Original Article

Detection of Brain Activation in Unresponsive Patients with Acute Brain Injury

List of authors.
  • Jan Claassen, M.D.,
  • Kevin Doyle, M.A.,
  • Adu Matory, B.A.,
  • Caroline Couch, B.A.,
  • Kelly M. Burger, B.A., R.E.E.G.T.,
  • Angela Velazquez, M.D.,
  • Joshua U. Okonkwo, M.D.,
  • Jean-Rémi King, Ph.D.,
  • Soojin Park, M.D.,
  • Sachin Agarwal, M.D.,
  • David Roh, M.D.,
  • Murad Megjhani, Ph.D.,
  • Andrey Eliseyev, Ph.D.,
  • E. Sander Connolly, M.D.,
  • and Benjamin Rohaut, M.D.

Abstract

Background

Brain activation in response to spoken motor commands can be detected by electroencephalography (EEG) in clinically unresponsive patients. The prevalence and prognostic importance of a dissociation between commanded motor behavior and brain activation in the first few days after brain injury are not well understood.

Methods

We studied a prospective, consecutive series of patients in a single intensive care unit who had acute brain injury from a variety of causes and who were unresponsive to spoken commands, including some patients with the ability to localize painful stimuli or to fixate on or track visual stimuli. Machine learning was applied to EEG recordings to detect brain activation in response to commands that patients move their hands. The functional outcome at 12 months was determined with the Glasgow Outcome Scale–Extended (GOS-E; levels range from 1 to 8, with higher levels indicating better outcomes).

Results

A total of 16 of 104 unresponsive patients (15%) had brain activation detected by EEG at a median of 4 days after injury. The condition in 8 of these 16 patients (50%) and in 23 of 88 patients (26%) without brain activation improved such that they were able to follow commands before discharge. At 12 months, 7 of 16 patients (44%) with brain activation and 12 of 84 patients (14%) without brain activation had a GOS-E level of 4 or higher, denoting the ability to function independently for 8 hours (odds ratio, 4.6; 95% confidence interval, 1.2 to 17.1).

Conclusions

A dissociation between the absence of behavioral responses to motor commands and the evidence of brain activation in response to these commands in EEG recordings was found in 15% of patients in a consecutive series of patients with acute brain injury. (Supported by the Dana Foundation and the James S. McDonnell Foundation.)

Introduction

Digital Object ThumbnailQUICK TAKE
EEG Activity in Clinically Unresponsive Patients
 01:47

Clinically unresponsive patients can have electroencephalographic (EEG) or magnetic resonance imaging (MRI) evidence of brain activation in response to spoken commands.1-4 A meta-analysis has reported that 14% of chronically unresponsive patients may have a dissociation between behavior and brain activation (cognitive–motor dissociation5) months or years after injury.6 However, the prevalence and prognostic relevance of this dissociation, if detected in the days soon after brain injury, are not well understood. The absence of an ability to follow commands shortly after brain injury may have an effect on decisions regarding the withdrawal of life-sustaining therapies.7,8

We studied the prevalence and prognostic importance of brain activation detected by EEG in response to spoken commands to perform a motor task. We used a machine-learning technique3 to analyze EEG recordings obtained at the bedside in a prospective cohort of unresponsive patients with acute brain injury in a single intensive care unit (ICU).

