Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2026 Aug 28.
Published in final edited form as: Nat Med. 2025 Mar 3;31(5):1578–1585. doi: 10.1038/s41591-025-03578-x

Sleep spindles as a predictor of cognitive motor dissociation and recovery of consciousness after acute brain injury

Elizabeth E Carroll 1,2,5, Qi Shen 1,5, Vedant Kansara 1, Nicole Casson 1, Andrew Michalak 1,2, Itamar Niesvizky-Kogan 1,2, Jaehyung Lim 1,2, Amy Postelnik 1,2, Matthew J Viereck 1, Satoshi Egawa 1, Joshua Kahan 1,2, Jerina C Carmona 1, Lucie Kruger 1, You Lim Song 1, Angela Velazquez 1, Catherine A Schevon 1, E Sander Connolly 2,3, Shivani Ghoshal 1,2, Sachin Agarwal 1,2, David Roh 1,2, Soojin Park 1,2,4, Paul Kent 1,2, Jan Claassen 1,2,✉
PMCID: PMC13519939  NIHMSID: NIHMS2200828  PMID: 40033114

Abstract

Cognitive motor dissociation (CMD) can improve the accuracy to predict recovery of behaviorally unresponsive patients with acute brain injury, but acquisition and analysis of task-based electroencephalography (EEG) are technically challenging. N2 sleep patterns, such as sleep spindles on EEG, have been associated with good outcomes, rely on similar thalamocortical networks as consciousness and could provide less technically challenging complementary outcome predictors. In this prospective observational cohort study of 226 acutely brain injured patients, well-formed sleep spindles (WFSS) were more likely present in those with CMD than in those without CMD, often preceding the detection of CMD. WFSS were associated with a shorter time to recovery of consciousness, and both CMD and WFSS independently predicted recovery of independence, controlling for age, admission neurological status and injury type. WFSS are seen in approximately every third behaviorally unresponsive patient after acute brain injury, frequently precede detection of CMD and are a promising complementary predictor for recovery of consciousness and functional independence.

Prolonged impairment of consciousness is common after acute brain injury1,2, greatly impacting healthcare systems worldwide3–5. Recovery trajectories of these patients are often uncertain6–10, and delayed recovery is increasingly seen6,8,10–13. Although predicting recovery is imprecise11,13–15, outcomes are often determined by early withdrawal of life-sustaining therapy (WLST)16–21. As such, accurate prognostication is crucial. Beyond behavioral assessments, advanced magnetic resonance imaging (MRI) and electroencephalography (EEG) allow for quantification of brain function at rest22,23 and in response to stimuli23–27. Repeat assessments are critical as patient trajectories early after brain injury fluctuate. EEG is uniquely poised to provide these as measures are obtained at the bedside using widely available technology.

Although the clinical examination has historically been the standard for assessment and definition of the disorders of consciousness (DoC) spectrum28–31, research over the past decade has revealed that covert cognitive capacity hidden from the standard clinical examination exists32,33. Brain activation to motor commands detected on functional MRI or EEG in behaviorally unresponsive patients, also known as cognitive motor dissociation (CMD)32, is seen in 15–25% of patients with acute and chronic brain injury25–27,34,35. CMD was identified as an independent predictor of shorter time to long-term functional recovery27. Currently, CMD implementation is limited by the complexity of data acquisition and analysis36,37, which require time-locked auditory stimulation and EEG analysis using machine learning algorithms. In contrast, resting EEG measures are immediately available at low cost, allowing widespread implementation. In addition to evaluation of various aspects of the resting-state EEG background, presence or absence of sleep features on resting-state EEG in patients with DoC has also been of interest38,39. In preliminary studies, N2 sleep patterns on EEG have been associated with severity of consciousness impairment40,41 and with recovery of consciousness and higher cognitive function42. Underlying mechanisms of CMD, recovery of consciousness and restoration of sleep after acute brain injury are still under investigation, but structural integrity of thalamocortical projections has been implicated in all three12,33.

The objective of the present study was two-fold: first, to determine if N2 sleep patterns improve the accuracy to predict recovery of independence at 12 months after severe brain injury; second, to determine if N2 sleep patterns predict CMD status and recovery of consciousness.

Results

Overall cohort

Between 1 July 2014 and 30 September 2023, 226 (21%) of 1,042 screened patients were enrolled (Extended Data Fig. 1). Included patients more frequently had an admission Glasgow Coma Scale (GCS) score of less than 8 and an admission diagnosis of intracerebral hemorrhage (ICH), traumatic brain injury (TBI) or subarachnoid hemorrhage (SAH) (Supplementary Table 1).

Included cohort

Mean age of included patients was 63 ± 17 years, and 45% were female (Table 1). The most common admission diagnoses were ICH (88 (39%)), cardiac arrest (36 (16%)), SAH (32 (14%)) and TBI/subdural hematoma (27 (12%)). Median admission GCS score was 6 (interquartile range (IQR) 4–8), and, at the time of enrollment, the median Coma Recovery Scale-Revised (CRS-R)28 score was 2 (1–5). The median post-injury day of the first EEG was 4 (2–8). From 577 EEG studies that qualified for screening, we excluded 47 (8%) because more than 4 hours of EEG were missing or uninterpretable. N2 sleep patterns were assessed on 530 overnight EEGs, with a median number of EEGs per patient of 2 (1–3). Thirty-three of the 226 included patients (15%) were CMD positive, which was first detected on median post-injury day 4 (2–8). No differences were observed in age, sex, race, primary language, primary admission diagnosis, admission GCS and time elapsed between injury and first EEG recording for those with or without CMD (Table 2).

Table 1 ∣.

Characteristics of the 226 patients

Characteristics Value
Age—years 62.8 ± 16.8
Female sex—no. (%) 101 (45)
Cause of acute brain injury—no. (%)
 Cardiac arrest 36 (16)
 ICH 88 (39)
 TBI or subdural hematoma 27 (12)
 SAH 32 (14)
 Other 43 (19)
Behavioral assessments
 Median GCS score at admission (IQR)a 6 (4–8)
 Median CRS-R score (IQR)b
 Median 2 (1–5)
 Worst 1 (0–3)
 Best 4 (1–9)
Ethnicity—no. (%)
 Asian 11 (5)
 Black or African American 66 (29)
 White 107 (47)
 More than one race 20 (9)
 Unknown or not reported 22 (10)
EEG recordings to assess for N2 sleep patterns
 Median EEG recordings per patient (IQR)—no. 2 (1–3)
 Median time from onset of acute brain injury to first recording (IQR)—days 4 (2–8)
Patient’s primary language
 English 164 (73)
 Spanish 62 (27)
Sleep features
 Sleep spindles
 Well formed 71 (31)
 Rudimentary 58 (26)
 Symmetric sleep spindles 88 (39)
 Synchronous sleep spindles 107 (47)
 Vertex waves 83 (37)
 K-complexes 82 (36)
 CMD status 33 (15)
Open in a new tab
a

Scores on the GCS range from 3 to 15, with higher scores indicating less neurologic dysfunction.

b

Scores on the CRS-R range from 0 to 23, with higher scores indicating a higher level of consciousness.

Table 2 ∣.

