Chapter 13: Continuous Electroencephalogram Monitoring in the Intensive Care Unit

There is a mismatch between the neurologic examination and imaging findings; therefore, alternate causes for altered mental status must be explored. The frontal left hypodensity with a surrounding hyperdensity should not account for such a degree of obtundation. Depending on the clinical scenario, this may take the form of obtaining further history or diagnostic tests.

The goal of neuromonitoring is to identify secondary brain injury as early as possible and prevent permanent injury by triggering timely interventions. Ideally, such monitoring should be highly sensitive and specific, noninvasive, widely available and relatively inexpensive, pose no risks to patients, have high inter- and intra-rater reliability, and have good temporal and spatial resolution. Limitations of cEEG monitoring include high cost, vulnerability to artifact and medications, poor spatial resolution, and poor inter- and intra-rater reliability. On the other hand, it is noninvasive (as long as it is limited to surface EEG monitoring), has great temporal resolution, and allows assessment of neuronal activity.

Applications of cEEG in the ICU

1. To rule out subclinical or nonconvulsive seizures in patients with:

   • Persistently impaired mental status after a convulsive seizure or convulsive status epilepticus (SE)

   • Ongoing or frequently recurring movements, including isolated eye movements, that may or may not represent seizures

   • Impaired mental status and a history of epilepsy

   • Unexplained impaired brain function after basic evaluation (labs and imaging)

   • Acute brain injury with stupor or coma, including post anoxic coma, and sepsis

   • Unexplained fluctuations in mental status, including those secondary to hypoxia and increased intracranial pressure (ie, hydrocephalus, intracranial hemorrhage, and intraventricular hemorrhage)

2. To characterize paroxysmal clinical events, including posturing, rigidity, tremors, chewing, or even autonomic spells such as sudden hypertension, tachycardia, bradycardia, or apnea

3. Epileptiform activity or periodic discharges on initial EEG (often performed for not more than 30 minutes)

4. To detect cerebral ischemia (ie, subarachnoid hemorrhage patients at risk for delayed cerebral ischemia)

5. To guide medication titration and quantify seizure frequency in patients with refractory status epilepticus (RSE)

Nonconvulsive seizures (NCSs) and nonconvulsive status epilepticus (NCSE) are frequently seen in the ICU setting. These can present similar to the patient described in this case. Even in the medical ICU population after excluding all patients with any clinical suspicion of seizures, approximately 8% of comatose patients have electrographic seizures.^1,2 Electrographic seizures can be detected by cEEG in 20% to 48% and NCSE in 14% in the aftermath of clinically successfully treated convulsive SE.^3,4 In the acute brain injury setting, electrographic seizures are even more frequent but the incidence clearly depends on the underlying etiology. However, no population-based studies are available to date; thus, the real incidence is unknown. In the NICU setting, subclinical seizures were seen in 18% of a cohort of 570 consecutive patients who underwent cEEG for the detection of subclinical seizures or to evaluate an unexplained decrease in consciousness (Table 13-1). In this study, seizures were more frequently recorded in younger patients with a history of epilepsy, patients who had convulsive seizures prior to start of cEEG monitoring, and those who were comatose at the time cEEG was started. When stratified by etiology, either periodic epileptiform discharges (PEDs), NCSs, or NCSE can be seen in one-quarter to one-half of those undergoing monitoring (Table 13-2). While the majority of seizures will be recorded in the first 24 hours, frequently 48 hours or more of cEEG are needed to record the first seizure in comatose patients (Figure 13-3).⁵