Methods

Patients

From July 2014 through September 2017, we prospectively screened all patients who were admitted with acute brain injury to the neuroscience ICU of our hospital; screening was performed within 3 days after admission. Patients were screened for the absence of the ability to follow spoken commands — for example, “stick out your tongue” or “show me two fingers with your right hand” (details are provided in the Supplementary Appendix, available with the full text of this article at NEJM.org). In keeping with our routine practice and in accordance with guidelines regarding EEG monitoring of patients in the ICU,9 unresponsive patients either were monitored by continuous EEG or were anticipated to be connected to monitoring within 12 hours after screening, unless imminent death was expected. We enrolled all patients who were in a coma, vegetative state, or minimally conscious state–minus (defined as unresponsiveness with preserved visual fixation, visual pursuit, or localization to noxious stimuli); who had an acute brain injury of any type; and who were undergoing or were expected to undergo imminent continuous EEG monitoring. The presence of the minimally conscious state–minus was determined with the use of the Coma Recovery Scale–Revised (CRS-R,10 a six-dimension, 23-point scale of hierarchically arranged items [with no cutoff score used for enrollment]), which we assessed among patients who were not receiving deep sedation or neuromuscular blockade. The exclusion criteria were an age of less than 18 years, a preexisting disorder of consciousness before the onset of the acute brain injury that resulted in the current admission, pregnancy, deafness before the acute brain injury, clinical recovery of the ability to follow commands before enrollment, patients or families who did not want to participate in the study, or logistic reasons (details are provided in the Supplementary Appendix).

Patients, families, and treating physicians were unaware of the results of the EEG recordings, and these results were not made available to treating clinicians in relation to decisions regarding the withdrawal of care. Demographic data and data on complications that occurred during the hospital stay and on outcomes were prospectively collected. In addition, we recorded EEGs from 10 healthy volunteers with a mean age of 31 years, using the same EEG protocol as in the patients (see the Supplementary Appendix).

Study Oversight

The study was approved for patients and healthy volunteers by the local institutional review board. Written informed consent was obtained from the patients’ surrogates and from the healthy volunteers; patients who recovered consciousness were given the opportunity to withdraw from the study.

The first and last authors are responsible for the study design and drafting of the manuscript. There was no industry involvement in or support for the study. The authors vouch for the accuracy and completeness of the data and for the fidelity of the trial to the protocol, available at NEJM.org. The results are reported in accordance with the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines for reporting observational studies.11

Study Procedures

Daily neurologic examinations, including a clinical assessment of the ability or inability of the patient to follow spoken commands (“stick out your tongue,” “show me two fingers with your right hand,” and “wiggle your toes”), were performed during morning rounds, and the results were recorded.12 Each EEG assessment was preceded by a clinical examination that included the CRS-R10 in order to categorize the clinical state of consciousness at the time of the recording (details are provided in the Supplementary Appendix).

Functional outcome was assessed with the Glasgow Outcome Scale–Extended (GOS-E; levels range from 0 to 8, with higher levels indicating better outcomes), with data obtained in a structured telephone interview at 12 months after the injury.13,14 Both the patient and the interviewer who performed the outcome assessments were unaware of the results of the above-noted examinations during routine rounds and were unaware of the EEG categorization. Outcomes were dichotomized at a GOS-E level of 4, a level that signifies the ability to be left up to 8 hours during the day without assistance.

At the time of the EEG and clinical assessments, all patients were evaluated to ensure the absence of the following complications: seizures, hyperglycemia (serum glucose level, >11.1 mmol per liter [>200 mg per deciliter]), hyponatremia or hypernatremia (serum sodium level <133 and >150 mmol per liter, respectively), and renal or fulminant liver failure. For the daily neurologic assessment, sedated patients underwent interruption or reduction of sedation if it was deemed safe by the attending physician during rounds.15 Both the behavioral and the EEG assessments were performed during interruption of sedation whenever possible. To account for cases in which stopping sedation was unsafe, during the study we developed a post hoc method to explore the effect of sedative and analgesic medications on EEG responses by collecting information on the doses of administered medications at the time of the assessments as well as the cumulative doses received within the two preceding elimination half-lives of each agent.3 Sedation was categorized as “minimal” for discontinuous (e.g., single-push) administration and as “low” or “moderate” according to the cumulative doses administered through continuous drip during the two previous half-lives (details are provided in Table S3 in the Supplementary Appendix).

Motor Command Protocol

Motor Command Protocol and Data Processing.

