NCSz refer to discrete electrographic seizures. A precise definition of what constitutes NCSE vs NCSz is not well established, and some studies merge patients with NCSz and NCSE; others merge NCSz with ictal–interictal patterns, described in the next section. Despite this, there is clear evidence to suggest that isolated NCSz, even those that occur in a small region of the brain recorded only using intracortical recordings, result in increases in heart rate and blood pressure and may result in increased intracranial pressure and metabolic demand with reciprocal changes in brain tissue oxygen and regional cerebral blood flow [21]. After SAH, NCSz have been independently associated with poor functional outcome defined as mRS 4–6 at 3 months, even after controlling for age, bleed severity, delayed cerebral ischemia, and inflammatory response [22]. After intracerebral hemorrhage (ICH), NCSz appeared to worsen neurological function and increase midline shift, although overall functional outcome at hospital discharge was not affected [23, 24]. Interestingly, nonconvulsive seizures appear to occur in patients with an intermediate-severity brain injury; while patients with seizures are at higher risk for poor outcome in general, they tend to do better than those who have markedly severe background attenuation or lack reactivity in patients with SAH or with severe sepsis [19, 22].

Critically ill patients may exhibit a variety of periodic or rhythmic patterns during cEEG that do not fulfill typical definitions of NCSz [25]. The interactions between neuronal injury and metabolic dysfunction manifest as abnormal patterns on cEEG which cannot be easily classified, falling somewhere on an ictal–interictal continuum [26]. Periodic discharges (PDs) are commonly associated with acute brain injury and may be acute, prolonged [27], or even chronic [28]. PDs are clearly associated with electrographic seizures [27, 29–31], but their relationship with outcome is less clear. In the medical ICU, the presence of PDs or NCSz has retrospectively been associated with poor outcome defined as GOS 1–3 at hospital discharge [32], although in a prospective study of medical ICU patients with severe sepsis, PDs or NCSz had no impact on 1-year survival or functional and cognitive outcomes [19]. In the setting of central nervous system infection, 48% had either PDs or NCSz with sixfold odds increase for poor outcome at discharge, defined as severe disability or worse [33]. In the surgical ICU, PDs were not independently associated with hospital discharge outcome, although 97 % of patients with PDs lasting longer than 24 h had poor outcome compared with 58 % of patients with transient PDs [17]. In comatose neurological ICU patients, however, prolonged PDs lasting more than 5 days in a row were found to have no impact on mortality [27] compared to intermittent or no PDs.

Of the PDs, lateralized periodic discharges (LPDs) are the best studied. Case series have demonstrated instances of increased regional glucose metabolism or blood flow in the region of LPDs, mirroring changes seen during seizures. After GCSE, LPDs on cEEG have been associated with poor outcome: more than half of those with LPDs died or were dependent at discharge [34]. After intracranial hemorrhage and SAH, LPDs were found to be independent predictors of poor outcome, defined as GOS 1–2 at discharge or mRS 4–6 at 3 months, respectively [14, 24] (see Fig. 1). Although overall mortality rates in patients with LPDs are between 25 and 41 %, much of the mortality associated with LPDs appears to be related to the underlying etiology, and case–control studies are lacking.

Generalized periodic discharges (GPDs), which include triphasic waves, are associated with mortality rates between 30 and 47 %, but when age, etiology, and level of arousal were controlled, there was no association with poor outcome, was found defined as death or vegetative state at hospital discharge [29]. Bilateral independent periodic discharges (BIPDs) have been highly associated with poor outcome, with a mortality of 61 % compared to 29 % in patients with LPDs. Nearly three-quarters of patients with BIPDs were comatose (vs 24 % of patients with LPDs), and with such a small sample size, it is likely that injury severity accounted for much of the mortality [31]. Further studies of BIPDs are needed to fully assess their prognostic importance.

Burst suppression (BS) has long been associated with diffuse brain injuries and is caused by a loss of normal cortical inhibition with hyperexcitable bursting interrupted by refractory periods [35]. Prognosis is strongly dependent on etiology: traditionally, BS has been associated with poor outcome after cardiac arrest but in other cases may be seen transiently during deep anesthesia. Asymmetric bursts in particular may point to structural lesions in the corpus callosum [36, 37]. Background suppression, on the other hand, refers to either a low-voltage record of predominantly theta or delta frequencies without reactivity or an isoelectric recording [38]. The prognosis associated with this pattern is typically poor once the effects of sedative medications or metabolic dysfunction (e.g., severe sepsis, uremia, hypercalcemia, thyroid or adrenal failure, hypoglycemia) are ruled out [39]. However, much of the literature on burst suppression and background suppression is focused on anoxic brain injury, and the majority are from the pre-hypothermia era.

