Neurocrit Care

https://doi.org/10.1007/s12028-023-01816-z

ORIGINAL WORK

Multimodal Prediction of 3‑ and 12‑Month

Outcomes in ICU Patients with Acute Disorders
of Consciousness
Moshgan Amiri1, Federico Raimondo2,3 , Patrick M. Fisher4,5, Melita Cacic Hribljan6†, Annette Sidaros1,6†,
Marwan H. Othman1, Ivan Zibrandtsen1,6, Ove Bergdal7, Maria Louise Fabritius8, Adam Espe Hansen9,10,
Christian Hassager10,11, Joan Lilja S. Højgaard1, Helene Ravnholt Jensen8, Niels Vendelbo Knudsen8,
Emilie Lund Laursen1, Jacob E. Møller11, Vardan Nersesjan12, Miki Nicolic6, Sigurdur Thor Sigurdsson8,
Jacobo D. Sitt13, Christine Sølling8, Karen Lise Welling8 , Lisette M. Willumsen7, John Hauerberg7,
Vibeke Andrée Larsen9, Martin Ejler Fabricius6,10, Gitte Moos Knudsen4,10, Jesper Kjærgaard10,11,
Kirsten Møller8,10 and Daniel Kondziella1,10*

© 2023 The Author(s)

Abstract
Background: In intensive care unit (ICU) patients with coma and other disorders of consciousness (DoC), outcome
prediction is key to decision-making regarding prognostication, neurorehabilitation, and management of family
expectations. Current prediction algorithms are largely based on chronic DoC, whereas multimodal data from acute
DoC are scarce. Therefore, the Consciousness in Neurocritical Care Cohort Study Using Electroencephalography and
Functional Magnetic Resonance Imaging (i.e. CONNECT-ME; ClinicalTrials.gov identifier: NCT02644265) investigates
ICU patients with acute DoC due to traumatic and nontraumatic brain injuries, using electroencephalography (EEG)
(resting-state and passive paradigms), functional magnetic resonance imaging (fMRI) (resting-state) and systematic
clinical examinations.
Methods: We previously presented results for a subset of patients (n = 87) concerning prediction of consciousness
levels in the ICU. Now we report 3- and 12-month outcomes in an extended cohort (n = 123). Favorable outcome was
defined as a modified Rankin Scale score ≤ 3, a cerebral performance category score ≤ 2, and a Glasgow Outcome
Scale Extended score ≥ 4. EEG features included visual grading, automated spectral categorization, and support vec-
tor machine consciousness classifier. fMRI features included functional connectivity measures from six resting-state
networks. Random forest and support vector machine were applied to EEG and fMRI features to predict outcomes.
Here, random forest results are presented as areas under the curve (AUC) of receiver operating characteristic curves
or accuracy. Cox proportional regression with in-hospital death as a competing risk was used to assess independent
clinical predictors of time to favorable outcome.

*Correspondence: [email protected]
†
Melita Cacic Hribljan and Annette Sidaros have contributed equally to
this work.
1
Department of Neurology, Copenhagen University Hospital -
Rigshospitalet, Blegdamsvej 9, 2100 Copenhagen, Denmark
Full list of author information is available at the end of the article
 Results: Between April 2016 and July 2021, we enrolled 123 patients (mean age 51 years, 42% women). Of 82 (66%)
ICU survivors, 3- and 12-month outcomes were available for 79 (96%) and 77 (94%), respectively. EEG features pre-
dicted both 3-month (AUC 0.79 [95% confidence interval (CI) 0.77–0.82]) and 12-month (AUC 0.74 [95% CI 0.71–0.77])
outcomes. fMRI features appeared to predict 3-month outcome (accuracy 0.69–0.78) both alone and when combined
with some EEG features (accuracies 0.73–0.84) but not 12-month outcome (larger sample sizes needed). Independent
clinical predictors of time to favorable outcome were younger age (hazard ratio [HR] 1.04 [95% CI 1.02–1.06]), trau-
matic brain injury (HR 1.94 [95% CI 1.04–3.61]), command-following abilities at admission (HR 2.70 [95% CI 1.40–5.23]),
initial brain imaging without severe pathological findings (HR 2.42 [95% CI 1.12–5.22]), improving consciousness in
the ICU (HR 5.76 [95% CI 2.41–15.51]), and favorable visual-graded EEG (HR 2.47 [95% CI 1.46–4.19]).
Conclusions: Our results indicate that EEG and fMRI features and readily available clinical data predict short-term
outcome of patients with acute DoC and that EEG also predicts 12-month outcome after ICU discharge.
Keywords: Coma, Consciousness, Functional magnetic resonance imaging, Electroencephalography, Intensive care
unit