Each block in the motor command protocol consisted of eight trials alternating between the instructions “keep opening and closing your right (left) hand” and “stop opening and closing your right (left) hand” (Panel A). The 10 seconds of electroencephalographic (EEG) recording after the instructions were given were extracted and segmented in five epochs, each 2 seconds long, for further analysis (Panel B). This procedure resulted in 480 epochs in patients (5 epochs×2 instructions×8 trials×6 blocks) and 240 epochs in controls (5 epochs×2 instructions×8 trials×3 blocks). Power spectral density (PSD) analysis was applied to the obtained EEG matrix in four frequency bands (δ [1 to 3 Hz], θ [4 to 7 Hz], α [8 to 13 Hz], and β [14 to 30 Hz]) (Panel C). The resulting features were used to train and test a support vector machine (SVM). The classification performance of the SVM for a given recording was assessed as the area under the receiver-operating-characteristic curve (AUC).

Spoken command instructions during EEG recording alternated between “keep opening and closing your right hand” and “stop opening and closing your right hand.”16 A total of six blocks each with eight consecutive trials of “keep opening …” and “stop opening …” commands were recorded (Figure 1A). For each patient, we recorded three blocks in which the patient was asked to move the right hand and three blocks in which the patient was asked to move the left hand (recordings of the right-hand and left-hand blocks were alternated; see the Supplementary Appendix). The total duration of the motor command session was approximately 25 minutes.

EEG Acquisition and Processing

Digital bedside EEG monitoring was performed with a standard 21-electrode montage.9 EEG recording quality (e.g., lead maintenance and movement artifact) was determined by bedside visual observation at the time of recording in addition to twice-daily lead maintenance by EEG technicians (see the Supplementary Appendix). For each EEG recording, power in predefined frequency ranges was calculated3,12,17 and used to train a machine-learning algorithm (support vector machine [SVM] with a linear kernel) to distinguish between the EEG responses that followed the commands “keep opening …” and “stop opening ….”

Statistical Analysis

The performance of the machine-learning algorithm for each EEG recording was estimated as the area under the receiver-operating-characteristic curve (AUC). To evaluate the significance of the AUC, a one-tailed permutation test was performed (training and evaluation of the classifier 500 times after random shuffling of the “keep opening …” and “stop opening …” commands18,19). Recordings were considered to show evidence of brain activation that was temporally concordant with spoken commands if the AUC was significantly greater than 0.5 (corresponding to the level that would be expected by chance), after application of the Benjamini–Hochberg false-discovery-rate method in cases of multiple recordings in a given patient.20,21 All EEG analyses were performed with the use of open-source packages, including MNE-Python (www.martinos.org/mne/stable/index.html)22 and Scikit-learn (http://scikit-learn.org/stable/index.html).23

Categorical variables were expressed as numbers and percentages and were compared with the use of Fisher’s exact or ordinal chi-square tests, as appropriate. Continuous variables were expressed as medians and interquartile ranges or as means and standard deviations, as appropriate, and were compared with the use of Wilcoxon signed-rank tests. All tests (other than the permutation test applied on the SVM output) were two-sided. Statistical analyses were performed with R statistical software, version 3.4.1 (R Project for Statistical Computing).24

Results

Patients and Volunteers

Characteristics of the 104 Patients. Enrollment and Follow-up.

The group of 126 patients who were excluded from the analysis because they followed commands before enrollment includes patients with reproducible movements in response to commands or with intentional communication defined according to the Coma Recovery Scale–Revised. Glasgow Outcome Scale–Extended (GOS-E) levels range from 1 to 8, with higher levels indicating better outcomes; a level of 4 indicates the ability to be left up to 8 hours during the day without assistance. CMD denotes cognitive–motor dissociation.

A total of 401 unresponsive patients with acute brain injury were screened, and 104 met the criteria for inclusion and were enrolled (Table 1 and Figure 2). Enrolled patients were similar to those who were not enrolled with respect to age, sex, and admission Glasgow Coma Scale score (Table S1 in the Supplementary Appendix).

