Abstract
Standardized EEG Interpretation Accurately Predicts Prognosis after Cardiac Arrest
Westhall E, Rossetti AO, van Rootselaar AF, Wesenberg Kjaer T, Horn J, Ullén S, Friberg H, Nielsen N, Rosén I, Åneman A, Erlinge D, Gasche Y, Hassager C, Hovdenes J, Kjaergaard J, Kuiper M, Pellis T, Stammet P, Wanscher M, Wetterslev J, Wise MP, Cronberg T; TTM-trial investigators. Neurology 2016;86:1482–1490.
OBJECTIVE: To identify reliable predictors of outcome in comatose patients after cardiac arrest using a single routine EEG and standardized interpretation according to the terminology proposed by the American Clinical Neurophysiology Society. METHODS: In this cohort study, 4 EEG specialists, blinded to outcome, evaluated prospectively recorded EEGs in the Target Temperature Management trial (TTM trial) that randomized patients to 33°C vs 36°C. Routine EEG was performed in patients still comatose after rewarming. EEGs were classified into highly malignant (suppression, suppression with periodic discharges, burst-suppression), malignant (periodic or rhythmic patterns, pathological or nonreactive background), and benign EEG (absence of malignant features). Poor outcome was defined as best Cerebral Performance Category score 3–5 until 180 days. RESULTS: Eight TTM sites randomized 202 patients. EEGs were recorded in 103 patients at a median 77 hours after cardiac arrest; 37% had a highly malignant EEG and all had a poor outcome (specificity 100%, sensitivity 50%). Any malignant EEG feature had a low specificity to predict poor prognosis (48%) but if 2 malignant EEG features were present specificity increased to 96% (p < 0.001). Specificity and sensitivity were not significantly affected by targeted temperature or sedation. A benign EEG was found in 1% of the patients with a poor outcome. CONCLUSIONS: Highly malignant EEG after rewarming reliably predicted poor outcome in half of patients without false predictions. An isolated finding of a single malignant feature did not predict poor outcome whereas a benign EEG was highly predictive of a good outcome.
Commentary
With publication of the 2005 American Heart Association guidelines for cardiopulmonary resuscitation, “induced hypothermia protocol” after cardiac arrest has gradually become a standard of care. Specific local guidelines and care pathways were developed to help with implementation and to streamline such protocols. These protocols, based on the results of two randomized clinical trials (1, 2), can be paraphrased as induction of mild hypothermia (32–34°C) for a period of up to 24 hours in any patient who has suffered an out-of-hospital cardiac arrest; slow rewarming to normal temperature occurs thereafter. Continuous EEG monitoring in patients undergoing induced therapeutic hypothermia is also standard, typically initiated at the time of hypothermia induction and continued for up to 24 hours after the completion of rewarming (3, 4). However, the role of EEG in prognostication after a hypoxic-ischemic event is not entirely clear, documented in the 2006 guideline published by the American Academy of Neurology as “generalized suppression to ≤20μV, burst suppression pattern with generalized epileptiform activity, or generalized periodic complexes on a flat background are strongly but not invariably associated with poor outcome,” with level C recommendation of insufficient accuracy (5). This recommendation is based on prospective and retrospective studies but does not take into account the neuroprotective effects of therapeutic hypothermia, as the guideline was published prior to such studies becoming available (5). Some have questioned performing continuous EEG studies in all patients enrolled in therapeutic hypothermia protocols (6); recent retrospective and prospective studies address this issue to provide some needed clarification, but they do not address all of the existing clinical uncertainties.