In addition to looking at the raw EEG data, the EEG recordings can be subjected to quantitative analysis. A number of different quantification methods are available with most commercially available EEG software. Quantification techniques include the calculation of power spectra by means of fast Fourier transform (FFT), which then can be used to calculate ratios of fast over slow activity or other parameters. These can be displayed as numbers or graphically as compressed spectral arrays (CSAs), histograms, or as staggered arrays, which have the potential to unmask subtle background changes when trended over time. The qEEG parameters may include total power, frequency activity totals (eg, total or percentage alpha power), spectral edge frequencies (eg, frequency below which 50% of the EEG resides), amplitude-integrated interval, frequency ratios (eg, alpha-to-delta ratio), and brain symmetry index.⁶ Other more automated EEG data reduction display formats include the cerebral function monitor (CFAM), EEG density modulation, automated analysis of segmented EEG (AAS-EEG), and bispectral index (BIS) monitor.^7-11 However, qEEG should always be interpreted in the context of the raw EEG since artifacts, medications, and other clinical events may cause very similar findings to epileptiform findings (Figure 13-5). This is particularly true for software packages that display “user-friendly” composite scores, and any outputs that are based on unpublished algorithms (“proprietary information”) should be viewed with great caution. The qEEG parameters are particularly useful for seizure quantification and to estimate effects of seizure treatment once the qEEG “footprint” of a seizure has been identified.

In addition to seizure quantification, qEEG is useful for trending patterns over time as the transitions may not be as evident on review of the raw data. The qEEG and raw EEG in Figure 13-6 illustrate resolution of NCSE over an extended period of time.

In addition to the quantitative measures described above, seizure detection programs using specialized EEG signal–processing software can be used for screening large cEEG datasets, allowing the reviewer to hone in on marked segments (Figure 13-7, top row). These programs may be based on a simple FFT-based frequency analysis, may recognize specific waveform patterns, or incorporate a complex learning algorithm that recognizes more combinations and sequential developments of typical waveforms. Based on the FFT analysis of EEG, CSA graphs can be generated to determine the occurrence of subclinical seizures once the “CSA signature” or footprint of a seizure in an individual patient has been determined.¹² This can be used to quickly screen a 24-hour recording and quantify the frequency of seizures.

Others have investigated the utility of the cerebral function analyzing monitor (CFAM) to detect seizures, but sensitivity was particularly low for partial seizures.¹² Algorithms were limited by a high false-positive rate with good sensitivity.¹² While promising, commercially available automated seizure-detection software has been developed for use in epilepsy-monitoring units and programmed to identify seizures from healthy patients. However, seizures in brain-injured and often medically sick comatose patients have rather different patterns, which are typically less organized, have a slower maximum frequency, and are longer in duration with unclear on and off settings. Thus, in order to utilize seizure-detection programs in the NeuroICU, new software will need to be developed and programmed to more accurately identify complex ICU ictal patterns. Automated seizure-detection software based on multichannel, digitized real-time FFT (2 seconds per epoch, 2 minutes averaging) and trends of total EEG power has been used by some groups with reasonable success but has not been verified on an independent dataset.^11,12

Yes. Symptomatic vasospasm and the associated delayed cerebral ischemia (DCI) are common complications that occur in 20% to 40% of patients with SAH.¹³ A number of treatments are available, but making the diagnosis of DCI can be challenging, particularly in those with a limited neurologic examination. Ideally, timely recognition of ischemia leads to interventions that prevent infarction. In addition to recording ictal activity, cEEG may detect decreased cerebral perfusion in the setting of vasospasm after SAH. EEG is very sensitive to ischemia and may reveal changes at a reversible stage of reduced cerebral blood flow (CBF) and neuronal dysfunction (25 to 30 mL/100 g per minute; Table 13-1). The raw EEG progresses through predictable stages with increasing degrees of hypoperfusion, including loss of fast activity, increased slowing, and background attenuation^14,15

