ORIGINAL RESEARCH
Automated Auditory Mismatch Negativity Paradigm Improves Coma Prognostic Accuracy After Cardiac Arrest and Therapeutic Hypothermia Andrea O. Rossetti,* Athina Tzovara,†‡ Micah M. Murray,*†‡ Marzia De Lucia,†‡ and Mauro Oddo§
Purpose: EEG and somatosensory evoked potential are highly predictive of poor outcome after cardiac arrest; their accuracy for good recovery is however low. We evaluated whether addition of an automated mismatch negativity–based auditory discrimination paradigm (ADP) to EEG and somatosensory evoked potential improves prediction of awakening. Methods: EEG and ADP were prospectively recorded in 30 adults during therapeutic hypothermia and in normothermia. We studied the progression of auditory discrimination on single-trial multivariate analyses from therapeutic hypothermia to normothermia, and its correlation to outcome at 3 months, assessed with cerebral performance categories. Results: At 3 months, 18 of 30 patients (60%) survived; 5 had severe neurologic impairment (cerebral performance categories ¼ 3) and 13 had good recovery (cerebral performance categories ¼ 1–2). All 10 subjects showing improvements of auditory discrimination from therapeutic hypothermia to normothermia regained consciousness: ADP was 100% predictive for awakening. The addition of ADP significantly improved mortality prediction (area under the curve, 0.77 for standard model including clinical examination, EEG, somatosensory evoked potential, versus 0.86 after adding ADP, P ¼ 0.02). Conclusions: This automated ADP significantly improves early coma prognostic accuracy after cardiac arrest and therapeutic hypothermia. The progression of auditory discrimination is strongly predictive of favorable recovery and appears complementary to existing prognosticators of poor outcome. Before routine implementation, validation on larger cohorts is warranted. Key Words: Hypoxic-ischemic encephalopathy, Cognitive evoked potential, Outcome, Prognosis. (J Clin Neurophysiol 2014;31: 356–361)
H
ypoxic-ischemic encephalopathy is the most frequent cause of coma in intensive care units, with a rising prevalence over the last several years (Weiss et al., 2012). Outcome prognostication represents a frequent challenge for clinicians (Wijdicks et al., 2006; Zandbergen et al., 2006). This aspect has been increasingly investigated following routine implementation of therapeutic
From the *Department of Clinical Neurosciences, CHUV and University of Lausanne, Lausanne, Switzerland; †EEG Brain Mapping Core, Center for Biomedical Imaging (CIBM), CHUV and University of Lausanne, Lausanne, Switzerland; Departments of ‡Radiology and §Intensive Care Medicine, CHUV and University of Lausanne, Lausanne, Switzerland. The Swiss National Science Foundation provides financial support to A.O. Rossetti (CR32I3_143780), M.M. Murray (320030–149982), M.D. Lucia (K-33K1_122518/1), and M. Oddo (320030_138191). Address correspondence and reprint requests to Andrea O. Rossetti, MD, Service de Neurologie, CHUV-BH07, CH-1011 Lausanne, Switzerland; e-mail: [email protected]. Copyright Ó 2014 by the American Clinical Neurophysiology Society
ISSN: 0736-0258/14/3104-0356
356
hypothermia (TH), using standard validated variables such as clinical evaluation after rewarming (especially brainstem reflexes, early myoclonus), EEG (particularly its background activity and reactivity), and early cortical responses of somatosensory evoked potentials (SSEPs) (Al Thenayan et al., 2008; Bouwes et al., 2012; Fugate et al., 2010; Rossetti et al., 2010a; Samaniego et al., 2011b). These items are reliable for predicting poor outcome, but their accuracy in predicting awakening and good functional recovery has been reported to be much lower. In this context, the auditory mismatch negativity (MMN), a differential auditory evoked potential between responses to infrequent deviant sounds embedded within a series of frequent standard sounds (Fischer et al., 2006), represents an emerging promising tool. For example, in a meta-analysis, the MMN has been shown to better predict awakening than the N100 component (Daltrozzo et al., 2007), and it has been shown to be preserved in a minority of patients in vegetative or minimally conscious state several months to years after the initial insult (Fischer et al., 2010). However, the predictive value of the MMN has not been assessed consistently during the very early stage of coma, and to the best of our knowledge, it was never analyzed in patients treated with TH. Recent findings from our group have shown that very early after cardiac arrest (CA) under TH, multivariate single-trial EEG decoding can be applied in a semiautomated manner to MMN data collected from a clinical EEG montage in comatose patients (Tzovara et al., 2013). The aim of the present study was to examine if the addition of this MMN-based auditory discrimination paradigm (ADP) to the standard prognostic tools improves early outcome prediction in postanoxic coma.
METHODS Patients and General Management Between December 2009 and July 2011, we prospectively enrolled in our ongoing registry 95 consecutive comatose adults admitted to the Department of Intensive Care Medicine of our center after CA and treated according to recent recommendations (American Heart Association, 2005) and our protocol (Oddo et al., 2008). Patients underwent mild TH to 338C for 24 hours followed by passive rewarming. Midazolam (0.1 mg$kg21$hour21) and fentanyl (1.5 mg$kg21$hour21) were administered for sedation–analgesia, and vecuronium (0.1 mg/kg boluses) was administered for shivering. Subjects without any sign of EEG activity (isoelectric recording) after reaching a body temperature of at least 368C were excluded. We only considered for further analysis those who were investigated during the first 24 to 48 hours with the ADP paradigm in addition to the standard clinical evaluation (see below). This study was approved by our institutional review board.
Journal of Clinical Neurophysiology Volume 31, Number 4, August 2014
Journal of Clinical Neurophysiology Volume 31, Number 4, August 2014
Standard Prognostic Tools All clinical variables were collected prospectively. Continuous video-EEGs (VIASYS NeuroCare, Madison, WI) were acquired from 21 electrodes positioned according to the international 10 to 20 system both during TH and after rewarming to a temperature .358C and off sedation (apart from patients requiring at times low-dose benzodiazepines for epileptiform EEGs), at least 36 hours after admission. Continuous recordings lasted for at least 60 minutes; 12 patients were recorded for more than 24 hours until rewarming. EEG was sampled at 250 Hz and was referenced according to a bipolar montage; impedances were kept below 10 kU. The same EEG system was also used for the ADP, but at a sampling of 1024 Hz and using FPz as the reference electrode. Background reactivity was tested at bedside, as detailed previously, using repetitive auditory, visual, and nociceptive stimuli at least 6 hours after CA (Rossetti et al., 2010a, 2010b). EEG recordings were visually interpreted by two EEG-certified neurologists, and findings were categorized for the presence or absence of three variables: (1) EEG background reactivity, defined as an activity $ 10 mV (regardless of frequency), and any clear and reproducible change in amplitude or frequency on stimulation, excluding “stimulusinduced rhythmic, periodic, or irritative discharges” and muscle artifacts; (2) spontaneous discontinuous pattern, defined as an EEG background interrupted by flat periods; (3) epileptiform activity, defined as any periodic or rhythmic spikes, sharp waves, spike waves, or rhythmic waves evolving in amplitude, frequency, or field (Rossetti et al., 2010a, 2010b, 2012). Patients with clinical or EEG evidence of epileptiform activity were treated with nonsedating intravenous antiepileptic drugs (valproate, levetiracetam), started as soon as possible (even during TH) and maintained for at least 24 hours. Repeated neurologic examination and median-nerve SSEP were performed after rewarming. As illustrated previously (Rossetti et al., 2010a), brainstem reflexes (pupillary, oculocephalic, corneal) were categorized as all present versus one or more absent, and myoclonus occurrence was recorded.