Demographic, clinical and EEG features associated with CMD (n = 226)

CMD (n = 33) Non-CMD (n = 193) OR (95% CI) P value
Age < 64 years 15 (45%) 98 (51%) 1.0 (0.9–1.1) 0.57
Female sex 17 (52%) 84 (44%) 0.7 (0.3–1.5) 0.40
Language 0.09
 English 28 (85%) 136 (71%) Ref
 Spanish 5 (15%) 57 (30%) 0.4 (0.1–1.1)
Cause of brain injury 0.95
 Non-structure 8 (24%) 52 (27%) Ref
 Non-traumatic structural brain injury and not subdural hematoma 21 (64%) 118 (61%) 1.0 (0.9–1.1)
 TBI or subdural hematoma 4 (12%) 23 (12%) 1.0 (0.9–1.2)
Behavioral assessments
 Median GCS score at admission 6 (4–7) 6 (3–8) 0.7 (0.3–1.4) 0.32
 CRS-R score
 Median 2 (1–5) 2 (1–5) 1.0 (0.5–2.0) 0.93
 Worst 1 (0–4) 1 (0–3) 1.0 (0.5–2.0) 0.98
 Best 6 (3–9) 4 (1–9) 1.7 (0.8–3.7) 0.15
Sleep features
 WFSS (present yes/no) 17 (52%) 54 (28%) 1.2 (1.1–1.4) 0.005
 Sleep spindles (ordinal scale) 0.009
 None 7 (21%) 90 (59%) Ref
 Rudimentary 9 (27%) 49 (32%) 2.4 (0.8–7.0)
 Well formed 17 (52%) 54 (28%) 4.0 (1.6–11.0)
 Symmetric sleep spindles 19 (58%) 69 (36%) 1.1 (1.0–1.2) 0.02
 Synchronous sleep spindles 20 (61%) 87 (45%) 1.1 (1.0–1.2) 0.11
 Vertex waves 16 (48%) 67 (35%) 1.8 (0.9–3.8) 0.12
 K-complexes 16 (48%) 66 (35%) 1.8 (0.8–3.7) 0.13
Post-injury day EEG used for sleep staging 4 (2–6) 4 (2–8) 0.3 (0–3.1) 0.31
Open in a new tab

Data are reported as mean ± s.d. or median (IQR or n (%)) as appropriate. Percentages may not total 100 because of rounding. CI denotes confidence interval. ORs for the categorical variables were computed using Fisher’s exact test. Data for continuous and ordinal variables were split according to the median. Bolded values denote statistically significant P values in the multivariate analysis (P < 0.05). Ref, reference.

N2 sleep patterns

Eighty-three of 226 patients (37%) had vertex waves detected; 82 (36%) had K-complexes; and 58 (26%) had rudimentary sleep spindles and 71 (31%) had WFSS (Table 1 and Extended Data Figs. 2 and 3). A strong association was observed between presence of sleep spindles and K-complexes (P < 0.0001, odds ratio (OR) = 49.7 (22.4–109.9)) as well as between sleep spindles and vertex waves (P < 0.0001, OR = 25.4 (12.8–50.4)) (Extended Data Fig. 3). Inter-rater reliability on 45 randomly selected EEGs from 45 unique patients was excellent for the presence of sleep spindles (kappa score = 0.93), K-complexes (0.84), vertex waves (0.87) and symmetric (0.83) and synchronous (1.00) sleep spindles.

N2 sleep patterns and CMD

CMD-positive patients more frequently had N2 sleep patterns present than CMD-negative patients (Table 2). WFSS were more likely to be present in CMD-positive patients than in CMD-negative patients (52% versus 28%, OR = 2.7 (1.3–5.9)).

Sequencing of sleep spindles and CMD testing

WFSS frequently preceded the detection of CMD (14 of 17 patients; time to WFSS was significantly shorter than time to CMD, P < 0.0001; Fig. 1 and Extended Data Table 1). Among CMD-positive patients with WFSS (n = 17), in four (24%) sleep spindles were detected more than 24 hours, in 10 (59%) on the evening before and in one (6%) on the evening of testing positive for CMD (in all these patients, the EEG identifying sleep spindles was the first available EEG assessment). Only in two (12%) patients were WFSS first detected more than 24 hours after testing positive for CMD (Extended Data Figs. 4 and 5).

Fig. 1 ∣. Sequencing of WFSS, CMD and recovery of consciousness.

Fig. 1 ∣

Open in a new tab

a, Time to detection of WFSS was significantly shorter than time to CMD in the overall cohort. b–d, Time to recovery of consciousness (MCS+/EMCS) was shorter for those with CMD (b) and those without CMD (c) and especially for non-CMD patients with WFSS (d). e, Among patients with WFSS, 41% of 17 patients with CMD and 24% of 54 patients without CMD recovered consciousness. Among patients without WFSS, 19% of 16 patients with CMD and 14% of 139 patients without CMD recovered consciousness. Patients who recovered consciousness (defined by MCS+/EMCS) are kept in red, and first detection of WFSS, CMD and MCS+/EMCS is indicated by a green diamond, a blue triangle and a red dot, respectively. The solid line denotes the Kaplan–Meier estimate of overall survival, and the shaded region represents the 95% pointwise confidence intervals, calculated using Greenwood’s formula. Statistical significance was assessed using the log-rank test (via the survdiff function in R). The test was conducted as a two-sided test, and no adjustments for multiple comparisons were applied. The exact P value for a is 6 × 10−15 and for d is 2 × 10−16.

In total, 184 patients had CMD testing and overnight EEG available within 10 days of the injury. In only three (5%) of 58 patients with WFSS, these were first detected after day 10. The association of WFSS with CMD remained significant in this subcohort of 184 patients (Supplementary Table 2). WFSS once detected persisted in 19 of 31 patients (61%) with follow-up EEGs (this was seen for patients both with and without CMD, 66% versus 67%, respectively). Recovery of consciousness (minimally conscious state plus (MCS+) or exit MCS (EMCS), 25% versus 47%, not significant (NS)) and long-term independence (Glasgow Outcome Scale-Extended (GOS-E) ≥4, 33% versus 32%, NS) were seen at similar rates in the 12 remaining patients who had follow-up EEGs but did not have WFSS, when compared to those with EEG and WFSS.

Sedation and EEG findings

Among the 530 EEG recordings that were reviewed, sedative medication was not present in 210 (39%), was classified as ‘minimal’ or ‘low’ sedation in 219 (41%) and was classified as ‘moderate’ sedation in 101 (19%) EEGs. ‘Moderate’ (P = 0.0002, OR = 0.4 (0.3–0.7)) and ‘minimal’ or ‘low’ (P < 0.0001, OR = 0.5 (0.3–0.7)) levels of sedation were both associated with a lower likelihood of WFSS compared to no sedation (Extended Data Fig. 6). No association was observed between CMD status and sedation (P = 0.9, OR 0.97 (0.6–1.7)). In stratified analysis, WFSS were associated with CMD positivity on the day after EEG recordings in those without sedation (P = 0.06, OR = 1.1 (1.0–1.3)) and in those with ‘minimal’ or ‘low’ sedation levels (P = 0.002, OR = 1.2 (1.1–1.4)) but not in those receiving ‘moderate’ sedation levels (P = 0.4, OR = 1.1 (0.9–1.4)). After excluding assessments on moderate sedation (n = 101) from the analysis, WFSS were still more likely to be present in CMD-positive patients compared to CMD-negative patients (45% versus 25%, OR = 2.6 (1.1–5.9)).

Parameters to define recovery of consciousness

When evaluating time to CRS-R ≥ 8 versus time to MCS+/EMCS, we found that time to CRS-R ≥ 8 was significantly earlier than time to MCS+/ECMS (Extended Data Figs. 7 and 8).