Introduction in patients with DoC that are not apparent through

Since the first report in 2006 of a patient with cogni- clinical examinations [10]. Task-based fMRI and EEG
tive motor dissociation [1], the challenge of identifying paradigms (i.e., active paradigms) are highly specific
patients with brain injury with residual consciousness in identifying these patients [18, 19], but they may not
and predicting their long-term recovery has stimulated always accurately detect residual consciousness because
a new field of research. This, however, mostly concerns of insufficient arousal levels, lack of sustained attention,
patients with subacute or chronic disorders of conscious- and fluctuating awareness, which is particularly chal-
ness (DoC) in rehabilitation facilities [2, 3]. lenging with patients with acute DoC. Therefore, fMRI
Each year 2 of 1000 people fall into a coma and are and EEG paradigms that involve resting-state (i.e.,
admitted to an intensive care unit (ICU) [4], with the key patients receive no stimulations) or passive paradigms
questions being: Who regains consciousness, and who (e.g., addressing the patient by their name or passive
will make a good functional outcome? Accurate predic- eye opening) may be more suitable when assessing
tion of long-term functional outcomes of patients with patients with acute DoC.
acute DoC, including coma, is a major challenge, espe- In previous work, we established that resting-state
cially during the early phase in the ICU [5]. Although EEG, EEG with external stimulations, and resting-state
some DoC survivors enter a state of prolonged unrespon- fMRI can accurately predict consciousness levels in
sive wakefulness, many recover within weeks to months, patients with acute DoC during ICU admission [20]. Cor-
and a few patients with DoC may show signs of recovery roborating our findings, multimodal approaches were
even years after their brain injury [3, 6]. Accurate prog- recommended in a recent review of neuroimaging-based
nostication is hence essential for decision-making in the outcome prediction of patients with DoC [21]. However,
ICU, including decisions about therapeutic management, prognostication of functional recovery of acute DoC is
withdrawal of life-sustaining therapy [7–9], resource typically limited to unimodal approaches and certain
allocation and rehabilitation, and management of fam- patient subcategories [6, 14, 19, 22]. Only one study
ily expectations. The first step to improve prognostica- reported 6-month outcome of patients with acute DoC
tion of patients with acute DoC is accurate determination with severe traumatic brain injury (TBI) assessed with
of their levels of consciousness [10]. This is important both EEG and fMRI [23]. Research reporting the poten-
because patients with even minimal clinical signs of tial of multimodal approaches to predict both early and
residual consciousness [11, 12] have more favorable long- late functional outcomes of patients with acute DoC in
term outcomes (as do those with covert consciousness the ICU across a wide range of brain injuries is, to our
[13–15]). However, determining consciousness levels knowledge, nonexistent.
by routine clinical examinations alone is imprecise [16] To bridge this knowledge gap, we investigated whether
because intermittent signs of consciousness are often a multimodal approach consisting of EEG with resting
missed when sensitive systematic ratings scales are omit- and passive stimulation paradigms, resting-state fMRI,
ted [3, 10, 17]. and repeated systematic clinical evaluations could accu-
Advanced methods such as functional magnetic rately predict functional outcomes of patients with acute
resonance imaging (fMRI) and electroencephalogra- DoC with TBI and various nontraumatic brain injuries 3
phy (EEG) can reveal covert signs of consciousness and 12 months after ICU discharge.
Methods systematic clinical approach that included sub-elements
The Consciousness in Neurocritical Care Cohort Study of the Coma Recovery Scale Revised [17], with the addi-
using EEG and fMRI (CONNECT-ME) (ClinicalTrials. tion of the Glasgow Coma Scale (GCS) [25], and the
gov identifier: NCT02644265) is a tertiary-center pro- Full Outline of Unresponsiveness [26]. Furthermore,
spective, observational diagnostic phase IIb cohort study. daily routine neurological examinations were performed
Detailed methods of data acquisition and analysis are by the attending team of physicians, and results were
described in the study protocol [24] and a recent article accessed from the electronic health records. We classified
[20]. Results concerning the prediction of consciousness patients according to their level of consciousness into the
levels in a subset of patients (n = 87) at ICU discharge following categories, applying clinical examination tech-
have been published elsewhere [20]. Here, we evaluated niques as previously described in detail [20]:
3- and 12-month functional outcomes in an extended
patient cohort (n = 123). In the following sections, we •  Coma [27, 28]
provide a brief overview of the methods. Figure 1 shows •  Unresponsive wakefulness syndrome (UWS) [29]:
the flow of patients through the study. only reflex behavior, such as spontaneous eye open-
ing
Patients and Study Design •  Minimally conscious state (MCS − / +) [30, 31]:
We prospectively included patients admitted to one of MCS − , definite signs of nonreflex behavior, such
the four ICUs (excluding the neonatal ICU) at Rigshos- as visual pursuit, localization to noxious stimuli, or
pitalet (Campus Blegdamsvej), Copenhagen University relevant emotional response; MCS + , ability to fol-
Hospital, Copenhagen, Denmark, between April 2016 low simple commands repeatedly but not necessarily
and July 2021 and collected demographics, clinical status, consistently
and data regarding previous medical history. We included •  Emergence from MCS [32]: reliable two-way com-
ICU patients aged ≥ 16 years with acute DoC (time from munication or functional object use
brain injury < 31 days) and Danish- or English-language •  Locked-in syndrome (LIS) [33]: consistent and reli-
proficiency who had a clinical indication for structural able communication by rudimentary eye opening
brain magnetic resonance imaging (MRI) ordered by the
treating physician. Clinical examinations, EEG, and fMRI EEG
were all performed at the time of enrollment into the Standard 19/25 channel bedside video-EEG (NicoletOne,
study and within a 24-h window or as close to this time Natus Medical Inc., Middleton, WI) was recorded with
window as possible. We aimed for unsedated patients or electrodes placed according to the international 10/20
for the lowest possible sedation levels if patients could not system [34]. All EEGs contained a 10-min resting-state
be fully weaned from sedation. As previously described segment, and for reactivity assessment, a segment with
[20], sedation levels were graded as “none or minimal” stimulations included eye opening; calling the patient
(i.e., absence of intravenous fentanyl, remifentanil, propo- by their name; noxious stimuli applied as pressure to the
fol, midazolam, sodium thiopental, or sevoflurane), “low earlobes, fingertips, and sternum; and sensory tactile
to moderate” (fentanyl < 500 µg/h or < 200 µg/h combined stimuli applied with a cotton swap to the nostrils.
with propofol, remifentanil < 1,000 µg/h or < 250 µg/h EEGs were assessed in three different ways, as
combined with propofol, propofol < 100 mg/h, mida- described previously [20]:
zolam < 10 mg/h, sevoflurane < 3%) or “high or very
high” (propofol ≥ 100 mg/h, fentanyl ≥ 500 µg/h 1. Manual visual analysis by two experienced board-
or ≥ 200 µg/h combined with propofol, remifenta- certified neurophysiologists (Melita Cacic Hribljan
nil ≥ 1,000 µg/h or ≥ 250 µg/h combined with propofol, and Annette Sidaros; disagreements resolved by
midazolam ≥ 10 mg/h, sevoflurane ≥ 3%, or any dosage Martin Ejler Fabricius) scoring the EEGs according to
of sodium thiopental). Patients with contraindications for the Synek scale [35] (level I to V with increasing level
MRI, major premorbid neurological deficits (e.g., men- indicating increasing pathology)
tal retardation, aphasia, or deafness), and/or acute life- 2. ABCD spectral analysis by MA and IZ (disagree-
threatening conditions with immediate risk of clinical ments resolved by DK) as described by Forgacs et al.
deterioration were excluded. [36] (with category A indicating complete cortico-
thalamic disruption and category D indicating full
Classification of Consciousness Levels recovery of corticothalamic circuit; segments not
Briefly, consciousness levels were determined at the clearly falling under A, B, C, or D were classified as
time of study enrollment and at ICU discharge using a “non-ABCD”)
Fig. 1 Study flowchart, data assessment strategy and death in ICU. A. A total of 123 patients with acute DoC were included, of whom 41 died
during ICU admission. Of the 82 patients discharged alive, 10 (12%) patients were discharged directly to their own home, 20 (24%) to other care
facilities such as nursing homes, and the remaining 52 (63%) to a high-level neurorehabilitation facility. Three-month follow-up data was available
from 79 (96%) patients, and 12-month follow-up data from 77 (94%) patients. B. Full sets of 3- and 12-month follow-up data were available for 77
(94%) patients. EEG recordings were available from all patients (blue box), while fMRI resting-state sequences were available from 45 (58%) patients
(purple box). EEGs were analyzed with three different approaches; (1) visual manual analysis and scoring according to the Synek scale, (2) auto-
mated spectral analysis according to the ABCD model, and (3) a machine learning based SVM consciousness classifier resulting in the probability of
being at least in a minimal conscious state (P(MCS)) and 68 EEG markers derived from segments of resting-state EEG (EEG markers-r). Two different
machine learning algorithms (i.e., random forest and SVM) were used to conduct seven different predictive models based on EEG features (i.e.,
models I to VIII) and three different models including fMRI features with or without EEG features (i.e., models IX to XI). Models including fMRI features
were assessed with additional LOO-CV procedure due to the limited number of available samples. C. This part depicts the proportion of patients in
coma or UWS who either awoke to at least MCS- (i.e., regained consciousness to some degree) or died during ICU admission. At time 0 (admission
to the ICU) none of the patients were awake (0%) and all were alive (100%). The red line shows the proportion of patients who died in the ICU,
and the green line shows the proportion of patients who awoke from coma or UWS in the ICU. During ICU admission, a total of 41 patients (33%)
died, while 82 (67%) survived, of whom 73 (59%) awoke prior to ICU discharge. The area between the red and green line indicates the proportion
of patients (7%) who remained in coma or UWS at ICU discharge. aIncluding eight patients who died prior to 3-month follow-up; bIncluding 13
patients who died prior to 12-month follow-up; * all EEG models were also tested with same-sample data for head-to-head comparison (see also
Table 3). Abbreviations: ICU = intensive care unit, EEG = electroencephalography, fMRI = functional magnetic resonance imaging, SVM = support
vector machine, DMN = default mode network, SN = salience network, FPN = frontoparietal network, AN =auditory network, SMN = somatosen-
sory network, VN = visual network, LOO-CV = leave-one-out cross-validation, FC = functional connectivity, UWS = unresponsive wakefulness
syndrome.
 3. A support vector machine (SVM)–based con- recovery, as our study cohort consists of a heterogene-
sciousness classifier [37] predicting the probabil- ous group of patients regarding the cause of brain injury
ity of the patient’s consciousness level being at least (i.e., stroke, TBI, cardiac arrest, and other neurologi-
MCS − (P(MCS)) from 68 EEG markers derived cal and medical causes). Functional outcome was deter-
separately from EEG resting segments and segments mined from electronic health records typically based on
with stimulations. structural assessments by experienced nursing staff at the
high-level rehabilitation facility most surviving patients
All investigators performing EEG visual and spectral were discharged to. If sufficient data were not available
analyses were unaware of patient outcomes. For other from health records, patients, family members, or other
details regarding data cleaning, preprocessing, and spe- caregivers were contacted by telephone. Favorable out-
cific EEG features included in the aforementioned meth- come was defined as a combination of an mRS score ≤ 3
ods, please refer to our previous publication [20]. (indicating that patients can walk unassisted), a GOS-E
score ≥ 4 (indicating that patients can take care of them-
fMRI selves alone for at least 8 h at home), and a CPC score ≤ 2
A 10-min resting-state scan session with a T2*-weighted (indicating that patients are conscious and independent
echo-planar imaging blood oxygen level–dependent of others for activities of daily living). Patients who died
fMRI sequence was performed on 1.5- or 3-Tesla MRI after hospital discharge were included in the primary out-
scanners (Siemens, Erlangen, Germany) with 20- or come analysis, whereas patients who died during the ICU
64-channel head coils, respectively. Preprocessing of stay were excluded, as were patients lost to follow-up.
fMRI data was performed using SPM12 in MATLAB
v2019a (https://​www.​fil.​ion.​ucl.​ac.​uk/​spm/​softw​are/​ Machine Learning Algorithms and Predictive Models
spm12/) according to our previously described method Two machine learning algorithms were used: random
[20]. Briefly, denoised regional time series were extracted, forest and SVM. This ensured exploiting both linear
and region-to-region functional connectivity was esti- and nonlinear interactions. Algorithms were trained to
mated by calculating the timewise correlation coefficient predict binary outcome at 3- and 12-months’ follow-
(Pearson’s ρ) between each pair of regional time series up. Model performance was estimated using stratified
and applying Fisher’s r-to-z transformation to the corre- fivefold cross-validation (repeated ten times). A special
lation coefficient. A total of 21 within- and between-net- cross-validation scheme (leave-one-out cross-validation
work functional connectivity measures were calculated [LOO-CV] [41]) was used to evaluate the potential of
as the average functional connectivity across the set of fMRI features, as the limited fMRI samples available
region-to-region pairs for six resting-state networks (i.e., from patients with follow-up outcome did not allow us
the default mode network, frontoparietal network, audi- to obtain reliable estimates with fivefold cross-validation.
tory network, salience network, sensorimotor network, Algorithm hyperparameters were selected using nested
and visual network). Investigators assessing fMRI data cross-validation and a grid-search procedure. Both uni-
were unaware of patient outcomes. modal models based on single features (EEG or fMRI fea-
tures) and multimodal models based on a combination
Follow‑up Data of several features (e.g., combination of EEG and fMRI
We used three outcome scales to assess functional out- features) were developed with main outcome measures
come at 3 and 12 months after ICU discharge: (1) the as binary targets. Furthermore, we conducted a clinical
modified Rankin Scale (mRS) [38], (2) the Glasgow Out- model based on the following clinical characteristics: (1)
come Scale Extended (GOS-E) [39], and (3) the cerebral improving consciousness levels during the ICU stay (i.e.,
performance category (CPC) [40] (Box S1). The mRS is higher level of consciousness at ICU discharge compared
used for evaluation of recovery in stroke patients with with study enrollment), (2) sex, (3) age at admission,
a focus on the patient’s ability to walk with or without (4) preadmission comorbidities, (5) TBI as the cause of
assistance [38]. The GOS-E is an overall functional out- injury leading to admission, and (6) command-follow-
come scale frequently used to collect follow-up data of ing abilities at admission. In total, 12 different predic-
patients with TBI and includes other aspects of func- tive models (I–XII) were developed and tested with each
tional recovery, such as the ability to work and socialize algorithm based on EEG, fMRI, and clinical features.
and the level of emotional deficits [39]. Finally, the CPC Same-sample models were tested for head-to-head
is an evaluation tool to assess the level of recovery of car- comparison of EEG features and clinical features but
diac arrest patients, with regaining of consciousness con- could not be tested with fMRI features because of the low
sidered a main aspect [40]. By including all three scales, number of available patients with the full set of EEG fea-
we aimed at evaluating different aspects of functional tures, fMRI features, and outcome measures. Prediction
performance of models evaluated with fivefold cross-val- (File-nr.:H-16040845). Written consent was waived
idation were assessed with area under the curve (AUC) because all data were acquired during routine clinical
of the receiver operating characteristic curve, sensitiv- workup. CONNECT-ME is registered with ClinicalTrials.
ity, and positive predictive value (PPV), whereas perfor- org (identifier: NCT02644265).
mance of the LOO-CV models including fMRI features
was assessed with the accuracy measure (ratio of cor- Results
rectly predicted samples over total samples). Pairwise Demographics and Clinical Characteristics
comparison of the AUC of same-sample EEG and clini- We included 123 patients (mean age 51 ± 19 years; 51
cal models were performed using the corrected t-test [42%] women), of whom 82 (67%) were discharged alive
(two-sided) for comparing machine learning models from the ICU (Fig. 1 and Table 1). Of the 41 deaths in
[42]. P values were corrected for multiple comparisons the ICU, 37 (90%) occurred after withdrawal of life-sus-
using Bonferroni. AUC, sensitivity, and PPV estimates taining therapy. Advanced age, preadmission comorbid-
are reported as mean (95% confidence interval [CI]), ity, cardiac arrest as the cause of ICU admission (odds
and accuracies are reported as a number between 0 and ratio [95% CI] 10.4 [2.46–78.3]), lower GCS motor
1. The models hence predict the precision with which score at admission, lower total GCS and Full Outline
favorable outcomes can be distinguished from unfavora- of Unresponsiveness scores at study enrollment, lower
ble outcomes. All machine learning analyses were done consciousness levels at study enrollment, and shorter
using Julearn and scikit-learn [43]. duration of ICU admission were all significantly associ-
ated with death in the ICU (all P < 0.05; Table 1). EEG
Outcome Measures was available from 122 (99%) patients, whereas fMRI
Our primary target outcome was the binary outcome at was available from 67 (54%) patients. The proportion of
3- and 12-months’ follow-up. Time to favorable outcome patients with fMRI did not differ between those who died
was considered a secondary outcome. in the ICU and patients discharged alive.

