Intensive care units (ICUs) employ physiologic monitoring equipment to detect potential problems, aid in rapid diagnosis, and guide therapy. Although the equipment available in most general ICUs allows sophisticated monitoring of cardiac and pulmonary variables, the most important organ for functional outcome, the brain, is typically monitored only by physical examination. Additional monitoring devices are used in special circumstances. Intracranial pressure (ICP) monitoring has been available since the 1960s and brain tissue oxygen and brain temperature monitoring since the 1990s. Microdialysis of the brain extracellular fluid can be performed, but the time lag between the equilibration of dialysate and the generation of results prevents changes in management in real time. Other devices, such as probes that provide continuous regional measurements of cerebral blood flow, are becoming available in a few research centers. By far the most widely available techniques for monitoring the brain, however, are electrophysiologic, primarily electroencephalography (EEG), but also including evoked potentials (EPs) and event-related potentials (ERPs).

The most common current indication for electrophysiologic monitoring is refractory status epilepticus (RSE).¹ Once a patient fails to awaken after first-line therapy for status epilepticus (SE), EEG monitoring becomes necessary because a substantial percentage of patients will enter a state of electrical-mechanical dissociation in which motor activity ceases but the EEG continues to show ongoing seizure activity. It is no more conscionable to attempt treatment of RSE without EEG monitoring than to treat complex cardiac arrhythmias without electrocardiographic monitoring. In the coming years, neurologists should ensure that such monitoring is available in all centers that treat such patients.

Performing EEGs in ICUs using standard EEG equipment has long been viewed as a challenge for both the technologist and the interpreter. The electrical environment of the ICU is commonly viewed as hostile because the large number of devices produces electrical interference, mechanical artifacts, or both. Attention to detail, however, such as keeping electrode wires parallel to each other and away from cables carrying mains current can reduce many artifacts. Technologists may not be able to eliminate all physical artifacts, such as those produced by mechanical ventilation, but they can provide the interpreter with accurate information about the timing of extracerebral events so that they are not misinterpreted as being of brain origin. Simultaneous video recordings are helpful for identifying artifacts and avoiding misinterpretation. In the modern era, fears about compromising patient electrical safety are largely unfounded because all devices attached to a patient are electrically isolated, and the ground electrodes placed on the patient’s skin are virtual grounds that are not physically connected to the equipment chassis or the earth. Thus, there is no longer a prohibition against placing more than one ground electrode on a patient (assuming that all of the devices are properly isolated). ICU staff members involved in patient care still need to understand the possibility of microshock hazard, but they should also be aware that contemporary vascular devices that could provide a current path through the heart are also electrically isolated.

The large numbers of tubes, catheters, and other devices attached to the patient may seem intimidating to EEG technologists accustomed to working in the EEG laboratory, but with assistance from the patient’s nurse, it is almost always possible to achieve good electrode attachment with acceptable scalp impedances. Many patients will have scalp wounds or monitoring devices that may require some creativity in electrode placement. In such circumstances, it is often not possible to position all electrodes at standard International 10–20 System locations. If the interpreter is aware of the alternate location, this does not pose a significant problem. For brief studies, saline paste is adequate unless the patient is very diaphoretic. For longer recordings, electrode attachment using collodion or similar substances is preferred. However, EEG electrodes produce artifacts in computed tomographic (CT) and magnetic resonance (MR) images, and the presence of any wire loops inside the MR magnet may cause inductive heating, with the attendant risk of burns. Although several groups are developing CT- and MR-compatible EEG electrodes,^16,19 there are still no readily available and practical MRI-safe and non–image-distorting electrodes.

Several types of devices are available that can record EEG activity in some manner. The standard against which these should be judged is the regular digital EEG machine, recording activity from at least the 18 usual locations of the International 10–20 System. The device employed should store the raw EEG signal recorded against a common reference electrode so that montages of choice can be reformatted off-line. Simultaneous recording of the electrocardiogram (ECG) is crucial, and the system should also be capable of recording other polygraphic variables such as airflow, chest movement, electromyographic (EMG) signals, and eye movements. Video recording capability is important for the identification of artifacts, especially during prolonged recordings when a technologist will likely not be present for the bulk of the recording. A reliable system for recording medications and other interventions is also a necessity. Having the data automatically downloaded to a network server for analysis and archiving is very useful.

A variety of less capable devices have been developed with a view toward use in both the operating room and
the ICU. One of the earliest, the cerebral function monitor, provides a single channel of EEG for visual analysis. The signal is derived from two electrodes, usually placed over the parietal regions, and filtered to diminish low (<2 Hz) and higher (>15 Hz) components. The device displays the amplitude of the signal remaining after filtering. The tracing is displayed on a chart recorder at a low paper speed. The effect of seizures on this signal is unpredictable, and the manufacturer (Lectromed) does not suggest its use for seizure monitoring.