CMD, WFSS and recovery of consciousness (MCS+/EMCS)

MCS+/EMCS was seen in 10 patients (30%) with CMD and in 32 patients (17%) without CMD, first detected on post-injury days 9 (6–11) and 8 (6–13), respectively. In a multivariate model, WFSS predicted time to MCS+/EMCS (P = 0.03, hazard ratio (HR) = 1.1 (1.0–1.3)) when considered together with admission GCS, brain injury etiology and age as covariates, but CMD did not (Table 3). This association of WFSS with MCS+/EMCS remained significant when considering only WFSS detected by day 10 (Supplementary Table 3). WFSS were detected earlier than MCS+/EMCS for patients with CMD (P = 0.05) and without CMD (P = 0.03; Fig. 1 and Supplementary Table 3). These findings were similar when replacing MCS+/EMCS with CRS-R ≥ 8 (Supplementary Table 4 and Extended Data Fig. 8).

Table 3 ∣.

Prediction of recovery of consciousness (MCS+/EMCS, seen for 42 patients) by hospital discharge and independence (GOS-E ≥ 4, seen for 38 patients) by 12 months by presence of WFSS and CMD status (n = 226)

Time to MCS+/EMCS Time to GOS-E ≥ 4
HR (95% CI) P value HR (95% CI) P value
Age < 64 years 1.2 (0.6–2.2) 0.56 1.9 (1.0–3.9) 0.06
Admission GCS > 6 1.2 (0.6–2.3) 0.55 2.3 (1.1–4.6) 0.02
Etiology 0.03 0.26
 Non-structure Ref Ref
 Structural, non-traumatic 3.4 (1.2–9.8) 1.3 (0.5–3.3)
 Structural, traumatic 2.4 (0.6–9.3) 2.4 (0.8–7.3)
EEG measures
 WFSS 1.1 (1.0–1.3) 0.03 1.2 (1.0–1.3) 0.02
 CMD 1.6 (0.8–3.3) 0.19 2.6 (1.2–5.6) 0.01
Open in a new tab

Bolded values denote statistically significant P values in the multivariate analysis (P < 0.05). Statistical significance for factors associated with recovery was evaluated using a Fine–Gray subdistribution hazards model. P values were derived from the model’s Wald test (two-sided), and no multiple comparison corrections were applied.

CMD, WFSS and recovery of independence (GOS-E ≥ 4)

By 12 months, the best GOS-E was 8 in one patient (0.4%), 7 in three patients (1%), 6 in six patients (3%), 5 in six patients (3%), 4 in 22 patients (10%), 3 in 65 patients (29%), 2 in 39 patients (17%) and 1 in 18 patients (8%), with 66 patients (29%) who underwent WLST. Presence of WFSS (P = 0.02, HR = 1.2 (1.0–1.3)) and positive CMD status (P = 0.01, HR = 2.6 (1.2–5.6)) independently predicted time to 12-month GOS-E ≥ 4, when considering age, admission GCS and etiology as covariates (Table 3). This association of WFSS with time to GOS-E ≥ 4 remained significant when considering only WFSS recorded by day 10 (Supplementary Table 3).

Accuracy of predicting GOS-E ≥ 4 at hospital discharge and at 3 months did not improve by adding WFSS to CMD and established predictors. However, prediction accuracy was higher but not statistically different at 6 months and significantly higher at 12 months (Fig. 2, left panel; area under the curve (AUC) from 0.79 to 0.89, P = 0.0001). Randomly splitting the dataset confirmed that the performance of a model predicting GOS-E ≥ 4 at 12 months improved by adding WFSS to a median AUC of 0.76 (0.72–0.80; Fig. 2, right panel).

Fig. 2 ∣. Prediction of 12-month independence (GOS-E ≥ 4) using a clinical model with CMD and sleep spindles.

Fig. 2 ∣

Open in a new tab

An improvement was observed in the performance of the model predicting outcomes at all follow-up timepoints by adding CMD and WFSS (left panel). At 12 months, the increase was significant (AUC from 0.79 to 0.89, P = 0.0001). Randomly splitting the dataset of 226 samples confirmed that the performance of a model predicting GOS-E ≥ 4 at 12 months improved by adding WFSS to a median AUC of 0.76 (0.72–0.80; right panel). To estimate model performance, we generated a series (100 repetitions) of prediction models training on randomly selected 80% of the 226 samples to test performance on the 20% hold-out set (each whisker plot representing 100 AUCs, with the horizontal line representing the median, the lower and upper box edges representing the 25th and 75th percentiles and the whiskers extending to data points within 1.5× the IQR from the box edges; dots beyond the whiskers represent outliers). Statistical significance for comparisons was determined using a paired two-sided Student’s t-test. No adjustments for multiple comparisons were applied. **** denotes P < 0.0001 (the exact P value for the comparison between M1 versus M2 is P = 6 × 10−5 and, for M2 versus M3, P = 1 × 10−10). GOS-E levels range from 1 to 8, with higher levels indicating better outcomes. DC, discharge from the hospital.

Overall, patients with WFSS had better outcomes than those without (Table 3), which is particularly true for those without CMD (P = 0.001; Fig. 3). For patients with CMD, those with rudimentary sleep spindles had better outcomes than those without (P < 0.0001). For patients with rudimentary spindles (P = 0.01) or no sleep spindles (P < 0.0001), those with CMD had better outcomes than those without CMD, but no difference was observed for those with WFSS.

Fig. 3 ∣. Sleep spindles and CMD status in relation to outcome prediction.

Fig. 3 ∣

Open in a new tab

CMD-positive patients with rudimentary sleep spindles (P = 0.01) or no sleep spindles (P < 0.0001) had better outcomes than those without CMD. CMD patients with rudimentary sleep spindles had better outcome than those without sleep spindles (P < 0.0001), and non-CMD patients with WFSS had a better outcome than those with rudimentary WFSS (P = 0.001). Improved outcome prediction is denoted by the arrow, pointing in the direction of the cohort of patients with a better outcome. The gray line with blunted ends represents a lack of difference in outcome between the two groups.

After excluding all assessments on moderate sedation (remaining were 429 assessments in 196 patients), we found that WFSS remained an independent predictor of time to MCS+/EMCS and time to GOS-E ≥ 4 (Supplementary Table 5).

Discussion

Among 226 hospitalized patients early after brain injury, WFSS were seen in a third (n = 71, 31%) of patients and frequently preceded the detection of CMD. Sleep spindles predicted recovery of consciousness (defined as MCS+/EMCS as well as CRS-R ≥ 8) and 1-year recovery of independence (GOS-E ≥ 4) together with CMD status after controlling for established predictors. This builds on prior studies that suggested an association between preserved N2 sleep architecture, in particular sleep spindles, and improved outcomes39–47. Our findings support the use of WFSS as low-cost, scalable biomarkers for recovery of consciousness and functional independence. We illustrate here the complementary nature of sleep spindles and CMD status, with sleep spindles not only predicting CMD status but also improving outcome prediction when analyzed in conjunction with CMD (Fig. 2). Evaluation for sleep spindles as a part of a multimodal neuroprognostication assessment may further strengthen outcome prediction23. Detection of sleep spindles may support screening to identify patients with high yield for CMD testing27,35.

Sleep spindles are considered a surface manifestation of synchronized neuronal oscillations arising from deep within the thalamus48. Computational models, low-resolution brain electromagnetic tomography and simultaneous magnetoencephalography/EEG49–51, localize sleep spindle generators and pacemakers to GABAergic thalamic reticular nucleus, with cortical projections to the prefrontal cortex and precuneus48,50. In patients with TBI with diffuse axonal injury, localization for spindle generators remained in the thalamus despite changes in frequency, amplitude and source strength of sleep spindles49.