Statistical Analysis Functional Outcome and Time to Favorable Outcome

Quantitative data are expressed as mean ± SD or Of the 82 patients discharged alive from the ICU, func-
median (interquartile range), and group comparisons tional outcome was available from 79 (96%) at 3 months
were assessed with Student’s t-test, the Mann–Whitney and from 77 (94%) at 12 months (Fig. 1). Thirteen
U-test, or the Kruskal–Wallis test. Categorical data are patients (16%) died prior to the 12-month follow-up, of
expressed as numbers (percentages) and were compared whom eight were dead by the 3-month follow-up. Of the
using the χ2 test or Fisher’s exact test. Cox proportional 79 patients with 3-month follow-up data, 26 (33%) had an
hazards regression model with in-hospital death consid- mRS score of ≤ 3, 24 (30%) had a CPC score of ≤ 2, and
ered as a competing risk was used for the assessment of 33 (42%) had a GOS-E score of ≥ 4. Of the 77 patients
important predictors of time to favorable outcome. Mul- with 12-month follow-up data, 32 (42%) had an mRS
ticollinearity analysis was performed, and variable infla- score of ≤ 3, 33 (43%) had a CPC score of ≤ 2, and 44
tion factor was assessed to avoid high level of correlation (57%) had a GOS-E score of ≥ 4. Overall, 24 (30%) of the
between the variables in the regression model. Results 79 patients had favorable outcome (i.e., favorable func-
are expressed as hazard ratios (HRs) with corresponding tional outcome according to all three outcome scales) at
95% CIs and P values. The statistical software R version 3 months, and 31 (40%) of 77 had favorable outcome at
4.2.0 was used for statistical analysis. 12 months. Patients with an unfavorable outcome at both
3 and 12 months were more likely to be discharged from
hospital to a high-level rehabilitation facility or another
Data Availability care facility, such as a nursing home, rather than to their
fMRI data cannot be made fully anonymous and are not own home. Clinical characteristics and comparison of
publicly available. Other data will be shared upon reason- patients with favorable and unfavorable 3- and 12-month
able request. The code used in the predictive models is functional outcomes are shown in Table 2. As illus-
available at https://​github.​com/​fraim​ondo/​conne​ctme-​ trated by Fig. 2, variables independently predicting time
follo​wup. to favorable outcome were younger age (HR 1.04 [95%
CI 1.02–1.06]), TBI as the cause of ICU admission (HR
Ethics 1.94 [95% CI 1.04–3.61]), ability to follow commands at
This study was approved by the Danish Data Protec- admission (HR 2.70 [95% CI 1.40–5.23]), improving con-
tion Agency (RH-2016-191, I-Suite nr:04760) and the sciousness level during the stay in the ICU (HR 5.76 [95%
Ethics Committee of the Capital Region of Denmark CI 2.14–15.51]), and initial brain imaging without severe
Table 1 Demographics and clinical characteristics of study population and comparison of patients discharged alive
with patients who died in the ICU
Basic characteristics All (n = 123) Discharged alive Death in ICU n = 41) P
from ICU n = 82)

Age, years 51 (19) 48 (19) 58 (18) 0.004

Female sex 51 (42%) 36 (44%) 15 (37%) 0.447
Prehospital ­comorbiditiesa
Any 88 (72%) 54 (66%) 34 (83%) 0.048
Cardiopulmonary 46 (37%) 28 (34%) 18 (44%) 0.301
Neurological 30 (24%) 24 (29%) 13 (32%) 0.780
Psychiatric 23 (19%) 17 (21%) 6 (15%) 0.431
 ­Otherb 60 (49%) 29 (35%) 22 (54%) 0.057
mRS prior to admission 0 (0–1) 0 (0–1) 0 (0–1.5) 0.366
Cause of admission
Traumatic brain injury 35 (29%) 23 (28%) 12 (29%) 0.883
Ischemic stroke 12 (10%) 6 (7%) 6 (15%) 0.223
Cardiac arrest 11 (9%) 2 (2%) 9 (22%) 0.001
Subarachnoid hemorrhage 7 (6%) 7 (9%) 0 (0%) –
Intracerebral hemorrhage 9 (7%) 7 (9%) 2 (5%) 0.504
Epilepsy 6 (5%) 5 (6%) 1 (2%) 0.428
Other, ­neurologyc 20 (16%) 15 (18%) 5 (12%) 0.407
Other, medical/ ­surgicald 23 (19%) 17 (21%) 6 (15%) 0.431
Injury onset to study enrolment, days 15 (14) 17 (15) 11 (11) 0.053
Severe pathology on brain imaging at a­ dmissione 48 (39%) 31 (38%) 17 (42%) 0.698
fMRI available 67 (54%) 48 (58%) 19 (46%) 0.209
Behavioral scales
GCS motor score at admission 1 (1–6) 4 (1–6) 1 (1–5) 0.026
GCS total score at enrolment 6 (5–9) 7 (6–10) 5 (3–6) < 0.001
FOUR total score at enrolment 8 (6–10) 9 (7–12) 6 (5–8) < 0.001
Command following at admission 48 (39%) 38 (46%) 10 (24%) 0.031
Consciousness level at study enrolment
Coma/UWS 73 (59%) 35 (43%) 38 (93%) < 0.001
MCS−or above 50 (41%) 47 (57%) 3 (7%) –
Duration of ICU admission, days 31 (26) 35 (28) 22 (16) 0.006
Cause of death in ICU, WLST 37 (30%) – 37 (90%) –
Discharged from hospital to (n = 82)
Home 10 (8%) 10 (12%) – –
Rehabilitation facility 52 (42%) 52 (63%) – –
Other care facility 20 (16%) 20 (24%) – –
3-month functional outcome – –
Favorable 24 (20%) 24 (29%) – –
Unfavorable 55 (45%) 55 (67%) – –
No data 3 (2%) 3 (4%) – –
12-month functional outcome – –
Favorable 31 (25%) 31 (38%) – –
Unfavorable 46 (37%) 46 (56%) – –
No data 5 (4%) 5 (6%) – –
Data are presented as n (%), mean (SD), and median (interquartile range). Bold values indicate statistical significance
Abbreviations: ICU = intensive care unit, mRS = modified Rankin Scale, fMRI = functional magnetic resonance imaging, GCS = Glasgow Coma Scale, FOUR = Full
Outline of Unresponsiveness Score, UWS = unresponsive wakefulness syndrome, MCS = minimally conscious state, WLST = withdrawal of life-sustaing therapy.
a
Patients could have more than one comorbidity
b
Other comorbidities included diabetes mellitus, other endocrine, cancer, fibromyalgia, inflammatory bowel disease, MGUS, cirrhosis, osteoporosis, chronic
nephropathy, and arthritis
Table 1 (continued)
c
Other causes, neurology included autoimmune encephalitis, brain tumor, hydrocephalus and shunt revision, meningoencephalitis, autoimmune encephalitis, global
cerebral oedema, cerebral venous thrombosis, myasthenic crisis, and anoxic ischemic brain damage due to drowning or strangulation
d
Other causes, medical or surgical included hypoglycemia or hyperglycemia, acute respiratory failure, aortic dissection or ruptured aortic aneurysm, perforated
diverticulitis and sepsis, ileus and sepsis, pulmonary embolism, and carbon monoxide poisoning
e
Severe pathology on brain imaging was defined as Fisher grade ≥ 3 (for subarachnoid hemorrhage), Marshall classification ≥ 3 (for TBI), hemorrhage volume ≥ 30 mL
(for intracerebral hemorrhage), strategic brainstem lesions (for ischemic stroke or infratentorial hemorrhage), any visible sign of anoxic brain injury on CT scan (for
cardiac arrest), global cortical edema (for patients with brain edema), midline compression, compression of basal cisterns and/or visible signs of hydrocephalus (for
patients with any type of brain tumor)