The healthy volunteers all had EEG evidence of brain activation in response to motor commands. We obtained a total of 240 EEG recordings from the 104 patients (median number of recordings per patient, 2; interquartile range, 1 to 3) a median of 6 days (interquartile range, 3 to 10) after the injury. Of the 240 EEG recordings, 126 (52%) were acquired while patients were comatose, 54 (22%) while patients were in the vegetative state, and 60 (25%) while patients were in the minimally conscious state–minus category.

Comparison of Patients According to Cognitive–Motor Dissociation Status. Temporal Pattern in Healthy Volunteers and in Patients with and Patients without Cognitive–Motor Dissociation.

The y axis indicates the decoding prediction, based on the EEG response, that a given epoch corresponds to a “move” instruction (higher number) or a “rest” instruction (lower number). The graph is shown for descriptive purposes only; the displayed averaged decoding prediction curves are related to the AUCs used to diagnose cognitive–motor dissociation.25

Of the 104 patients, 16 (15%) had cognitive–motor dissociation detected on at least one recording. This dissociation was detected at a median of 4.0 days (interquartile range, 2.0 to 5.3) after admission to the ICU. Among these 16 patients, the causes of the acute brain injury were subarachnoid hemorrhage (5 patients), traumatic brain injury (3), intracerebral hemorrhage (4), cardiac arrest (2), neurosarcoidosis (1), and bupropion overdose (1) (Table 2). The group-classification performance of the machine-learning algorithm among patients and healthy volunteers over time is shown in Figure 3.

The condition in 8 (50%) of the patients with cognitive–motor dissociation improved such that they were able to follow the spoken commands used during the daily clinical assessment by the time of hospital discharge, 6.0 days (interquartile range, 4.5 to 8.3) after cognitive–motor dissociation was first documented by EEG (Table S2 in the Supplementary Appendix). The condition in 2 additional patients (12%) with cognitive–motor dissociation improved after hospital discharge such that they were able to follow the spoken commands used during the daily clinical assessment. In comparison, 26% of the patients who did not have cognitive–motor dissociation (23 patients) were able to follow commands before hospital discharge (median ICU day, 12.0; interquartile range, 9.5 to 19.5).

Outcomes

The GOS-E level at 12 months after the acute brain injury was obtained for 100 of the 104 patients (4 patients were lost to follow-up). A total of 7 of 16 patients (44%) with and 12 of 84 patients (14%) without cognitive–motor dissociation had a GOS-E level of 4 or greater (odds ratio, 4.6; 95% confidence interval [CI], 1.2 to 17.1), and 6 (38%) with cognitive–motor dissociation and 50 (60%) without cognitive–motor dissociation were dead at 12 months. Among the 6 patients with cognitive–motor dissociation who were dead at 12 months, 4 had died in the context of withdrawal of life-sustaining therapy (Table 2, and Table S2 in the Supplementary Appendix). After the 28 patients who underwent withdrawal of life-sustaining therapy were removed from the analysis, cognitive–motor dissociation remained predictive of a GOS-E level of 4 or greater (odds ratio, 5.4; 95% CI, 1.2 to 26.0).

Discussion

We found that 15% of patients in a consecutive series of patients with acute brain injury who were clinically unresponsive — some of whom had motor localization to pain stimuli, visual fixation, or visual tracking — had evidence of brain activation in response to spoken motor commands, as determined on the basis of EEG activity. This dissociation between behavior and EEG responses to spoken motor commands has been referred to as cognitive–motor dissociation.5 In our study, this state was seen more frequently in patients with trauma or brain hemorrhages than in patients with hypoxic–ischemic injury, but it was detected in patients with other acute brain injuries and, in some instances, in patients who had been lightly sedated. The frequency of cognitive–motor dissociation and the prognostic associations in our single-center descriptive study require validation in larger, multicenter studies that are powered to detect differences in long-term outcomes.