Broad withdrawal of excitatory synaptic activity due to direct cortical injury or reduced input from thalamic neurons was implicated as a common pathophysiological mechanism for DoC12. With restoration of excitatory neurotransmission as part of cognitive processing across these cortico-cortical and thalamocortical projections, recovery of consciousness can be seen12. Similarly, some degree of sparing of cortico-cortical and thalamocortical pathways is required for sleep spindles to be present40,48. Prior studies implicated intact thalamocortical projections and intact cortico-cortical connectivity as prerequisites for CMD33,52, possibly requiring cohesive underlying functional networks53. This link between consciousness and the thalamocortical networks is further supported by the ABCD framework—a classification system based on EEG power spectra (categories A to D, indicating increasing connectivity)54. These categories correspond to various levels of functional thalamocortical network activity and were demonstrated to correlate with outcome in post–cardiac arrest patients with coma55.

Pharmacologic studies corroborate further a link between sleep physiology and consciousness. In some patients with DoC from brain injury, zolpidem, which, in healthy individuals, promotes sleep, may induce improved levels of consciousness56,57. This paradoxical response was associated with increased cerebral blood flow after zolpidem administration58. It is hypothesized that zolpidem’s inhibition of the globus pallidus interna, which, under normal circumstances, disinhibits the central thalamus, promotes normalization of central thalamic activity58,59.

In the present study, N2 sleep patterns were seen in a third of patients (71/226, 31%) and were particularly common in patients with CMD (17/33, 52%). WFSS were detected before or on the same day as the first positive CMD test in 14 out of 17 (82%) CMD-positive patients who also had WFSS. Time to detection of WFSS was significantly shorter than time to detection of CMD (Fig. 1). Once patients had sleep spindles on EEG, these typically persisted (61%) on follow-up assessments, indicating that they are not transient, regardless of persistence of CMD testing status or fluctuating CRS-R results. This supports the notion that sleep spindles may allow for more robust assessments of functional connectivity in the thalamocortical network than clinical arousal or CMD40.

Predicting which patients are more likely to recover independence is imperative to guide families in decision-making26,27 and possibly to identify patients who may benefit most from early rehabilitation. setting had higher rates of recovery of independence than non-CMD patients26,27. Using CMD alone, false-negative predictions may be seen in 7.5–8.6% of patients, as most patients are CMD negative (75–86%)25–27,34,35. Multimodal prognostication including EEG metrics may be particularly powerful23. Here, we confirmed that CMD independently predicts time to 12-month recovery of independence26,27 and, in addition, that this prediction can be further strengthened by considering WFSS (Fig. 2). Also strengthened by the addition of WFSS is prediction of time to recovery of consciousness; CMD-negative patients were found to have a shorter time to recovery of consciousness if WFSS were detected when compared to CMD-negative patients without WFSS (Fig. 1). Using sleep patterns is a particularly attractive approach as the data can be obtained passively in any hospital or rehabilitation setting and may not require investments in novel technology. Using the CRS-R to derive an endpoint of recovery of consciousness can be approached in more liberal or more restrictive approaches, a few of which are illustrated here. Dichotomizing the total CRS-R score at ≥8 captures a relatively large group of patients with possible conscious awareness, providing a good balance between sensitivity and specificity, but provides little insight into the performance on individual subscales29. A more conservative approach classifies consciousness based on performance in the CRS-R subscales, capturing consciousness with (MCS+ or EMCS) or without recovery of language function (including minimally conscious state minus (MCS−) as well as MCS+ and EMCS)28,29. The association of WFSS with recovery was confirmed with all of these endpoints, strengthening our confidence in the observation.

It should be noted that, in our cohort of 226 patients, 19 of the 139 patients who did not have WFSS and were CMD negative did recover consciousness. Although WFSS may help decrease the false-negative rate of CMD, it is imperative to consider the presence of both WFSS and CMD not as the sole predictors of recovery of consciousness but, instead, as aspects of a multimodal approach to prognostication23, alongside additional modalities, such as structural and functional brain imaging60. In addition to type of modality used, timing of the testing should be considered. As the ideal time for testing has yet to be determined, repeat testing in the weeks after initial evaluation may help to capture patients who are more likely to recover consciousness who were missed on the initial evaluation.

Our data suggest that WFSS frequently precede the detection of CMD, but, given the possibility of false-negative CMD tests, we cannot exclude the possibility that a patient was already conscious at the time of detecting WFSS without being able to clinically express this, and the rather demanding CMD test was a false-negative result. This further supports our argument that CMD testing should be complemented by additional tests that are cognitively less challenging and may be able to detect lower levels of consciousness than CMD testing can.

We acknowledge several limitations to our study. First, our cohort included a variety of underlying brain injuries; however, this is representative of a typical neurological intensive care unit (ICU) population. Second, WLST confounds the natural history of acute brain injuries. To address this, patients who underwent WLST or were lost to follow-up were censored61, and death was treated as a competing risk. Third, the number of EEG assessments performed and time at which they were performed was often dictated by factors outside of study control, including availability of the EEG technician to place and fix EEG leads; inability to place EEG electrodes immediately postoperatively due to bandages; diagnostic tests or treatments that required discontinuation of EEG leads (for example, MRI and surgical procedures); changes in goals of care to comfort and WLST; or death. Additionally, continuous assessments of EEG metrics could have yielded a higher rate and possibly earlier detection of sleep features or CMD status but would have been logistically challenging. Interestingly, WFSS were predictive of 12-month recovery regardless of CMD status and were almost always detected before CMD positivity. These results may suggest a temporal sequence of WFSS, followed by positive CMD testing and, finally, recovery of behavioral responsiveness. It is possible that CMD-negative patients with delayed sleep features would have tested positive for CMD on later follow-ups that were not performed in the study. Fourth, sedation may confound any behavioral or EEG-supported assessment of consciousness, and, unsurprisingly, sedation level was associated with the rate of N2 sleep patterns. However, sedation level was not associated with CMD status, and we were able to detect WFSS across all levels of sedation (Extended Data Fig. 6). Notably, as in prior studies26,27, we excluded patients with barbiturate coma (Extended Data Fig. 1). An additional limitation includes the lack of formal hearing tests for all patients; we instead used a surrogate report of hearing difficulties and inspected the ear canals of those patients presenting with trauma for obstruction. Furthermore, in our study, we used EEG epochs that were obtained as a part of our routine continuous EEG testing in the ICU. As such, we did not use additional electrodes that are often used in sleep studies as recommended by the American Academy of Sleep Medicine62, including electromyography chin electrodes and electro-oculogram electrodes. In addition, we did not capture the frequency or numerosity of sleep spindles on EEG but suggest that this may be an excellent future direction of study. Scarcity of trained specialists to interpret EEG is another limitation to consider63. To overcome this limitation, we would propose the use of training modules available for EEG interpretation, such as those available through the American Clinical Neurophysiology Society64. A future approach that could be considered is development of an automated process to allow for quantification of sleep spindles. This has already been demonstrated to be efficacious in sleep spindle detection in children65 and is a promising option for automatic interpretation of adult EEG in resource-limited settings. Finally, sleep assessments in the ICU are challenging given frequent stimulation for neurological examinations and other disruptions (that is, alarms). The typical circadian rhythm is disrupted in these patients66, and it is possible that the time period of 19:00 to 7:00 on EEG chosen for review may have missed N2 sleep patterns outside of the set time window.

In conclusion, presence of WFSS combined with CMD status improves outcome prediction accuracy compared to CMD status alone.

Online content

Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41591-025-03578-x.

Methods

Patients

Between July 2014 and September 2023, we prospectively screened all patients admitted with acute brain injury to Columbia University Irving Medical Center, New-York-Presbyterian Hospital (New York, NY, USA). The cohort included 226 patients from an ongoing prospective cohort study, of which the results of 184 patients were previously published (93 in Claassen et al.26 and 184 in Egbebike et al.27). Within 3 days of hospital admission, all patients (aged 18 years or older) who were admitted to the neurologic ICU were screened according to underlying admission diagnosis to determine the presence of acute brain injury (Extended Data Fig. 9). A staff of research coordinators in turn screened the daily admission log for any unconscious patient admitted with acute brain injury who met enrollment criteria (see below).