EEG was first used to detect ischemia during carotid endarterectomy (CEA) and is still used to monitor for ischemia during cross-clamping (Figure 13-11). In acute ischemic stroke, reocclusion occurs in about 34% after successful recanalization with tissue plasminogen activator (tPA).¹⁶ Quantitative EEG parameters such as the brain symmetry index (BSI) and the acute delta change index correlate with the degree of cerebral perfusion after acute ischemic stroke (AIS) measures by diffusion and perfusion-weighted imaging MRI.¹⁷ Some studies suggest that the qEEG parameter known as regional attenuation without delta (RAWOD) has a distinctive early pattern reflecting hemispheric infarction and may help to anticipate which patients might benefit from thrombolytic therapy.¹⁸ Furthermore, EEG parameters such as BSI correlate with the clinical examination measured by the National Institute of Health Stroke Scale of outcome after stroke.^19,20 Quantitative EEG is often used in the operating room at our institution for ischemia detection during CEA as illustrated by the case shown in Figure 13-11 in which a woman underwent clamping of the internal carotid artery (ICA) and had pronounced attenuation of the ipsilateral hemisphere suggesting an incomplete circle of Willis and/or collateralization. To minimize intraoperative ischemia, a shunt between the common carotid artery and ICA was placed for the surgery.

A few investigations reported on the applicability of qEEG parameters for the detection of DCI and/or vasospasm after a SAH. While the ideal qEEG parameter for assessing ischemia is a matter of debate, most include some ratio of fast-to-slow activity.²¹ Variability of relative alpha and post stimulation ADRs have been shown to correlate with DCI in good- and poor-grade SAH patients. Changes in relative alpha variability have been detected up to 2 days before the onset of clinical symptoms.¹⁵ Among 34 patients with poor grade, a SAH monitored with cEEG, a single drop greater than 50% (sensitivity 89% and specificity 84%, PPV 67% and NPV 96%) and six consecutive recordings with 10% or more reduction in the baseline ADR (sensitivity 100%, specificity 76%, PPV 60% and NPV 100%) correlated well with the development of DCI.²¹ Other qEEG methods to characterize ischemia include the compressed spectral array (CSA), suppression-burst ratio, amplitude-integrated EEG, or the asymmetry indices.¹⁵

This patient underwent cEEG monitoring and developed new-onset right hemispheric slowing on postbleed day 7. As seen in the figure (Figure 13-9), the cEEG ADR progressively decreased after day 6, particularly in the right anterior region, to settle into a steady trough level later that night, reflective of the decrease in faster frequencies and slowing over the right hemisphere (Figure 13-10). These changes occurred several hours before TCD velocities were obtained and clinical symptoms were noted. A transient increase of the right anterior and posterior ADRs were noted after the infusion of intra-arterial vasodilators during angiogram. However, the increase was transient and several hours later the ADR decreased again, reflecting progression of her vasospasm.

A review of the quantitative and raw EEG with hemodynamic parameters is shown in Figure 13-13.

EEG monitoring can be especially useful in evaluating the response to interventions that aim to depress neuronal activity and brain metabolism. As described in Chapter 12, cEEG is essential in guiding successful treatment of refractory status epilepticus. It gives clinicians an end point that allows adjustment of IV medications to maintain enough suppression of the EEG background while minimizing adverse effects. Pharmacologic coma remains a tool for the treatment of refractory ICP elevations, especially when occurring acutely. Although effective in reducing ICP, its use is coupled with a high risk of systemic complications, including hypotension, renal failure, cardiac depression, and hepatic dysfunction. The goal is to use the minimum dose required to achieve the goals of ICP control and decrease metabolism.²²

Continuous EEG may also provide information about the level of sedation in the critically ill, especially in the setting of neuromuscular blockade. A number of commercially available qEEG-based tools such as the Bispectral Index,²³ Patient State Index,²⁴ and Entropy Monitor and Narcotrend²⁵ have been used in operating rooms and ICUs for decades to monitor sedation levels. These single-purpose devices use proprietary algorithms and have been shown to be very inconsistent, particularly when confronted with EEGs showing seizures or ischemia. For the time being one can only ask for extreme caution when using these “black-box” analytical methods in the acute brain injury or ICU setting.