pathological findings (HR 2.42 [95% CI 1.12–5.22]). Fur- [95% CI 0.70–0.76], AUC​P(MCS) 0.53 [95% CI 0.45–0.61])
thermore, favorable visual EEG grading (i.e., Synek score (Table 3 and Fig. 4, model Ia compared to model IIa and
I or II) (HR 2.47 [95% CI 1.46–4.19]) was also an inde- IIIa).
pendent predictor of time to favorable outcome (Fig. 2). All but one model (model VII) based on different
combinations of EEG features (Fig. 3 and Table 3, mod-
Machine Learning Predictive Models els IV, V, VI, VIII and IVa, Va, VIa, VIIIa) could predict
In the following sections are results from random for- functional outcome at both 3- and 12-months’ follow-
est predictive models, and results from SVM models, up. The best combination of AUC, PPV, and sensitivity
statistical analysis of pairwise comparison, and feature for prediction of both outcomes was achieved with the
importance analysis are presented in the Supplementary model based on the combination of Synek score, ABCD
Material (Table S1-S2 and Figs. S1-S3). categories, and EEG markers derived from resting EEG
segments (3-month outcome: AUC 0.80 [95% CI 0.76–
EEG Features and Functional Outcome 0.82], PPV 0.58 [95% CI 0.48–0.70], sensitivity 0.36 [95%
EEG Synek scores were determined for all 77 patients CI 0.29–0.41]; 12-month outcome: AUC 0.73 [95% CI
who had available 3- and 12-month outcomes. ABCD 0.67–0.81], PPV 0.67 [95% CI 0.58–0.81], sensitivity 0.54
scores could be determined for 66 of the 77 patients (the [95% CI 0.45–0.63]) (Fig. 3 and Table 3, model V). When
remaining 11 were classified non-ABCD), and P(MCS) comparing the combined EEG same-sample models, all
could be determined for 68 patients. Sedation levels the models performed equally well (Table 3 and Fig. 3,
were high or very high in 6 of the 77 patients (8%) dur- models IVa–VIIIa). Detailed results from the statistical
ing EEG recording, with no statistically significant effects analysis of the pairwise comparison of same-sample EEG
on functional outcomes (Table 2). Of the predictive mod- models are presented in Table S2. Abbreviations: ROC =
els based on individual EEG features (i.e., Synek score, receiver operating curve, AUC = area under the curve.
ABCD categories, and P(MCS)), only the Synek score
could predict functional outcome at both 3 months (AUC fMRI Functional Connectivity and Functional Outcome
0.67 [95% CI 0.65–0.70], PPV 0.43 [95% CI 0.34–0.54], fMRI features were available for 45 of the 77 patients with
sensitivity 0.46 [95% CI 0.41–0.52]) and 12 months (AUC both 3- and 12-month outcomes, and 10 of the 45 (22%)
0.66 [95% CI 0.59–0.69], PPV 0.65 [95% CI 0.57–0.73], received high or very high levels of sedation during the
sensitivity 0.42 [95% CI 0.41–0.44]). The models based scan, with no statistically significant effects on functional
on ABCD categories could not predict 3-month outcome outcomes (Table 2). Because of a limited number of sam-
(AUC 0.38 [95% CI 0.34–0.47], PPV 0.13 [95% CI 0.04– ples with both fMRI data and outcome measures, predic-
0.22], sensitivity 0.24 [95% CI 0.07–0.46]) but could pre- tive models including fMRI functional connectivity (FC)
dict 12-month outcome (AUC 0.58 [95% CI 0.50–0.64], were tested with the LOO-CV procedure (Fig. 1, Table 3,
PPV 0.61 [95% CI 0.44–0.80], sensitivity 0.21 [95% CI and Table S1, models IX-XI). fMRI FC measures tested
0.17–0.29]), whereas the models based on P(MCS) could with both random forest and SVM algorithms showed
not 12-month outcome (AUC 0.54 [95% CI 0.44–0.63], evidence suggesting that predicting 3-month outcome is
PPV 0.41 [95% CI 0.31–0.59], sensitivity 0.26 [95% CI possible (random forest model IX: accuracy 0.69; SVM
0.18–0.32]) (Table 3 and Fig. 3, models I to III). Head-to- model IX: accuracy 0.78) but not 12-month outcome
head comparison of the same-sample models based on (random forest model IX: accuracy 0.47; SVM model IX:
individual EEG features showed that models based on the accuracy 0.47). More samples are required to confirm
Synek score outperformed the ABCD model in predict- and correctly estimate the performance of such models.
ing 3-month outcome (AUC​Synek 0.70 [95% CI 0.68–0.73],
AUC​ABCD 0.38 [95% CI 0.22–0.51]) and the P(MCS)
model in predicting 12-month outcome (AUC​Synek 0.72
Table 2 Comparison of patients with favorable and unfavorable 3- and 12-month functional outcomes
Basic characteristics 3-month outcome (n = 79) 12-month outcome (n = 77)
Favorable (n = 24) Unfa‑ P Favorable (n = 31) Unfavorable (n = 46) P
vorable
(n = 55)

Age, years 43 (19) 49 (19) 0.189 38 (16) 54 (18) < 0.001

Female sex 11 (46%) 24 (44%) 0.858 14 (45%) 21 (46%) 0.968
Preadmission ­comorbiditiesa
Any 16 (67%) 36 (66%) 0.927 17 (55%) 35 (76%) 0.059
Cardiopulmonary 7 (29%) 19 (35%) 0.658 6 (19%) 20 (44%) 0.03
Neurological 3 (13%) 21 (38%) 0.022 5 (16%) 19 (42%) 0.021
Psychiatric 10 (42%) 7 (13%) 0.007 11 (36%) 6 (13%) 0.026
b
Other ­medical 8 (33%) 20 (36%) 0.809 7 (23%) 21 (46%) 0.042
mRS prior to admission 0 (0–0.5) 0 (0–1) 0.098 0 (0–0) 1 (0–2) 0.005
Clinical characteristics
Cause of admission
Traumatic brain injury 6 (25%) 16 (29%) 0.729 10 (32%) 11 (24%) 0.434
Ischemic stroke 1 (4%) 5 (9%) 0,509 2 (6%) 4 (9%) 0.758
Cardiac arrest 0 (0%) 2 (4%) 1 0 (0%) 2 (4%) 0.513
Subarachnoid hemorrhage 0 (0%) 7 (13%) 0.094 1 (3%) 6 (13%) 0.165
Intracerebral hemorrhage 0 (0%) 7 (13%) 0.094 0 (0%) 7 (15%) 0.037
Epilepsy 2 (8%) 3 (5%) 0,644 1 (3%) 4 (9%) 0.395
Other, ­neurologyc 8 (33%) 7 (13%) 0.045 8 (26%) 7 (15%) 0.27
Other, medical/ ­surgicald 7 (29%) 8 (15%) 0.15 9 (29%) 5 (11%) 0.054
Severe pathology on brain imaging at 3 (13%) 28 (51%) 0.001 6 (19%) 25 (54%) 0.002
­admissione
fMRI available 11 (46%) 34 (62%) 0.153 18 (58%) 27 (59%) 0.827
Sedation level during fMRI 0.146
None or minimal 4 (36%) 12 (35%) 5 (28%) 11 (41%)
Low or moderate 4 (36%) 13 (38%) 6 (33%) 11 (41%)
High or very high 3 (27%) 7 (21%) 7 (39%) 3 (11%)
Unknown 0 (0%) 2 (6%) 0 (0%) 2 (7%)
Sedation level during EEG 0.758 0.26
None or minimal 17 (71%) 43 (78%) 21 (68%) 37 (80%)
Low or moderate 5 (21%) 8 (15%) 6 (19%) 7 (15%)
High or very high 2 (8%) 4 (7%) 4 (13%) 2 (4%)
Behavioral scales
GCS motor score at admission 6 (4–6) 3 (1–6) 0.013 5 (1–6) 4 (1–6) 0.5
GCS total score at enrolment 6.5 (5–10) 7 (6–10) 0.492 8 (5–10) 7 (6–9.75) 0.762
FOUR total score at enrolment 8.5 (7–12) 9 (7–12) 0.306 9 (7–12) 8.5 (7–12) 0.243
Command following at admission 18 (75%) 20 (36%) 0.004 18 (58%) 19 (41%) 0.226
Consciousness level at study enrolment 0.878 0.388
Coma/UWS 10 (42%) 24 (44%) – 11 (36%) 21 (46%) –
MCS−or above 14 (58%) 31 (56%) – 20 (65%) 25 (54%) –
Consciousness level at ICU discharge 0.104 0.023
Coma/UWS 1 (4%) 10 (18%) – 1 (3%) 10 (22%) –
MCS−or above 23 (96%) 45 (82%) – 30 (97%) 36 (78%) –
Duration of ICU admission, days 24 (20) 41 (31) 0.022 31 (30) 39 (28) 1.01 (0.99–1.03)
Discharged from hospital to < 0.001 0.009
Home 9 (38%) 1 (2%) – 8 (26%) 1 (2%) –
Rehabilitation facility 6 (25%) 10 (19%) – 5 (16%) 11 (24%) –
Other care facility 9 (38%) 43 (80%) – 18 (58%) 33 (73%) –
Table 2 (continued)
Data are presented as n (%), mean (SD), and median (interquartile range). Bold values indicate statistical significance
Abbreviations: mRS = modified Rankin Scale, fMRI = functional magnetic resonance imaging, EEG = electroencephalography, GCS = Glasgow Coma Scale, FOUR =
Full Outline of Unresponsiveness Score, UWS = unresponsive wakefulness syndrome, MCS = minimally conscious state, ICU = intensive care unit.
a
Patients could have more than one comorbidity
b
Other comorbidities included diabetes mellitus, other endocrine, cancer, fibromyalgia, inflammatory bowel disease, MGUS, cirrhosis, osteoporosis, chronic
nephropathy and arthritis
c
Other causes, neurology included autoimmune encephalitis, brain tumor, hydrocephalus and shunt revision, meningoencephalitis, autoimmune encephalitis, global
cerebral oedema, cerebral venous thrombosis, myasthenic crisis, and anoxic ischemic brain damage due to drowning or strangulation
d
Other causes, medical or surgical included hypoglycemia or hyperglycemia, acute respiratory failure, aortic dissection or ruptured aortic aneurysm, perforated
diverticulitis and sepsis, ileus and sepsis, pulmonary embolism, and carbon monoxide poisoning
e
Severe pathology on brain imaging was defined as Fisher grade ≥ 3 (for subarachnoid hemorrhage), Marshall classification ≥ 3 (for traumatic brain injury),
hemorrhage volume ≥ 30 mL (for intracerebral hemorrhage), strategic brainstem lesions (for ischemic stroke or infratentorial hemorrhage), any visible sign of anoxic
brain injury on CT scan (for cardiac arrest), global cortical edema (for patients with brain edema), midline compression, compression of basal cisterns and/or visible
signs of hydrocephalus (for patients with any type of brain tumor)