Our findings support those of previous studies that have shown that EEG or functional MRI can in some cases reflect activation of parts of the brain in response to spoken commands in unresponsive patients,1,4,26 but whether the detected signal represents recognition or comprehension of commands is uncertain. Patients who had an EEG response to spoken commands more often had later recovery than those who did not have this pattern. It is possible that these patients had overall greater functional integrity of the brain stem, thalamus, and cortex and of the connections among these structures, similar to findings in previous studies in which metabolic measurements such as fluorodeoxyglucose positron-emission tomography were used.27

In chronic brain injury, cognitive–motor dissociation has been studied most often in patients with traumatic brain injury2,4,16,17,28,29 and has been estimated to have a prevalence of 14%.6 Our study shows that cognitive–motor dissociation can be detected in the ICU early after brain injury in a similar percentage of patients.3 The use of methods such as functional MRI to detect cognitive–motor dissociation may result in more frequent detection than EEG, but functional MRI is challenging to perform in a critical care setting.30

Limitations of our study include the varied causes of brain injuries among the patients. In addition, the withdrawal of life-sustaining therapies confounds studies of the natural history of acute brain injury.31 However, even after the exclusion of patients who underwent withdrawal of life-sustaining therapy, our study continued to show a difference in long-term functional outcomes between patients with and patients without cognitive–motor dissociation. We did not perform in-person 12-month follow-up assessments but recorded functional outcomes (GOS-E) by telephone. Sedation is a potential confounder in the classification of patients as being in a comatose, vegetative, or minimally conscious state; however, we were able to detect brain activation in response to motor commands in some lightly sedated patients.

In conclusion, early after brain injury, 15% of clinically unresponsive patients who did not follow commands had EEG evidence of brain activation in response to spoken motor commands recorded at the bedside in the ICU.

Funding and Disclosures

Supported by the Dana Foundation (to Dr. Claassen) and the James S. McDonnell Foundation. The study was conceptualized with the use of instruments that were developed and standardized as part of an international collaborative funded by James S. McDonnell Foundation (principal investigator, Nicholas D. Schiff; details are provided in the Supplementary Appendix). Dr. Rohaut received postdoctoral grants from Amicale des Anciens Internes des Hôpitaux de Paris and Syndicat des Chefs de Cliniques et Assistants des Hôpitaux de Paris, Assistance Publique–Hôpitaux de Paris, and the Philippe Foundation.

Disclosure forms provided by the authors are available with the full text of this article at NEJM.org.

This article was updated on June 27, 2019, at NEJM.org.

A data sharing statement provided by the authors is available with the full text of this article at NEJM.org.

We thank the nurses, attending physicians, fellows, and neurology and neurosurgery residents of the Neuroscience ICU and Epilepsy Division for their overall support of this project; Denis A. Engemann and Federico Raimondo for their support with MNE-Python; Amelia K. Boehme for statistical advice; Jennifer A. Egbebike and Anna A. Calderon for additional data collection; and Nicholas D. Schiff for providing the auditory recording of spoken commands.

Author Affiliations

From the Departments of Neurology (J.C., K.D., A.M., C.C., K.M.B., A.V., J.U.O., S.P., S.A., D.R., M.M., A.E., B.R.) and Neurosurgery (E.S.C.), Columbia University, and the Department of Psychology, New York University (J.-R.K.) — both in New York.

Address reprint requests to Dr. Claassen at the Neurological Institute, Columbia University, 177 Fort Washington Ave., MHB 8 Center, Rm. 300, New York, NY 10032, or at .

Supplementary Material

References (31)

  1. 1. Owen AM, Coleman MR, Boly M, Davis MH, Laureys S, Pickard JD. Detecting awareness in the vegetative state. Science 2006;313:1402-1402.

  2. 2. Goldfine AM, Victor JD, Conte MM, Bardin JC, Schiff ND. Determination of awareness in patients with severe brain injury using EEG power spectral analysis. Clin Neurophysiol 2011;122:2157-2168.

  3. 3. Edlow BL, Chatelle C, Spencer CA, et al. Early detection of consciousness in patients with acute severe traumatic brain injury. Brain 2017;140:2399-2414.