Auditory Evoked Potentials (AEPs) All patients were not known for preexisting severe auditory impairments. EEG recordings for the ADP were performed during TH and were repeated after rewarming, under normothermic (NT) conditions, usually immediately after the clinical examination. We used an MMN paradigm involving a series of standard sounds (70% of trials) intermixed with deviant sounds, in terms of duration, location, or pitch (each occurring in 10% of trials, after Todd et al. 2008). Pure sinusoidal tones (16-bit stereo, sampled at 44.1 kHz) were presented through earphones, with a constant interstimulus interval of 700 milliseconds. Standard sounds were tones of 1,000 Hz, 100-millisecond duration, and 0 ms interaural time difference. Pitch deviants were delivered at 1,200 Hz, duration deviants lasted 150 milliseconds, and location deviants had an interaural time difference of 700 ms (left ear leading). All other features were otherwise identical to standard sounds. At stimulus onset and offset, we applied a 10-millisecond linear amplitude envelope to avoid clicks. In total, 3 blocks of trials were presented, each consisting of 500 sounds.
Multivariate Decoding of AEPs For ADP analysis, we re-referenced EEG data to a common average reference, with a band-pass filter of 0.1 to 40 Hz, extracting peri-stimulus trials ranging 50 milliseconds before the sound onset up to 500 milliseconds after the sound onset. Trials containing Copyright Ó 2014 by the American Clinical Neurophysiology Society
MMN in Postanoxic Coma
artifacts were rejected on a criterion of 6100 mV, applied offline on all electrodes. Noisy electrodes were interpolated using threedimensional splines (Perrin et al., 1987). This preprocessing was performed using Cartool (Brunet et al., 2011). Using a multivariate decoding algorithm optimized for each patient (Tzovara et al., 2013), we assessed the presence or absence of auditory discrimination at the single EEG response level. This algorithm provides a data-driven way to quantify the difference in responses to standard versus deviant sounds, by modeling the configuration of voltage topographies at the scalp separately for each patient and recording (i.e., during TH and during NT) using one part of the trials (training data set). The computed models are then used to decode sound categories on another set of trials (test data set). The advantages of this algorithm are that it can be applied automatically, it provides a quantitative measure of EEG responses, and it takes into account responses from all electrodes, thereby preventing misinterpretations of the EEG results inherent to analyses of amplitude modulations at single electrode traces (Murray et al., 2008; Tzovara et al., 2012 for discussion). Our previous results (Tzovara et al., 2013) showed that during early coma, information on outcome is not provided by the presence of an MMN response at single electrodes but rather by the progression of discrimination performance from TH to NT. Specifically, for each patient and EEG recording, we quantified auditory discrimination by decoding single-trial AEPs to standard versus each type of deviant sounds separately (three comparisons) and then considered the average decoding performance across the three comparisons. We subtracted the average decoding performance during NT from that obtained during TH, and we considered a positive difference as an improvement in auditory discrimination. These analyses were performed blindly to patient outcome and to clinical assessment and were not available to clinicians at any time during clinical or other electrophysiological tests.
Decisions on Intensive Care Withdrawal and Outcome Assessment The involved physicians had full access to the clinical examination, EEGs, and SSEP, but hypothermic EEG findings, serum neuron-specific enolase, and auditory discrimination results were not used for withdrawal of intensive care support. Our approach (Rossetti et al., 2010a) prescribes that intensive care support is withdrawn if two or more of the following are present: incomplete recovery of brainstem reflexes, early myoclonus, unreactive normothermic background EEG, and bilaterally absent early cortical SSEPs. In all surviving patients, functional outcome was assessed at 3 months through a telephone interview and categorized according to the Glasgow-Pittsburgh cerebral performance categories (CPC) (1 ¼ good recovery, 2 ¼ moderate disability, 3 ¼ severe disability with dependency for daily life activity, 4 ¼ vegetative state, and 5 ¼ death; Booth et al., 2004); it was dichotomized as survival (CPC 1–3) versus vegetative state or death (CPC 4–5).
Data Analysis Comparisons between potential predictors and outcome were performed using t-tests or x2 tests as appropriate. Predictive values and accuracy were estimated using binomial confidence intervals. The predictive performance of the model detailed above with four items (i.e., brainstem reflexes, early myoclonus, EEG reactivity, cortical SSEP) was compared with that of the same model with the addition of the MMN paradigm (i.e., five items). The respective 357
Journal of Clinical Neurophysiology Volume 31, Number 4, August 2014
A. O. Rossetti et al.
areas under the receiver operating characteristic curves were compared with a nonparametric approach. Significance threshold was set at P , 0.05. Calculations were performed with the Stata software, version 9 (Stata Corp LP, College Station, TX).
RESULTS Of the 95 patients admitted during the study period, 30 (32%) underwent the MMN paradigm. Compared with the 65 excluded subjects, the studied cohort did not show any statistical difference regarding age (62.6 6 13.3 versus 61.8 6 12.7 years, P ¼ 0.79, ttest), female gender (14/65 vs. 10/30, P ¼ 0.22, x2 test), prevalence of patients with ventricular fibrillation as initial rhythm (39/65 vs. 10/30, P ¼ 0.53, x2 test), time to return of spontaneous circulation (ROSC) (24.3 6 14.5 vs. 20.1 6 13.1 minutes, P ¼ 0.18, t-test), time from CA and recording of the first EEG during TH (15.7 6 5.8 vs. 15.1 6 5.3 hours, P ¼ 0.67, t-test), and mortality at 3 months (31/ 65 vs. 12/30, P ¼ 0.49, x2 test). No major preexisting or acute focal brain lesion was found on brain computed tomographies performed within the first week. Figure 1 illustrates procedures, results of EEG and auditory discrimination progression, and clinical outcome. Of the 30 patients undergoing ADP, 18 (60%) survived at 3 months: 5 (17%) with severe neurologic impairment (CPC ¼ 3), and 13 (43%) with a good outcome (CPC ¼ 1–2). The principal demographic and clinical characteristics of these 30 patients, stratified by outcome, are summarized in Table 1. Of note, a relatively high proportion of patients with nonventricular fibrillation rhythms (asystole or pulseless electrical activity) survived. Standard prognostic tools for the assessment of postanoxic coma prognosis, with the addition of the ADP, are illustrated in Table 2. Cortical SSEPs were initially absent in one survivor (CPC ¼ 3). The highest predictive accuracy was found for EEG background reactivity (both during TH and NT), followed by the progression of auditory discrimination. An example of AEPs
recorded during NT for two exemplar patients is shown in Fig. 2 for qualitative purposes. Significant differences in responses to standard versus deviant sounds (time-point by time-point t-test across trials, P , 0.05) were identified at frontal electrodes for patients irrespective of their outcome (see Fig. 2A for a patient who later awoke, and Fig. 2B for one who did not). As observed previously (Tzovara et al., 2013), the presence of ADP alone was not predictive of the patients’ outcomes. By contrast, the evolution of auditory discrimination from TH to NT had a positive predictive value of 100% for survival: all patients (n ¼ 10) with improvement in auditory discrimination awoke and survived. Among these 10 patients, 7 reached a good functional outcome, and 3 a CPC ¼ 3 at 3 months. Of note, all also had hypothermic and normothermic EEGs reactive to stimuli (and with no epileptiform transients in hypothermia). Discontinuous EEG during hypothermia was not correlated with progression of auditory discrimination. Midazolam was administered at 0.1 to 0.2 mg$kg21$hour21 during the NT recording to 2 patients; in both, cortical SSEPs were observed bilaterally. One was still on benzodiazepines because the TH EEG was initially erroneously interpreted as epileptiform. ADP improved and he survived. The other subject belonged to the five patients having epileptiform EEG activity during TH (in all, ADP showed lack of improvement; all died); he was the only one with preserved SSEPs. Figure 3 shows the predictive performance of the four-item multimodal algorithm compared with the same algorithm with the addition of the ADP; the latter led to a significantly improved performance (P ¼ 0.02). In particular, 7 of 12 nonsurvivors were identified through the 4-item multimodal algorithm to predict unfavorable outcome, while the ADP predicted 10 of 18 survivors.
