Unmasking Covert Language Processing in the Intensive Care Unit with Electroencephalography

For patients with severe brain injuries in the intensive care unit (ICU), consciousness can be present in the absence of purposeful responses on bedside examination. ¹ , ² If consciousness goes undetected, clinicians can render an inaccurate, pessimistic prognosis, increasing the risk that life‐sustaining therapy is withdrawn for a patient who had the potential for neurological recovery. Withdrawal of life‐sustaining therapy is a leading cause of death in ICU patients with severe traumatic brain injury and hypoxic–ischemic injury, ³ , ⁴ and early re‐emergence of consciousness is a primary determinant of decisions regarding withdrawal of life‐sustaining therapy. ⁵ Detection of consciousness in the ICU thus has time‐sensitive, life‐or‐death consequences.

The prognostic relevance of early detection of consciousness in the ICU has been demonstrated primarily using behavioral examination. ⁶ The first signs of re‐emergence of consciousness are typically visual fixation, gaze tracking, and pain localization, ⁷ but multiple confounders limit the behavioral examination, including pain, immobility from polytrauma, and disruption of central or peripheral motor pathways. Furthermore, patients may require continuous sedation or tolerate only brief, intermittent examinations owing to elevations in intracranial pressure, ventilator dyssynchrony, or bronchospasm. To circumvent these limitations, advanced electrophysiologic and neuroimaging tests have been developed to detect covert consciousness (ie, cognitive motor dissociation ⁸ ), which is present in approximately 15% of patients ⁹ whose behavioral examination suggests a vegetative state (also known as unresponsive wakefulness syndrome). Task‐based electroencephalography (EEG) and functional magnetic resonance imaging (fMRI) protocols initially developed for patients with subacute‐to‐chronic disorders of consciousness (DoC) ¹⁰ , ¹¹ , ¹² , ¹³ are now being deployed in the ICU for patients with acute DoC. ¹ , ² The recent observation that early detection of covert consciousness using task‐based EEG predicts 1‐year functional recovery ² suggests that covert consciousness has similar prognostic relevance to overt consciousness in the ICU.

Nonetheless, task‐based EEG and fMRI protocols pose substantial cognitive demands on critically ill patients, as evidenced by high false‐negative rates in patients who are able to follow commands on behavioral examination but not on task‐based EEG or fMRI. ¹ Even in healthy volunteers, the false‐negative rate for task‐based EEG and fMRI motor imagery protocols (ie, “imagine opening and closing your hand”) is as high as 25%. ¹ , ¹² This high false‐negative rate raises the possibility that consciousness might evade detection by behavioral examination and task‐based EEG or fMRI protocols in some critically ill patients. Moreover, task‐based EEG and fMRI require infrastructure and personnel that are currently unavailable at most hospitals, limiting their clinical utility and generalizability.

The cognitive demand, high false‐negative rate, and limited generalizability of task‐based protocols provide motivation for the ground‐breaking study by Sokoliuk and colleagues ¹⁴ published in this issue of Annals of Neurology. The authors developed an EEG protocol that probes language function under passive stimulation, free of any task instruction. Specifically, the EEG protocol interrogates electrophysiologic responses to three levels of stimulus integration (single words, phrases, and whole sentences) according to frequency tagging in spectral EEG decomposition space. Crucially, all supra‐single word information was removed from the stimuli (eg, there is no prosody effect related to phrase or sentence structure) to isolate the syntactico‐semantic level of integration. This hierarchical language EEG protocol was performed in 28 critically ill patients with severe traumatic brain injury who were unresponsive on bedside examination. The key finding of the study is that EEG responses to language in the ICU were correlated with 3‐ and 6‐month outcomes, as assessed by the Glasgow Outcome Scale‐Extended. The EEG responses accounted for variance in outcomes beyond that accounted for by standard clinical predictors, such as the Glasgow Coma Scale score and head computed tomography scan. These results provide initial evidence that the response of a patient to a language stimulus in the ICU might predict long‐term functional outcomes.

