Early Multimodal Outcome Prediction After Cardiac Arrest in Patients Treated With Hypothermia* Mauro Oddo, MD1; Andrea O. Rossetti, MD2
Objectives: Therapeutic hypothermia and pharmacological sedation may influence outcome prediction after cardiac arrest. The use of a multimodal approach, including clinical examination, electroencephalography, somatosensory-evoked potentials, and serum neuron-specific enolase, is recommended; however, no study examined the comparative performance of these predictors or addressed their optimal combination. Design: Prospective cohort study. Setting: Adult ICU of an academic hospital. Patients: One hundred thirty-four consecutive adults treated with therapeutic hypothermia after cardiac arrest. Measurements and Main Results: Variables related to the cardiac arrest (cardiac rhythm, time to return of spontaneous circulation), clinical examination (brainstem reflexes and myoclonus), electroencephalography reactivity during therapeutic hypothermia, somatosensory-evoked potentials, and serum neuron-specific enolase. Models to predict clinical outcome at 3 months (assessed using the Cerebral Performance Categories: 5 = death; 3–5 = poor recovery) were evaluated using ordinal logistic regressions and receiving operator characteristic curves. Seventy-two patients (54%) had a poor outcome (of whom, 62 died), and 62 had a good outcome. Multivariable ordinal logistic regression identified absence of electroencephalography reactivity (p < 0.001), incomplete recovery of brainstem reflexes in normothermia (p = 0.013), and neuron-specific enolase higher than 33 μg/L (p = 0.029), but *See also p. 1535. 1 Department of Intensive Care Medicine, CHUV-Lausanne University Hospital and Faculty of Biology and Medicine, Lausanne, Switzerland. 2 Department of Clinical Neurosciences, CHUV-Lausanne University Hospital and Faculty of Biology and Medicine, Lausanne, Switzerland. Dr. Oddo conceived and designed the study; drafted and revised the manuscript; and analyzed, interpreted, and acquired the data. Dr. Rossetti conceived and designed the study; drafted and revised the manuscript; analyzed, interpreted, and acquired the data; did statistical analysis; and coordinated the study. Dr. Oddo received support for article research from Swiss National Science Foundation (grant 320030_138191). He consulted for Bard Medical Integra Neurosciences and lectured for Bard Medical Integra Neurosciences. His institution received grant support from the Swiss National Science Foundation. Dr. Rossetti and his institution received support from Swiss National Science Foundation (grant CR32I3_143780). For information regarding this article, E-mail: [email protected] Copyright © 2014 by the Society of Critical Care Medicine and Lippincott Williams & Wilkins DOI: 10.1097/CCM.0000000000000211
1340
www.ccmjournal.org
not somatosensory-evoked potentials, as independent predictors of poor outcome. The combination of clinical examination, electroencephalography reactivity, and neuron-specific enolase yielded the best predictive performance (receiving operator characteristic areas: 0.89 for mortality and 0.88 for poor outcome), with 100% positive predictive value. Addition of somatosensory-evoked potentials to this model did not improve prognostic accuracy. Conclusions: Combination of clinical examination, electroencephalography reactivity, and serum neuron-specific enolase offers the best outcome predictive performance for prognostication of early postanoxic coma, whereas somatosensory-evoked potentials do not add any complementary information. Although prognostication of poor outcome seems excellent, future studies are needed to further improve prediction of good prognosis, which still remains inaccurate. (Crit Care Med 2014; 42:1340–1347) Key Words: coma; electroencephalography; neuron-specific enolase; prognosis; somatosensory-evoked potentials
C
oma after cardiac arrest (CA) is a leading cause of admission in ICUs and has a high rate of mortality and morbidity. Therapeutic hypothermia (TH) is increasingly used for neuroprotection in this context (1–3). Several recent studies have shown that TH and the related pharmacological sedation may influence some prognosticators commonly used in this setting as compared with patients not undergoing hypothermia. Particularly, motor response may recover over a longer time frame, and serum biomarkers such as neuron-specific enolase (NSE) can display cutoff levels for poor outcome considerably higher than previously thought (4–10). In contrast, electrophysiological investigations with electroencephalography (EEG), particularly focusing on continuous patterns and background reactivity during and after TH (11–13), and somatosensory-evoked potentials (SSEPs) (7) seem to provide greater prognostic accuracy in this setting. Based on this evidence, current recommendations suggest the use of a multimodal approach to minimize false-positive prediction of poor outcome (14–16), and raise concerns in relying on American Academy of Neurology prognostic guidelines (17) that were assessed before the TH became a standard of care. Despite such recommendations, however, none of the recent studies on this topic by our and other groups (6, 7, 18–21) has June 2014 • Volume 42 • Number 6
Feature Articles
critically evaluated the prognostic performance of the different clinical, electrophysiological, and serological tests, alone or in combination, nor attempted to identify the multimodal algorithm with the highest prognostic accuracy in this setting. We therefore undertook this analysis to identify those variables among baseline demographic data, clinical examination, hypothermic EEG, normothermic SSEP, and NSE, which might offer the best performance in predicting outcome of post-CA coma.
MATERIAL AND METHODS Patients and Setting This prospective cohort study was conducted at the Department of Intensive Care Medicine, CHUV-Lausanne University Hospital, an adult medical-surgical ICU; it was approved by the institutional ethics committee, and waiver of consent was allowed since all examinations were part of standard patient care. The population consisted of patients older than 16 years admitted for coma following out-of-hospital CA between December 2009 and April 2013, treated with TH, and not brain dead within 48 hours. Patients were managed according to our previously published protocol (TH is considered in adults with cardiac and noncardiac etiologies, regardless of the initial rhythm, unless presenting a marked hemodynamic instability, or a patient’s “do-not-resuscitate” directive; about 85% of all patients reaching alive our hospital after CA receive TH) (5, 11), and in line with the American Heart Association guidelines (16). TH of 33°C ± 1°C was induced as soon as possible using ice packs plus ice-cold isotonic solutions, followed by the application of a surface cooling device with computerized automatic temperature control (Arctic Sun 2000 TTM; Bard Medical, Louisville, CO), maintained for 24 hours. Standardized sedation (midazolam, 0.1 mg/kg/hr), analgesia (fentanyl, 1.5 μg/kg/hr), and neuromuscular blockade (rocuronium boluses, 0.6 mg/kg, if shivering) were administered IV during TH and discontinued after passive rewarming at a core temperature above 35°C. Electrophysiological Procedures and Neurological Assessments Video-EEGs (Viasys Neurocare, Madison, WI) were performed during TH, using 21 electrodes according to the international 10–20 system. EEG was continuously recorded for 24–48 hours until July 2010 (20), and afterward intermittently, for 20–30 minutes. Background reactivity was tested at the bedside, at least 6 hours after CA, as previously reported (11, 20), by applying repetitive auditory, visual, and nociceptive stimuli. The same procedure was repeated during the early phase following rewarming, in normothermic conditions, within 72 hours from CA. In case of repeated testing at different time points, the best result was used for analysis. EEG findings were categorized by certified EEG interpreters according to the presence/absence of the following: 1) background reactivity: activity more than or equal to 10 μV (regardless of frequency) and any clear and reproducible change in amplitude or frequency upon Critical Care Medicine
stimulation, excluding “stimulus-induced rhythmic, periodic, or ictal discharges” (22, 23) and muscle artifacts; 2) spontaneous discontinuous pattern: background interrupted by diffusely suppressed periods of at least 10% of the recording (24); and 3) epileptiform activity: any periodic or rhythmic spikes, sharp waves, spike waves, or rhythmic waves (5, 20). At normothermia and off sedation, within 72 hours from CA, patients were repetitively examined by a certified neurologist, and the best patients’ performances were considered. Brainstem reflexes (pupillary, oculocephalic, and corneal) were categorized as all present versus at least one absent; the motor response was not taken into account for this study in view of its reduced accuracy due to TH and related sedation (4–7). Myoclonus was considered if appearing within 24 hours after discontinuation of sedative drugs. Treatment of epileptiform EEG activity and myoclonus consisted mostly in the administration of nonsedating antiepileptic agents (mainly levetiracetam and valproate) and, when necessary, sedative agents (mainly propofol), as previously reported (25). Bilateral SSEPs were recorded during early normothermia; the cortical N20 responses on SSEP were categorized as present or bilaterally absent (no clear negative deflection at 18–25 ms followed by a positive wave on both sides). NSE is released in the serum after CA with a half-life of about 24 hours (26, 27); it was collected 1–3 times within the first 72 hours following the CA. To prevent hemolysis and false-positive test results, samples were handled manually and measured with an automated immunofluorescent assay (Thermo Scientific Brahms NSE Kryptor Immunoassay). In the vast majority of patients, at least one probe was collected at 48 hours, and for the present analysis, we considered the peak value, exploring cutoffs at 33 μg/L (17) and higher, to reach a 100% specificity for poor outcome (7–9). Decisions on Withdrawal of Care In line with our previous study (5), supportive care withdrawal was considered upon the occurrence of two or more of four criteria after more than 72 hours after CA, in normothermia and off sedation: unreactive EEG background in normothermia, treatment-resistant myoclonus, bilateral absence of N20 in SSEP in normothermia, and incomplete return of the three principal brainstem reflexes. This was based on a consensus between intensivists and consultant neurologists involved in patient care. Serum NSE levels and the EEG obtained during TH were not taken into consideration for this decision. Data Collection and Outcome Assessment Demographical and clinical variables were collected prospectively using Utstein’s style recommendations (28). Etiology of CA was dichotomized as cardiac versus noncardiac, and the initial CA rhythm was categorized as ventricular fibrillation (VF) versus non-VF (including asystole and pulseless electrical activity). CA duration was estimated by calculating the time to return of spontaneous circulation and collected at the time of patient admission to the ICU. Neurological outcome at 3 months was prospectively assessed through a semistructured www.ccmjournal.org
1341
Oddo and Rossetti
telephone interview using Glasgow-Pittsburgh Cerebral Performance categories (CPC) (29): CPC 1 indicates full recovery, CPC 2 moderate disability, and CPC 3 severe disability. Subjects in CPC 4 are comatose or in vegetative state and those with CPC 5 died. In accordance to the vast majority of studies on the topic, good neurological recovery was defined as CPC 1 or 2. Statistical Analysis To explore outcome predictors, chi-square test or two-tailed Fisher exact test for categorical variables, and parametric or nonparametric analysis of variance for continuous variables, were used as needed; significant (p < 0.05) variables were entered into a multivariable backward ordinal logistic regression, to identify independent outcome predictors. Each potential prognosticator of poor outcome (CPC 3–5) or mortality (CPC 5) was also assessed with C statistics, using the area under the receiving operator characteristic (ROC) curve in consideration of dichotomous outcomes, applying nonparametric comparisons. We further examined the performance of prognostic models constructed using combinations of the single most robust outcome predictors. The qualitatively best outcome model was assessed for sensitivity, specificity, positive predictive value (PPV), negative predictive value (with binomial 95% CI), and unweighted accuracy, for poor outcome and mortality. Statistical analyses were performed with a STATA software, version 9 (College Station, TX).
