now known to be partly stimulus-evocable by somatosensory, auditory, and noxious input, softening the older framing of them as purely endogenous - though this does not change their maturational timing.

Two further age-specific transients round out the picture. Temporal theta bursts - rhythmic temporal theta, sometimes described as 'sawtooth' or temporal theta coalescent - are a normal feature of the very premature brain, prominent around 26 to 31 weeks PMA with their sharpest expression near 29 to 31 weeks, fading by about 34 weeks; they should be kept conceptually distinct from the separate occipital monomorphic delta or sawtooth activity of the very premature occiput, which some sources conflate with them. Encoches frontales (frontal sharp transients) are normal, often bisynchronous frontal sharp waves associated with the trace alternant of the late preterm and term infant, prominent from around 34 weeks with a resolution tail extending toward 46 to 49 weeks. The reader of a neonatal EEG is thus engaged in a kind of electrical dating of the brain - estimating postmenstrual age from the morphology and then comparing that estimate to the known age, with a discrepancy itself constituting a finding.

Neonatal seizures and electroclinical dissociation

Neonatal seizures are unlike seizures at any later age, and the difference is fundamental: in the neonate the link between the electrographic seizure and the clinical seizure is loose. This is the principle of electroclinical dissociation (also called electroclinical uncoupling). Many neonatal seizures are electrographic-only - clear ictal discharges on EEG with no visible clinical accompaniment - and conversely some clinically suspicious events have no EEG correlate and are not seizures at all. The magnitude is striking: in careful video-cEEG cohorts only about a third of electrographic seizures had any clinical correlate, implying that roughly two-thirds were electrographic-only, and in one such study only about 9 percent of seizures were both clinically expressed and correctly identified at the bedside. A conservative summary is that something on the order of 50 to 85 percent of neonatal seizures are electrographic-only, with a central estimate near 60 to 70 percent, and that the subclinical proportion rises further after antiseizure medication, which can abolish the clinical signs while the electrographic seizures continue - the very phenomenon of uncoupling.

Because the immature cortex is poorly myelinated and incompletely connected, ictal activity often fails to produce the organized motor manifestations seen later, and the practical consequence is profound: in the neonate the EEG is the arbiter of whether a seizure is occurring. The 2021 ILAE classification of neonatal seizures formalizes exactly this, defining the neonatal seizure as an electrographic event and routing classification through whether a clinical sign is present (electroclinical) or absent (an explicit electrographic-only category), with clinical seizures further described as motor, non-motor, or sequential. This is a genuine conceptual shift from a semiology-first to an EEG-confirmed framework, and it should not be confused with the separate 2022 ILAE classification of epilepsy syndromes with neonatal and infantile onset.

When clinical signs do occur they are characteristically subtle: ocular phenomena (sustained eye deviation, fluttering), oral-buccal-lingual movements, 'bicycling' or 'swimming' limb movements, autonomic changes, and apnea. The reliability with which a clinical event predicts an electrographic seizure varies sharply by type. Focal clonic events are the most consistently EEG-confirmed and are reliably epileptic, and focal tonic events are usually epileptic as well; by contrast, generalized tonic posturing is frequently not an electrographic seizure but a brainstem-release phenomenon - resembling decerebrate or decorticate posturing, often stimulus-provoked and suppressible - and subtle automatisms and generalized myoclonus are inconsistently epileptic. Under the 2021 EEG-based definition, a generalized tonic-posturing event with no ictal EEG correlate is not classified as a seizure at all, which reinforces rather than overturns the classic Mizrahi and Kellaway teaching. The morphology of the true neonatal electrographic seizure is itself distinctive: a discrete, evolving discharge with a defined beginning, a change in frequency and amplitude over its course, and an end - frequently focal or multifocal, even migrating between regions within a single infant, reflecting the immature brain's inability to sustain a generalized synchronous discharge.

The clinical stakes of this monitoring are heightened by the fact that neonatal seizures are overwhelmingly acute symptomatic - provoked by an identifiable acute brain insult rather than reflecting an established epilepsy - and the EEG both detects the seizures and helps characterize the underlying process. Hypoxic-ischemic encephalopathy is the single most common cause, followed by acute ischemic stroke (which classically produces focal seizures with a focal background abnormality over the infarct), intracranial hemorrhage, and central nervous system infection; metabolic disturbances such as hypoglycemia and hypocalcaemia, and the inborn errors and genetic and structural causes that underlie the rarer neonatal-onset epilepsies, round out the differential. Because the seizure burden in these infants is frequently high and largely subclinical, and because emerging evidence links cumulative electrographic seizure burden to worse neurodevelopmental outcome, the contemporary standard is to initiate continuous EEG (or aEEG with raw-tracing confirmation) early in any neonate at risk and to treat electrographic seizures whether or not they are clinically apparent. The EEG thus serves a dual role at the bedside: it is simultaneously the diagnostic arbiter of seizures and a continuously updated readout of the severity and evolution of the encephalopathy that caused them.