  4. 4. Monti MM, Vanhaudenhuyse A, Coleman MR, et al. Willful modulation of brain activity in disorders of consciousness. N Engl J Med 2010;362:579-589.

  5. 5. Schiff ND. Cognitive motor dissociation following severe brain injuries. JAMA Neurol 2015;72:1413-1415.

  6. 6. Kondziella D, Friberg CK, Frokjaer VG, Fabricius M, Møller K. Preserved consciousness in vegetative and minimal conscious states: systematic review and meta-analysis. J Neurol Neurosurg Psychiatry 2016;87:485-492.

  7. 7. Turgeon AF, Lauzier F, Simard J-F, et al. Mortality associated with withdrawal of life-sustaining therapy for patients with severe traumatic brain injury: a Canadian multicentre cohort study. CMAJ 2011;183:1581-1588.

  8. 8. Elmer J, Torres C, Aufderheide TP, et al. Association of early withdrawal of life-sustaining therapy for perceived neurological prognosis with mortality after cardiac arrest. Resuscitation 2016;102:127-135.

  9. 9. Claassen J, Taccone FS, Horn P, Holtkamp M, Stocchetti N, Oddo M. Recommendations on the use of EEG monitoring in critically ill patients: consensus statement from the neurointensive care section of the ESICM. Intensive Care Med 2013;39:1337-1351.

  10. 10. Giacino JT, Kalmar K, Whyte J. The JFK Coma Recovery Scale-Revised: measurement characteristics and diagnostic utility. Arch Phys Med Rehabil 2004;85:2020-2029.

  11. 11. von Elm E, Altman DG, Egger M, Pocock SJ, Gøtzsche PC, Vandenbroucke JP. The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement: guidelines for reporting observational studies. Int J Surg 2014;12:1495-1499.

  12. 12. Claassen J, Velazquez A, Meyers E, et al. Bedside quantitative electroencephalography improves assessment of consciousness in comatose subarachnoid hemorrhage patients. Ann Neurol 2016;80:541-553.

  13. 13. Jennett B, Snoek J, Bond MR, Brooks N. Disability after severe head injury: observations on the use of the Glasgow Outcome Scale. J Neurol Neurosurg Psychiatry 1981;44:285-293.

  14. 14. Lu J, Marmarou A, Lapane K, Turf E, Wilson L. A method for reducing misclassification in the extended Glasgow Outcome Score. J Neurotrauma 2010;27:843-852.

  15. 15. Oddo M, Crippa IA, Mehta S, et al. Optimizing sedation in patients with acute brain injury. Crit Care 2016;20:128-128.

  16. 16. Curley WH, Forgacs PB, Voss HU, Conte MM, Schiff ND. Characterization of EEG signals revealing covert cognition in the injured brain. Brain 2018;141:1404-1421.

  17. 17. Cruse D, Chennu S, Chatelle C, et al. Bedside detection of awareness in the vegetative state: a cohort study. Lancet 2011;378:2088-2094.

  18. 18. Noirhomme Q, Lesenfants D, Gomez F, et al. Biased binomial assessment of cross-validated estimation of classification accuracies illustrated in diagnosis predictions. Neuroimage Clin 2014;4:687-694.

  19. 19. Good P. Permutation, parametric, and bootstrap tests of hypotheses. 3rd ed. New York: Springer Science+Business Media, 2005.

  20. 20. Ojala M, Garriga GC. Permutation tests for studying classifier performance. J Mach Learn Res 2010;11:1833-1863.

  21. 21. Benjamini Y, Hochberg Y. Controlling the false discovery rate: a practical and powerful approach to multiple testing. J R Stat Soc [B] 1995;57:289-300.

  22. 22. Gramfort A, Luessi M, Larson E, et al. MEG and EEG data analysis with MNE-Python. Front Neurosci 2013;7:267-267.

  23. 23. Pedregosa F, Varoquaux G, Gramfort A, et al. Scikit-learn: machine learning in Python. J Mach Learn Res 2011;12:2825-2830.