DISCUSSION
This study shows for the first time that assessing auditory discrimination during an ADP increases prognostic accuracy in
FIG. 1. Flow chart summarizing the main procedures, results, and outcome. ADP, auditory discrimination paradigm; MMN, mismatch negativity; SSEPs, somatosensory evoked potentials. 358
Copyright Ó 2014 by the American Clinical Neurophysiology Society
Journal of Clinical Neurophysiology Volume 31, Number 4, August 2014
TABLE 1. Clinical Characteristics of 30 Comatose Patients With Hypoxic-Ischemic Encephalopathy According to Mortality at 3 Months Alive (N ¼ 18) Dead (N ¼ 12) Age, mean 6 SD Female gender, n (%) Initial rhythm asystole or pulseless electrical activity, n (%) Time to return of spontaneous circulation, mean 6 SD
61.8 6 12.5 7 (39) 7 (39)
61.7 6 13.5 3 (25) 3 (25)
16.9 6 8.9
24.8 6 17.1
SD, standard deviation.
comatose patients with hypoxic-ischemic encephalopathy treated with TH, owing to the excellent predictive performance for awakening of the progression of auditory discrimination over time (100% positive predictive value). Current standard prognostic tools are powerful in predicting poor prognosis but are much less reliable at predicting return of consciousness. If confirmed in a larger assessment, our data may have potentially important clinical implications, because they could provide clinicians with a novel tool to improve prediction of good recovery, and thus represent an ideal complement to standard prognostic tools. Because hypothermia or the related pharmacological sedation have been shown to potentially bias several clinical (Al Thenayan et al., 2008; Bouwes et al., 2012; Rossetti et al., 2010a; Samaniego et al., 2011b), serological (Daubin et al., 2011; Fugate et al., 2010; Steffen et al., 2010), and electrophysiological (Leithner et al., 2010) predictors, refined multimodal approaches are increasingly used. However, all predictors currently applied in clinical practice are relatively reliable for foreseeing death or vegetative state, but not at identifying patients who will certainly wake. For example, the bilateral absence of early cortical SSEP is one of the most robust parameters to predict poor outcome. However, if SSEPs are present, only about two thirds of patients will wake (Rossetti et al., 2010a). Some years ago, the role of event-related potentials in the prediction of good outcome was highlighted (Fischer et al., 2006). Nonetheless, those patients were recorded before the TH era and after a mean of 8 days (and up to 56 days) after CA, leaving unclear whether similar results could also be expected during the very early stages of coma. The detection of event-related potential differences
TABLE 2.
MMN in Postanoxic Coma
between standard and deviant sounds, even for nonsurvivors, shows that the presence or absence of an MMN per se is not directly predictive of outcome during early coma; indeed, both patients in Fig. 2 showed a significant difference in response to standard versus duration deviant sounds at the single electrode level, corresponding to an MMN component. The evolution of auditory discrimination from TH to NT was instead informative of the chances of return of consciousness; a discrepancy that is likely due to the early time of the recording and the TH treatment. Furthermore, standard MMN analysis requires the detection of a robust N1 component of the auditory evoked potential, resulting in some cases in the exclusions of up to 33% of patients from the final outcome analysis (Fischer et al., 1999). These aspects probably explain why the MMN has been very scarcely reported as a prognostic tool in postanoxic subjects treated with TH (Suppiej et al., 2009). Here, we examined the value of an automated multivariate decoding algorithm that allows inclusion of all patient data. Our results are encouraging with respect to the introduction of an MMN-based ADP within the standard clinical tests at an early coma stage. It is important to mention that our analysis does not require a priori criteria of inclusion for assessing auditory discrimination, and can be applied automatically to any patient within 72 hours from CA. Patients are investigated with the ADP during clinical EEG, using the same set of electrodes and recording machine. The added value of the ADP in postanoxic coma prognosis, related to the positive predictive value of 100% for awakening beyond vegetative state, was quantified here as a gain of at least 10% of the predictive performance, representing a marked improvement in clinical terms, as shown by the receiver operating characteristic areas of 85%. Interestingly, EEG reactivity to stimuli (mostly nociceptive) seems a necessarydbut not sufficientdprerequisite to have an improvement in auditory discrimination; this may relate to the fact that EEG reactivity to pain, which can be observed also in patients in a vegetative state, likely involves more limited brain connections than auditory discrimination assayed by the ADP. One patient initially showed bilaterally absent SSEP, but he awoke and reached a CPC ¼ 3 at 3 months. As all other prognosticators (hypothermic and normothermic EEG showing reactivity, early recovery of brainstem reflexes, absence of myoclonus, neuron-specific enolase (NSE) ,20 mmol/L) were not congruent with this SSEP finding, we repeated the recording after 48 hours and found SSEP “recovery.” This underscores the importance of
Predictive Values of Each Prognostic Tool Among the 30 Patients Alive (N ¼ 18), n (%)
Incomplete recovery of brainstem reflexes* Early myoclonus Hypothermic EEG nonreactive Normothermic EEG nonreactive Cortical SSEP bilaterally absent Improvement in auditory discrimination
4 (22) 0 0 0 1 (6) 10 (55)
Dead (N ¼ 12), n (%) 6 1 7 7 4
(50) (8) (58) (58) (33) 0
PV for Awakening (95% CI) 0.70 0.62 0.78 0.78 0.68 1.0†
(0.46–0.88) (0.42–0.79) (0.56–0.93) (0.56–0.93) (0.46–0.85) (0.69–1.0)
PV for Mortality (95% CI) 0.60 1.0† 1.0† 1.0† 0.8 0.60
(0.26–0.88) (0.02–1.0) (0.59–1.00) (0.59–1.00) (0.28–0.99) (0.36–0.81)
Accuracy (95% CI) 0.67 0.63 0.83 0.83 0.70 0.73
(0.47–0.83) (0.44–0.80) (0.65–0.94) (0.65–0.94) (0.51–0.85) (0.54–0.88)
CI, confidence interval; PV, predictive value; SSEP, somatosensory evoked potentials. *Pupillary, oculocephalic, corneal. †Absolute prediction.
Copyright Ó 2014 by the American Clinical Neurophysiology Society
359
A. O. Rossetti et al.
Journal of Clinical Neurophysiology Volume 31, Number 4, August 2014
FIG. 2. Average AEPs in response to standard (solid gray lines) and duration deviant sounds (dashed gray lines) for an exemplar patient who later awoke (A) and one who did not (B) under normothermia. The difference in event-related potential responses to the duration deviant minus standard sounds at a frontal electrode (Fz) is indicated by the red line. Periods over which this difference was statistically significant based on a time-point by time-point t-test are marked on the x-axis in red. Both patients had such a period around 220 milliseconds after the sound onset, corresponding to a mismatch negativity component.
multimodal assessments and the dangers of interrupting supportive care in the presence of only one bad prognostic sign (Oddo and Rossetti, 2011; Samaniego et al., 2011a); similar cases have been reported (Bender et al., 2012; Leithner et al., 2010). This study has some limitations. First, the sample size is relatively small; therefore, our results should be regarded as preliminary. Second, it was conducted at a single center. While this implies some caution for the generalization of the results, the internal validity is strengthened. Third, the timing of EEG recordings (and therefore the ADP paradigm) was not homogenous. This heterogeneity reflects clinical practice, and all recording pairs were nonetheless completed within 72 hours after CA. Fourth, benzodiazepines were still administered in 2 of 30 patients during NT recording; in view of the convergent results of the other prognostic
parameters in the patient who died (early myoclonus, nonreactive hypothermic EEG with a “seizure-suppression” pattern, lack of awakening on subsequent weaning of the sedation), it appears unlikely that the use of benzodiazepines during NT biased the results. An important strength of our analysis is represented by the fact that ADP results were not available at the time of decision on interruption of care in the intensive care unit, and auditory discrimination performance was analyzed masked to patients’ outcomes. Before this algorithm may be routinely used in clinical practice, a larger cohort should be assessed: in our analysis, improvement in the ADP was observed in 10 of 30 patients. To obtain a lower limit of the 95% confidence interval of the predictive value at 0.9, and assuming a similar proportion of improvement in auditory discrimination, a cohort of approximately 120 patients should be investigated.