Inclusion and exclusion criteria

Inclusion criteria were as follows: (1) diagnosis of primary ICH (for example, hypertension and anticoagulation); (2) lobar (frontal, temporal, partial or occipital) or subcortical (thalamic, capsular or basal ganglia) location of ICH verified by computerized tomography and/ or MRI; (3) a neurologist independent of the study team applied the CRS-R to determine if the patient was in a coma, vegetative state or MCS− (defined as unresponsive with preserved visual fixation/pursuit or localization to noxious stimuli); (4) age 18 years or older; and (5) pre-morbid fluency in English or Spanish.

Exclusion criteria were as follows: (1) secondary cause of ICH (for example, trauma, neoplasm, vascular malformation and hemorrhagic transformation of ischemic stroke); (2) ICH location in the cerebellum or brainstem; (3) clinically recovered command-following before enrollment; (4) ongoing seizures not controlled with anti-epileptic medication; (5) severe cardiorespiratory compromise; (6) pre-existing DoC; (7) pre-morbid aphasia; (8) deafness (that is, history per chart or family or evidence of or suspected injury to both ear canals); (9) pregnancy; (10) imprisonment; (11) logistical reasons hindering enrollment (that is, research team not available or stimulation apparatus not working); and (12) refusal to participate in the study on the part of the patient or healthcare agent.

Patients were not excluded from screening or enrollment into this study based on sex or race/ethnicity. The study was limited to adults (age 18 years or older), as pediatric patients under the age of 18 who need intensive care treatment are admitted to a different hospital. Age, sex and race/ethnicity were determined based on self-report by the patient or the healthcare agent.

All clinicians and research staff responsible for data collection, including outcomes assessments, were blinded to results of the CMD testing.

Study procedures

EEG recording.

Per ICU management protocol, all unresponsive patients underwent continuous EEG, unless death was imminently expected. EEG was recorded using a digital bedside video EEG monitoring system (Natus Medical, Inc.; low-pass filter = 70 Hz, high-pass filter = 1 Hz, sampling rate = 200 Hz). A standard 21-electrode montage following the international 10–20 system was used. EEG recording also included continuous electrocardiogram monitoring, as is standard in our institution’s continuous EEG monitoring protocol. Additional electrodes used in formal sleep studies, including electromyography chin electrodes and electro-oculogram electrodes, are not a part of our continuous EEG monitoring protocol and, as such, were not used in this study.

Behavioral assessments.

Serial assessments of consciousness were performed using the CRS-R28 scale, including preceding each CMD assessment. Based on performance on the CRS-R subscales, each assessment was classified as coma, vegetative state, MCS−, MCS+ or EMCS, applying criteria of the CRS-R manual29. To allow for the most rigorous analysis, MCS+/EMCS by day 30 after injury was chosen as the primary endpoint for recovery of consciousness, but results for the alternate endpoints (CRS-R ≥ 8 and MCS/EMCS) are provided in the extended data and supplementary materials.

Motor command protocol.

During EEG recordings, instructions alternated between ‘keep opening and closing your right hand’ and ‘stop opening and closing your right hand’67. A total of six blocks each with eight consecutive trials of ‘keep opening…’ and ‘stop opening…’ commands were recorded. Commands were provided in the patient’s native tongue, English or Spanish, via single-use headphones. For each patient, we recorded three blocks in which the patient was asked to move the right hand and three blocks for the left hand, alternating right-hand and left-hand commands. Total duration of motor command sessions was approximately 25 minutes.

Sleep staging of the EEG.

EEG clips recorded from 19:00 the evening preceding CMD testing to 7:00 the morning of CMD testing were evaluated by six neurophysiologists for the presence of N2 sleep patterns. Recordings were excluded for artifacts rendering the EEG uninterpretable or if more than 4 hours of the selected time block were missing. If EEG was not available from the evening preceding CMD testing, EEG recordings preceding and closest to the requested period were reviewed instead. Sleep architecture evaluated included K-complexes, vertex waves and sleep spindles. K-complexes were defined as a large amplitude waveform with a negative deflection immediately followed by a positive component68. Vertex waves were defined as sharp transient waves maximum at the vertex, with an amplitude of 50–150 mV (ref. 68). Sleep spindles were defined as moderate amplitude, fronto-centrally maximal waves with frequency of 11–15 Hz (ref. 69). Sleep spindles were described as absent, present but rudimentary if abnormal or atypical appearing (those lacking the typical spindle ‘shape’) or present and well formed70,71. Additionally, we assessed synchronicity and symmetry of N2 sleep patterns and presence or absence of K-complexes and vertex waves.

Inter-rater reliability.

N2 sleep patterns were independently scored by two neurophysiologists for a subset of 20% of the patients. EEGs were randomly picked from the entire patient cohort using a random number generator. Neurophysiologists were blinded to the scoring from the other neurophysiologist, to all clinical data and to CMD status. Inter-rater reliability of N2 sleep patterns was determined by applying Cohen’s kappa statistics72. Scores over 0.80 were classified as excellent agreement, 0.61–0.80 as good agreement, 0.41–0.60 as moderate agreement and below 0.40 as poor agreement73.

Patient outcomes.

Functional outcomes were assessed with the GOS-E at 12 months after injury by structured telephone interview with the patient or with the caregiver if the patient was unable to participate. GOS-E scores ≥4 correspond to the ability to be left up to 8 hours without assistance.

Sedation.

Sedation was categorized according to doses averaged over two previous half-lives (midazolam 24 hours; fentanyl 8 hours; ketamine 6 hours; dexmedetomidine 4 hours; propofol 2 hours). Categories included the following: ‘none’, no sedation; ‘minimal’, only pushes of sedatives allowed; ‘low’, continuously administrated doses of midazolam (≤0.15 mg kg−1 h−1), propofol (≤4 mg kg−1 h−1), fentanyl (≤2 μg kg−1 h−1) or dexmedetomidine (any dose); and ‘moderate’, continuously administrated doses of midazolam (>0.15 mg kg−1 h−1), propofol (>4 mg kg−1 h−1), fentanyl (>2 μg kg−1 h−1) or ketamine (any dose)26,74.

Statistical analysis

Logistic regression was conducted to evaluate the relationship among sleep features, CMD and covariates at the patient level. Additionally, a random effects model was applied to account for repeated measures, exploring associations between CMD and sleep features, correcting for covariates (that is, sedation) at the assessment level (R package GLM version 4.1-2). Independent predictors of time to recovery of consciousness were identified using survival analysis (R package survival version 3.2-7). Kaplan–Meier estimates were plotted to compare the cumulative event rates for time to WFSS, CMD and recovery of consciousness, and a log-rank test was performed to compare cumulative event rate curves (R package survminer version 0.4.8). Independent predictors of time to GOS-E ≥ 4 were identified using a sub-distribution hazard model27, with death as a competing risk, and patients who underwent WLST or were lost to follow-up were censored61 (R package tidycmprsk version 1.0.0). We then split the data into an original26 and a secondary cohort27 to quantify the improvement of predicting GOS-E ≥ 4, adding WFSS to clinical predictors with or without CMD using a risk regression model75. Further validation of the entire cohort was performed with 100 repetitions using an 80% training and 20% testing split, using stratified sampling to address class imbalance (R package riskRegression version 2023.03.22).