RESULTS From December 2009 to April 2013 (40 mo), 183 consecutive adults were treated for coma after CA with TH at our center. All had full recording of historical data, were examined clinically, and had normothermic EEG, while 49 lacked at least one of the following examinations: hypothermic EEG, SSEP, or serum NSE. We, therefore, restricted our analysis to the 134 subjects who received the complete set of early tests (this cohort overlaps with 59 patients who were recently described focusing on EEG and NSE) (11). As compared with the 49 excluded subjects, the 134 patients did not differ in terms of mean age (62.4 ± 12.5 vs 62.7 ± 15.1; p = 0.44, t test), gender (35 of 134 vs 11 of 49 women; p = 0.61, chi-square test), and mortality at 3 months (62 of 139 vs 22 of 49; p = 0.87, chi-square test). Of the 134 studied patients, 29 (21%) had continuous EEG for 16–48 hours, whereas the remaining 105 had routine intermittent EEG studies. Although the EEG was started earlier in patients who had continuous EEG (13.0 ± 5.8 vs 18. ± 5.9 hr in subjects who had intermittent EEG, p < 0.001, t test), the prevalence of patients with a nonreactive EEG background was similar between the two groups (11 of 29 [38%] vs 36 of 105 [34%], p = 0.72, chi-square test), as was mortality at 3 months (13 of 29 [45%] vs 49 of 105 [47%], p = 0.86, chi-square test). Table 1 shows clinical and electrophysiological variables assessed within 72 hours after CA, according to CPC at 3 months. Mortality occurred in 45%, whereas 86% of the survivors reached a good outcome; of note, no patient remained in a vegetative state (CPC 4). Although baseline demographics
Table 1. Clinical and Electrophysiological Characteristics of 134 Patients Treated With Therapeutic Hypothermia After Cardiac Arrest, Stratified for Their Functional Outcome at 3 Months Variable
CPC 1 (40 Patients)
Mean age, yr (± sd)
60.1 (± 12.7) 62.2 (± 10.3) 62.6 (± 11.2) 64.0 (± 13.3)
Female gender
CPC 2 (22 Patients)
CPC 3 (10 Patients)
CPC 5 (62 Patients)
p
Statistical Test
0.555
ANOVA
12
6
2
15
0.896
Fisher
Noncardiac etiology
3
5
1
18
0.043
Fisher
Non–ventricular fibrillation rhythm
6
8
4
28
0.012
Fisher
Time to return of spontaneous circulation, min (± sd)
19.8 (± 11.7) 15.0 (± 9.8)
21.0 (± 11.3) 26.1 (± 15.1)
< 0.01
ANOVA
Incomplete brainstem reflexesa
2
2
3
36
< 0.01
Fisher
Myoclonus
1
1
1
20
< 0.01
Fisher
Hypothermic electroencephalography nonreactive
0
1
0
46
< 0.01
Fisher
Bilaterally absent cortical somatosensory-evoked potentials
0
0
1
27
< 0.01
Fisher
Serum NSE > 33 μg/Lb
5
3
4
37
< 0.01
Fisher
Serum NSE > 75 μg/L
0
0
0
27
< 0.01
Fisher
b
CPC = Cerebral Performance Category, ANOVA = analysis of variance, NSE = neuron-specific enolase. a Pupillary, oculocephalic, corneal. b Peak value at 24–48 hr. CPC 1–2 implies good prognosis; CPC 5 implies death.
1342
www.ccmjournal.org
June 2014 • Volume 42 • Number 6
Feature Articles
Figure 1. Prognostic performance for prediction of 3-month clinical outcome (mortality, left; poor neurological recovery, right), assessed using the receiver operator characteristic (ROC) curves of the model including clinical evaluation (brainstem reflexes, early myoclonus), hypothermic electroencephalography reactivity, and neuron-specific enolase higher than 33 μg/L.
were not discriminatory, acute CA variables were significantly different among the outcome groups; however, the most robust predictors in univariable analyses were clinical examination and electrophysiological data, as well as NSE serum levels (for which a cutoff of 75 μg/L was also assessed, since beyond this threshold no patient awoke in our cohort). Ordinal logistic regression identified as independent predictors: absent hypothermic EEG reactivity (coefficient, 4.36; 95% CI, 2.27–6.46; p < 0.001), incomplete return of brainstem reflexes (coefficient, 1.10; 95% CI, 0.23–1.96; p = 0.013), and serum NSE more than 33 μg/L (coefficient, 1.09; 95% CI, 0.11–2.07; p = 0.029). Predictive performances for mortality (CPC 5) and poor outcome (CPC 3–5) were assessed using C statistics for clinical examination (combining occurrence of incomplete brainstem reflexes and occurrence of myoclonus in a 0–2 score), EEG reactivity, SSEP, and peak serum NSE at different cutoffs. Table 2 summarizes the areas under the ROC curves: EEG and clinical examination stand out for both conditions and (at least
for poor outcome) NSE at the 33 μg/L cutoff seemed better than at 75 μg/L. Several predictive models, using NSE dichotomized at 33 μg/L, were tested (Table 3); for both outcomes, the model with the highest prognostic accuracy included clinical examination, EEG reactivity, and serum NSE. Of note, prognostication was not improved neither by addition of SSEP nor by its use instead of NSE. Figure 1 illustrates the ROC curves of the best predictive models for the two main outcomes, and shows that 100% specificity in forecasting poor outcome and mortality is reached using a cutoff of 0–2 versus 3–4. Table 4 gives the predictive performance of the model with this cutoff: although the PPV (and specificity) is optimal, anticipation of a good clinical outcome is less reliable. Of the 62 deceased patients, nine showed all four poor predictive variables (lack of at least one brainstem reflex, myoclonus, NSE more than 33 μg/L, and nonreactive hypothermic EEG), and 24 had three of them; none of the surviving patients had more than two items (false-positive rate = 0). The
Table 2. Areas Under the Receiving Operator Characteristic Curves for Each Examined Single Functional Outcome Predictor Mortality (CPC 5) Variable
ROC Area
95% CI
Clinical examinationa
0.78
0.71–0.84
Hypothermic electroencephalography nonreactive
0.86
Bilaterally absent cortical somatosensoryevoked potentials
Poor Outcome (CPC 3–5) p
ROC Area
95% CI
0.77
0.70–0.83
0.81–0.92
0.81
0.75–0.87
0.71
0.64–0.77
0.69
0.64–0.75
Serum NSE > 33 μg/L
0.72
0.64–0.79
0.72
0.65–0.79
Serum NSE > 75 μg/L
0.72
0.66–0.78
0.68
0.63–0.74
< 0.001
p
< 0.001
CPC = Clinical Performance Category, ROC = receiving operator characteristic, NSE = neuron-specific enolase. a Incomplete brainstem reflexes (including pupillary, oculocephalic, corneal) and myoclonus.