ACKNOWLEDGMENTS The authors thank Peter Kaplan, MD, for valuable discussions at a preliminary stage of the study, and Malin Maeder-Ingvar, MD, Christine Stähli, RN, the EEG fellows and technicians, and the ICU fellows, for their help in data collection. REFERENCES
FIG. 3. Receiver operating characteristic (ROC) curves for prediction of survival. Outcome prediction with model 1 (blue line) includes standard prognostic tools (normothermic EEG reactivity, brainstem reflexes, early myoclonus, somatosensory evoked potentials) versus model 2, which adds mismatch negativity-P data to model 1; P ¼ 0.02 for comparison of model 1 with model 2. 360
American Heart Association. 2005 American Heart Association Guidelines for Cardiopulmonary Resuscitation and Emergency Cardiovascular Care. Circulation 2005;112:IV1–IV203. Al Thenayan E, Savard M, Sharpe M, et al. Predictors of poor neurologic outcome after induced mild hypothermia following cardiac arrest. Neurology 2008;71:1535–1537. Bender A, Howell K, Frey M, et al. Bilateral loss of cortical SSEP responses is compatible with good outcome after cardiac arrest. J Neurol 2012;259:2481– 2483. Booth CM, Boone RH, Tomlinson G, Detsky AS. Is this patient dead, vegetative, or severely neurologically impaired? assessing outcome for comatose survivors of cardiac arrest. JAMA 2004;291:870–879. Bouwes A, Binnekade JM, Kuiper MA, et al. Prognosis of coma after therapeutic hypothermia: a prospective cohort study. Ann Neurol 2012;71:206–212. Brunet D, Murray MM, Michel CM. Spatiotemporal analysis of multichannel EEG: CARTOOL. Comput Intell Neurosci 2011;2011:813870. Daltrozzo J, Wioland N, Mutschler V, Kotchoubey B. Predicting coma and other low responsive patients outcome using event-related brain potentials: a metaanalysis. Clin Neurophysiol 2007;118:606–614.
Copyright Ó 2014 by the American Clinical Neurophysiology Society
Journal of Clinical Neurophysiology Volume 31, Number 4, August 2014
Daubin C, Quentin C, Allouche S, et al. Serum neuron-specific enolase as predictor of outcome in comatose cardiac-arrest survivors: a prospective cohort study. BMC Cardiovasc Disord 2011;11:48. Fischer C, Morlet D, Bouchet P, et al. Mismatch negativity and late auditory evoked potentials in comatose patients. Clin Neurophysiol 1999;110:1601– 1610. Fischer C, Luaute J, Nemoz C, et al. Improved prediction of awakening or nonawakening from severe anoxic coma using tree-based classification analysis. Crit Care Med 2006;34:1520–1524. Fischer C, Luaute J, Morlet D. Event-related potentials (MMN and novelty P3) in permanent vegetative or minimally conscious states. Clin Neurophysiol 2010;121:1032–1042. Fugate JE, Wijdicks EF, Mandrekar J, et al. Predictors of neurologic outcome in hypothermia after cardiac arrest. Ann Neurol 2010;68:907–914. Leithner C, Ploner CJ, Hasper D, Storm C. Does hypothermia influence the predictive value of bilateral absent N20 after cardiac arrest? Neurology 2010;74:965–969. Murray MM, Brunet D, Michel CM. Topographic ERP analyses: a step-by-step tutorial review. Brain Topogr 2008;20:249–264. Oddo M, Rossetti AO. Predicting neurological outcome after cardiac arrest. Curr Opin Crit Care 2011;17:254–259. Oddo M, Ribordy V, Feihl F, et al. Early predictors of outcome in comatose survivors of ventricular fibrillation and non-ventricular fibrillation cardiac arrest treated with hypothermia: a prospective study. Crit Care Med 2008;36:2296–2301. Perrin F, Pernier J, Bertrand O, et al. Mapping of scalp potentials by surface spline interpolation. Electroencephalogr Clin Neurophysiol 1987;66:75–81. Rossetti AO, Oddo M, Logroscino G, Kaplan PW. Prognostication after cardiac arrest and hypothermia: a prospective study. Ann Neurol 2010a;67:301– 307. Rossetti AO, Urbano LA, Delodder F, et al. Prognostic value of continuous EEG monitoring during therapeutic hypothermia after cardiac arrest. Crit Care 2010b;14:R173.
Copyright Ó 2014 by the American Clinical Neurophysiology Society
MMN in Postanoxic Coma
Rossetti AO, Carrera E, Oddo M. Early EEG correlates of neuronal injury after brain anoxia. Neurology 2012;78:796–802. Samaniego EA, Persoon S, Wijman CA. Prognosis after cardiac arrest and hypothermia: a new paradigm. Curr Neurol Neurosci Rep 2011a;11:111–119. Samaniego EA, Mlynash M, Caulfield AF, et al. Sedation confounds outcome prediction in cardiac arrest survivors treated with hypothermia. Neurocrit Care 2011b;15:113–119. Steffen IG, Hasper D, Ploner CJ, et al. Mild therapeutic hypothermia alters neuron specific enolase as an outcome predictor after resuscitation: 97 prospective hypothermia patients compared to 133 historical non-hypothermia patients. Crit Care 2010;14:R69. Suppiej A, Cappellari A, Cogo PE. Prognostic role of somatosensory and auditory evoked potentials in paediatric hypoxic-ischemic encephalopathy managed with hypothermia: an illustrative case. Neurophysiol Clin 2009;39:101–105. Todd J, Michie PT, Schall U, et al. Deviant matters: duration, frequency, and intensity deviants reveal different patterns of mismatch negativity reduction in early and late schizophrenia. Biol Psychiatry 2008;63:58–64. Tzovara A, Murray MM, Plomp P, et al. Decoding stimulus-related information from single-trial EEG responses based on voltage topographies. Pattern Recognition 2012;45:2109–2122. Tzovara A, Rossetti AO, Spierer L, et al. Progression of auditory discrimination based on neural decoding predicts awakening from coma. Brain 2013;136:81–89. Weiss N, Regard L, Vidal C, et al. Causes of coma and their evolution in the medical intensive care unit. J Neurol 2012;259:1474–1477. Wijdicks EF, Hijdra A, Young GB, et al. Practice parameter: prediction of outcome in comatose survivors after cardiopulmonary resuscitation (an evidence-based review): report of the Quality Standards Subcommittee of the American Academy of Neurology. Neurology 2006;67:203–210. Zandbergen EG, Hijdra A, Koelman JH, et al. Prediction of poor outcome within the first 3 days of postanoxic coma. Neurology 2006;66:62–68.