This elegant and promising study capitalizes on four previous findings. First, Tononi and colleagues ¹⁵ used frequency tagging in a binocular rivalry (ie, they used 1 time–frequency for each eye) while recording magnetoencephalography signals to probe cortical networks oscillating at the frequency of the consciously perceived stimulus. Second, fMRI and EEG studies conducted in healthy controls and in patients with DoC showed that hierarchical protocols can be used to investigate cortical networks engaged in processing linguistic structures, ranging from single words to phrases and sentences, in both the visual and auditory modalities. ¹⁶ , ¹⁷ , ¹⁸ , ¹⁹ Third, fMRI studies revealed that probing residual linguistic abilities can improve the accuracy of detecting and predicting recovery of consciousness in patients with DoC. ¹⁶ However, most of these studies, in particular those using EEG, were limited by the weakness of the reported effects. ¹⁸ , ¹⁹ Fourth, Ding and colleagues ²⁰ combined frequency tagging and hierarchical linguistic structures to design the protocol used in the present study. The present results are also strengthened by a recent independent study that reported robust EEG results using a similar approach. ²¹

From a neuroscientific perspective, the new findings by Sokoliuk and colleagues ¹⁴ raise important questions about the neural correlates of consciousness in the human brain. ²² Although the unmasking of cortical responses to integrated and complex linguistic stimuli clearly demonstrates the existence of high‐level cognitive functions, it does not translate unequivocally in terms of conscious versus unconscious processing and state. ²³ , ²⁴ Activation of language networks for stimuli inaccessible to conscious self‐report has been shown reliably in healthy conscious participants (eg, visual masking or attentional blink paradigms) and in patients with neurological conditions, such as left unilateral neglect. ²⁵ More generally, the existence of cortical processing is a necessary but insufficient condition for consciousness. Cortically mediated behaviors and cognitions are not necessarily conscious, either in healthy conscious participants or in patients with DoC. ²⁶ This is why a passive neuronal response is insufficient or equivocal, whereas active, volitional modulation of brain activity provides direct and unequivocal proof of consciousness. Indeed, the concept of covert consciousness (ie, cognitive motor dissociation) emerged from task‐based fMRI and EEG studies that demonstrated active, willful modulation of brain activity. ¹⁰ , ¹¹ , ¹² , ¹³ In contrast, the passive response of the brain to a language stimulus in patients with DoC ¹ , ¹⁶ , ²⁷ or in healthy subjects undergoing anesthesia ¹⁷ has been interpreted historically as providing evidence of perception, but not necessarily of conscious comprehension.

From a clinical‐translational perspective, the present findings raise the possibility that EEG‐based tests for “covert cortical processing” ²⁸ might have a role in ICU prognostication. Although the results of this single‐center study will need to be replicated in larger, multi‐center studies, the test proposed by Sokoliuk and colleagues ¹⁴ is safe, low cost, and feasible to perform at the bedside in the ICU. Even if test characteristics such as sensitivity and specificity for outcome prediction have yet to be elucidated fully, it is not too soon to consider the clinical and ethical implications of implementing this prognostic technique, in addition to similar EEG‐ and fMRI‐based techniques, in the ICU. ⁵ , ²⁹ , ³⁰ The limitations of current prognostic models for individual patients with acute severe traumatic brain injury are well established, ²⁸ , ³¹ because demographic, examination, and conventional neuroimaging data do not account for the large variance in outcomes. Given the time sensitivity and life‐or‐death stakes of early prognostication in the ICU, the potential clinical role of advanced prognostic techniques is being debated actively. ³²

Implementation of advanced electrophysiologic and neuroimaging techniques in the diagnostic classification of states of consciousness is rapidly gaining acceptance, culminating in the incorporation of these techniques into recently published clinical guidelines for patients with DoC. ³³ , ³⁴ , ³⁵ In this context, the new study by Sokoliuk and colleagues ¹⁴ highlights the urgent need to create a new taxonomy for assessing states of consciousness in patients with DoC that combines expert behavioral examination with advanced measures of brain activity.

Author Contributions

The authors contributed equally to this work.

Potential Conflicts of Interest

Nothing to report.