Critical Care Medicine
www.ccmjournal.org
1343
Oddo and Rossetti
Table 3. Areas Under the Receiving Operator Characteristic Curves of Several Models Combining Different Outcome Predictors Mortality (CPC 5) Variable
Poor Outcome (CPC 3–5)
ROC Area
95% CI
ROC Area
95% CI
Clinical examinationa + EEG
0.87
0.81–0.93
0.84
0.78–0.90
Clinical examinationa + NSE
0.83
0.76–0.89
0.83
0.77–0.90
EEG + NSE
0.87
0.81–0.93
0.84
0.78–0.90
Clinical examination + EEG + NSE
0.89
0.83–0.94
0.88
0.82–0.93
Clinical examination + EEG + SSEP
0.87
0.81–0.93
0.84
0.78–0.90
Clinical examination + EEG + NSE + SSEP
0.88
0.83–0.94
0.88
0.82–0.93
a
a
a
CPC = Clinical Performance Category, ROC = receiving operator characteristic, EEG = background reactivity on hypothermic electroencephalography, NSE = neuron-specific enolase > 33 μg/L, SSEP = somatosensory-evoked potentials. a Incomplete brainstem reflexes (including pupillary, oculocephalic, corneal) and myoclonus.
Table 4. Predictive Performance of the Model With the Highest Predictive Value, Including Clinical Examination (Brainstem Reflexes and Myoclonus), Electroencephalography Background Reactivity, and Serum Neuron-Specific Enolase > 33 μg/L Mortality (CPC 5) Variable
Poor Outcome (CPC 3–5)
Point Estimate
95% CI
Point Estimate
95% CI
Sensitivity
0.53
0.40–0.66
0.49
0.37–0.61
Specificity
1.00
0.95–1.00
1.00
0.94–1.00
Positive predictive value
1.00
0.89–1.00
1.00
0.90–1.00
Negative predictive value
0.71
0.61–0.80
0.62
0.52–0.72
Unweighted accuracy
0.77
0.75
CPC = Cerebral Performance Category.
sensitivity of this predictive model for mortality (53%) was thus higher than bilaterally absent SSEP alone (44%, with one false-positive at CPC = 3) and NSE more than 75 μg/L (44%, no false positive). Hypothermic EEG reactivity had a very high sensitivity (74%) but also one false positive (CPC = 2); of note, both clinical examination and NSE more than 33 μg/L showed several false-positive results (Table 1). Finally, our previously reported model that combined clinical examination, SSEP, and normothermic EEG reactivity (5) had a slightly higher sensitivity for mortality (37 of 62, 60%) but a false-positive rate of two of 72 (3%) (both surviving patients had CPC = 3, one showed absent brainstem reflexes and myoclonus in the acute setting, the other absent brainstem reflexes and SSEP). To obtain a false-positive rate of 0%, the sensitivity of that previous model would then be decreased to 26 of 62 (42%), thus 11% lower than the present one.
DISCUSSION This study provides class III evidence that clinical examination, hypothermic EEG background reactivity, and serum NSE are independent outcome predictors in comatose patients surviving CA, and that their combination provides the highest accuracy for prognostication of poor outcome and mortality, 1344
www.ccmjournal.org
offering 100% specificity. SSEPs do not provide any additional information when added to that model, and their sensitivity for mortality appears lower (one false-positive patient). Early outcome prognostication following CA and TH is receiving increasing attention; several studies of geographically different cohorts have addressed this topic, but none tested the predictive performance using multivariable analysis and including all presently recommended outcome predictors, that is, brainstem reflexes, EEG, SSEP, and NSE. The first article to point out that TH may interfere with prognostication showed in 37 patients that motor recovery is less reliable than brainstem reflexes (4); there was no information on electrophysiological and serological biomarkers. A study on 82 subjects (62% treated with TH) showed that motor responses may be influenced by sedation, more often used for a prolonged time in hypothermic patients (6); however, EEG data were not described and several patients lacked NSE and SSEP. Lower sedation policies may explain the findings of some groups supporting the reliability of motor signs at 3 days even after hypothermia (19); nevertheless, of the 103 subjects with TH, about 30% had missing clinical and serological data and even more than 80% lacked electrophysiological variables. Our previous study on 111 subjects did not June 2014 • Volume 42 • Number 6
Feature Articles
consider NSE, and about 10% were not examined with SSEP (5). A meta-analysis contended that SSEPs are not better than motor signs at day 3, but no NSE values were considered, and the study was limited to assessments performed before the TH era (30). An analysis on 34 patients in coma 4.5 days after CA showed that serum biomarkers correlate with other prognostic variables, but SSEP recordings were not performed in four of them (18). In 65 subjects undergoing TH, seven did not have EEG, and an unspecified number lacked SSEP (8). A multicenter study on 390 patients pointed out that NSE and motor scores should be applied with caution in this setting; EEG was not described, the majority of patients lacked brainstem reflexes analysis, and SSEPs were performed in part during hypothermia (7). A recent systematic review confirmed that neurological examination may be insufficient to accurately predict prognosis after CA and underscored the value of SSEP in this setting (10). In the present cohort, EEG background reactivity was the most reliable predictor of unfavorable outcome, with a sensitivity of 74% (specificity of 99%, given one false-positive result), confirming our previous observations on a subgroup of patients (11, 20). To our knowledge, no other group focused on this aspect during hypothermia, while the importance of EEG reactivity in normothermia has been described previously by us (5) (in a different set of patients) and others (31), and the importance of a continuous pattern for a favorable outcome was highlighted in conventional (11, 12) and amplitude-integrated EEG (13). Furthermore, it was recently found that EEG features correlate with NSE (11, 18), a marker of neuronal injury; nevertheless, EEG reactivity and NSE are independently related to outcome in the present analysis, suggesting that they provide complementary information; this is illustrated by the ROC areas in Tables 2 and 3, showing an increase after adding NSE to the EEG. The predictive performance of NSE in our cohort (ROC area of 0.72 for poor outcome and mortality at 33 μg/L; slightly less for mortality at 75 μg/L, which offers 100% specificity) appears somewhat lower than reported by other groups: 0.85 (cutoff 26 μg/L) (32), 0.84 (cutoff 26 μg/L) (33), 0.82 (cutoff 80 μg/L) (7), and 0.86 (cutoff 97 μg/L) (8). This probably reflects differences in patients’ cohorts and laboratory assays (14, 17). Clinical variables, especially brainstem reflexes, were the third independent outcome predictor. We chose to also include early myoclonus, since it had a relatively low false-positive rate in our previous study (5) and is easily recognizable in normothermia, at the time of brainstem reflexes examination. Brainstem structures are more resistant to hypoxic damage than the cerebral cortex and do not contribute directly to the EEG signal; therefore, they are complementary to serological and electrographic features. SSEPs, conversely, were not independently related to the clinical outcome and did not improve the predictive performance of other models. Evoked potentials, in fact, reflect a limited subset of cortical neuronal interactions, mostly involved in afferent pathways, as opposed to the wider interplay subtending the EEG signals (34). Critical Care Medicine
An important aspect of this analysis is that no test was absolutely reliable in predicting mortality, apart from a peak serum NSE more than 75 μg/L. Although the latter reflects previous studies by other groups (7–9), which found higher cutoffs than the one proposed by the AAN guidelines at 33 μg/L (17), this finding underscores the paramount importance of a multimodal assessment in this context, to minimize false-positive predictions of poor prognosis (14, 15, 35–37). In fact, the prognostication algorithm used in clinical practice (5) appears somewhat better than the proposed multimodal algorithm in terms of sensitivity for mortality; in this expanded set of patients, it shows however a false-positive rate of 3%. This illustrates that we were particularly cautious with withdrawal of intensive care in patients with discrepancy of test results, especially when EEG was reactive. Addition of NSE instead of SSEP, and of TH instead of normothermic EEG, enabled us to obtain an optimal false-positive rate of 0, at the expense however of a lower sensitivity. This study has some limitations. Being single center, its generalizability is not implicit, but the internal validity resulting from a uniform assessment of variables is strengthened. We restricted our analysis to patients with all variables of interest; the studied subjects, however, were entirely comparable to the excluded ones from a clinical point of view, suggesting that our findings may be generalized to similar populations. Only a subgroup of our patients had continuous EEG monitoring, but there is no evidence that this policy influenced patients’ outcome, as was pointed out recently in a separate study by our group (38). We did not consider brain MRI (39), as this technique was not routinely available within the first days following CA and is still used inconstantly and without definite variables (particularly regarding quantification of findings), as opposed to clinical examination, EEG, SSEP, and (more recently) NSE. Although serum NSE was assessed within a limited time frame following CA (24– 72 hr), due to laboratory availability it was not possible to measure it at fixed times nor to evaluate its kinetics. Finally, some extent of a self-fulfilling prophecy characterizes practically all cohort studies on this topic (5, 7, 17, 40). Related to the present analysis, decisions upon interruption of ICU care were based on clinical variables, normothermic EEG, and SSEP: importantly, hypothermic EEG reactivity and serum NSE, two of the three independent predictors forming the proposed model, were not directly involved in the prophecy, although it is impossible to completely rule out that clinicians might have used at times these data to corroborate their decisions.
