Can We Screen EEGs More Efficiently? Spectrographic Review of EEG Data

Commentary

Continuous EEG studies in the intensive care unit place the epileptologist or clinical neurophysiologist in a very different setting than the traditional epilepsy monitoring unit. The goal is not to provoke and capture seizures, but rather to rapidly and efficiently identify and subsequently prevent seizures. The increasing recognition of the relatively high prevalence of status epilepticus and seizures in the intensive care unit setting has led to increasing utilization of continuous EEG recordings (1, 2). This translates to a significant increase in the number of 24-hour studies that need to be reviewed and screened for clinically significant findings on any given day. While large epilepsy centers may have a significant number of individuals with sufficient expertise to visually screen the EEGs (e.g., attending physicians, epilepsy/neurophysiology fellows, or registered EEG technicians), that is not true for many centers. With ever-increasing attention on increased efficiency and increased quality, methods to screen prolonged EEG recordings more rapidly without sacrificing accuracy are increasingly important.

Spike and seizure detection algorithms are routinely utilized to supplement the visual screening of EEG, and multiple studies support their efficacy, but accuracy limitations have largely precluded their ability to replace visual screening (3). Although quantitative analysis utilizing a compressed spectral array of EEG data is not a new concept (4), it has only relatively recently been readily available in commercial EEG software systems. The detection rate for identifying seizures with compressed spectral analysis (CSA) has been studied in the past by the authors of the present study in a retrospective study of 113 continuous EEGs, which had 39 patients with a total of 1,190 seizures (5). They reported a detection rate of 89% for seizures and even higher detection rates for other epileptiform and nonepileptiform EEG findings. A similar median detection rate of 81.5% was reported in 553 seizures within 27 pediatric ICU EEG records using spectral array analysis (6). Despite the increasing availability and utilization of these quantitative methods, there are few data to address their overall impact on the clinical review process.

In the study by Moura and colleagues, the authors utilize a retrospective cohort of patients to assess not only the accuracy but also the efficiency of the CSA-guided review method. In the supplemental materials, they provide detailed descriptions of the procedure for review utilized in this study. The fellows started with a page-by-page visual analysis for the first 20 to 60 minutes of EEG followed by review using the CSA thereafter. They could review no more than 1 minute beyond (in either direction) the marking period identified utilizing CSA-guided review. The fellows also reviewed these studies with conventional visual review of the raw data. Comparison of these two review strategies performed by the same group of reviewers was the basis for assessing the time for review. The EEG data were also reviewed by one of two experienced attending physicians to provide the “gold standard” with regard to the abnormalities and seizures on the study. This comparison was utilized to assess the accuracy.

Of the 118 continuous EEGs reviewed, 40 (34%) had seizures. A total of 1,192 seizures were identified for the 2,092 hours of EEG reviewed. All patients with seizures were identified using the compressed spectral array review method, and 87.3% of all seizures were identified. The percentage of missed seizures was twice as high (13%) for studies with seizures less than 1 minute. All 19 cases that met their criteria for electrographic status epilepticus (either by duration over 5 minutes or having an hourly interseizure interval of less than 5 minutes) were identified with CSA-guided review. Sensitivities for detecting generalized slowing and periodic epileptiform discharges were 100%. Sensitivities for identifying rhythmic delta activity (97%), focal slowing (99%), and epileptiform discharges (89%) were similarly high. Significant time savings were seen with the CSA-guided review method; the time to review a 24-hour study went from an average of 38 minutes for standard visual review to 8 minutes for CSA-guided review.

Although this is a retrospective study from a single center, the authors demonstrated a significant improvement in efficiency with CSA-guided review with only a limited impact on the accuracy. They employed a five-channel montage of CSA data (left lateral, left parasagittal, right lateral, right parasagittal, and relative hemispheric asymmetry indices) that could be correlated directly to the raw EEG data (viewed on side-by-side monitors). As the authors also note, the optimal quantitative tools for review remain unknown.

The desire to have a set of tools that allow members of the health care team without EEG training to identify clinically significant abnormalities remains elusive. Although it is not as steep as conventional EEG review, there is a significant learning curve to CSA-guided review and, as the authors also point out, correlation with the raw EEG is necessary. Despite the encouraging results shown here, caution should be exercised before applying this strategy; the readers need to have experience correlating the quantitative data with the raw EEG data. These tools should not be viewed as a replacement for the traditional review process, but rather as a way to augment the process. It is often said that we should “work smarter, not harder.” Although this phrase is often misused, this study highlights one area where a smarter, targeted approach to continuous EEG review may save time. Before realizing this goal, further studies to replicate these findings and assess the optimal quantitative tools will be necessary.