361
Automated Auditory Mismatch Negativity Paradigm Improves Coma Prognostic Accuracy After Cardiac Arrest and Therapeutic Hypothermia Andrea O. Rossetti,* Athina Tzovara,†‡ Micah M. Murray,*†‡ Marzia De Lucia,†‡ and Mauro Oddo§
Purpose: EEG and somatosensory evoked potential are highly predictive of poor outcome after cardiac arrest; their accuracy for good recovery is however low. We evaluated whether addition of an automated mismatch negativity–based auditory discrimination paradigm (ADP) to EEG and somatosensory evoked potential improves prediction of awakening. Methods: EEG and ADP were prospectively recorded in 30 adults during therapeutic hypothermia and in normothermia. We studied the progression of auditory discrimination on single-trial multivariate analyses from therapeutic hypothermia to normothermia, and its correlation to outcome at 3 months, assessed with cerebral performance categories. Results: At 3 months, 18 of 30 patients (60%) survived; 5 had severe neurologic impairment (cerebral performance categories ¼ 3) and 13 had good recovery (cerebral performance categories ¼ 1–2). All 10 subjects showing improvements of auditory discrimination from therapeutic hypothermia to normothermia regained consciousness: ADP was 100% predictive for awakening. The addition of ADP significantly improved mortality prediction (area under the curve, 0.77 for standard model including clinical examination, EEG, somatosensory evoked potential, versus 0.86 after adding ADP, P ¼ 0.02). Conclusions: This automated ADP significantly improves early coma prognostic accuracy after cardiac arrest and therapeutic hypothermia. The progression of auditory discrimination is strongly predictive of favorable recovery and appears complementary to existing prognosticators of poor outcome. Before routine implementation, validation on larger cohorts is warranted. Key Words: Hypoxic-ischemic encephalopathy, Cognitive evoked potential, Outcome, Prognosis. (J Clin Neurophysiol 2014;31: 356–361)
H
ypoxic-ischemic encephalopathy is the most frequent cause of coma in intensive care units, with a rising prevalence over the last several years (Weiss et al., 2012). Outcome prognostication represents a frequent challenge for clinicians (Wijdicks et al., 2006; Zandbergen et al., 2006). This aspect has been increasingly investigated following routine implementation of therapeutic
From the *Department of Clinical Neurosciences, CHUV and University of Lausanne, Lausanne, Switzerland; †EEG Brain Mapping Core, Center for Biomedical Imaging (CIBM), CHUV and University of Lausanne, Lausanne, Switzerland; Departments of ‡Radiology and §Intensive Care Medicine, CHUV and University of Lausanne, Lausanne, Switzerland. The Swiss National Science Foundation provides financial support to A.O. Rossetti (CR32I3_143780), M.M. Murray (320030–149982), M.D. Lucia (K-33K1_122518/1), and M. Oddo (320030_138191). Address correspondence and reprint requests to Andrea O. Rossetti, MD, Service de Neurologie, CHUV-BH07, CH-1011 Lausanne, Switzerland; e-mail: [email protected]. Copyright Ó 2014 by the American Clinical Neurophysiology Society
ISSN: 0736-0258/14/3104-0356
356
hypothermia (TH), using standard validated variables such as clinical evaluation after rewarming (especially brainstem reflexes, early myoclonus), EEG (particularly its background activity and reactivity), and early cortical responses of somatosensory evoked potentials (SSEPs) (Al Thenayan et al., 2008; Bouwes et al., 2012; Fugate et al., 2010; Rossetti et al., 2010a; Samaniego et al., 2011b). These items are reliable for predicting poor outcome, but their accuracy in predicting awakening and good functional recovery has been reported to be much lower. In this context, the auditory mismatch negativity (MMN), a differential auditory evoked potential between responses to infrequent deviant sounds embedded within a series of frequent standard sounds (Fischer et al., 2006), represents an emerging promising tool. For example, in a meta-analysis, the MMN has been shown to better predict awakening than the N100 component (Daltrozzo et al., 2007), and it has been shown to be preserved in a minority of patients in vegetative or minimally conscious state several months to years after the initial insult (Fischer et al., 2010). However, the predictive value of the MMN has not been assessed consistently during the very early stage of coma, and to the best of our knowledge, it was never analyzed in patients treated with TH. Recent findings from our group have shown that very early after cardiac arrest (CA) under TH, multivariate single-trial EEG decoding can be applied in a semiautomated manner to MMN data collected from a clinical EEG montage in comatose patients (Tzovara et al., 2013). The aim of the present study was to examine if the addition of this MMN-based auditory discrimination paradigm (ADP) to the standard prognostic tools improves early outcome prediction in postanoxic coma.
METHODS Patients and General Management Between December 2009 and July 2011, we prospectively enrolled in our ongoing registry 95 consecutive comatose adults admitted to the Department of Intensive Care Medicine of our center after CA and treated according to recent recommendations (American Heart Association, 2005) and our protocol (Oddo et al., 2008). Patients underwent mild TH to 338C for 24 hours followed by passive rewarming. Midazolam (0.1 mg$kg21$hour21) and fentanyl (1.5 mg$kg21$hour21) were administered for sedation–analgesia, and vecuronium (0.1 mg/kg boluses) was administered for shivering. Subjects without any sign of EEG activity (isoelectric recording) after reaching a body temperature of at least 368C were excluded. We only considered for further analysis those who were investigated during the first 24 to 48 hours with the ADP paradigm in addition to the standard clinical evaluation (see below). This study was approved by our institutional review board.
Journal of Clinical Neurophysiology Volume 31, Number 4, August 2014
Journal of Clinical Neurophysiology Volume 31, Number 4, August 2014
Standard Prognostic Tools All clinical variables were collected prospectively. Continuous video-EEGs (VIASYS NeuroCare, Madison, WI) were acquired from 21 electrodes positioned according to the international 10 to 20 system both during TH and after rewarming to a temperature .358C and off sedation (apart from patients requiring at times low-dose benzodiazepines for epileptiform EEGs), at least 36 hours after admission. Continuous recordings lasted for at least 60 minutes; 12 patients were recorded for more than 24 hours until rewarming. EEG was sampled at 250 Hz and was referenced according to a bipolar montage; impedances were kept below 10 kU. The same EEG system was also used for the ADP, but at a sampling of 1024 Hz and using FPz as the reference electrode. Background reactivity was tested at bedside, as detailed previously, using repetitive auditory, visual, and nociceptive stimuli at least 6 hours after CA (Rossetti et al., 2010a, 2010b). EEG recordings were visually interpreted by two EEG-certified neurologists, and findings were categorized for the presence or absence of three variables: (1) EEG background reactivity, defined as an activity $ 10 mV (regardless of frequency), and any clear and reproducible change in amplitude or frequency on stimulation, excluding “stimulusinduced rhythmic, periodic, or irritative discharges” and muscle artifacts; (2) spontaneous discontinuous pattern, defined as an EEG background interrupted by flat periods; (3) epileptiform activity, defined as any periodic or rhythmic spikes, sharp waves, spike waves, or rhythmic waves evolving in amplitude, frequency, or field (Rossetti et al., 2010a, 2010b, 2012). Patients with clinical or EEG evidence of epileptiform activity were treated with nonsedating intravenous antiepileptic drugs (valproate, levetiracetam), started as soon as possible (even during TH) and maintained for at least 24 hours. Repeated neurologic examination and median-nerve SSEP were performed after rewarming. As illustrated previously (Rossetti et al., 2010a), brainstem reflexes (pupillary, oculocephalic, corneal) were categorized as all present versus one or more absent, and myoclonus occurrence was recorded.