CONCLUSION This multivariable analysis, to the best of our knowledge using for the first time a combination of all widely available and recognized post-CA outcome predictors, provides a robust model—including clinical data (brainstem reflexes and myoclonus), hypothermic EEG reactivity, and serum NSE—to prognosticate the clinical outcome of patients treated with TH after a CA. These findings should be confirmed by independent www.ccmjournal.org
1345
Oddo and Rossetti
groups. The model is accurate in predicting poor outcome, but prognostication of good outcome still remains less accurate with the currently used methods, as emphasis is at present primarily directed toward a PPV of 100% for poor outcome. Refinements of methods complementing current prognostication and targeting prediction of good recovery, such as event-related potentials (41–43), should therefore be encouraged in this setting.
ACKNOWLEDGMENTS We thank Christine Stähli, RN, and Tamara Suys, RN, as well as the EEG and ICU fellows for their help in data acquisition. Dr. Pedro Marques-Vidal kindly reviewed the statistical section.
REFERENCES
1. Bernard SA, Gray TW, Buist MD, et al: Treatment of comatose survivors of out-of-hospital cardiac arrest with induced hypothermia. N Engl J Med 2002; 346:557–563 2. Holzer M: Targeted temperature management for comatose survivors of cardiac arrest. N Engl J Med 2010; 363:1256–1264 3. Hypothermia after Cardiac Arrest Study Group: Mild therapeutic hypothermia to improve the neurologic outcome after cardiac arrest. N Engl J Med 2002; 346:549–556 4. 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 5. Rossetti AO, Oddo M, Logroscino G, et al: Prognostication after cardiac arrest and hypothermia: A prospective study. Ann Neurol 2010; 67:301–307 6. Samaniego EA, Mlynash M, Caulfield AF, et al: Sedation confounds outcome prediction in cardiac arrest survivors treated with hypothermia. Neurocrit Care 2011; 15:113–119 7. Bouwes A, Binnekade JM, Kuiper MA, et al: Prognosis of coma after therapeutic hypothermia: A prospective cohort study. Ann Neurol 2012; 71:206–212 8. 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 9. 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 10. Kamps MJ, Horn J, Oddo M, et al: Prognostication of neurologic outcome in cardiac arrest patients after mild therapeutic hypothermia: A meta-analysis of the current literature. Intensive Care Med 2013; 39:1671–1682 11. Rossetti AO, Carrera E, Oddo M: Early EEG correlates of neuronal injury after brain anoxia. Neurology 2012; 78:796–802 12. Cloostermans MC, van Meulen FB, Eertman CJ, et al: Continuous electroencephalography monitoring for early prediction of neurological outcome in postanoxic patients after cardiac arrest: A prospective cohort study. Crit Care Med 2012; 40:2867–2875 13. Rundgren M, Westhall E, Cronberg T, et al: Continuous amplitude-integrated electroencephalogram predicts outcome in hypothermia-treated cardiac arrest patients. Crit Care Med 2010; 38:1838–1844 14. Oddo M, Rossetti AO: Predicting neurological outcome after cardiac arrest. Curr Opin Crit Care 2011; 17:254–259 15. Samaniego EA, Persoon S, Wijman CA: Prognosis after cardiac arrest and hypothermia: A new paradigm. Curr Neurol Neurosci Rep 2011; 11:111–119 16. Peberdy MA, Callaway CW, Neumar RW, et al; American Heart Association: Part 9: Post-cardiac arrest care: 2010 American Heart Association Guidelines for Cardiopulmonary Resuscitation and Emergency Cardiovascular Care. Circulation 2010; 122:S768–S786
1346
www.ccmjournal.org
17. Wijdicks EF, Hijdra A, Young GB, et al; Quality Standards Subcommittee of the American Academy of Neurology: 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 18. Cronberg T, Rundgren M, Westhall E, et al: Neuron-specific enolase correlates with other prognostic markers after cardiac arrest. Neurology 2011; 77:623–630 19. Fugate JE, Wijdicks EF, Mandrekar J, et al: Predictors of neurologic outcome in hypothermia after cardiac arrest. Ann Neurol 2010; 68:907–914 20. Rossetti AO, Urbano LA, Delodder F, et al: Prognostic value of continuous EEG monitoring during therapeutic hypothermia after cardiac arrest. Crit Care 2010; 14:R173 21. Fugate JE, Wijdicks EF, White RD, et al: Does therapeutic hypothermia affect time to awakening in cardiac arrest survivors? Neurology 2011; 77:1346–1350 22. Alvarez V, Oddo M, Rossetti AO: Stimulus-induced rhythmic, periodic or ictal discharges (SIRPIDs) in comatose survivors of cardiac arrest: Characteristics and prognostic value. Clin Neurophysiol 2013; 124:204–208 23. Hirsch LJ, Claassen J, Mayer SA, et al: Stimulus-induced rhythmic, periodic, or ictal discharges (SIRPIDs): A common EEG phenomenon in the critically ill. Epilepsia 2004; 45:109–123 24. Hirsch LJ, LaRoche SM, Gaspard N, et al: American Clinical Neurophysiology Society’s Standardized Critical Care EEG Terminology: 2012 version. J Clin Neurophysiol 2013; 30:1–27 25. Rossetti AO, Oddo M, Liaudet L, et al: Predictors of awakening from postanoxic status epilepticus after therapeutic hypothermia. Neurology 2009; 72:744–749 26. Oksanen T, Tiainen M, Skrifvars MB, et al: Predictive power of serum NSE and OHCA score regarding 6-month neurologic outcome after out-of-hospital ventricular fibrillation and therapeutic hypothermia. Resuscitation 2009; 80:165–170 27. Shinozaki K, Oda S, Sadahiro T, et al: S-100B and neuron-specific enolase as predictors of neurological outcome in patients after cardiac arrest and return of spontaneous circulation: A systematic review. Crit Care 2009; 13:R121 28. Jacobs I, Nadkarni V, Bahr J, et al: Cardiac arrest and cardiopulmonary resuscitation outcome reports: Update and simplification of the Utstein templates for resuscitation registries: A statement for healthcare professionals from a task force of the International Liaison Committee on Resuscitation (American Heart Association, European Resuscitation Council, Australian Resuscitation Council, New Zealand Resuscitation Council, Heart and Stroke Foundation of Canada, InterAmerican Heart Foundation, Resuscitation Councils of Southern Africa). Circulation 2004; 110:3385–3397 29. Booth CM, Boone RH, Tomlinson G, et al: Is this patient dead, vegetative, or severely neurologically impaired? Assessing outcome for comatose survivors of cardiac arrest. JAMA 2004; 291:870–879 30. Lee YC, Phan TG, Jolley DJ, et al: Accuracy of clinical signs, SEP, and EEG in predicting outcome of hypoxic coma: A meta-analysis. Neurology 2010; 74:572–580 31. Thenayan EA, Savard M, Sharpe MD, et al: Electroencephalogram for prognosis after cardiac arrest. J Crit Care 2010; 25:300–304 32. Tiainen M, Roine RO, Pettilä V, et al: Serum neuron-specific enolase and S-100B protein in cardiac arrest patients treated with hypothermia. Stroke 2003; 34:2881–2886 33. Rundgren M, Karlsson T, Nielsen N, et al: Neuron specific enolase and S-100B as predictors of outcome after cardiac arrest and induced hypothermia. Resuscitation 2009; 80:784–789 34. van Putten MJ: The N20 in post-anoxic coma: Are you listening? Clin Neurophysiol 2012; 123:1460–1464 35. Young GB: Clinical practice. Neurologic prognosis after cardiac arrest. N Engl J Med 2009; 361:605–611 36. Josephson SA: Predicting neurologic outcomes after cardiac arrest: The crystal ball becomes cloudy. Ann Neurol 2010; 67:A5–A6 37. Rabinstein AA, Wijdicks EF: The value of EEG monitoring after cardiac arrest treated with hypothermia. Neurology 2012; 78:774–775 June 2014 • Volume 42 • Number 6
Feature Articles 38. Alvarez V, Sierra-Marcos A, Oddo M, et al: Yield of intermittent versus continuous EEG in comatose survivors of cardiac arrest treated with hypothermia. Crit Care 2013; 17:R190 39. Wijman CA, Mlynash M, Caulfield AF, et al: Prognostic value of brain diffusion-weighted imaging after cardiac arrest. Ann Neurol 2009; 65:394–402 40. Zandbergen EG, Hijdra A, Koelman JH, et al; PROPAC Study Group: Prediction of poor outcome within the first 3 days of postanoxic coma. Neurology 2006; 66:62–68
Critical Care Medicine
41. Fischer C, Luauté J, Némoz 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 42. 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 43. Steppacher I, Eickhoff S, Jordanov T, et al: N400 predicts recovery from disorders of consciousness. Ann Neurol 2013; 73: 594–602
www.ccmjournal.org
1347
Objectives: Therapeutic hypothermia and pharmacological sedation may influence outcome prediction after cardiac arrest. The use of a multimodal approach, including clinical examination, electroencephalography, somatosensory-evoked potentials, and serum neuron-specific enolase, is recommended; however, no study examined the comparative performance of these predictors or addressed their optimal combination. Design: Prospective cohort study. Setting: Adult ICU of an academic hospital. Patients: One hundred thirty-four consecutive adults treated with therapeutic hypothermia after cardiac arrest. Measurements and Main Results: Variables related to the cardiac arrest (cardiac rhythm, time to return of spontaneous circulation), clinical examination (brainstem reflexes and myoclonus), electroencephalography reactivity during therapeutic hypothermia, somatosensory-evoked potentials, and serum neuron-specific enolase. Models to predict clinical outcome at 3 months (assessed using the Cerebral Performance Categories: 5 = death; 3–5 = poor recovery) were evaluated using ordinal logistic regressions and receiving operator characteristic curves. Seventy-two patients (54%) had a poor outcome (of whom, 62 died), and 62 had a good outcome. Multivariable ordinal logistic regression identified absence of electroencephalography reactivity (p < 0.001), incomplete recovery of brainstem reflexes in normothermia (p = 0.013), and neuron-specific enolase higher than 33 μg/L (p = 0.029), but *See also p. 1535. 1 Department of Intensive Care Medicine, CHUV-Lausanne University Hospital and Faculty of Biology and Medicine, Lausanne, Switzerland. 2 Department of Clinical Neurosciences, CHUV-Lausanne University Hospital and Faculty of Biology and Medicine, Lausanne, Switzerland. Dr. Oddo conceived and designed the study; drafted and revised the manuscript; and analyzed, interpreted, and acquired the data. Dr. Rossetti conceived and designed the study; drafted and revised the manuscript; analyzed, interpreted, and acquired the data; did statistical analysis; and coordinated the study. Dr. Oddo received support for article research from Swiss National Science Foundation (grant 320030_138191). He consulted for Bard Medical Integra Neurosciences and lectured for Bard Medical Integra Neurosciences. His institution received grant support from the Swiss National Science Foundation. Dr. Rossetti and his institution received support from Swiss National Science Foundation (grant CR32I3_143780). For information regarding this article, E-mail: [email protected] Copyright © 2014 by the Society of Critical Care Medicine and Lippincott Williams & Wilkins DOI: 10.1097/CCM.0000000000000211
