Early Electroencephalography for Neurologic Prognostication After Cardiac Arrest: More Optimization Before Generalization?*
Studies using continuous EEG have shown that EEG had the best prognostic accuracy at 12 and 24 hours after ROSC, with a normal voltage, continuous EEG devoid of periodic or epileptiform discharges (“benign” EEG) at 12 hours being most often associated with good recovery and suppressed (< 10 μV) with periodic discharges or suppression-burst EEG (“highly malignant” EEG) at 24 hours after ROSC being strongly associated with lack of neurologic recovery (2–6). These recent findings have not been incorporated in available guidelines, which still recommend deferring the EEG to until at least 48 hours after ROSC (1). There is also a strong but imperfect association between lack of EEG reactivity and poor outcome, with a false positive rate (FPR) ranging from 0% to 15%, possibly due to a combination of variability in stimulation protocols and low interobserver agreement (4, 6, 7).
In this issue of Critical Care Medicine, Rossetti et al (8) assessed and compared the prognostic value of EEG background and reactivity assessment, early (during hypothermia) and later (after return to normothermia), in a large bicenter cohort. Similarly to prior studies, they used a standardized classification of EEG patterns based on the American Clinical Neurophysiology Society terminology for EEG in critically ill patients. In line with previous studies (2–6, 9), they found that an early “highly malignant” EEG was strongly associated with mortality, with an FPR of 1.5%. Lack of early EEG reactivity performed similarly. An early “benign” EEG and the presence of early EEG reactivity were both associated with good outcome with an accuracy of greater than 80%. Overall, early EEG assessment was more accurate than late assessment and outperformed all other predictors, including SSEP, presence of myoclonus, and neurologic examination. These results thus further demonstrate that an early EEG, performed during hypothermia and interpreted with standardized criteria, likely represents the best currently available predictor of neurologic outcome in survivors of CA and should be included in prognostication protocols. In principle, the bicenter design of the study also indicates that this standardized interpretation could be easily generalized to other centers. This is mitigated however by the fact that authors, as well as most groups involved in prior studies, have a longstanding experience in the interpretation of EEG of CA survivors. This generalizability thus remains to be formally demonstrated in a larger scale multicenter study, ideally including centers new to the technique.
Another limitation to mention is that most EEG studies were not purposely reviewed by the authors for the study but were retrospectively classified using data from prospective registries covering most, but not all, necessary features to appropriately define the different EEG categories. This may perhaps seem like a trivial issue, especially for non-neurophysiologists, but this may have affected the ability to identify some highly malignant patterns. For instance, the distinction between a suppressed (< 10 μV; highly malignant) and a low-voltage (10–20 μV; not highly malignant) EEG or between a suppression-burst (> 50% of suppressed EEG; highly malignant) and a discontinuous (10–50%; not highly malignant) could not be accurately done in this study, as acknowledged by the authors.