Auditory Evoked Potentials (AEPs) All patients were not known for preexisting severe auditory impairments. EEG recordings for the ADP were performed during TH and were repeated after rewarming, under normothermic (NT) conditions, usually immediately after the clinical examination. We used an MMN paradigm involving a series of standard sounds (70% of trials) intermixed with deviant sounds, in terms of duration, location, or pitch (each occurring in 10% of trials, after Todd et al. 2008). Pure sinusoidal tones (16-bit stereo, sampled at 44.1 kHz) were presented through earphones, with a constant interstimulus interval of 700 milliseconds. Standard sounds were tones of 1,000 Hz, 100-millisecond duration, and 0 ms interaural time difference. Pitch deviants were delivered at 1,200 Hz, duration deviants lasted 150 milliseconds, and location deviants had an interaural time difference of 700 ms (left ear leading). All other features were otherwise identical to standard sounds. At stimulus onset and offset, we applied a 10-millisecond linear amplitude envelope to avoid clicks. In total, 3 blocks of trials were presented, each consisting of 500 sounds.
Multivariate Decoding of AEPs For ADP analysis, we re-referenced EEG data to a common average reference, with a band-pass filter of 0.1 to 40 Hz, extracting peri-stimulus trials ranging 50 milliseconds before the sound onset up to 500 milliseconds after the sound onset. Trials containing Copyright Ó 2014 by the American Clinical Neurophysiology Society
MMN in Postanoxic Coma
artifacts were rejected on a criterion of 6100 mV, applied offline on all electrodes. Noisy electrodes were interpolated using threedimensional splines (Perrin et al., 1987). This preprocessing was performed using Cartool (Brunet et al., 2011). Using a multivariate decoding algorithm optimized for each patient (Tzovara et al., 2013), we assessed the presence or absence of auditory discrimination at the single EEG response level. This algorithm provides a data-driven way to quantify the difference in responses to standard versus deviant sounds, by modeling the configuration of voltage topographies at the scalp separately for each patient and recording (i.e., during TH and during NT) using one part of the trials (training data set). The computed models are then used to decode sound categories on another set of trials (test data set). The advantages of this algorithm are that it can be applied automatically, it provides a quantitative measure of EEG responses, and it takes into account responses from all electrodes, thereby preventing misinterpretations of the EEG results inherent to analyses of amplitude modulations at single electrode traces (Murray et al., 2008; Tzovara et al., 2012 for discussion). Our previous results (Tzovara et al., 2013) showed that during early coma, information on outcome is not provided by the presence of an MMN response at single electrodes but rather by the progression of discrimination performance from TH to NT. Specifically, for each patient and EEG recording, we quantified auditory discrimination by decoding single-trial AEPs to standard versus each type of deviant sounds separately (three comparisons) and then considered the average decoding performance across the three comparisons. We subtracted the average decoding performance during NT from that obtained during TH, and we considered a positive difference as an improvement in auditory discrimination. These analyses were performed blindly to patient outcome and to clinical assessment and were not available to clinicians at any time during clinical or other electrophysiological tests.
Decisions on Intensive Care Withdrawal and Outcome Assessment The involved physicians had full access to the clinical examination, EEGs, and SSEP, but hypothermic EEG findings, serum neuron-specific enolase, and auditory discrimination results were not used for withdrawal of intensive care support. Our approach (Rossetti et al., 2010a) prescribes that intensive care support is withdrawn if two or more of the following are present: incomplete recovery of brainstem reflexes, early myoclonus, unreactive normothermic background EEG, and bilaterally absent early cortical SSEPs. In all surviving patients, functional outcome was assessed at 3 months through a telephone interview and categorized according to the Glasgow-Pittsburgh cerebral performance categories (CPC) (1 ¼ good recovery, 2 ¼ moderate disability, 3 ¼ severe disability with dependency for daily life activity, 4 ¼ vegetative state, and 5 ¼ death; Booth et al., 2004); it was dichotomized as survival (CPC 1–3) versus vegetative state or death (CPC 4–5).
Data Analysis Comparisons between potential predictors and outcome were performed using t-tests or x2 tests as appropriate. Predictive values and accuracy were estimated using binomial confidence intervals. The predictive performance of the model detailed above with four items (i.e., brainstem reflexes, early myoclonus, EEG reactivity, cortical SSEP) was compared with that of the same model with the addition of the MMN paradigm (i.e., five items). The respective 357
Journal of Clinical Neurophysiology Volume 31, Number 4, August 2014
A. O. Rossetti et al.
areas under the receiver operating characteristic curves were compared with a nonparametric approach. Significance threshold was set at P , 0.05. Calculations were performed with the Stata software, version 9 (Stata Corp LP, College Station, TX).
RESULTS Of the 95 patients admitted during the study period, 30 (32%) underwent the MMN paradigm. Compared with the 65 excluded subjects, the studied cohort did not show any statistical difference regarding age (62.6 6 13.3 versus 61.8 6 12.7 years, P ¼ 0.79, ttest), female gender (14/65 vs. 10/30, P ¼ 0.22, x2 test), prevalence of patients with ventricular fibrillation as initial rhythm (39/65 vs. 10/30, P ¼ 0.53, x2 test), time to return of spontaneous circulation (ROSC) (24.3 6 14.5 vs. 20.1 6 13.1 minutes, P ¼ 0.18, t-test), time from CA and recording of the first EEG during TH (15.7 6 5.8 vs. 15.1 6 5.3 hours, P ¼ 0.67, t-test), and mortality at 3 months (31/ 65 vs. 12/30, P ¼ 0.49, x2 test). No major preexisting or acute focal brain lesion was found on brain computed tomographies performed within the first week. Figure 1 illustrates procedures, results of EEG and auditory discrimination progression, and clinical outcome. Of the 30 patients undergoing ADP, 18 (60%) survived at 3 months: 5 (17%) with severe neurologic impairment (CPC ¼ 3), and 13 (43%) with a good outcome (CPC ¼ 1–2). The principal demographic and clinical characteristics of these 30 patients, stratified by outcome, are summarized in Table 1. Of note, a relatively high proportion of patients with nonventricular fibrillation rhythms (asystole or pulseless electrical activity) survived. Standard prognostic tools for the assessment of postanoxic coma prognosis, with the addition of the ADP, are illustrated in Table 2. Cortical SSEPs were initially absent in one survivor (CPC ¼ 3). The highest predictive accuracy was found for EEG background reactivity (both during TH and NT), followed by the progression of auditory discrimination. An example of AEPs
recorded during NT for two exemplar patients is shown in Fig. 2 for qualitative purposes. Significant differences in responses to standard versus deviant sounds (time-point by time-point t-test across trials, P , 0.05) were identified at frontal electrodes for patients irrespective of their outcome (see Fig. 2A for a patient who later awoke, and Fig. 2B for one who did not). As observed previously (Tzovara et al., 2013), the presence of ADP alone was not predictive of the patients’ outcomes. By contrast, the evolution of auditory discrimination from TH to NT had a positive predictive value of 100% for survival: all patients (n ¼ 10) with improvement in auditory discrimination awoke and survived. Among these 10 patients, 7 reached a good functional outcome, and 3 a CPC ¼ 3 at 3 months. Of note, all also had hypothermic and normothermic EEGs reactive to stimuli (and with no epileptiform transients in hypothermia). Discontinuous EEG during hypothermia was not correlated with progression of auditory discrimination. Midazolam was administered at 0.1 to 0.2 mg$kg21$hour21 during the NT recording to 2 patients; in both, cortical SSEPs were observed bilaterally. One was still on benzodiazepines because the TH EEG was initially erroneously interpreted as epileptiform. ADP improved and he survived. The other subject belonged to the five patients having epileptiform EEG activity during TH (in all, ADP showed lack of improvement; all died); he was the only one with preserved SSEPs. Figure 3 shows the predictive performance of the four-item multimodal algorithm compared with the same algorithm with the addition of the ADP; the latter led to a significantly improved performance (P ¼ 0.02). In particular, 7 of 12 nonsurvivors were identified through the 4-item multimodal algorithm to predict unfavorable outcome, while the ADP predicted 10 of 18 survivors.