Later versions of the small EEG monitor include the amplitude-integrated EEG (aEEG) monitor (Brainz Instruments) and the BIS monitor (Aspect Medical). The aEEG device is designed for neonatal intensive care units; it displays and stores two channels of EEG along with the corresponding amplitude-integrated tracings at a lower sweep speed. The current version of the BIS monitor provides displays of EEG and EMG activity as well as a number calculated from the EEG, derived from a single bipolar frontopolar derivation. The algorithm for calculation of this “BIS number” is proprietary, but it appears to be heavily influenced by the presence of periodic suppressions in the EEG. According to the manufacturer, a normal waking EEG should produce a number of 100; surgical anesthesia produces a number between 60 and 40; a suppression-burst pattern causes the number to fall to near 20; and an isoelectric EEG yields a number of 0. Similar devices are now in development, and their ultimate role in ICU monitoring will require study and experience. None of these devices is intended to detect seizures; rather, their purpose is to assess and track EEG background activity.

The Moberg Neurotrac was an eight-channel EEG monitor that could display and store raw EEG along with a variety of computed parameters. The most useful of these was the compressed spectral array (CSA), discussed later. Its flexible display and its ability to store 24 hours of EEG on a removable hard drive made it a very useful device, but the slow microprocessors of the 1990s prevented its wide adoption.

Although continuous monitoring of all ICU patients with altered awareness may be the ideal, compromises are usually necessary because of too few monitoring devices, technologists, or electroencephalographers. Claassen and colleagues showed that nearly 20% of the ICU patients they monitored had seizures, and >90% (101/110) of these had purely nonconvulsive activity that would have been missed without EEG monitoring.⁸ Eighty-eight percent of these seizures were detected in the first 24 hours of monitoring and another 5% on the second day.

Recording 24 hours of continuous multichannel EEG with simultaneous video is now a trivial matter, but it poses substantial challenges in monitoring, interpretation, and data archiving. EEG monitoring implies that someone knowledgeable is observing either the raw EEG or the results of some computer program running in real time designed to detect events or trends of possible pathophysiologic significance. Such events may be seizures, but they could also be runs of focal slow waves, periods of voltage attenuation, or a change in a derived value such as the ratio of delta to alpha activity, which appears to be valuable in the prediction of vasospasm in patients with subarachnoid hemorrhage (SAH). Such observation needs to be done in real time to provide the intensive care staff with the timeliest information with which to manage the patient. However, becoming a skilled electroencephalographer requires years of training and experience. In addition, patients being monitored in ICUs often display EEG activity that is not commonly seen in other settings and may thus present interpretive difficulties.

Seizure detection software is sometimes used to assist the bedside staff in determining whether changes in the EEG are ictal in origin. Because such software was not developed with the ICU patient in mind, it may have difficulty with either the EEG consequences of many of the drugs administered in the ICU or with some of the unusual rhythmic artifacts that may occur in the unit. Furthermore, seizures in encephalopathic patients tend to have more gradual onsets and offsets and slower frequencies, characteristics that make them more difficult to distinguish from background EEG, particularly the abnormal backgrounds found in these patients. Thus, the output of this software cannot substitute for experienced analysis of the EEG. Other analytic approaches involve data reduction at the bedside using CSA or density spectral array (DSA) displays to aid ICU staff in detecting seizures. In contrast to routine EEG practice, CSA is often able to detect seizures in patients being treated with sedative drugs for SE or to improve their ability to tolerate mechanical ventilation. In such patients, the spectral edge frequency (the frequency envelope that includes 95% of the power in the EEG) is often in the range of 1 to 2 Hz. A seizure may shift the spectral edge frequency to 10 Hz, which causes an obvious change in the activity displayed, thus alerting the staff. One must have access to the raw EEG, however, to confirm that this was actually a seizure rather than an artifact or a transient emergence from sedation.

Because an experienced electroencephalographer is usually not available much of the time in most ICUs, many units rely on a combination of ICU physician and nursing interpretation at the bedside, with later off-line review by an electroencephalographer. The diagnostic accuracy of this approach is unknown. To decrease the time between data acquisition and interpretation, network transmission of the EEG from the ICU to the EEG laboratory, or the EEG reader’s office or home, is very useful. However, this off-line interpretative service only constitutes a retrospective check when therapies have been altered based on the local interpretation. Thus, it is incumbent for the critical care staff to become as proficient in EEG interpretation as they can.

The vast amount of data generated by continuous video-EEG monitoring creates archiving problems. The percentage of the acquired data that contains events of interest is usually small, but the time involved in selecting and archiving these events is often prohibitive. Although the cost to archive the data to DVD media is relatively low, the space eventually consumed to store the DVDs suggests that in routine practice some attempt should be made to preserve only clinical or electroencephalographic events of note. As memory storage capability continues to improve, it may soon become practical to simply store all EEG data on large servers.

The electroencephalographic appearance of SE varies with the type of SE and its duration. Unfortunately, attempts to classify the types of SE have not yet yielded a consensus. Table 1 presents one classification, which will be used in this chapter.