  24. 24. R Development Core Team. R: a language and environment for statistical computing. R Project for Statistical Computing. 2017 (http://www.R-project.org/).

  25. 25. Kriegeskorte N, Simmons WK, Bellgowan PSF, Baker CI. Circular analysis in systems neuroscience: the dangers of double dipping. Nat Neurosci 2009;12:535-540.

  26. 26. Naccache L. Psychology: is she conscious? Science 2006;313:1395-1396.

  27. 27. Stender J, Mortensen KN, Thibaut A, et al. The minimal energetic requirement of sustained awareness after brain injury. Curr Biol 2016;26:1494-1499.

  28. 28. Giacino JT, Katz DI, Schiff ND, et al. Practice guideline update recommendations summary: disorders of consciousness: report of the Guideline Development, Dissemination, and Implementation Subcommittee of the American Academy of Neurology; the American Congress of Rehabilitation Medicine; and the National Institute on Disability, Independent Living, and Rehabilitation Research. Neurology 2018;91:450-460.

  29. 29. Cruse D, Chennu S, Chatelle C, et al. Relationship between etiology and covert cognition in the minimally conscious state. Neurology 2012;78:816-822.

  30. 30. Weijer C, Bruni T, Gofton T, et al. Ethical considerations in functional magnetic resonance imaging research in acutely comatose patients. Brain 2016;139:292-299.

  31. 31. Rohaut B, Claassen J. Decision making in perceived devastating brain injury: a call to explore the impact of cognitive biases. Br J Anaesth 2018;120:5-9.

Citing Articles (182)

    Figures/Media

      Digital Object ThumbnailQUICK TAKE
      EEG Activity in Clinically Unresponsive Patients
       01:47

    1. Motor Command Protocol and Data Processing.
      Motor Command Protocol and Data Processing.

      Each block in the motor command protocol consisted of eight trials alternating between the instructions “keep opening and closing your right (left) hand” and “stop opening and closing your right (left) hand” (Panel A). The 10 seconds of electroencephalographic (EEG) recording after the instructions were given were extracted and segmented in five epochs, each 2 seconds long, for further analysis (Panel B). This procedure resulted in 480 epochs in patients (5 epochs×2 instructions×8 trials×6 blocks) and 240 epochs in controls (5 epochs×2 instructions×8 trials×3 blocks). Power spectral density (PSD) analysis was applied to the obtained EEG matrix in four frequency bands (δ [1 to 3 Hz], θ [4 to 7 Hz], α [8 to 13 Hz], and β [14 to 30 Hz]) (Panel C). The resulting features were used to train and test a support vector machine (SVM). The classification performance of the SVM for a given recording was assessed as the area under the receiver-operating-characteristic curve (AUC).

    2. Characteristics of the 104 Patients.*
      Characteristics of the 104 Patients.
    3. Enrollment and Follow-up.
      Enrollment and Follow-up.

      The group of 126 patients who were excluded from the analysis because they followed commands before enrollment includes patients with reproducible movements in response to commands or with intentional communication defined according to the Coma Recovery Scale–Revised. Glasgow Outcome Scale–Extended (GOS-E) levels range from 1 to 8, with higher levels indicating better outcomes; a level of 4 indicates the ability to be left up to 8 hours during the day without assistance. CMD denotes cognitive–motor dissociation.

    4. Comparison of Patients According to Cognitive–Motor Dissociation Status.*
      Comparison of Patients According to Cognitive–Motor Dissociation Status.
    5. Temporal Pattern in Healthy Volunteers and in Patients with and Patients without Cognitive–Motor Dissociation.
      Temporal Pattern in Healthy Volunteers and in Patients with and Patients without Cognitive–Motor Dissociation.

      The y axis indicates the decoding prediction, based on the EEG response, that a given epoch corresponds to a “move” instruction (higher number) or a “rest” instruction (lower number). The graph is shown for descriptive purposes only; the displayed averaged decoding prediction curves are related to the AUCs used to diagnose cognitive–motor dissociation.25