Fig. 2 Predictors of time to favorable outcome. This figure depicts independent variables predicting time to favorable outcome (i.e., GOS-E ≥ 4,
mRS ≤ 3 and CPC ≤ 2). Death in ICU (n = 41) was treated as a competing risk in a multivariate Cox proportional regression model. Younger age,
patients with TBI, ability to follow commands at admission, improving consciousness level during ICU, no severe pathological findings at admis-
sion brain imaging, and favorable visual grading of EEG (i.e., Synek score I or II) were all independent predictors of earlier recovery. *Of all 123
included patients, one patient without EEG was excluded from this analysis. #Severe pathological findings on brain imaging was defined as Fisher
grade ≥ 3 (for subarachnoid hemorrhage), Marshall classification ≥ 3 (for TBI), hemorrhage volume ≥ 30 mL (for intracerebral hemorrhage), strategic
hemorrhage or infarct in brainstem (for ischemic stroke or infratentorial hemorrhage), any visible sign of anoxic brain injury on CT scan (for cardiac
arrest), global cortical edema (for patients with brain edema), brain tumors with midline compression, compression of basal cisterns and/or signs of
hydrocephalus (for patients with any type of brain tumor). Abbreviations: TBI = traumatic brain injury. ICU = intensive care unit, EEG = electroen-
cephalography.

Combined EEG and fMRI Features and Functional Outcome Table S1, models X and XI), regardless of which algo-
We evaluated prediction of 3- and 12-month functional rithm was used.
outcomes with the combination of fMRI FC with the
Synek score (n = 45) or P(MCS) derived from EEG mark-
ers-r (n = 44), as depicted in Table 3 (models X and XI). Clinical Features and Functional Outcome
Both combined models showed evidence that predicting The clinical features used to conduct a prediction model
3-month outcome is possible, with accuracies between were available from all patients with 3- and 12-month
0.73 and 0.84, but not 12-month outcome (Table 3 and outcome data. The clinical model (Table 3, model XII)
could predict both 3- and 12-month outcome, with
the highest combination of AUC, PPV, and sensitivity
achieved for prediction of 12-month outcome (3-month
Table 3 Prediction performance of EEG, fMRI, and clinical features in predicting 3- and 12-month functional outcome
Model Features N 3-month outcome 12-month outcome
AUC​ Positive predic‑ Sensitivity AUC​ Positive predic‑ Sensitivity
tive value tive value

Random forest EEG models based on all available data

I Synek 77 0.67 [0.65–0.70] 0.43 [0.34–0.54] 0.46 [0.41–0.52] 0.66 [0.59–0.69] 0.65 [0.57–0.73] 0.42 [0.41–0.44]
II ABCD 66 0.38 [0.34–0.47] 0.13 [0.04–0.22] 0.24 [0.07–0.46] 0.58 [0.50–0.64] 0.61 [0.44–0.80] 0.21 [0.17–0.29]
III P(MCS) 68 0.66 [0.61–0.69] 0.33 [0.25–0.39] 0.64 [0.51–0.76] 0.54 [0.44–0.63] 0.41 [0.31–0.59] 0.26 [0.18–0.32]
IV Synek, ABCD 66 0.59 [0.55–0.65] 0.45 [0.29–0.59] 0.48 [0.45–0.54] 0.71 [0.62–0.76] 0.61 [0.51–0.74] 0.54 [0.44–0.65]
V Synek, ABCD, EEG 64 0.80 [0.76–0.82] 0.58 [0.48- 0.70] 0.36 [0.29–0.41] 0.73 [0.67–0.81] 0.67 [0.58–0.81] 0.54 [0.45–0.63]
markers-r
VI Synek, ABCD, 58 0.69 [0.60–0.75] 0.27 [0.024–0.47] 0.34 [0.06–0.46] 0.68 [0.59–0.73] 0.53 [0.42–0.63] 0.37 [0.25–0.46]
P(MCS)
VII Synek, P(MCS) 68 0.68 [0.65–0.71] 0.33 [0.22–0.42] 0.61 [0.41–0.75] 0.56 [0.45–0.62] 0.48 [0.40–0.58] 0.36 [0.29–0.41]
VIII Synek, ABCD, 58 0.74 [0.58–0.80] 0.38 [0.21–0.50] 0.21 [0.13–0.33] 0.65 [0.58–0.71] 0.51 [0.37–0.74] 0.41 [0.31–0.53]
P(MCS), EEG
markers-r
Random forest EEG models based on same sample data
Ia Synek 58 0.70 [0.68–0.73] 0.38 [0.27–0.50] 0.53 [0.47–0.59] 0.72 [0.70–0.76] 0.52 [0.30–0.71] 0.52 [0.28–0.63]
IIa ABCD 58 0.38 [0.22–0.51] 0.07 [0.00–0.15] 0.17 [0.02–0.29] 0.66 [0.63–0.68] 0.51 [0.36–0.63] 0.26 [0.23–0.28]
IIIa P(MCS) 58 0.64 [0.59–0.68] 0.25 [0.12–0.35] 0.49 [0.32–0.69] 0.53 [0.45–0.61] 0.45[0.31–0.62] 0.24 [0.15–0.36]
IVa Synek, ABCD 58 0.58 [0.43–0.68] 0.29 [0.14–0.47] 0.42 [0.37–0.49] 0.73 [0.69–0.79] 0.57 [0.44–0.74] 0.50 [0.35–0.67]
Va Synek, ABCD, EEG 58 0.72 [0.51–0.79] 0.29 [0.04–0.56] 0.18 [0.01–0.29] 0.66 [0.59–0.73] 0.51 [0.36–0.72] 0.41 [0.28–0.58]
markers-r
VIa Synek, ABCD, 58 0.69 [0.59–0.73] 0.26 [0.02–0.46] 0.33 [0.05–0.49] 0.68 [0.56–0.77] 0.54 [0.47–0.74] 0.39 [0.23–0.50]
P(MCS)
VIIa Synek, P(MCS) 58 0.69 [0.62–0.74] 0.28 [0.08–0.40] 0.46 [0.13–0.66] 0.64 [0.53–0.72] 0.56 [0.45–0.73] 0.38 [0.28–0.46]
VIIIa Synek, ABCD, 58 0.72 [0.53–0.79] 0.33 [0.14–0.51] 0.22 [0.13–0.36] 0.67 [0.58–0.74] 0.52 [0.33–0.68] 0.43 [0.27–0.56]
P(MCS), EEG
markers-r
Random forest fMRI model with LOO-CV procedure
IX fMRI FC 45 0.69 0.47
X fMRI FC, Synek 45 0.75 0.42
XI fMRI FC, P(MCS)rest 44 0.76 0.45
Random forest clinical model based on all available data
XII Clinical features 77 0.62 [0.55- 0.69] 0.46 [0.27- 0.63] 0.38 [0.25–0.58] 0.79 [0.77–0.83] 0.66 [0.63–0.70] 0.73 [0.68–0.77]
Random forest clinical model based on same sample data
XIIa Clinical features 58 0.68 [0.60–0.76] 0.22 [0.06–0.41] 0.19 [0.08–0.27] 0.82 [0.80–0.84] 0.72 [0.65–0.77] 0.74 [0.64–0.81]
EEG markers-r = 68 EEG markers derived from the EEG resting segments, P(MCS) = support vector machine classifier indicating probability of consciousness
derived from EEG markers from the full EEG. P(MCS)rest = support vector machine classifier indicating probability of consciousness derived from EEG markers from
the EEG resting segments. Clinical features = age, known preadmission comorbidity, primary injury leading to admission being traumatic brain injury, improving
consciousness during intensive care unit, and command following abilities at admission. Numbers in brackets indicate 95% CI. Bold values indicate AUCs or
accuracy ≥ 0.70
Abbreviations: AUC = area under the curve, EEG = electroencephalography, fMRI = functional magnetic resonance imaging, LOO-CV = leave-one-out cross-
validation. FC = functional connectivity