Categorical variables were reported as counts (percentages) and continuous variables as means (s.d.) or medians (IQR), depending on distribution. Associations between variables and outcomes were assessed using the Wilcoxon rank-sum test for continuous variables and chi-squared tests for categorical variables. All tests were two-tailed and were performed using R (version 4.0.3).

Ethics and inclusion statement

This study was approved by the Columbia University Institutional Review Board (IRB AAAS3574 and IRB AAAR3191). We obtained written informed consent from the patients’ healthcare proxy (HCP) or legally authorized representative (LAR). Patients who recovered consciousness at the time of follow-up were given the opportunity to withdraw from the study. There was no industry support for or involvement in the study, and all authors vouch for the completeness and accuracy of the data as well as fidelity to the study protocol.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Extended Data

Extended Data Fig. 1 ∣. Consort diagram.

Extended Data Fig. 1 ∣

Open in a new tab

EEG, electroencephalography; WLST, withdrawal of list sustaining treatment.

Extended Data Fig. 2 ∣. Samples of EEG recordings with K-complex followed by a well-formed sleep spindle (a), well formed (b & c), and rudimentary (c&d) sleep spindles (e).

Extended Data Fig. 2 ∣

Open in a new tab

The green box indicates K-complex followed by a well-formed sleep spindle (Extended Data Fig. 2A), the light blue box indicates well-formed sleep spindles (Extended Data Fig. 2B and 2C), and the purple box indicates rudimentary sleep spindles (Extended Data Fig. 2D and 2E).

Extended Data Fig. 3 ∣. Presence of K-complexes and Vertex waves and their relationship to sleep spindles in CMD and non-CMD patients.

Extended Data Fig. 3 ∣

Open in a new tab

Presence of K-complexes and vertex waves in the entire patient cohort (a), CMD positive patients (b), and CMD negative patients (c) was highly correlated with concomitant presence of sleep spindles.

Extended Data Fig. 4 ∣. Post-injury timeline of CMD testing and presence of sleep spindles in CMD positive patients.

Extended Data Fig. 4 ∣

Open in a new tab

Well-formed sleep spindles frequently preceded the detection of CMD.

Extended Data Fig. 5 ∣. Post-injury timeline of CMD testing and presence of sleep spindles in CMD negative patients.

Extended Data Fig. 5 ∣

Open in a new tab

Of those CMD negative patients with delayed sleep spindles, well-formed sleep spindles were still predictive of recovery of consciousness, and 12-month outcome. Similarly to CMD positive patients, when well-formed sleep spindles were present on EEG, they were likely to persist on subsequent EEG recordings.

Extended Data Fig. 6 ∣. Level of sedation at the time of EEG recording and associated presence and quality of sleep spindles (N = 530).

Extended Data Fig. 6 ∣

Open in a new tab

Patients receiving “moderate” (p = 0.0002, OR = 0.4 [0.3, 0.7]) and “minimal” or “low” levels of sedation (p < 0.0001 OR = 0.5 [0.3, 0.7]) were less likely to have well-formed sleep spindles present when compared to those with no sedation.

Extended Data Fig. 7 ∣. Time to recovery of consciousness defined by CRS-R ≥ 8 and MCS + /EMCS.

Extended Data Fig. 7 ∣

Open in a new tab

Time to CRS-R ≥ 8 is significantly shorter than time to MCS + / EMCS (p = 0.04).

Extended Data Fig. 8 ∣. Sequencing of WFSS, CMD, and CRS-R ≥ 8.

Extended Data Fig. 8 ∣

Open in a new tab

Time to detect WFSS is significantly shorter than time to detect CMD in the overall cohort (Panel a). Time to CRS-R ≥ 8 is shorter for those with CMD (Panel b) but not for those without CMD (Panel c) except for non-CMD patients with WFSS (Panel d). Amongst patients with WFSS, 35% (6 of 17) of patients with and 46% (25 of 54) of patients without CMD recovered consciousness (Panel e). Amongst patients without WFSS, 25% (4 of 16) of patients with and 21% (30 of 139) of patients without CMD recovered consciousness. Patients with CRS-R ≥ 8 are kept in orange and first detecting WFSS, CMD and CRS-R are indicated by a green diamond, blue triangle, and orange dot, respectively. CMD, cognitive motor dissociation; CRS-R, coma recovery scale-revised; well-formed sleep spindles, WFSS.

Extended Data Fig. 9 ∣. Study Timeline.

Extended Data Fig. 9 ∣

Open in a new tab

This schematic depicts the timeline of testing and results for an exemplary patient found to have well-formed sleep spindles, followed by positive CMD testing, and ultimately recovery of consciousness and long-term functional recovery.

Extended Data Table 1 ∣.

Average time to WFSS, positive CMD testing and recovery of consciousness (MCS+/EMCS)

Average time (in days from injury)
Group Patients with MCS+/EMCS To first detection of WFSS To first positive CMD test To first detecting MCS+/EMCS Death/WLST
WFSS+/CMD+ (N=17) 7(41%) 3.5 (2-3.5) 4 (2.5-5) 8 (7-11.5) 2 (12%)
WFSS+/CMD− (N=54) 13 (24%) 4.5 (2.5,10.5) NA 9 (5-13) 12 (22%)
WFSS−/CMD+ (N=16) 3 (19%) NA 7 (2.5-9.75) 10 (7-13.5) 7 (44%)
WFSS−/CMD− (N=139) 19 (14%) NA NA 8 (6-11.5) 52 (37%)
Open in a new tab

Recovery of consciousness is defined as detecting MCS+ or EMCS. EMCS by day 30 after injury, exit minimally conscious state; NA, not applicable.

Supplementary Material

Supplementary information

Supplementary information The online version contains supplementary material available at https://doi.org/10.1038/s41591-025-03578-x.

Acknowledgements

We thank the nurses, attending physicians, fellows and neurology residents of the neuroscience ICU for their overall support of this project. We are grateful to the National Institutes of Health (NIH)/National Institute of Neurological Disorders and Stroke (NS106014; LM011826) and the Clinical and Translational Science Awards (UL1TR001873 from the National Center for Advancing Translational Sciences/NIH) for support of this study.

Footnotes

Competing interests

The authors declare no competing interests. J. Claassen is a minority shareholder at iCE Neurosystems, but this amounts to less than $10,000 and less than 5% equity in the company. No technology from iCE Neurosystems was used for any of the study procedures, data acquisition or analysis presented here. None of the patients included in this study were managed using any technology from iCE Neurosystems.

Extended data is available for this paper at https://doi.org/10.1038/s41591-025-03578-x.

Peer review information Nature Medicine thanks Brian Edlow, Daniel Kondziella and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Primary Handling Editor: Jerome Staal, in collaboration with the Nature Medicine team.

Data availability

The primary patient-level data that support all the main analyses are available as part of Extended Data Figs. 4 and 5, including sleep data, CMD status, sedation data and outcomes data. Other individual patient data are not openly available due to reasons of sensitivity for patient privacy reasons.

Code availability

The code to analyze EEG for cognitive motor dissociation was previously shared (see Claassen et al.26). All other codes used based on publicly available R packages are non-proprietary (R version 4.0.3, R package GLM version 4.1-2, R package survival version 3.2-7, R-package survminer version 0.4.8, R package tidycmprsk version 1.0.0 and R package riskRegression version 2023.03.22).