Burst suppression may be induced therapeutically, as in refractory intracranial pressure or refractory status epilepticus. The prognosis in these cases is usually considered poor as a result of the underlying etiology, although maintenance of BS induced for treatment of refractory status epilepticus for >72 h may help improve overall seizure control and perhaps even survival [40, 41]. In one series of patients requiring pentobarbital for refractory status epilepticus, 90 % of whom had BS on cEEG, one in ten had minimal or no disability at 1 year [41], and in a series of patients with TBI requiring pentobarbital titrated to BS for refractory intracranial pressure, 24 % survived to good recovery defined as GOS 4–5 at 1 year [42].

Spontaneous BS may be observed in critically ill patients, many of whom require sedation and analgesia as part of their critical care. In neurological ICU patients with ICH, the presence of BS does not significantly portend poor outcome, defined as GOS 1–2 at hospital discharge, once clinical and radiologic predictors are controlled [24]. This appears to be true across neurological ICU patients, regardless of etiology [43]. In the operating room and in some ICUs, BIS monitoring (Covidien; Boulder, CO) is performed via a device designed to process the crossed bispectrum of raw frontal EEG into a simple-to-read number between 0 and 100. Using the BIS, anesthesiologists have reported on a large cohort of patients from two prior clinical trials, 28 % of whom experienced > 5 min of anesthesia-related BS during elective surgery under general anesthesia. Patients who experienced BS were older and had higher preoperative comorbidity. After controlling for these factors, those with both anesthesia-related BS and intraoperative hypotension had three times the risk for 90-day postoperative mortality [44]. Another study found that each hour spent in BS intraoperatively increased mortality at 1 year by 24 %, and each minute spent with a systolic blood pressure < 80 mmHg increased mortality at 1 year by 4 % [45]. Among medical ICU patients, 62 % of whom had sepsis, those with BS had double the hazard ratio for 6-month mortality than those without BS after controlling for age, comorbidities, and illness severity [46].

The background frequency content of the cEEG is typically reported descriptively: attenuated delta frequency slowing, predominant theta/delta frequencies, organized posterior dominant alpha, etc. Some frequency changes are important for the interpretation of cEEG in the critically ill, such as declining faster frequencies in the setting of inadequate cerebral perfusion, and particularly focal findings such as focal delta frequencies or asymmetry in the EEG related to stroke. Quantitative analysis of the frequency spectra provides useful objective information [47] and may have a role in predicting certain outcomes, as described in Chapter 15. However, the overall background frequency mix is largely nonspecific. Prior attempts to develop grading systems to reflect prognosis after diffuse brain injury have used overall background frequency patterns and their reactivity [6, 7]. However, with the use of continuously recorded EEG, the importance of dynamic parameters such as reactivity and state changes has been more heavily emphasized. As an example, the prognosis of frequency-dominant patterns such as alpha or theta coma, best studied after hypoxia, is predicated on their reactivity [48] and frequently evolves over several days to different patterns altogether [49].