outcome: AUC 0.62 [95% CI 0.55–0.69], PPV 0.46 [95% perform equally well to the EEG models for prediction
CI 0.27–0.63], sensitivity 0.38 [95% CI 0.25–0.58]; of 3-month outcome and slightly better for prediction
12-month outcome: AUC 0.79 [95% CI 0.77–0.83], of 12-month outcome.
PPV 0.66 [95% CI 0.63–0.70], sensitivity 0.73 [95% CI
0.68–0.77]). The same-sample model (Table 3, model Discussion
XIIa) showed the same pattern. When comparing the In this, to our knowledge, first prospective multimodal
same-sample clinical model (model XIIa) to EEG mod- cohort study including 123 ICU patients with acute
els (models Ia–VIIIa), the clinical model seemed to DoC from various underlying conditions, we show that
 Fig. 3 Random forest EEG models with maximum available data Fig. 4 Random forest EEG models with same-sample data pre-
predicting 3- and 12-month outcomes. Boxplots illustrating model dicting 3- and 12-month outcomes. Boxplots illustrating model
performances (AUCs) of RF-models based on EEG features predicting performances (AUCs) of machine learning models based on EEG
3-month (blue) and 12-month (orange) functional outcomes. Each features predicting 3-month (blue) and 12-month (orange) functional
model is based on the maximum amount of data available (see also outcomes. Each model is based on the same samples (n = 58) for
Fig. 1). Of the unimodal models (I-III), only model I based on the Synek head-to-head comparison of EEG features. Of the unimodal models
score could predict both 3- and 12-month outcomes. The highest (Ia-IIIa), model Ia based on Synek score outperformed model IIa based
AUC for predicting both outcomes (AUC​3-month 0.79 [0.77–0.82]; AUC​ on ABCD categories in predicting 3-month outcome (AUC​Synek 0.70
12-month 0.74 [0.71–0.77]) were obtained with the combined model (V)
[0.69–0.74] vs. AUC​ABCD 0.38 [0.31–0.45]). In predicting 12-month out-
based on combination of three EEG features (i.e., Synek score, ABCD come, model Ia outperformed model IIIa which was based on P(MCS)
categories and EEG markers-r derived from the SVM consciousness measures (AUC​Synek 0.70 [0.69–0.74] vs. AUC​P(MCS) 0.54 [0.50–0.59]). Of
classifier). Overall, this figure shows that while Synek score was the the combined models based on at least three EEG features (Va-VIIa),
only unimodal EEG-model that predicted both 3- and 12-month all models could predict 3- and 12-month outcomes, and none
functional outcomes, all models based on a combination of EEG outperformed the others. A similar pattern was observed for SVM
features (IV–VII) could predict both 3- and 12-month outcomes with machine learning models (see Fig. S2). Individual same-sample EEG
AUCs above chance level. A similar pattern was observed for SVM random forest models: Ia = Synek, IIa = ABCD, IIIa = P(MCS) C. Com-
machine learning models (see Fig. S1). Individual EEG random forest bined same-sample EEG random forest models: IVa = Synek + ABCD,
models: I = Synek, II = ABCD, III = P(MCS) C. Combined EEG random Va = Synek + ABCD + EEG markers-r, VIa = Synek + ABCD + P(MCS),
forest models: IV = Synek + ABCD, V = Synek + ABCD + EEG markers-r, VIIa = Synek + P(MCS) and VIIIa = Synek + ABCD + P(MCS) + EEG
VI = Synek + ABCD + P(MCS), VII = Synek + P(MCS) and VIII = Synek + markers-r. Abbreviations: ROC = receiver operating curve, AUC =
ABCD + P(MCS) + EEG markers-r area under the curve.