1340
www.ccmjournal.org
not somatosensory-evoked potentials, as independent predictors of poor outcome. The combination of clinical examination, electroencephalography reactivity, and neuron-specific enolase yielded the best predictive performance (receiving operator characteristic areas: 0.89 for mortality and 0.88 for poor outcome), with 100% positive predictive value. Addition of somatosensory-evoked potentials to this model did not improve prognostic accuracy. Conclusions: Combination of clinical examination, electroencephalography reactivity, and serum neuron-specific enolase offers the best outcome predictive performance for prognostication of early postanoxic coma, whereas somatosensory-evoked potentials do not add any complementary information. Although prognostication of poor outcome seems excellent, future studies are needed to further improve prediction of good prognosis, which still remains inaccurate. (Crit Care Med 2014; 42:1340–1347) Key Words: coma; electroencephalography; neuron-specific enolase; prognosis; somatosensory-evoked potentials
C
oma after cardiac arrest (CA) is a leading cause of admission in ICUs and has a high rate of mortality and morbidity. Therapeutic hypothermia (TH) is increasingly used for neuroprotection in this context (1–3). Several recent studies have shown that TH and the related pharmacological sedation may influence some prognosticators commonly used in this setting as compared with patients not undergoing hypothermia. Particularly, motor response may recover over a longer time frame, and serum biomarkers such as neuron-specific enolase (NSE) can display cutoff levels for poor outcome considerably higher than previously thought (4–10). In contrast, electrophysiological investigations with electroencephalography (EEG), particularly focusing on continuous patterns and background reactivity during and after TH (11–13), and somatosensory-evoked potentials (SSEPs) (7) seem to provide greater prognostic accuracy in this setting. Based on this evidence, current recommendations suggest the use of a multimodal approach to minimize false-positive prediction of poor outcome (14–16), and raise concerns in relying on American Academy of Neurology prognostic guidelines (17) that were assessed before the TH became a standard of care. Despite such recommendations, however, none of the recent studies on this topic by our and other groups (6, 7, 18–21) has June 2014 • Volume 42 • Number 6
Feature Articles
critically evaluated the prognostic performance of the different clinical, electrophysiological, and serological tests, alone or in combination, nor attempted to identify the multimodal algorithm with the highest prognostic accuracy in this setting. We therefore undertook this analysis to identify those variables among baseline demographic data, clinical examination, hypothermic EEG, normothermic SSEP, and NSE, which might offer the best performance in predicting outcome of post-CA coma.
MATERIAL AND METHODS Patients and Setting This prospective cohort study was conducted at the Department of Intensive Care Medicine, CHUV-Lausanne University Hospital, an adult medical-surgical ICU; it was approved by the institutional ethics committee, and waiver of consent was allowed since all examinations were part of standard patient care. The population consisted of patients older than 16 years admitted for coma following out-of-hospital CA between December 2009 and April 2013, treated with TH, and not brain dead within 48 hours. Patients were managed according to our previously published protocol (TH is considered in adults with cardiac and noncardiac etiologies, regardless of the initial rhythm, unless presenting a marked hemodynamic instability, or a patient’s “do-not-resuscitate” directive; about 85% of all patients reaching alive our hospital after CA receive TH) (5, 11), and in line with the American Heart Association guidelines (16). TH of 33°C ± 1°C was induced as soon as possible using ice packs plus ice-cold isotonic solutions, followed by the application of a surface cooling device with computerized automatic temperature control (Arctic Sun 2000 TTM; Bard Medical, Louisville, CO), maintained for 24 hours. Standardized sedation (midazolam, 0.1 mg/kg/hr), analgesia (fentanyl, 1.5 μg/kg/hr), and neuromuscular blockade (rocuronium boluses, 0.6 mg/kg, if shivering) were administered IV during TH and discontinued after passive rewarming at a core temperature above 35°C. Electrophysiological Procedures and Neurological Assessments Video-EEGs (Viasys Neurocare, Madison, WI) were performed during TH, using 21 electrodes according to the international 10–20 system. EEG was continuously recorded for 24–48 hours until July 2010 (20), and afterward intermittently, for 20–30 minutes. Background reactivity was tested at the bedside, at least 6 hours after CA, as previously reported (11, 20), by applying repetitive auditory, visual, and nociceptive stimuli. The same procedure was repeated during the early phase following rewarming, in normothermic conditions, within 72 hours from CA. In case of repeated testing at different time points, the best result was used for analysis. EEG findings were categorized by certified EEG interpreters according to the presence/absence of the following: 1) background reactivity: activity more than or equal to 10 μV (regardless of frequency) and any clear and reproducible change in amplitude or frequency upon Critical Care Medicine
stimulation, excluding “stimulus-induced rhythmic, periodic, or ictal discharges” (22, 23) and muscle artifacts; 2) spontaneous discontinuous pattern: background interrupted by diffusely suppressed periods of at least 10% of the recording (24); and 3) epileptiform activity: any periodic or rhythmic spikes, sharp waves, spike waves, or rhythmic waves (5, 20). At normothermia and off sedation, within 72 hours from CA, patients were repetitively examined by a certified neurologist, and the best patients’ performances were considered. Brainstem reflexes (pupillary, oculocephalic, and corneal) were categorized as all present versus at least one absent; the motor response was not taken into account for this study in view of its reduced accuracy due to TH and related sedation (4–7). Myoclonus was considered if appearing within 24 hours after discontinuation of sedative drugs. Treatment of epileptiform EEG activity and myoclonus consisted mostly in the administration of nonsedating antiepileptic agents (mainly levetiracetam and valproate) and, when necessary, sedative agents (mainly propofol), as previously reported (25). Bilateral SSEPs were recorded during early normothermia; the cortical N20 responses on SSEP were categorized as present or bilaterally absent (no clear negative deflection at 18–25 ms followed by a positive wave on both sides). NSE is released in the serum after CA with a half-life of about 24 hours (26, 27); it was collected 1–3 times within the first 72 hours following the CA. To prevent hemolysis and false-positive test results, samples were handled manually and measured with an automated immunofluorescent assay (Thermo Scientific Brahms NSE Kryptor Immunoassay). In the vast majority of patients, at least one probe was collected at 48 hours, and for the present analysis, we considered the peak value, exploring cutoffs at 33 μg/L (17) and higher, to reach a 100% specificity for poor outcome (7–9). Decisions on Withdrawal of Care In line with our previous study (5), supportive care withdrawal was considered upon the occurrence of two or more of four criteria after more than 72 hours after CA, in normothermia and off sedation: unreactive EEG background in normothermia, treatment-resistant myoclonus, bilateral absence of N20 in SSEP in normothermia, and incomplete return of the three principal brainstem reflexes. This was based on a consensus between intensivists and consultant neurologists involved in patient care. Serum NSE levels and the EEG obtained during TH were not taken into consideration for this decision. Data Collection and Outcome Assessment Demographical and clinical variables were collected prospectively using Utstein’s style recommendations (28). Etiology of CA was dichotomized as cardiac versus noncardiac, and the initial CA rhythm was categorized as ventricular fibrillation (VF) versus non-VF (including asystole and pulseless electrical activity). CA duration was estimated by calculating the time to return of spontaneous circulation and collected at the time of patient admission to the ICU. Neurological outcome at 3 months was prospectively assessed through a semistructured www.ccmjournal.org
1341
Oddo and Rossetti
telephone interview using Glasgow-Pittsburgh Cerebral Performance categories (CPC) (29): CPC 1 indicates full recovery, CPC 2 moderate disability, and CPC 3 severe disability. Subjects in CPC 4 are comatose or in vegetative state and those with CPC 5 died. In accordance to the vast majority of studies on the topic, good neurological recovery was defined as CPC 1 or 2. Statistical Analysis To explore outcome predictors, chi-square test or two-tailed Fisher exact test for categorical variables, and parametric or nonparametric analysis of variance for continuous variables, were used as needed; significant (p < 0.05) variables were entered into a multivariable backward ordinal logistic regression, to identify independent outcome predictors. Each potential prognosticator of poor outcome (CPC 3–5) or mortality (CPC 5) was also assessed with C statistics, using the area under the receiving operator characteristic (ROC) curve in consideration of dichotomous outcomes, applying nonparametric comparisons. We further examined the performance of prognostic models constructed using combinations of the single most robust outcome predictors. The qualitatively best outcome model was assessed for sensitivity, specificity, positive predictive value (PPV), negative predictive value (with binomial 95% CI), and unweighted accuracy, for poor outcome and mortality. Statistical analyses were performed with a STATA software, version 9 (College Station, TX).