Hypoxic injury, aEEG, and the confounders of cooling

After hypoxic-ischemic encephalopathy (HIE) the background EEG is one of the best early electrophysiologic predictors of outcome, graded by severity of background depression. A mildly abnormal background (mild excess discontinuity, transient asymmetries) carries a far better prognosis than a moderately abnormal one, while severe patterns - persistent burst-suppression, a low-voltage undifferentiated record, or electrocerebral inactivity - are associated with poor neurodevelopmental outcome. As with the adult comatose brain, reactivity and the recovery of background continuity and of sleep-state cycling over the first days are favorable signs, whereas a background that fails to improve, or fails to develop normal cycling, is ominous. The reappearance of organized sleep-wake cycling is one of the more reassuring milestones in the recovering neonate, and its presence within roughly the first day and a half in moderate HIE is a favorable sign.

In the neonatal ICU, conventional cEEG is frequently complemented by amplitude-integrated EEG (aEEG) - a reduced (one- or two-channel) montage that is filtered, rectified, amplitude-compressed on a semilogarithmic scale, and time-compressed to display hours per screen. Its strengths are the bedside display of the background trend, of sleep-wake cycling, and of the broad severity grade. The standard background classification (Hellstrom-Westas) runs from continuous normal voltage, through discontinuous normal voltage, burst-suppression, and continuous low voltage, to a flat (inactive) trace; a voltage-based scheme is also used. The limitations, however, are real and must be stated honestly: aEEG misses brief seizures (those under roughly 30 seconds), focal and low-amplitude seizures, and seizures distant from its limited montage, and it attenuates very slow activity; its reported seizure sensitivity is wide and operator-dependent, on the order of 40 to 85 percent, and interrater agreement can be poor. Conventional continuous video-EEG remains the gold standard, and aEEG is best used as a screening and trending adjunct whose flags are confirmed on the raw tracing.

Hypoxic patterns must be read through two confounders that echo the adult ICU. The first is therapeutic hypothermia, the standard of care for moderate-to-severe HIE in infants of at least 36 weeks: whole-body cooling to about 33.5 degrees Celsius for 72 hours, started within 6 hours of birth. Cooling itself, together with the sedation and analgesia that usually accompany it, depresses and can discontinue the EEG, and rewarming transiently increases discontinuity. The second is medication: phenobarbital loading causes background depression, lidocaine can produce discontinuity or burst-suppression, and benzodiazepines depress the preterm background especially. The crucial point is that the confounding is partial, not absolute: a burst-suppression pattern during active cooling and sedation cannot be assigned the grave meaning it would carry in a normothermic, unmedicated infant, yet the cooling-era literature still finds the background highly predictive once the timing is right - a severely abnormal background at around 24 hours, and after rewarming, remains strongly associated with death or impairment. The reliable window thus shifts later (roughly 24 to 72 hours and post-rewarming) rather than vanishing.

Two developments anchor the 2026 picture and are worth stating with their actual evidential weight. Cooling is not routinely recommended for mild HIE outside research, and trials of longer or deeper cooling have shown no benefit and possible harm, so 33.5 degrees for 72 hours remains the standard. Cooling has fared differently in different settings: in low- and middle-income countries the HELIX trial found not merely an absence of benefit but a statistically significant increase in mortality, so cooling is not recommended there. Adjunctive neuroprotection has largely disappointed - erythropoietin in the HEAL trial did not improve death or neurodevelopmental impairment and caused more serious adverse events, and xenon and melatonin remain investigational rather than standard care. The honest message for a physician-learner is that supportive care plus appropriately targeted hypothermia, monitored with EEG, remains the evidence-based core.

Finally, the neonatal EEG is the start of a developmental trajectory that the electroencephalographer follows for years. The discontinuity of prematurity yields to the continuity of term; the trace alternant of quiet sleep gives way over the first weeks to the continuous slow-wave sleep of infancy; delta brushes, temporal theta, and encoches frontales each resolve on their own schedule; and across infancy the posterior dominant rhythm emerges and accelerates - a slow posterior rhythm appears in the first months and climbs toward the mature alpha frequency through early childhood, reaching roughly 8 Hz by about age three - while sleep spindles, K-complexes, and vertex waves appear and organize. To read a neonatal EEG well is therefore to hold the entire maturational sequence in mind, because the single judgment that matters most - is this brain developing as it should? - can only be made against the lawful timetable of normal electrical maturation.