References

  • 1.Kondziella D. et al. European Academy of Neurology guideline on the diagnosis of coma and other disorders of consciousness. Eur. J. Neurol. 27, 739–740 (2020). [DOI] [PubMed] [Google Scholar]
  • 2.Kondziella D. et al. A precision medicine framework for classifying patients with disorders of consciousness: Advanced Classification of Consciousness Endotypes (ACCESS). Neurocrit. Care 35, 27–36 (2021). [DOI] [PubMed] [Google Scholar]
  • 3.Tsao CW. et al. Heart disease and stroke statistics—2023 update: a report from the American Heart Association. Circulation 147, e93–e621 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 4.Kittner SJ. et al. Ethnic and racial variation in intracerebral hemorrhage risk factors and risk factor burden. JAMA Netw. Open 4, e2121921 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 5.Maas AIR. et al. Traumatic brain injury: progress and challenges in prevention, clinical care, and research. Lancet Neurol. 21, 1004–1060 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 6.Russell ME, Hammond FM. & Murtaugh B. Prognosis and enhancement of recovery in disorders of consciousness. NeuroRehabilitation 54, 43–59 (2024). [DOI] [PubMed] [Google Scholar]
  • 7.Fitzgerald E. et al. Functional outcomes at 12 months for patients with traumatic brain injury, intracerebral haemorrhage and subarachnoid haemorrhage treated in an Australian neurocritical care unit: a prospective cohort study. Aust. Crit. Care 33, 497–503 (2020). [DOI] [PubMed] [Google Scholar]
  • 8.Karras CL. et al. Outcomes following penetrating brain injuries in military settings: a systematic review and meta-analysis. World Neurosurg. 166, 39–48 (2022). [DOI] [PubMed] [Google Scholar]
  • 9.Ponfick M, Wiederer R. & Nowak DA. Outcome of intensive care unit–dependent, tracheotomized patients with cerebrovascular diseases. J. Stroke Cerebrovasc. Dis. 24, 1527–1531 (2015). [DOI] [PubMed] [Google Scholar]
  • 10.Humble SS. et al. Prognosis of diffuse axonal injury with traumatic brain injury. J. Trauma Acute Care Surg. 85, 155–159 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 11.Hayamizu M. et al. Delayed neurologic improvement and long-term survival of patients with poor neurologic status after out-of-hospital cardiac arrest: a retrospective cohort study in Japan. Resuscitation 188, 109790 (2023). [DOI] [PubMed] [Google Scholar]
  • 12.Edlow BL, Claassen J, Schiff ND. & Greer DM. Recovery from disorders of consciousness: mechanisms, prognosis and emerging therapies. Nat. Rev. Neurol. 17, 135–156 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 13.Egawa S. et al. Long-term outcomes of patients with stroke predicted by clinicians to have no chance of meaningful recovery: a Japanese cohort study. Neurocrit. Care 38, 733–740 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 14.Dijkland SA. et al. Prognosis in moderate and severe traumatic brain injury: a systematic review of contemporary models and validation studies. J. Neurotrauma 37, 1–13 (2020). [DOI] [PubMed] [Google Scholar]
  • 15.Witsch J. et al. Prognostication after intracerebral hemorrhage: a review. Neurol. Res. Pract. 3, 22 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 16.Turgeon AF. et al. Mortality associated with withdrawal of life-sustaining therapy for patients with severe traumatic brain injury: a Canadian multicentre cohort study. CMAJ 183, 1581–1588 (2011). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 17.Elmer J. et al. Long-term survival benefit from treatment at a specialty center after cardiac arrest. Resuscitation 108, 48–53 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 18.Giacino JT. et al. Behavioral recovery and early decision making in patients with prolonged disturbance in consciousness after traumatic brain injury. J. Neurotrauma 37, 357–365 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 19.Giacino JT. 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 91, 450–460 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 20.Alkhachroum A. et al. Withdrawal of life-sustaining treatment mediates mortality in patients with intracerebral hemorrhage with impaired consciousness. Stroke 52, 3891–3898 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 21.Alkhachroum A. et al. Association of acute alteration of consciousness in patients with acute ischemic stroke with outcomes and early withdrawal of care. Neurology 98, e1470–e1478 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 22.Amiri M. et al. Multimodal prediction of residual consciousness in the intensive care unit: the CONNECT-ME study. Brain 146, 50–64 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 23.Rohaut B. et al. Multimodal assessment improves neuroprognosis performance in clinically unresponsive critical-care patients with brain injury. Nat. Med. 30, 2349–2355 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 24.Owen AM. et al. Detecting awareness in the vegetative state. Science 313, 1402 (2006). [DOI] [PubMed] [Google Scholar]
  • 25.Edlow BL. et al. Early detection of consciousness in patients with acute severe traumatic brain injury. Brain 140, 2399–2414 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 26.Claassen J. et al. Detection of brain activation in unresponsive patients with acute brain injury. N. Engl. J. Med. 380, 2497–2505 (2019). [DOI] [PubMed] [Google Scholar]
  • 27.Egbebike J. et al. Cognitive-motor dissociation and time to functional recovery in patients with acute brain injury in the USA: a prospective observational cohort study. Lancet Neurol. 21, 704–713 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 28.Giacino JT, Kalmar K. & Whyte J. The JFK Coma Recovery Scale-Revised: measurement characteristics and diagnostic utility. Arch. Phys. Med. Rehabil. 85, 2020–2029 (2004). [DOI] [PubMed] [Google Scholar]
  • 29.Bodien YG, Carlowicz CA, Chatelle C. & Giacino JT. Sensitivity and specificity of the Coma Recovery Scale–Revised total score in detection of conscious awareness. Arch. Phys. Med. Rehabil. 97, 490–492 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 30.Giacino JT. et al. The minimally conscious state: definition and diagnostic criteria. Neurology 58, 349–353 (2002). [DOI] [PubMed] [Google Scholar]
  • 31.Giacino JT. et al. Comprehensive systematic review update 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. Arch. Phys. Med. Rehabil. 99, P1710–P1719 (2018). [DOI] [PubMed] [Google Scholar]
  • 32.Schiff ND. Cognitive motor dissociation following severe brain injuries. JAMA Neurol. 72, 1413–1415 (2015). [DOI] [PubMed] [Google Scholar]
  • 33.Fernández-Espejo D, Rossit S. & Owen AM. A thalamocortical mechanism for the absence of overt motor behavior in covertly aware patients. JAMA Neurol. 72, 1442–1450 (2015). [DOI] [PubMed] [Google Scholar]
  • 34.Sanders WR. et al. Recovery potential in patients who died after withdrawal of life-sustaining treatment: a TRACK-TBI propensity score analysis. J. Neurotrauma 41, 2336–2348 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 35.Bodien YG. et al. Cognitive motor dissociation in disorders of consciousness. N. Engl. J. Med. 391, 598–608 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 36.Claassen J. et al. Cognitive motor dissociation: gap analysis and future directions. Neurocrit. Care 40, 81–98 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 37.Jacobson SD. et al. Impact of aphasia on brain activation to motor commands in patients with acute intracerebral hemorrhage. Neurocrit. Care 10.1007/s12028-024-02086-z (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 38.Raciti L. et al. Sleep in disorders of consciousness: a brief overview on a still under investigated issue. Brain Sci. 13, 275 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 39.Sandsmark DK. et al. Sleep features on continuous electroencephalography predict rehabilitation outcomes after severe traumatic brain injury. J. Head Trauma Rehabil. 31, 101–107 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 40.Gottshall JL. & Rossi Sebastiano D. Sleep in disorders of consciousness: diagnostic, prognostic, and therapeutic considerations. Curr. Opin. Neurol. 33, 684–690 (2020). [DOI] [PubMed] [Google Scholar]