RESULTS From December 2009 to April 2013 (40 mo), 183 consecutive adults were treated for coma after CA with TH at our center. All had full recording of historical data, were examined clinically, and had normothermic EEG, while 49 lacked at least one of the following examinations: hypothermic EEG, SSEP, or serum NSE. We, therefore, restricted our analysis to the 134 subjects who received the complete set of early tests (this cohort overlaps with 59 patients who were recently described focusing on EEG and NSE) (11). As compared with the 49 excluded subjects, the 134 patients did not differ in terms of mean age (62.4 ± 12.5 vs 62.7 ± 15.1; p = 0.44, t test), gender (35 of 134 vs 11 of 49 women; p = 0.61, chi-square test), and mortality at 3 months (62 of 139 vs 22 of 49; p = 0.87, chi-square test). Of the 134 studied patients, 29 (21%) had continuous EEG for 16–48 hours, whereas the remaining 105 had routine intermittent EEG studies. Although the EEG was started earlier in patients who had continuous EEG (13.0 ± 5.8 vs 18. ± 5.9 hr in subjects who had intermittent EEG, p < 0.001, t test), the prevalence of patients with a nonreactive EEG background was similar between the two groups (11 of 29 [38%] vs 36 of 105 [34%], p = 0.72, chi-square test), as was mortality at 3 months (13 of 29 [45%] vs 49 of 105 [47%], p = 0.86, chi-square test). Table 1 shows clinical and electrophysiological variables assessed within 72 hours after CA, according to CPC at 3 months. Mortality occurred in 45%, whereas 86% of the survivors reached a good outcome; of note, no patient remained in a vegetative state (CPC 4). Although baseline demographics
Table 1. Clinical and Electrophysiological Characteristics of 134 Patients Treated With Therapeutic Hypothermia After Cardiac Arrest, Stratified for Their Functional Outcome at 3 Months Variable
CPC 1 (40 Patients)
Mean age, yr (± sd)
60.1 (± 12.7) 62.2 (± 10.3) 62.6 (± 11.2) 64.0 (± 13.3)
Female gender
CPC 2 (22 Patients)
CPC 3 (10 Patients)
CPC 5 (62 Patients)
p
Statistical Test
0.555
ANOVA
12
6
2
15
0.896
Fisher
Noncardiac etiology
3
5
1
18
0.043
Fisher
Non–ventricular fibrillation rhythm
6
8
4
28
0.012
Fisher
Time to return of spontaneous circulation, min (± sd)
19.8 (± 11.7) 15.0 (± 9.8)
21.0 (± 11.3) 26.1 (± 15.1)
< 0.01
ANOVA
Incomplete brainstem reflexesa
2
2
3
36
< 0.01
Fisher
Myoclonus
1
1
1
20
< 0.01
Fisher
Hypothermic electroencephalography nonreactive
0
1
0
46
< 0.01
Fisher
Bilaterally absent cortical somatosensory-evoked potentials
0
0
1
27
< 0.01
Fisher
Serum NSE > 33 μg/Lb
5
3
4
37
< 0.01
Fisher
Serum NSE > 75 μg/L
0
0
0
27
< 0.01
Fisher
b
CPC = Cerebral Performance Category, ANOVA = analysis of variance, NSE = neuron-specific enolase. a Pupillary, oculocephalic, corneal. b Peak value at 24–48 hr. CPC 1–2 implies good prognosis; CPC 5 implies death.
1342
www.ccmjournal.org
June 2014 • Volume 42 • Number 6
Feature Articles
Figure 1. Prognostic performance for prediction of 3-month clinical outcome (mortality, left; poor neurological recovery, right), assessed using the receiver operator characteristic (ROC) curves of the model including clinical evaluation (brainstem reflexes, early myoclonus), hypothermic electroencephalography reactivity, and neuron-specific enolase higher than 33 μg/L.
were not discriminatory, acute CA variables were significantly different among the outcome groups; however, the most robust predictors in univariable analyses were clinical examination and electrophysiological data, as well as NSE serum levels (for which a cutoff of 75 μg/L was also assessed, since beyond this threshold no patient awoke in our cohort). Ordinal logistic regression identified as independent predictors: absent hypothermic EEG reactivity (coefficient, 4.36; 95% CI, 2.27–6.46; p < 0.001), incomplete return of brainstem reflexes (coefficient, 1.10; 95% CI, 0.23–1.96; p = 0.013), and serum NSE more than 33 μg/L (coefficient, 1.09; 95% CI, 0.11–2.07; p = 0.029). Predictive performances for mortality (CPC 5) and poor outcome (CPC 3–5) were assessed using C statistics for clinical examination (combining occurrence of incomplete brainstem reflexes and occurrence of myoclonus in a 0–2 score), EEG reactivity, SSEP, and peak serum NSE at different cutoffs. Table 2 summarizes the areas under the ROC curves: EEG and clinical examination stand out for both conditions and (at least
for poor outcome) NSE at the 33 μg/L cutoff seemed better than at 75 μg/L. Several predictive models, using NSE dichotomized at 33 μg/L, were tested (Table 3); for both outcomes, the model with the highest prognostic accuracy included clinical examination, EEG reactivity, and serum NSE. Of note, prognostication was not improved neither by addition of SSEP nor by its use instead of NSE. Figure 1 illustrates the ROC curves of the best predictive models for the two main outcomes, and shows that 100% specificity in forecasting poor outcome and mortality is reached using a cutoff of 0–2 versus 3–4. Table 4 gives the predictive performance of the model with this cutoff: although the PPV (and specificity) is optimal, anticipation of a good clinical outcome is less reliable. Of the 62 deceased patients, nine showed all four poor predictive variables (lack of at least one brainstem reflex, myoclonus, NSE more than 33 μg/L, and nonreactive hypothermic EEG), and 24 had three of them; none of the surviving patients had more than two items (false-positive rate = 0). The
Table 2. Areas Under the Receiving Operator Characteristic Curves for Each Examined Single Functional Outcome Predictor Mortality (CPC 5) Variable
ROC Area
95% CI
Clinical examinationa
0.78
0.71–0.84
Hypothermic electroencephalography nonreactive
0.86
Bilaterally absent cortical somatosensoryevoked potentials
Poor Outcome (CPC 3–5) p
ROC Area
95% CI
0.77
0.70–0.83
0.81–0.92
0.81
0.75–0.87
0.71
0.64–0.77
0.69
0.64–0.75
Serum NSE > 33 μg/L
0.72
0.64–0.79
0.72
0.65–0.79
Serum NSE > 75 μg/L
0.72
0.66–0.78
0.68
0.63–0.74
< 0.001
p
< 0.001
CPC = Clinical Performance Category, ROC = receiving operator characteristic, NSE = neuron-specific enolase. a Incomplete brainstem reflexes (including pupillary, oculocephalic, corneal) and myoclonus.