DISCUSSION
This study shows for the first time that assessing auditory discrimination during an ADP increases prognostic accuracy in
FIG. 1. Flow chart summarizing the main procedures, results, and outcome. ADP, auditory discrimination paradigm; MMN, mismatch negativity; SSEPs, somatosensory evoked potentials. 358
Copyright Ó 2014 by the American Clinical Neurophysiology Society
Journal of Clinical Neurophysiology Volume 31, Number 4, August 2014
TABLE 1. Clinical Characteristics of 30 Comatose Patients With Hypoxic-Ischemic Encephalopathy According to Mortality at 3 Months Alive (N ¼ 18) Dead (N ¼ 12) Age, mean 6 SD Female gender, n (%) Initial rhythm asystole or pulseless electrical activity, n (%) Time to return of spontaneous circulation, mean 6 SD
61.8 6 12.5 7 (39) 7 (39)
61.7 6 13.5 3 (25) 3 (25)
16.9 6 8.9
24.8 6 17.1
SD, standard deviation.
comatose patients with hypoxic-ischemic encephalopathy treated with TH, owing to the excellent predictive performance for awakening of the progression of auditory discrimination over time (100% positive predictive value). Current standard prognostic tools are powerful in predicting poor prognosis but are much less reliable at predicting return of consciousness. If confirmed in a larger assessment, our data may have potentially important clinical implications, because they could provide clinicians with a novel tool to improve prediction of good recovery, and thus represent an ideal complement to standard prognostic tools. Because hypothermia or the related pharmacological sedation have been shown to potentially bias several clinical (Al Thenayan et al., 2008; Bouwes et al., 2012; Rossetti et al., 2010a; Samaniego et al., 2011b), serological (Daubin et al., 2011; Fugate et al., 2010; Steffen et al., 2010), and electrophysiological (Leithner et al., 2010) predictors, refined multimodal approaches are increasingly used. However, all predictors currently applied in clinical practice are relatively reliable for foreseeing death or vegetative state, but not at identifying patients who will certainly wake. For example, the bilateral absence of early cortical SSEP is one of the most robust parameters to predict poor outcome. However, if SSEPs are present, only about two thirds of patients will wake (Rossetti et al., 2010a). Some years ago, the role of event-related potentials in the prediction of good outcome was highlighted (Fischer et al., 2006). Nonetheless, those patients were recorded before the TH era and after a mean of 8 days (and up to 56 days) after CA, leaving unclear whether similar results could also be expected during the very early stages of coma. The detection of event-related potential differences
TABLE 2.
MMN in Postanoxic Coma
between standard and deviant sounds, even for nonsurvivors, shows that the presence or absence of an MMN per se is not directly predictive of outcome during early coma; indeed, both patients in Fig. 2 showed a significant difference in response to standard versus duration deviant sounds at the single electrode level, corresponding to an MMN component. The evolution of auditory discrimination from TH to NT was instead informative of the chances of return of consciousness; a discrepancy that is likely due to the early time of the recording and the TH treatment. Furthermore, standard MMN analysis requires the detection of a robust N1 component of the auditory evoked potential, resulting in some cases in the exclusions of up to 33% of patients from the final outcome analysis (Fischer et al., 1999). These aspects probably explain why the MMN has been very scarcely reported as a prognostic tool in postanoxic subjects treated with TH (Suppiej et al., 2009). Here, we examined the value of an automated multivariate decoding algorithm that allows inclusion of all patient data. Our results are encouraging with respect to the introduction of an MMN-based ADP within the standard clinical tests at an early coma stage. It is important to mention that our analysis does not require a priori criteria of inclusion for assessing auditory discrimination, and can be applied automatically to any patient within 72 hours from CA. Patients are investigated with the ADP during clinical EEG, using the same set of electrodes and recording machine. The added value of the ADP in postanoxic coma prognosis, related to the positive predictive value of 100% for awakening beyond vegetative state, was quantified here as a gain of at least 10% of the predictive performance, representing a marked improvement in clinical terms, as shown by the receiver operating characteristic areas of 85%. Interestingly, EEG reactivity to stimuli (mostly nociceptive) seems a necessarydbut not sufficientdprerequisite to have an improvement in auditory discrimination; this may relate to the fact that EEG reactivity to pain, which can be observed also in patients in a vegetative state, likely involves more limited brain connections than auditory discrimination assayed by the ADP. One patient initially showed bilaterally absent SSEP, but he awoke and reached a CPC ¼ 3 at 3 months. As all other prognosticators (hypothermic and normothermic EEG showing reactivity, early recovery of brainstem reflexes, absence of myoclonus, neuron-specific enolase (NSE) ,20 mmol/L) were not congruent with this SSEP finding, we repeated the recording after 48 hours and found SSEP “recovery.” This underscores the importance of
Predictive Values of Each Prognostic Tool Among the 30 Patients Alive (N ¼ 18), n (%)
Incomplete recovery of brainstem reflexes* Early myoclonus Hypothermic EEG nonreactive Normothermic EEG nonreactive Cortical SSEP bilaterally absent Improvement in auditory discrimination
4 (22) 0 0 0 1 (6) 10 (55)
Dead (N ¼ 12), n (%) 6 1 7 7 4
(50) (8) (58) (58) (33) 0
PV for Awakening (95% CI) 0.70 0.62 0.78 0.78 0.68 1.0†
(0.46–0.88) (0.42–0.79) (0.56–0.93) (0.56–0.93) (0.46–0.85) (0.69–1.0)
PV for Mortality (95% CI) 0.60 1.0† 1.0† 1.0† 0.8 0.60
(0.26–0.88) (0.02–1.0) (0.59–1.00) (0.59–1.00) (0.28–0.99) (0.36–0.81)
Accuracy (95% CI) 0.67 0.63 0.83 0.83 0.70 0.73
(0.47–0.83) (0.44–0.80) (0.65–0.94) (0.65–0.94) (0.51–0.85) (0.54–0.88)
CI, confidence interval; PV, predictive value; SSEP, somatosensory evoked potentials. *Pupillary, oculocephalic, corneal. †Absolute prediction.
Copyright Ó 2014 by the American Clinical Neurophysiology Society
359
A. O. Rossetti et al.
Journal of Clinical Neurophysiology Volume 31, Number 4, August 2014
FIG. 2. Average AEPs in response to standard (solid gray lines) and duration deviant sounds (dashed gray lines) for an exemplar patient who later awoke (A) and one who did not (B) under normothermia. The difference in event-related potential responses to the duration deviant minus standard sounds at a frontal electrode (Fz) is indicated by the red line. Periods over which this difference was statistically significant based on a time-point by time-point t-test are marked on the x-axis in red. Both patients had such a period around 220 milliseconds after the sound onset, corresponding to a mismatch negativity component.
multimodal assessments and the dangers of interrupting supportive care in the presence of only one bad prognostic sign (Oddo and Rossetti, 2011; Samaniego et al., 2011a); similar cases have been reported (Bender et al., 2012; Leithner et al., 2010). This study has some limitations. First, the sample size is relatively small; therefore, our results should be regarded as preliminary. Second, it was conducted at a single center. While this implies some caution for the generalization of the results, the internal validity is strengthened. Third, the timing of EEG recordings (and therefore the ADP paradigm) was not homogenous. This heterogeneity reflects clinical practice, and all recording pairs were nonetheless completed within 72 hours after CA. Fourth, benzodiazepines were still administered in 2 of 30 patients during NT recording; in view of the convergent results of the other prognostic
parameters in the patient who died (early myoclonus, nonreactive hypothermic EEG with a “seizure-suppression” pattern, lack of awakening on subsequent weaning of the sedation), it appears unlikely that the use of benzodiazepines during NT biased the results. An important strength of our analysis is represented by the fact that ADP results were not available at the time of decision on interruption of care in the intensive care unit, and auditory discrimination performance was analyzed masked to patients’ outcomes. Before this algorithm may be routinely used in clinical practice, a larger cohort should be assessed: in our analysis, improvement in the ADP was observed in 10 of 30 patients. To obtain a lower limit of the 95% confidence interval of the predictive value at 0.9, and assuming a similar proportion of improvement in auditory discrimination, a cohort of approximately 120 patients should be investigated.