  • 41.Rossi Sebastiano D. et al. Sleep patterns associated with the severity of impairment in a large cohort of patients with chronic disorders of consciousness. Clin. Neurophysiol. 129, 687–693 (2018). [DOI] [PubMed] [Google Scholar]
  • 42.Duclos C. et al. Parallel recovery of consciousness and sleep in acute traumatic brain injury. Neurology 88, 268–275 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 43.Forgacs PB. et al. Preservation of electroencephalographic organization in patients with impaired consciousness and imaging-based evidence of command-following. Ann. Neurol. 76, 869–879 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 44.Grigg-Damberger MM, Hussein O. & Kulik T. Sleep spindles and K-complexes are favorable prognostic biomarkers in critically ill patients. J. Clin. Neurophysiol. 39, 372–382 (2022). [DOI] [PubMed] [Google Scholar]
  • 45.Van Der Lande GJM, Blume C. & Annen J. Sleep and circadian disturbance in disorders of consciousness: current methods and the way towards clinical implementation. Semin. Neurol. 42, 283–298 (2022). [DOI] [PubMed] [Google Scholar]
  • 46.Arnaldi D. et al. The prognostic value of sleep patterns in disorders of consciousness in the sub-acute phase. Clin. Neurophysiol. 127, 1445–1451 (2016). [DOI] [PubMed] [Google Scholar]
  • 47.Yang XA. et al. Prognostic roles of sleep electroencephalography pattern and circadian rhythm biomarkers in the recovery of consciousness in patients with coma: a prospective cohort study. Sleep Med. 69, 204–212 (2020). [DOI] [PubMed] [Google Scholar]
  • 48.Fernandez LMJ. & Lüthi A. Sleep spindles: mechanisms and functions. Physiol. Rev. 100, 805–868 (2020). [DOI] [PubMed] [Google Scholar]
  • 49.Urakami Y. Relationship between sleep spindles and clinical recovery in patients with traumatic brain injury: a simultaneous EEG and MEG study. Clin. EEG Neurosci. 43, 39–47 (2012). [DOI] [PubMed] [Google Scholar]
  • 50.De Gennaro L. & Ferrara M. Sleep spindles: an overview. Sleep Med. Rev. 7, 423–440 (2003). [DOI] [PubMed] [Google Scholar]
  • 51.Anderer P. et al. Low-resolution brain electromagnetic tomography revealed simultaneously active frontal and parietal sleep spindle sources in the human cortex. Neuroscience 103, 581–592 (2001). [DOI] [PubMed] [Google Scholar]
  • 52.Franzova E. et al. Injury patterns associated with cognitive motor dissociation. Brain 146, 4645–4658 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 53.Zhu H. et al. Spectral-switching analysis reveals real-time neuronal network representations of concurrent spontaneous naturalistic behaviors in human brain. Preprint at bioRxiv 10.1101/2024.07.08.600416 (2024). [DOI] [Google Scholar]
  • 54.Schiff ND. in Brain Function and Responsiveness in Disorders of Consciousness (eds Monti MM. & Sannita WG) 195–204 (Springer, 2016). [Google Scholar]
  • 55.Forgacs PB. et al. Dynamic regimes of neocortical activity linked to corticothalamic integrity correlate with outcomes in acute anoxic brain injury after cardiac arrest. Ann. Clin. Transl. Neurol. 4, 119–129 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 56.Du B. et al. Zolpidem arouses patients in vegetative state after brain injury: quantitative evaluation and indications. Am. J. Med. Sci. 347, 178–182 (2014). [DOI] [PubMed] [Google Scholar]
  • 57.Whyte J. & Myers R. Incidence of clinically significant responses to zolpidem among patients with disorders of consciousness: a preliminary placebo controlled trial. Am. J. Phys. Med. Rehabil. 88, 410–418 (2009). [DOI] [PubMed] [Google Scholar]
  • 58.Sutton JA. & Clauss RP. A review of the evidence of zolpidem efficacy in neurological disability after brain damage due to stroke, trauma and hypoxia: a justification of further clinical trials. Brain Inj. 31, 1019–1027 (2017). [DOI] [PubMed] [Google Scholar]
  • 59.Schiff ND. Recovery of consciousness after brain injury: a mesocircuit hypothesis. Trends Neurosci. 33, 1–9 (2010). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 60.Stender J. et al. Diagnostic precision of PET imaging and functional MRI in disorders of consciousness: a clinical validation study. Lancet 384, 514–522 (2014). [DOI] [PubMed] [Google Scholar]
  • 61.Fine J. & Gray R. A proportional hazards model for the subdistribution of a competing risk. J. Am. Stat. Assoc. 94, 496–509 (1999). [Google Scholar]
  • 62.Berry RB. et al. The AASM Manual for the Scoring of Sleep and Associated Events: Rules, Terminology and Technical Specifications version 2.6 (American Academy of Sleep Medicine, 2020). [Google Scholar]
  • 63.Majersik JJ. et al. A shortage of neurologists—we must act now: a report from the AAN 2019 Transforming Leaders Program. Neurology 96, 1122–1134 (2021). [DOI] [PubMed] [Google Scholar]
  • 64.Hirsch LJ. et al. American Clinical Neurophysiology Society’s standardized critical care EEG terminology: 2021 version. J. Clin. Neurophysiol. 38, 1–29 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 65.Kwon H. et al. Sleep spindles in the healthy brain from birth through 18 years. Sleep 46, zsad017 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 66.Felten M. et al. Circadian rhythm disruption in critically ill patients. Acta Physiol. 238, e13962 (2023). [DOI] [PubMed] [Google Scholar]
  • 67.Curley WH, Forgacs PB, Voss HU, Conte MM. & Schiff ND. Characterization of EEG signals revealing covert cognition in the injured brain. Brain 141, 1404–1421 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 68.Wolpert EA. A Manual of Standardized Terminology, Techniques and Scoring System for Sleep Stages of Human Subjects. Arch. Gen. Psychiatry 20, 246–247 (1969). [Google Scholar]
  • 69.Purcell SM. et al. Characterizing sleep spindles in 11,630 individuals from the National Sleep Research Resource. Nat. Commun. 8, 15930 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 70.Valente M. et al. Sleep organization pattern as a prognostic marker at the subacute stage of post-traumatic coma. Clin. Neurophysiol. 113, 1798–1805 (2002). [DOI] [PubMed] [Google Scholar]
  • 71.Ambrogio C, Koebnick J, Quan SF, Ranieri M. & Parthasarathy S. Assessment of sleep in ventilator-supported critically III patients. Sleep 31, 1559–1568 (2008). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 72.McHugh ML. Interrater reliability: the kappa statistic. Biochem. Med. 22, 276–282 (2012). [PMC free article] [PubMed] [Google Scholar]
  • 73.Landis J. & Koch G. The measurement of observer agreement for categorical data. Biometrics 33, 159–174 (1977). [PubMed] [Google Scholar]
  • 74.Claassen J. et al. Bedside quantitative electroencephalography improves assessment of consciousness in comatose subarachnoid hemorrhage patients. Ann. Neurol. 80, 541–553 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 75.Blanche P. et al. Quantifying and comparing dynamic predictive accuracy of joint models for longitudinal marker and time-to-event in presence of censoring and competing risks. Biometrics 71, 102–113 (2015). [DOI] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supplementary information
NIHMS2200828-supplement-Supplementary_information.pdf (350.7KB, pdf)

Data Availability Statement

The primary patient-level data that support all the main analyses are available as part of Extended Data Figs. 4 and 5, including sleep data, CMD status, sedation data and outcomes data. Other individual patient data are not openly available due to reasons of sensitivity for patient privacy reasons.

The code to analyze EEG for cognitive motor dissociation was previously shared (see Claassen et al.26). All other codes used based on publicly available R packages are non-proprietary (R version 4.0.3, R package GLM version 4.1-2, R package survival version 3.2-7, R-package survminer version 0.4.8, R package tidycmprsk version 1.0.0 and R package riskRegression version 2023.03.22).

RESOURCES