Critical Care Medicine
www.ccmjournal.org
1343
Oddo and Rossetti
Table 3. Areas Under the Receiving Operator Characteristic Curves of Several Models Combining Different Outcome Predictors Mortality (CPC 5) Variable
Poor Outcome (CPC 3–5)
ROC Area
95% CI
ROC Area
95% CI
Clinical examinationa + EEG
0.87
0.81–0.93
0.84
0.78–0.90
Clinical examinationa + NSE
0.83
0.76–0.89
0.83
0.77–0.90
EEG + NSE
0.87
0.81–0.93
0.84
0.78–0.90
Clinical examination + EEG + NSE
0.89
0.83–0.94
0.88
0.82–0.93
Clinical examination + EEG + SSEP
0.87
0.81–0.93
0.84
0.78–0.90
Clinical examination + EEG + NSE + SSEP
0.88
0.83–0.94
0.88
0.82–0.93
a
a
a
CPC = Clinical Performance Category, ROC = receiving operator characteristic, EEG = background reactivity on hypothermic electroencephalography, NSE = neuron-specific enolase > 33 μg/L, SSEP = somatosensory-evoked potentials. a Incomplete brainstem reflexes (including pupillary, oculocephalic, corneal) and myoclonus.
Table 4. Predictive Performance of the Model With the Highest Predictive Value, Including Clinical Examination (Brainstem Reflexes and Myoclonus), Electroencephalography Background Reactivity, and Serum Neuron-Specific Enolase > 33 μg/L Mortality (CPC 5) Variable
Poor Outcome (CPC 3–5)
Point Estimate
95% CI
Point Estimate
95% CI
Sensitivity
0.53
0.40–0.66
0.49
0.37–0.61
Specificity
1.00
0.95–1.00
1.00
0.94–1.00
Positive predictive value
1.00
0.89–1.00
1.00
0.90–1.00
Negative predictive value
0.71
0.61–0.80
0.62
0.52–0.72
Unweighted accuracy
0.77
0.75
CPC = Cerebral Performance Category.
sensitivity of this predictive model for mortality (53%) was thus higher than bilaterally absent SSEP alone (44%, with one false-positive at CPC = 3) and NSE more than 75 μg/L (44%, no false positive). Hypothermic EEG reactivity had a very high sensitivity (74%) but also one false positive (CPC = 2); of note, both clinical examination and NSE more than 33 μg/L showed several false-positive results (Table 1). Finally, our previously reported model that combined clinical examination, SSEP, and normothermic EEG reactivity (5) had a slightly higher sensitivity for mortality (37 of 62, 60%) but a false-positive rate of two of 72 (3%) (both surviving patients had CPC = 3, one showed absent brainstem reflexes and myoclonus in the acute setting, the other absent brainstem reflexes and SSEP). To obtain a false-positive rate of 0%, the sensitivity of that previous model would then be decreased to 26 of 62 (42%), thus 11% lower than the present one.
DISCUSSION This study provides class III evidence that clinical examination, hypothermic EEG background reactivity, and serum NSE are independent outcome predictors in comatose patients surviving CA, and that their combination provides the highest accuracy for prognostication of poor outcome and mortality, 1344
www.ccmjournal.org
offering 100% specificity. SSEPs do not provide any additional information when added to that model, and their sensitivity for mortality appears lower (one false-positive patient). Early outcome prognostication following CA and TH is receiving increasing attention; several studies of geographically different cohorts have addressed this topic, but none tested the predictive performance using multivariable analysis and including all presently recommended outcome predictors, that is, brainstem reflexes, EEG, SSEP, and NSE. The first article to point out that TH may interfere with prognostication showed in 37 patients that motor recovery is less reliable than brainstem reflexes (4); there was no information on electrophysiological and serological biomarkers. A study on 82 subjects (62% treated with TH) showed that motor responses may be influenced by sedation, more often used for a prolonged time in hypothermic patients (6); however, EEG data were not described and several patients lacked NSE and SSEP. Lower sedation policies may explain the findings of some groups supporting the reliability of motor signs at 3 days even after hypothermia (19); nevertheless, of the 103 subjects with TH, about 30% had missing clinical and serological data and even more than 80% lacked electrophysiological variables. Our previous study on 111 subjects did not June 2014 • Volume 42 • Number 6
Feature Articles
consider NSE, and about 10% were not examined with SSEP (5). A meta-analysis contended that SSEPs are not better than motor signs at day 3, but no NSE values were considered, and the study was limited to assessments performed before the TH era (30). An analysis on 34 patients in coma 4.5 days after CA showed that serum biomarkers correlate with other prognostic variables, but SSEP recordings were not performed in four of them (18). In 65 subjects undergoing TH, seven did not have EEG, and an unspecified number lacked SSEP (8). A multicenter study on 390 patients pointed out that NSE and motor scores should be applied with caution in this setting; EEG was not described, the majority of patients lacked brainstem reflexes analysis, and SSEPs were performed in part during hypothermia (7). A recent systematic review confirmed that neurological examination may be insufficient to accurately predict prognosis after CA and underscored the value of SSEP in this setting (10). In the present cohort, EEG background reactivity was the most reliable predictor of unfavorable outcome, with a sensitivity of 74% (specificity of 99%, given one false-positive result), confirming our previous observations on a subgroup of patients (11, 20). To our knowledge, no other group focused on this aspect during hypothermia, while the importance of EEG reactivity in normothermia has been described previously by us (5) (in a different set of patients) and others (31), and the importance of a continuous pattern for a favorable outcome was highlighted in conventional (11, 12) and amplitude-integrated EEG (13). Furthermore, it was recently found that EEG features correlate with NSE (11, 18), a marker of neuronal injury; nevertheless, EEG reactivity and NSE are independently related to outcome in the present analysis, suggesting that they provide complementary information; this is illustrated by the ROC areas in Tables 2 and 3, showing an increase after adding NSE to the EEG. The predictive performance of NSE in our cohort (ROC area of 0.72 for poor outcome and mortality at 33 μg/L; slightly less for mortality at 75 μg/L, which offers 100% specificity) appears somewhat lower than reported by other groups: 0.85 (cutoff 26 μg/L) (32), 0.84 (cutoff 26 μg/L) (33), 0.82 (cutoff 80 μg/L) (7), and 0.86 (cutoff 97 μg/L) (8). This probably reflects differences in patients’ cohorts and laboratory assays (14, 17). Clinical variables, especially brainstem reflexes, were the third independent outcome predictor. We chose to also include early myoclonus, since it had a relatively low false-positive rate in our previous study (5) and is easily recognizable in normothermia, at the time of brainstem reflexes examination. Brainstem structures are more resistant to hypoxic damage than the cerebral cortex and do not contribute directly to the EEG signal; therefore, they are complementary to serological and electrographic features. SSEPs, conversely, were not independently related to the clinical outcome and did not improve the predictive performance of other models. Evoked potentials, in fact, reflect a limited subset of cortical neuronal interactions, mostly involved in afferent pathways, as opposed to the wider interplay subtending the EEG signals (34). Critical Care Medicine
An important aspect of this analysis is that no test was absolutely reliable in predicting mortality, apart from a peak serum NSE more than 75 μg/L. Although the latter reflects previous studies by other groups (7–9), which found higher cutoffs than the one proposed by the AAN guidelines at 33 μg/L (17), this finding underscores the paramount importance of a multimodal assessment in this context, to minimize false-positive predictions of poor prognosis (14, 15, 35–37). In fact, the prognostication algorithm used in clinical practice (5) appears somewhat better than the proposed multimodal algorithm in terms of sensitivity for mortality; in this expanded set of patients, it shows however a false-positive rate of 3%. This illustrates that we were particularly cautious with withdrawal of intensive care in patients with discrepancy of test results, especially when EEG was reactive. Addition of NSE instead of SSEP, and of TH instead of normothermic EEG, enabled us to obtain an optimal false-positive rate of 0, at the expense however of a lower sensitivity. This study has some limitations. Being single center, its generalizability is not implicit, but the internal validity resulting from a uniform assessment of variables is strengthened. We restricted our analysis to patients with all variables of interest; the studied subjects, however, were entirely comparable to the excluded ones from a clinical point of view, suggesting that our findings may be generalized to similar populations. Only a subgroup of our patients had continuous EEG monitoring, but there is no evidence that this policy influenced patients’ outcome, as was pointed out recently in a separate study by our group (38). We did not consider brain MRI (39), as this technique was not routinely available within the first days following CA and is still used inconstantly and without definite variables (particularly regarding quantification of findings), as opposed to clinical examination, EEG, SSEP, and (more recently) NSE. Although serum NSE was assessed within a limited time frame following CA (24– 72 hr), due to laboratory availability it was not possible to measure it at fixed times nor to evaluate its kinetics. Finally, some extent of a self-fulfilling prophecy characterizes practically all cohort studies on this topic (5, 7, 17, 40). Related to the present analysis, decisions upon interruption of ICU care were based on clinical variables, normothermic EEG, and SSEP: importantly, hypothermic EEG reactivity and serum NSE, two of the three independent predictors forming the proposed model, were not directly involved in the prophecy, although it is impossible to completely rule out that clinicians might have used at times these data to corroborate their decisions.