ACKNOWLEDGMENTS The authors thank Peter Kaplan, MD, for valuable discussions at a preliminary stage of the study, and Malin Maeder-Ingvar, MD, Christine Stähli, RN, the EEG fellows and technicians, and the ICU fellows, for their help in data collection. REFERENCES
FIG. 3. Receiver operating characteristic (ROC) curves for prediction of survival. Outcome prediction with model 1 (blue line) includes standard prognostic tools (normothermic EEG reactivity, brainstem reflexes, early myoclonus, somatosensory evoked potentials) versus model 2, which adds mismatch negativity-P data to model 1; P ¼ 0.02 for comparison of model 1 with model 2. 360
American Heart Association. 2005 American Heart Association Guidelines for Cardiopulmonary Resuscitation and Emergency Cardiovascular Care. Circulation 2005;112:IV1–IV203. Al Thenayan E, Savard M, Sharpe M, et al. Predictors of poor neurologic outcome after induced mild hypothermia following cardiac arrest. Neurology 2008;71:1535–1537. Bender A, Howell K, Frey M, et al. Bilateral loss of cortical SSEP responses is compatible with good outcome after cardiac arrest. J Neurol 2012;259:2481– 2483. Booth CM, Boone RH, Tomlinson G, Detsky AS. Is this patient dead, vegetative, or severely neurologically impaired? assessing outcome for comatose survivors of cardiac arrest. JAMA 2004;291:870–879. Bouwes A, Binnekade JM, Kuiper MA, et al. Prognosis of coma after therapeutic hypothermia: a prospective cohort study. Ann Neurol 2012;71:206–212. Brunet D, Murray MM, Michel CM. Spatiotemporal analysis of multichannel EEG: CARTOOL. Comput Intell Neurosci 2011;2011:813870. Daltrozzo J, Wioland N, Mutschler V, Kotchoubey B. Predicting coma and other low responsive patients outcome using event-related brain potentials: a metaanalysis. Clin Neurophysiol 2007;118:606–614.
Copyright Ó 2014 by the American Clinical Neurophysiology Society
Journal of Clinical Neurophysiology Volume 31, Number 4, August 2014
Daubin C, Quentin C, Allouche S, et al. Serum neuron-specific enolase as predictor of outcome in comatose cardiac-arrest survivors: a prospective cohort study. BMC Cardiovasc Disord 2011;11:48. Fischer C, Morlet D, Bouchet P, et al. Mismatch negativity and late auditory evoked potentials in comatose patients. Clin Neurophysiol 1999;110:1601– 1610. Fischer C, Luaute J, Nemoz C, et al. Improved prediction of awakening or nonawakening from severe anoxic coma using tree-based classification analysis. Crit Care Med 2006;34:1520–1524. Fischer C, Luaute J, Morlet D. Event-related potentials (MMN and novelty P3) in permanent vegetative or minimally conscious states. Clin Neurophysiol 2010;121:1032–1042. Fugate JE, Wijdicks EF, Mandrekar J, et al. Predictors of neurologic outcome in hypothermia after cardiac arrest. Ann Neurol 2010;68:907–914. Leithner C, Ploner CJ, Hasper D, Storm C. Does hypothermia influence the predictive value of bilateral absent N20 after cardiac arrest? Neurology 2010;74:965–969. Murray MM, Brunet D, Michel CM. Topographic ERP analyses: a step-by-step tutorial review. Brain Topogr 2008;20:249–264. Oddo M, Rossetti AO. Predicting neurological outcome after cardiac arrest. Curr Opin Crit Care 2011;17:254–259. Oddo M, Ribordy V, Feihl F, et al. Early predictors of outcome in comatose survivors of ventricular fibrillation and non-ventricular fibrillation cardiac arrest treated with hypothermia: a prospective study. Crit Care Med 2008;36:2296–2301. Perrin F, Pernier J, Bertrand O, et al. Mapping of scalp potentials by surface spline interpolation. Electroencephalogr Clin Neurophysiol 1987;66:75–81. Rossetti AO, Oddo M, Logroscino G, Kaplan PW. Prognostication after cardiac arrest and hypothermia: a prospective study. Ann Neurol 2010a;67:301– 307. Rossetti AO, Urbano LA, Delodder F, et al. Prognostic value of continuous EEG monitoring during therapeutic hypothermia after cardiac arrest. Crit Care 2010b;14:R173.
Copyright Ó 2014 by the American Clinical Neurophysiology Society
MMN in Postanoxic Coma
Rossetti AO, Carrera E, Oddo M. Early EEG correlates of neuronal injury after brain anoxia. Neurology 2012;78:796–802. Samaniego EA, Persoon S, Wijman CA. Prognosis after cardiac arrest and hypothermia: a new paradigm. Curr Neurol Neurosci Rep 2011a;11:111–119. Samaniego EA, Mlynash M, Caulfield AF, et al. Sedation confounds outcome prediction in cardiac arrest survivors treated with hypothermia. Neurocrit Care 2011b;15:113–119. Steffen IG, Hasper D, Ploner CJ, et al. Mild therapeutic hypothermia alters neuron specific enolase as an outcome predictor after resuscitation: 97 prospective hypothermia patients compared to 133 historical non-hypothermia patients. Crit Care 2010;14:R69. Suppiej A, Cappellari A, Cogo PE. Prognostic role of somatosensory and auditory evoked potentials in paediatric hypoxic-ischemic encephalopathy managed with hypothermia: an illustrative case. Neurophysiol Clin 2009;39:101–105. Todd J, Michie PT, Schall U, et al. Deviant matters: duration, frequency, and intensity deviants reveal different patterns of mismatch negativity reduction in early and late schizophrenia. Biol Psychiatry 2008;63:58–64. Tzovara A, Murray MM, Plomp P, et al. Decoding stimulus-related information from single-trial EEG responses based on voltage topographies. Pattern Recognition 2012;45:2109–2122. Tzovara A, Rossetti AO, Spierer L, et al. Progression of auditory discrimination based on neural decoding predicts awakening from coma. Brain 2013;136:81–89. Weiss N, Regard L, Vidal C, et al. Causes of coma and their evolution in the medical intensive care unit. J Neurol 2012;259:1474–1477. Wijdicks EF, Hijdra A, Young GB, et al. Practice parameter: prediction of outcome in comatose survivors after cardiopulmonary resuscitation (an evidence-based review): report of the Quality Standards Subcommittee of the American Academy of Neurology. Neurology 2006;67:203–210. Zandbergen EG, Hijdra A, Koelman JH, et al. Prediction of poor outcome within the first 3 days of postanoxic coma. Neurology 2006;66:62–68.
361