machine learning algorithms applied to EEG and fMRI that underly machine learning models is crucial, data
features obtained soon after ICU admission can assist in quantity is also important because data sets with many
the prediction of 3-month functional outcome, whereas variables but limited number of samples introduce high
12-month outcome can only be predicted by EEG fea- level of variance, rendering the models imbalanced [44].
tures. We also show that the model based on clinical fea- Despite our relatively large population of patients with
tures can predict both outcomes, with highest accuracy acute DoC, our results, especially those including fMRI
for predicting 12-month functional outcome. Thus, we features, should therefore be interpreted with caution
have confirmed readily available independent predictive until further validation from ongoing multicenter stud-
clinical variables of time to favorable recovery, with the ies [45]. These factors may also explain the relatively low
clinical model performing overall as good as EEG models PPV and sensitivities despite high AUCs of the combined
in predicting both outcomes. EEG models, which were based on data from patients
EEG features in combination, as well as the EEG with a complete data set including all EEG features
Synek score as an individual model, predicted both 3- (n = 58). High levels of sedation can have a significant
and 12-month functional outcomes (Fig. 3 and Table 3), impact on resting EEG measures and may affect the accu-
whereas all models based on fMRI FC measures could racy of EEG models used for prediction. However, in our
only predict 3-month outcome (Table 3). EEG recordings cohort, only six patients (8%) with both 3- and 12-month
were available from all 77 patients with outcome meas- outcome measures available were under high levels of
ures at both 3 and 12 months, whereas we only had fMRI sedation during their EEG recordings, and therefore we
sequences from 45 of these patients, thus resulting in a do not consider sedation a significant factor affecting
substantially reduced amount of data available for the our results. However, 10 of 45 patients (22%) with fMRI
fMRI feature models. Although the quality of the data sequences received high levels of sedation during their
scans, which may have had an impact on the data, but prognosis and thus withdrawal of life-sustaining therapy
we did not find any statistically significant differences in (Table 1).
sedation levels when comparing patients with favorable Models including fMRI features were tested with a
and unfavorable outcomes. LOO-CV procedure because of the limited number of
Despite the aforementioned limitations, we could show available samples. Results indicate that fMRI FC both
that most EEG features predicted both early and late alone and in combination with some EEG features may
functional outcomes individually and in various combi- be useful to predict early functional outcome at 3 months
nations (Fig. 3 and Table 3). This is an important finding (Table 3) but not (yet) late outcome at 12 months. The
because EEG is much more available bedside in the ICU LOO-CV procedure limits data waste and is therefore
than advanced neuroimaging, such as fMRI, and EEG primarily used for small data sets, but a major limitation
features like ours can be easily implemented in an ICU is that the results are prone to optimistic interpretation
setting. and therefore need external validation in larger data sets
When comparing the individual EEG features head-to- [41].
head with the same-sample models (Fig. 4 and Table 3), In the first article from CONNECT-ME [20], we found
we found that the Synek score outperformed the ABCD that EEG and fMRI features predicted levels of con-
categories for the prediction of short-term outcome sciousness of patients with acute DoC at the time of ICU
and the SVM classifier derived P(MCS) for the prediction discharge. Importantly, EEG and fMRI were performed
of long-term outcome. This finding may be explained by without active consciousness paradigms; thus, patients
the fact that the Synek score was assessed manually by likely had different degrees of residual consciousness
two board-certified electroencephalographers with many (e.g., including those who could not have participated in
years of experience with ICU EEG, whereas the ABCD active paradigms [10]). Collectively, our findings indicate
and P(MCS) features were initially developed in more that both EEG and fMRI have the potential not only to
homogenous patient groups (i.e., homogenous cardiac predict level of consciousness during ICU admission [20]
arrest [33] and chronic DoC cohorts [36, 37] vs. acute but also to predict functional outcome of patients with
DoC cohort with heterogeneous brain injuries) than ours. brain injury of various causes resulting in acute DoC in
Furthermore, visual analysis of EEGs is routinely used the early phase of hospitalization and (EEG, at least) up
for prognostication in ICU populations like the present to 1 year after discharge from the ICU.
cohort, which may also explain the higher performance In line with a recent study about recovery trajectories
of the models based on the Synek score. Still, we could of patients with cognitive motor dissociation [14], we
show that combining different EEG features resulted in additionally identified readily available clinical features
the best predictive performance of the models, regardless as independent predictors of time to favorable functional
of the algorithm used (Table 3 and Table S1). These are outcome (Fig. 2). In our heterogenous patient cohort
important findings because most ICU sites with patients reflecting a real-life ICU setting, we confirmed that TBI
with acute DoC do not have the resources to perform is related to earlier recovery. Furthermore, patients who
advanced EEG assessment using machine learning clas- were younger, could follow commands at ICU admis-
sifiers. These sites can thus safely rely on experienced sion, had no severe pathological findings on initial brain
electroencephalographers using established criteria for imaging, and showed improving consciousness level
visual EEG analyses instead. If the necessary electroen- in the ICU also recovered earlier. Similarly, patients
cephalographer expertise is unavailable, however, exter- with favorable functional outcomes at 3 and 12 months
nal data-driven analysis of EEGs may become a suitable were more likely to be discharged directly to their own
option for those sites in the near future. home, whereas patients with unfavorable outcome were
We also show that EEG models are overall compara- more often discharged to rehabilitation facilities and
ble to a model based on clinical features for prediction nursing homes (Table 2). This is explained by the fact
of 3-month outcome, while performing slightly worse for that patients with more severe injuries needed a higher
prediction of 12-month outcome. It is not surprising that level of care and were thus discharged to facilities with a
the clinical model performs well in predicting especially higher level of rehabilitation resources. All these findings
long-term outcome of this patient group when consider- can help clinicians when guiding patient families about
ing that readily available clinical features play a signifi- the prospects of recovery, including the time it takes to
cant role in end-of-life decision-making in the ICU. Thus, achieve a good recovery.
the patients who survive in the ICU are a selected group Several limitations need to be considered. As a sin-
of patients expected to perform better based at least par- gle-center study, CONNECT-ME is susceptible to sam-
tially on their clinical characteristics than those who died pling bias. Our follow-up data were primarily collected
in the ICU, where most deaths were due to expected poor through electronic health records based on notes from
trained nursing staff who routinely collect functional out- suggesting this might be of lesser importance to the
come data from ICU patients, especially those discharged overall results. Our study population is a heterogeneous
to high facility rehabilitation centers. Because most of group of patients with various causes of DoC, rendering
the follow-up data were not collected firsthand by the subgroup analysis unreliable because of the low num-
research team, we acknowledge there is a risk of bias. ber of patients in each group. Thus, further validation is
Taking this into consideration, we chose a composite needed to confirm our findings.
binary outcome measure (i.e., favorable vs. unfavorable) On the positive side, our findings are generalizable to
instead of an in-depth analysis of the respective outcome a real-life ICU setting and patients with acute DoC with
scales. various causes of brain injury. We also evaluated func-
A relatively large number of patients (33%) died in the tional outcome in our cohort by using three different out-
ICU, most because of withdrawal of life-sustaining ther- come scales designed for stroke (mRS) [38], TBI (GOS-E)
apy because of a presumed poor prognosis. Although the [47], and cardiac arrest (CPC) [40] patients to account
current study included 123 patients, data from only 77 for the heterogenicity of our patients. Owing to logistical
patients were available for the final analysis of 12-month challenges and resources needed for advanced data anal-
outcomes. Thus, the remaining cohort with available yses, to our knowledge, no previous EEG/fMRI study has
follow-up data consisted of patients who were expected managed to investigate acute DoC in a larger ICU cohort
to regain better functional outcome. This skewed the or with a longer follow-up than ours.
data set used in the machine learning models. The pre-
dictive performance of these models may hence have
been biased in that they lacked the (potential) clinical Conclusions
trajectories of patients who had life-sustaining therapy We show that EEG early during ICU admission predicted
withdrawn. To account for this bias to some extent, in both 3- and 12-month functional outcomes of patients
our analysis of independent variables related to time to with acute DoC with various causes of brain injury and
favorable outcome, we included in-hospital death as a that fMRI resting-state measures might be useful to pre-
competing risk in the multivariate Cox proportional dict 3-month outcome. Furthermore, young age, TBI,
hazards regression model. Still, death due to withdrawal initial brain imaging without severe pathological find-
of life-sustaining therapy in the ICU remains an impor- ings, ability to follow commands during ICU admission,
tant limitation and cannot be fully accounted for when improving consciousness level during the ICU stay, and
studying ICU patients with acute severe brain injury and favorable visual EEG grading all independently predicted
DoC. Furthermore, the heterogeneity of the brain injuries shorter time to favorable functional outcome. In sum-
studied made subgroup analysis other than TBI vs. non- mary, we suggest that combining EEG- and fMRI-based
TBI impractical because of the small numbers in each machine learning models with readily available clinical
subgroup. data allows for short-term outcome prediction of patients
Because EEG is more available in the ICU than fMRI, with coma and other acute DoC and potentially can pre-
it is routinely used for prognostication of patients with dict long-term outcome up to 1 year after ICU discharge.
acute DoC, especially of those admitted post cardiac Supplementary Information
arrest [46]. Excluding patients who died in the ICU may The online version contains supplementary material available at https://​doi.​
org/​10.​1007/​s12028-​023-​01816-z.
therefore have decreased the performance of the EEG
models as well. Additionally, of the three methods used
for EEG analysis, the visual scoring and ABCD scale Author details
1
Department of Neurology, Copenhagen University Hospital - Rigshospitalet,
are subjective and may introduce bias even though the Blegdamsvej 9, 2100 Copenhagen, Denmark. 2 Brain and Behaviour, Institute
investigators analyzing EEG in our study were blinded to of Neuroscience and Medicine, Research Center Jülich, Jülich, Germany.
outcomes. 3
Institute of Systems Neuroscience, Medical Faculty, Heinrich Heine University
Düsseldorf, Düsseldorf, Germany. 4 Neurobiology Research Unit, Copenhagen
MRI scans are logistically very challenging to obtain University Hospital - Rigshospitalet, Copenhagen, Denmark. 5 Department
in the ICU and are thus less often performed than EEG, of Drug Design and Pharmacology, University of Copenhagen, Copenhagen,
which might be yet another selection bias, affecting the Denmark. 6 Department of Neurophysiology, Copenhagen University Hospital
- Rigshospitalet, Copenhagen, Denmark. 7 Department of Neurosurgery,