CONCLUSION This multivariable analysis, to the best of our knowledge using for the first time a combination of all widely available and recognized post-CA outcome predictors, provides a robust model—including clinical data (brainstem reflexes and myoclonus), hypothermic EEG reactivity, and serum NSE—to prognosticate the clinical outcome of patients treated with TH after a CA. These findings should be confirmed by independent www.ccmjournal.org
1345
Oddo and Rossetti
groups. The model is accurate in predicting poor outcome, but prognostication of good outcome still remains less accurate with the currently used methods, as emphasis is at present primarily directed toward a PPV of 100% for poor outcome. Refinements of methods complementing current prognostication and targeting prediction of good recovery, such as event-related potentials (41–43), should therefore be encouraged in this setting.
ACKNOWLEDGMENTS We thank Christine Stähli, RN, and Tamara Suys, RN, as well as the EEG and ICU fellows for their help in data acquisition. Dr. Pedro Marques-Vidal kindly reviewed the statistical section.
REFERENCES
1. Bernard SA, Gray TW, Buist MD, et al: Treatment of comatose survivors of out-of-hospital cardiac arrest with induced hypothermia. N Engl J Med 2002; 346:557–563 2. Holzer M: Targeted temperature management for comatose survivors of cardiac arrest. N Engl J Med 2010; 363:1256–1264 3. Hypothermia after Cardiac Arrest Study Group: Mild therapeutic hypothermia to improve the neurologic outcome after cardiac arrest. N Engl J Med 2002; 346:549–556 4. 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 5. Rossetti AO, Oddo M, Logroscino G, et al: Prognostication after cardiac arrest and hypothermia: A prospective study. Ann Neurol 2010; 67:301–307 6. Samaniego EA, Mlynash M, Caulfield AF, et al: Sedation confounds outcome prediction in cardiac arrest survivors treated with hypothermia. Neurocrit Care 2011; 15:113–119 7. Bouwes A, Binnekade JM, Kuiper MA, et al: Prognosis of coma after therapeutic hypothermia: A prospective cohort study. Ann Neurol 2012; 71:206–212 8. 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 9. 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 10. Kamps MJ, Horn J, Oddo M, et al: Prognostication of neurologic outcome in cardiac arrest patients after mild therapeutic hypothermia: A meta-analysis of the current literature. Intensive Care Med 2013; 39:1671–1682 11. Rossetti AO, Carrera E, Oddo M: Early EEG correlates of neuronal injury after brain anoxia. Neurology 2012; 78:796–802 12. Cloostermans MC, van Meulen FB, Eertman CJ, et al: Continuous electroencephalography monitoring for early prediction of neurological outcome in postanoxic patients after cardiac arrest: A prospective cohort study. Crit Care Med 2012; 40:2867–2875 13. Rundgren M, Westhall E, Cronberg T, et al: Continuous amplitude-integrated electroencephalogram predicts outcome in hypothermia-treated cardiac arrest patients. Crit Care Med 2010; 38:1838–1844 14. Oddo M, Rossetti AO: Predicting neurological outcome after cardiac arrest. Curr Opin Crit Care 2011; 17:254–259 15. Samaniego EA, Persoon S, Wijman CA: Prognosis after cardiac arrest and hypothermia: A new paradigm. Curr Neurol Neurosci Rep 2011; 11:111–119 16. Peberdy MA, Callaway CW, Neumar RW, et al; American Heart Association: Part 9: Post-cardiac arrest care: 2010 American Heart Association Guidelines for Cardiopulmonary Resuscitation and Emergency Cardiovascular Care. Circulation 2010; 122:S768–S786
1346
www.ccmjournal.org
17. Wijdicks EF, Hijdra A, Young GB, et al; Quality Standards Subcommittee of the American Academy of Neurology: 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 18. Cronberg T, Rundgren M, Westhall E, et al: Neuron-specific enolase correlates with other prognostic markers after cardiac arrest. Neurology 2011; 77:623–630 19. Fugate JE, Wijdicks EF, Mandrekar J, et al: Predictors of neurologic outcome in hypothermia after cardiac arrest. Ann Neurol 2010; 68:907–914 20. Rossetti AO, Urbano LA, Delodder F, et al: Prognostic value of continuous EEG monitoring during therapeutic hypothermia after cardiac arrest. Crit Care 2010; 14:R173 21. Fugate JE, Wijdicks EF, White RD, et al: Does therapeutic hypothermia affect time to awakening in cardiac arrest survivors? Neurology 2011; 77:1346–1350 22. Alvarez V, Oddo M, Rossetti AO: Stimulus-induced rhythmic, periodic or ictal discharges (SIRPIDs) in comatose survivors of cardiac arrest: Characteristics and prognostic value. Clin Neurophysiol 2013; 124:204–208 23. Hirsch LJ, Claassen J, Mayer SA, et al: Stimulus-induced rhythmic, periodic, or ictal discharges (SIRPIDs): A common EEG phenomenon in the critically ill. Epilepsia 2004; 45:109–123 24. Hirsch LJ, LaRoche SM, Gaspard N, et al: American Clinical Neurophysiology Society’s Standardized Critical Care EEG Terminology: 2012 version. J Clin Neurophysiol 2013; 30:1–27 25. Rossetti AO, Oddo M, Liaudet L, et al: Predictors of awakening from postanoxic status epilepticus after therapeutic hypothermia. Neurology 2009; 72:744–749 26. Oksanen T, Tiainen M, Skrifvars MB, et al: Predictive power of serum NSE and OHCA score regarding 6-month neurologic outcome after out-of-hospital ventricular fibrillation and therapeutic hypothermia. Resuscitation 2009; 80:165–170 27. Shinozaki K, Oda S, Sadahiro T, et al: S-100B and neuron-specific enolase as predictors of neurological outcome in patients after cardiac arrest and return of spontaneous circulation: A systematic review. Crit Care 2009; 13:R121 28. Jacobs I, Nadkarni V, Bahr J, et al: Cardiac arrest and cardiopulmonary resuscitation outcome reports: Update and simplification of the Utstein templates for resuscitation registries: A statement for healthcare professionals from a task force of the International Liaison Committee on Resuscitation (American Heart Association, European Resuscitation Council, Australian Resuscitation Council, New Zealand Resuscitation Council, Heart and Stroke Foundation of Canada, InterAmerican Heart Foundation, Resuscitation Councils of Southern Africa). Circulation 2004; 110:3385–3397 29. Booth CM, Boone RH, Tomlinson G, et al: Is this patient dead, vegetative, or severely neurologically impaired? Assessing outcome for comatose survivors of cardiac arrest. JAMA 2004; 291:870–879 30. Lee YC, Phan TG, Jolley DJ, et al: Accuracy of clinical signs, SEP, and EEG in predicting outcome of hypoxic coma: A meta-analysis. Neurology 2010; 74:572–580 31. Thenayan EA, Savard M, Sharpe MD, et al: Electroencephalogram for prognosis after cardiac arrest. J Crit Care 2010; 25:300–304 32. Tiainen M, Roine RO, Pettilä V, et al: Serum neuron-specific enolase and S-100B protein in cardiac arrest patients treated with hypothermia. Stroke 2003; 34:2881–2886 33. Rundgren M, Karlsson T, Nielsen N, et al: Neuron specific enolase and S-100B as predictors of outcome after cardiac arrest and induced hypothermia. Resuscitation 2009; 80:784–789 34. van Putten MJ: The N20 in post-anoxic coma: Are you listening? Clin Neurophysiol 2012; 123:1460–1464 35. Young GB: Clinical practice. Neurologic prognosis after cardiac arrest. N Engl J Med 2009; 361:605–611 36. Josephson SA: Predicting neurologic outcomes after cardiac arrest: The crystal ball becomes cloudy. Ann Neurol 2010; 67:A5–A6 37. Rabinstein AA, Wijdicks EF: The value of EEG monitoring after cardiac arrest treated with hypothermia. Neurology 2012; 78:774–775 June 2014 • Volume 42 • Number 6
Feature Articles 38. Alvarez V, Sierra-Marcos A, Oddo M, et al: Yield of intermittent versus continuous EEG in comatose survivors of cardiac arrest treated with hypothermia. Crit Care 2013; 17:R190 39. Wijman CA, Mlynash M, Caulfield AF, et al: Prognostic value of brain diffusion-weighted imaging after cardiac arrest. Ann Neurol 2009; 65:394–402 40. Zandbergen EG, Hijdra A, Koelman JH, et al; PROPAC Study Group: Prediction of poor outcome within the first 3 days of postanoxic coma. Neurology 2006; 66:62–68
Critical Care Medicine
41. Fischer C, Luauté J, Némoz 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 42. 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 43. Steppacher I, Eickhoff S, Jordanov T, et al: N400 predicts recovery from disorders of consciousness. Ann Neurol 2013; 73: 594–602
www.ccmjournal.org
1347