Reactivity refers to a reproducible change in the EEG background as a result of stimulation (see Fig. 2). Although reactivity has received attention as a prognostic marker after cardiac arrest, the reproducibility of this finding is confounded by a lack of standardization of stimulus type and stimulus strength, inter-rater agreement on EEG changes, the effects of confounders such as sedation, and the timing of stimulation during the sleep–wake cycle. The kappa between expert raters for background reactivity in one study of routine EEG segments from patients with a clear “awake” state ranged broadly 0.33–0.95 [50]. Of practical importance in the ICU, the method of stimulation for a given patient needs to take into account other injuries, for example, hearing impairment from canal obstruction following trauma or surgery or sensory loss after cervical spinal cord injury. Caution should be used when applying background EEG reactivity to prognostic decision-making in the clinical setting. Within 48–72 h of TBI, one study used auditory clicks and nasal septum stimulation to elicit one of several categories of EEG reactivity: slow waves, attenuation, no reactivity, and uncertain. Increasing slow-wave activity (so-called paradoxical reactivity) was associated with GOS 4–5 (good recovery) at 18 months post-injury in 90 %. In those without EEG reactivity, 93 % had severe disability or worse; notably, 78 % with uncertain reactivity had good recovery or moderate disability [51]. After SAH, all patients with a poor outcome, defined as mRS >3 at 3 months, lacked EEG reactivity to any alerting stimuli; however, more than half of patients’ reactivity was not reported [14]. After ICH, reactivity was not found to be an independent predictor of outcome in multivariate analysis [24]. In patients with severe sepsis, a standardized stimulation protocol was used to define reactivity, which was absent in patients with continuous sedation, circulatory shock, and coma. Patients without reactivity had a mean survival time of 3 months compared with 8 months for those with EEG reactivity, although 13 % of survivors at 1 year had lacked reactivity in the ICU, and no differences were seen in functional outcome between those with and without EEG reactivity [19].

In a special case of reactivity, stimulation may produce ictal–interictal discharges. These stimulus-induced rhythmic, periodic, or ictal discharges (SIRPIDs) have been described in 22 % of patients undergoing cEEG on an inpatient neurology service [52]. SIRPIDs are thought to be related to dysregulated afferent input into hyperexcitable cortex and appear to be associated with the development of seizures. They are seen significantly more frequently after ICH, convulsive status epilepticus, and TBI [52, 53]. After ICH, focal SIRPIDs are independently associated with poor outcome, defined as GOS 1–2 at hospital discharge [24]; after SAH, SIRPIDs do not appear to be predictive of outcome [14]. The impact of SIRPIDs on outcome in general has not been adequately studied.

Variability refers to spontaneous changes in EEG over time, as opposed to reactive changes to a stimulus. A lack of variability – a “monotonous” EEG – has been described after brain injury using raw EEG [54] and using compressed digital spectral array [55] in association with poor outcome defined as death or unresponsive wakefulness. The presence of variability may refer to a simple biphasic EEG that alternates from one abnormal pattern to another over time, to more complex and definitive state changes with a behavioral correlate, or to variability in specific background frequencies during EEG. In a foundational study of states in comatose critically ill patients, all patients with a biphasic or invariant nighttime recording died [54]. In patients with TBI, variability has been quantified using the percentage of 6–14 Hz frequencies in the background EEG. The degree of variability in the first 72 h after TBI significantly correlated with discharge and 6-month GOS, and when added to a prognostic model of outcome that included standard clinical covariates, the alpha variability significantly improved the predictive ability of the model [56, 57]. After SAH, the presence or absence of state changes or spontaneous variability does not appear to independently predict outcome [14]. However, the alpha variability declines in patients with SAH a mean of 2.9 days prior to the onset of vasospasm [58].

Cyclic alternating patterns (CAP) are a specific form of variability described after brain injury. CAP refers to periodic bursts of delta frequencies in the absence of stimulation (hence, not reactivity). These patterns mimic normal sleep microarchitecture and appear independent of other alternating physiologic patterns, such as Cheyne–Stokes breathing. In one series of patients with CAP, 7 of 11 patients experienced good outcome (complete recovery or minimal dependency) after a follow-up period of several months [59] (see Fig. 3). In another cohort 1–2 weeks after traumatic injury, CAP was described in six patients, none of whom had a good outcome. This may have been related to the persistence or delayed presentation of CAP or a lack of overlying sleep transients which were not reported in the former series [60].

Sleep is frequently abnormal in critically ill patients with or without brain injury [61, 62]. In one series, only 30 % of critically ill patients had normal sleep at any point during cEEG [61]. The majority of patients in the ICU have state changes that do not fit the standard definitions of sleep as defined by the American Academy of Sleep Medicine. The term “atypical sleep” has been used to describe periods of behavioral or electrographic sleep-like states in these patients, who lack typical sleep transients such as K-complexes or sleep spindles [63]. After TBI, patients with typical sleep features have favorable prognosis: the presence of K-complexes or sleep spindles had a positive predictive value of 86 % for no or only mild disability at more than 1-year follow-up [60], and increasing spindles have been seen prior to recovery of consciousness and have been correlated to the duration of coma [54].