The study by Westhall et al. has several strengths and adds substantially to our ability to prognosticate in patients subjected to induced therapeutic hypothermia after cardiac arrest. The strengths of this study include prospective design with the EEG readers blinded to the clinical status of the patient, the use of standardized EEG reading terminology with strict definitions of the patterns observed on the collected EEGs, the relatively simple EEG classification scheme (“highly malignant,” “malignant,” “benign”), standardized assessment of reactivity, and an assessment of outcome at 180 days after the initial event. However, weaknesses of this study also need to be recognized: 1) Lack of blinding of the care providers to the EEG results, possibly leading to early or premature decisions to withdraw care (“self-fulfilling prophecy”); 2) availability of the EEG only in a small subsample from the original study (103/950); 3) short duration of the EEG (routine 20 minutes EEG was performed as opposed to continuous EEG monitoring typically practiced in major U.S. medical centers); and 4) the subjective nature of EEG interpretation expressed here as variability in the agreement between experienced reviewers in their interpretation of the EEG features. Their findings are not unexpected—“highly malignant” EEG patterns had 100% sensitivity and 50% specificity for predicting poor outcome. In contrast to the highly malignant patterns, the “benign” EEG patterns predicted good outcome in 93% of patients. Unfortunately, the EEG was not very helpful in the group of patients with intermediate or “malignant” EEG patterns—the group for which we typically need the most help in predicting outcomes. Frequent false positives and false negatives were noted, making prognostication in this group of patients difficult.
So, how do we reconcile the results of the study by Westhall et al. with other data from the literature? Prognostication is frequently the responsibility of a neurologist or neurointensivist who is designated with the task of helping the family and other healthcare providers make sensible decisions regarding the intensity of care and end-of-life decisions. In addition to serial examinations, ancillary tests (EEG, serum neuron-specific enolase, and somatosensory evoked potentials [8, 9]) play an increasing role in the decision-making process and help clinicians and families make these decisions (5, 7). Since the study in question focused on EEG, let's also focus on this single aspect of care in patients who have suffered from cardiac arrest.
One of the earlier studies prospectively combined EEG reactivity, somatosensory evoked potentials, and neurological examination (brain stem reflexes and presence/absence of myoclonus) to determine that unreactive EEG was incompatible with a good neurological outcome at 3 to 6 months and that combining any 2 out of 4 studied features accurately predicted a poor long-term outcome (positive predictive value = 1.00) (8). While these authors collected other EEG variables, they did not enter them into regression to determine their predictive value. A retrospective study rated EEG features from grade 1 (mild) to grade 3 (severe) to show that grades 1 and 3 correlated with outcomes, but grade 3 EEG predicted a poor outcome correctly in only ~80% of patients (3). In general, studies show that malignant EEG features including lack of reactivity, burst suppression, periodic or rhythmic patterns, and seizures carry poor prognosis with very few of these patients achieving meaningful long-term recovery (3, 4).
Thus, the study by Westhall et al. underscores the limitations of EEG in the ICU setting for predicting outcomes. As stated above, it is not terribly difficult for the clinician to predict the outcome of a patient with highly malignant or benign EEG patterns based on the clinical status and EEG features. The difficulty lies in the patients who have some combination of the benign and malignant features. Here, not surprisingly, the study by Westhall et al. falls short—almost half of the participants had a combination of benign and malignant features and, depending on the number of malignant features, the sensitivity and specificity for predicting poor outcomes varied between 76 to 99 and 48 to 96 percent, respectively. So, in the end, the usefulness of EEG as a single tool for predicting outcomes needs to be questioned. It is likely that some algorithm that combines EEG features and other variables needs to be developed and validated for the use in this setting. The study also does not address the question whether continuous EEG needs to be performed in all patients entered in the induced hypothermia protocols—something that will need to be examined in the future. Finally, this study raises a question of how we should translate their findings into practice; since this group of highly trained and experienced readers showed variability in their interpretation of EEGs, how will less experienced and trained providers perform in the community?
In the end, this study adds to the growing body of evidence that some features of EEG can help to predict the outcome of induced hypothermia some of the time. But it also confirms that, in many cases, the EEG features alone are not as useful as we would like them. The question of whether EEG should be ordered indiscriminately for all patients as part of the induced hypothermia protocol remains open.