Preliminary studies from our group have suggested that a small depth electrode inserted near the cortex in severely brain-injured patients (acute brain injury resulting in coma at risk for secondary brain injury) may improve the signal-to-noise ratio of EEG (ie, shivering obliterating surface recordings), clarify suspicious but not clearly epileptiform patterns (ie, rhythmic slowing without clear evolution), detect seizures not seen on the surface, and detect changes that indicate secondary complications(ie, ischemia).^25,26 The significance of depth-only findings is currently unclear and should not lead to management changes without any additional data to corroborate the impression.²⁷

In this case, the depth electrode showed the potential to improve the signal-to-noise ratio and identification of a secondary complication; that is, hemorrhagic transformation of an acute ischemic stroke. Additionally, depth electrode monitoring may allow clarification of unclear surface EEG patterns or show ictal patterns that are not evident on the scalp EEG recording (Figures 13-18 and 13-19).

A number of studies have observed high-frequency oscillations using intracortical EEG recordings in normal subjects and patients with epilepsy.^28-32 This phenomenon has also been termed microseizures and cannot be recorded by regular surface electrodes. Interestingly, these microseizures are more frequent in brain regions that generated seizures and also sporadically evolved into large-scale clinical seizures. The relationship to depth-only seizures in patients undergoing depth electrode monitoring, and if this is present in patients with acute brain injury, is currently unknown.

Cortical spreading depression as well as slow and prolonged peri-injury depolarizations lasting minutes have been reported in a number of recent studies in patients with acute brain injury.^33-36 Both patterns are thought to indicate compromised metabolism. Clinically, they may be used to detect infarct enlargement in patients with MCA infarction³⁵ and DCI after aSAH.³⁴

Abnormal epileptiform EEG patterns with or without clinical correlates including periodic discharges and seizures can be frequently seen after cardiac arrest. Periodic epileptiform patterns, seizures, loss of reactivity and absence of normal sleep architecture, suppression burst, flat background^37-38 have been associated with poor outcome after cardiac arrest. In the era of induced hypothermia after cardiac arrest,³⁹ the prognostic implications of EEG findings have to be revisited, and anecdotally it is clear that good outcomes are possible even in patients with status epilepticus after cardiac arrest.

Yes. Studies reporting the seizure frequencies in non-neurologic ICUs are limited. Over a 2-year period, 12.3% of 1758 admitted medical ICU patients experienced some type of neurologic complications. Among those, seizures (28%) were the second most common complication after metabolic encephalopathy and usually occurred in the context of metabolic disarrangement.⁴⁰ These include, but are not limited to, hyponatremia, hypo- or hyperglycemia, hypocalcemia, drug intoxication or withdrawal uremia, and hepatic failure and hypertensive encephalopathy.⁴¹ Septic encephalopathy is the most common type of metabolic encephalopathy encountered in the medical ICU and is found in up to 70% of patients.^42,43 Recently, a review of 257 patients admitted to the medical ICU and submitted to cEEG monitoring found NCSs in 11% of the cohort of 19% of those with sepsis. In fact, sepsis was an independent predictor of NCS, and the presence of electrographic seizures was associated with poor outcome.^44,45 The true overlap between seizures and sepsis will need to be studied prospectively.

Obstacles to more widespread use of this technique include expense of equipment and personnel, 24-hour availability of technicians to maintain high-quality recordings, and the presence of encephalographers available to review the studies in real time.⁴⁶ Few studies have investigated the clinical impact of cEEG-driven interventions and health care economics associated with cEEG monitoring. The key to developing an efficient and proficient cEEG service is being able to train nurses and medical staff to recognize both basic EEG patterns and their changes over time as well as common artifacts encountered in the ICU setting (Figure 13-20). Common artifacts may mimic seizures, including chest percussion, oscillating beds, ventilators, chewing, and rhythmic movements. Other artifacts in the ICU include electrical artifact (60 Hz) from dialysis, extracorporeal membrane oxygenation (ECMO), pumps/drips, heating/cooling devices, and electric beds. Identification of artifacts is greatly aided by use of video EEG. In the light of increased availability of CT- and MRI-compatible electrodes as well as prospective studies assessing the impact on outcomes, such monitoring is crucial and will help make cEEG a valuable tool in the armamentarium of neuromonitoring devices.⁴⁷