fMRI models owing to exclusion of patients without Copenhagen University Hospital - Rigshospitalet, Copenhagen, Denmark.
available fMRI. However, in our cohort, we found no 8
Department of Neuroanaesthesiology, Copenhagen University Hospital - Rig-
statistically significant difference in the frequency with shospitalet, Copenhagen, Denmark. 9 Department of Radiology, Copenhagen
University Hospital - Rigshospitalet, Copenhagen, Denmark. 10 Department
which fMRI was performed when comparing patients of Clinical Medicine, University of Copenhagen, Copenhagen, Denmark.
who died in the ICU with those who were discharged 11
Department of Cardiology, Copenhagen University Hospital - Rigshospitalet,
alive (Table 1) or when comparing patients with favorable Copenhagen, Denmark. 12 Biological and Precision Psychiatry, Copenhagen
Research Center for Mental Health, Copenhagen University Hospital, Copen-
outcome with those with unfavorable outcome (Table 2), hagen, Denmark. 13 Institut du Cerveau ‑ Paris Brain Institute, Inserm, Centre
nationl de la recherche scientifique, Assistance Publique ‑ Hôpitaux de Paris, 3. Edlow BL, Claassen J, Schiff ND, Greer DM. Recovery from disorders of
Sorbonne Université, Hôpital de La Pitié Salpêtrière, Paris, France. consciousness: mechanisms, prognosis and emerging therapies. Nat Rev
Neurol. 2021;17(3):135–56.
Acknowledgements 4. Kondziella D, Amiri M, Othman MH, et al. Incidence and prevalence of
We are very grateful to the patients and their families who participated in coma in the UK and the USA. Brain Commun. 2022. https://​doi.​org/​10.​
this study. Additionally, we thank the staff at the Departments of Clinical 1093/​brain​comms/​fcac1​88.
Neurophysiology, Radiology, Cardiology, and Neurology and ICUs (sections 5. Hammond FM, Katta-Charles S, Russell MB, et al. Research needs for
6062, 2143, 4131 and 4141) at Rigshospitalet, Copenhagen University Hospital, prognostic modeling and trajectory analysis in patients with disorders of
for their cooperation and support. A special thanks to neurointensive care consciousness. Neurocrit Care. 2021;35(Suppl 1):55–67.
nurse Ida Møller for her help with screening patients and to neurophysiology 6. Kowalski RG, Flora Hammond M, et al. Recovery of consciousness and
technicians Puk Overgaard Christiansen, Rikke Dresner, Julie Dyppel, Dorthe functional outcome in moderate and severe traumatic brain injury sup-
Husum Hansen, Allan Rene Hansen, Lars Hvidtfeldt, Sundus Said Muse Issa, plemental content. JAMA Neurol. 2021;78(5):548–57.
Trine Jessen, Camilla Hedegaard Mundt Larsen, Pia Christina Stade-Larsen, 7. Izzy S, Compton R, Carandang R, Hall W, Muehlschlegel S. Self-fulfilling
Marianne Sonnichsen, and Parisa Yazdanyar for their help with EEG recordings. prophecies through withdrawal of care: Do they exist in traumatic brain
Figure 1 was created with Biorender.com. injury, too? Neurocrit Care. 2013;19(3):347–63.
8. Hemphill JC, White DB. Clinical nihilism in neuro-emergencies. Emerg
Author contributions Med Clin North Am. 2009;27(1):27.
MA: data acquisition, data analysis and interpretation, initial draft prepara- 9. Turgeon AF, Dorrance K, Archambault P, et al. Factors influencing deci-
tion, editing, and final draft approval. FR, PF, MCH, AS, IZ: data analysis and sions by critical care physicians to withdraw life-sustaining treatments
interpretation, editing, and final draft approval. MO, OB, MLF, AEH, CH, JLSH, in critically ill adult patients with severe traumatic brain injury. CMAJ.
HRJ, NVK, ELL, JEM, VN, MN, STS, JDS, CS, KLW, LMW, JH, VAL, MEF, GMK, JK: data 2019;191(24):E652–63.
acquisition, editing, and final draft approval. DK (primary investigator): study 10. Kondziella D, Friberg CK, Frokjaer VG, Fabricius M, Møller K. Preserved con-
conception, data acquisition, data analysis and interpretation, initial draft sciousness in vegetative and minimal conscious states: systematic review
preparation, editing, and final draft approval. and meta-analysis. J Neurol Neurosurg Psychiatry. 2016;87(5):485–92.
11. Faugeras F, Rohaut B, Valente M, et al. Survival and consciousness recov-
Source of support ery are better in the minimally conscious state than in the vegetative
Open access funding provided by Royal Library, Copenhagen University state. Brain Inj. 2018;32(1):72–7.
Library. Funding was supplied by Region Hovedstadens Forskningsfond, Lun- 12. The Vegetative and Minimally Conscious States: A Comparison: The Jour-
dbeck Foundation, Rigshospitalets Forskningspuljer, and Offerfonden. Material nal of Head Trauma Rehabilitation [Internet]. [cited 2022 Nov 9];Available
execution, content, and results are the sole responsibility of the authors. The from: https://​journ​als.​lww.​com/​headt​rauma​rehab/​Abstr​act/​1997/​08000/​
assessments and views expressed in the material are the authors’ own and are The_​Veget​ative_​and_​Minim​ally_​Consc​ious_​State​s__A.​5.​aspx
not necessarily shared by the funders. 13. Pan J, Xie Q, Qin P, et al. Prognosis for patients with cognitive
motor dissociation identified by brain-computer interface. Brain.
Conflict of interest 2020;143(3):1177–89.
The authors report no conflicts of interest. 14. Egbebike J, Shen Q, Doyle K, et al. Cognitive-motor dissociation and
time to functional recovery in patients with acute brain injury in
Ethical Approval the USA: a prospective observational cohort study. Lancet Neurol.
This study was approved by the Danish Data Protection Agency (RH-2016-191, 2022;21(8):704–13.
I-Suite nr:04760) and the Ethics Committee of the Capital Region of Denmark 15. Jöhr J, Halimi F, Pasquier J, Pincherle A, Schiff N, Diserens K. Recovery in
(File-nr.:H-16040845). cognitive motor dissociation after severe brain injury: a cohort study.
PLoS ONE. 2020;15(2):e0228474.
Open Access 16. Schnakers C, Vanhaudenhuyse A, Giacino J, et al. Diagnostic accuracy of
This article is licensed under a Creative Commons Attribution 4.0 International the vegetative and minimally conscious state: clinical consensus versus
License, which permits use, sharing, adaptation, distribution and reproduction standardized neurobehavioral assessment. BMC Neurol. 2009;9:1–5.
in any medium or format, as long as you give appropriate credit to the original 17. Giacino JT, Kalmar K, Whyte J. The JFK coma recovery scale-revised: meas-
author(s) and the source, provide a link to the Creative Commons licence, and urement characteristics and diagnostic utility. Arch Phys Med Rehabil.
indicate if changes were made. The images or other third party material in this 2004;85(12):2020–9.
article are included in the article’s Creative Commons licence, unless indicated 18. Owen AM, Coleman MR, Boly M, Davis MH, Laureys S, Pickard JD. Detect-
otherwise in a credit line to the material. If material is not included in the ing awareness in the vegetative state. Science (80-). 2006;313(5792):1402.
article’s Creative Commons licence and your intended use is not permitted 19. Claassen J, Doyle K, Matory A, et al. Detection of brain activation
by statutory regulation or exceeds the permitted use, you will need to obtain in unresponsive patients with acute brain injury. N Engl J Med.
permission directly from the copyright holder. To view a copy of this licence, 2019;380(26):2497–505.
visit http://​creat​iveco​mmons.​org/​licen​ses/​by/4.​0/. 20. Amiri M, Fisher PM, Raimondo F, et al. Multimodal prediction of residual
consciousness in the intensive care unit: the CONNECT-ME study. Brain.
2023;146(1):50–64.
Publisher’s Note 21. Sanz LRD, Thibaut A, Edlow BL, Laureys S, Gosseries O. Update on
Springer Nature remains neutral with regard to jurisdictional claims in pub- neuroimaging in disorders of consciousness. Curr Opin Neurol.
lished maps and institutional affiliations. 2021;34(4):488–96.
22. Velly L, Perlbarg V, Boulier T, et al. Use of brain diffusion tensor imaging
Received: 10 May 2023 Accepted: 21 July 2023 for the prediction of long-term neurological outcomes in patients after
cardiac arrest: a multicentre, international, prospective, observational,
cohort study. Lancet Neurol. 2018;17(4):317–26.
23. Edlow BL, Chatelle C, Spencer CA, et al. Early detection of conscious-
ness in patients with acute severe traumatic brain injury. Brain.
2017;140(9):2399–414.
References
24. Skibsted AP, Amiri M, Fisher PM, et al. Consciousness in neurocritical care
1. Owen AM, Coleman MR, Boly M, Davis MH, Laureys S, Pickard JD. Detect-
cohort study using fMRI and EEG (CONNECT-ME): protocol for a longitu-
ing awareness in the vegetative state. Science. 2006;313(5792):1402.
dinal prospective study and a tertiary clinical care service. Front Neurol.
2. Ballanti S, Campagnini S, Liuzzi P, et al. EEG-based methods for recovery
2018. https://​doi.​org/​10.​3389/​fneur.​2018.​01012.
prognosis of patients with disorders of consciousness: a systematic
25. Teasdale G, Jennett B. Assessment of coma and impaired consciousness.
review. Clin Neurophysiol. 2022;144:98–114.
Lancet. 1974;304(7872):81–4.
26. Wijdicks EFM, Bamlet WR, Maramattom BV, Manno EM, McClelland 38. Van Swieten JC, Koudstaal PJ, Visser MC, Schouten H, Van Gijn J. Inter-
RL. Validation of a new coma scale: the FOUR score. Ann Neurol. observer agreement for the assessment of handicap in stroke patients.
2005;58(4):585–93. Stroke. 1988;19(5):604–7.
27. Posner JB, Saper CB, Schiff ND, Claassen J. Plum and Posner’s diagnosis of 39. Wilson L, Boase K, Nelson LD, et al. A manual for the glasgow outcome
stupor and coma. 5th ed. New York: Oxford University Press, Inc; 2019. scale-extended interview. J Neurotrauma. 2021;38(17):2435.
28. Kondziella D, Frontera JA. Pearls & oy-sters: eyes-open coma. Neurology. 40. Balouris SA, Raina KD, Rittenberger JC, Callaway CW, Rogers JC, Holm MB.
2021;96(18):864–7. Development and validation of the cerebral performance categories-
29. Laureys S, Celesia GG, Cohadon F, et al. Unresponsive wakefulness extended (CPC-E). Resuscitation. 2015;94:98–105.
syndrome: a new name for the vegetative state or apallic syndrome. BMC 41. Yeung AWK, More S, Wu J, Eickhoff SB. Reporting details of neuroimaging
Med. 2010;8(1):68. studies on individual traits prediction: a literature survey. Neuroimage.
30. Giacino JT, Ashwal S, Childs N, et al. The minimally conscious state. Neu- 2022;256:119275.
rology. 2002;58(3):349–53. 42. Nadeau C, Bengio Y. Inference for the generalization error. Mach Learn.
31. Bruno MA, Vanhaudenhuyse A, Thibaut A, Moonen G, Laureys S. From 2003;52:239–81. https://​doi.​org/​10.​1023/A:​10240​68626​366.
unresponsive wakefulness to minimally conscious PLUS and functional 43. Pedregosa FABIANPEDREGOSA F, Michel V, Grisel OLIVIERGRISEL O, et al.
locked-in syndromes: recent advances in our understanding of disorders Scikit-learn: Machine Learning in Python Gaël Varoquaux Bertrand Thirion
of consciousness. J Neurol. 2011;258(7):1373–84. Vincent Dubourg Alexandre Passos Pedregosa, Varoquaux, Gramfort et al.
32. Nakase-Richardson R, Yablon SA, Sherer M, Evans CC, Nick TG. Serial yes/ Matthieu Perrot. J Mach Learn Res 2011;12:2825–30.
no reliability after traumatic brain injury: implications regarding the 44. Chen PHC, Liu Y, Peng L. How to develop machine learning models for
operational criteria for emergence from the minimally conscious state. J healthcare. Nat Mater. 2019;18(5):410–4.
Neurol Neurosurg Psychiatry. 2008;79(2):216–8. 45. Willacker L, Raiser TM, Bassi M, et al. PerBrain: a multimodal approach to
33. Kondziella D. Roald Dahl and the complete locked-in syndrome: “Cold personalized tracking of evolving state-of-consciousness in brain-injured
dead body, living brain.” J Neurol Sci. 2017;379:276–8. patients: protocol of an international, multicentric, observational study.
34. Acharya JN, Acharya VJ. Overview of EEG montages and principles of BMC Neurol. 2022. https://​doi.​org/​10.​1186/​s12883-​022-​02958-x.
localization. J Clin Neurophysiol. 2019;36(5):325–9. 46. Sandroni C, Cariou A, Cavallaro F, et al. Prognostication in comatose sur-
35. Synek VM. Prognostically important EEG coma patterns in diffuse vivors of cardiac arrest: an advisory statement from the European Resusci-
anoxic and traumatic encephalopathies in adults. J Clin Neurophysiol. tation Council and the European Society of Intensive Care Medicine.
1988;5(2):161–74. Resuscitation. 2014;85(12):1779–89.
36. Forgacs PB, Frey HP, Velazquez A, et al. Dynamic regimes of neocorti- 47. McMillan T, Wilson L, Ponsford J, Levin H, Teasdale G, Bond M. The
cal activity linked to corticothalamic integrity correlate with outcomes glasgow outcome scale-40 years of application and refinement. Nat Rev
in acute anoxic brain injury after cardiac arrest. Ann Clin Transl Neurol. Neurol. 2016;12(8):477–85.
2017;4(2):119–29.
37. Engemann DA, Raimondo F, King JR, et al. Robust EEG-based cross-
site and cross-protocol classification of states of consciousness. Brain.
2018;141(11):3179–92.