Clinically unresponsive patients with acute brain injury activate a brain-computer interface

Brain activation to verbal commands can be detected by EEG in clinically unresponsive patients. These patients with cognitive motor dissociation (CMD) are unable to communicate with the environment despite preserved consciousness. Advanced brain-computer interfaces (BCI) are promising to support motor function in patients with motor neuron disease or spinal cord injury. Our primary objective was to determine if clinically unresponsive patients can control a self-paced, one-way BCI early after acute brain injury. Secondarily, we assessed the association of BCI control with short- and long-term outcomes. We conducted a prospective single-centre cohort study in 41 clinically unresponsive, critically ill and acutely brain injured patients, as well as in a convenience sample of 8 healthy volunteers and 22 conscious critically ill patients. The self-paced BCI analysed EEG in real time and modulated frequency and pitch of a tone that patients controlled by attempting a motor task: patients were provided with instructions 'to make the beep sound faster, keep opening and closing your right hand' alternating with 'to make the beep sound slower, stop opening and closing your right hand'. CMD was detected by applying machine learning to EEG recorded while patients listened to motor commands. Clinical outcomes were recovery of consciousness by hospital discharge, assessed with the Coma Recovery Scale-Revised, and recovery of functional independence at 12 months, defined as a Glasgow Outcome Scale-Extended score of 4 or higher (denoting the ability to function independently for 8 h or more). Additionally, we invited conscious patients as well as family members of conscious and clinically unresponsive patients to complete a survey on perceived utility and challenges associated with the BCI. In this study, 13 clinically unresponsive patients (32%) achieved BCI control and 8 of 41 (20%) were CMD-positive. All healthy volunteers and nine conscious patients (41%) were able to control the BCI. Median signal delay for BCI control was 49.8 [interquartile range (IQR) 39.9-75.6] seconds for clinically unresponsive patients, 17.1 (IQR 7.8-19.9) for conscious patients and 15.5 (IQR 12.9-21.2) for healthy volunteers. In the group of 41 clinically unresponsive patients, 4 of 13 patients with (31%) and 1 of 28 without BCI control (4%) recovered clinical consciousness at hospital discharge [odds ratio (OR) 21.2; 95% confidence interval (CI) 2.3-930.7]. At 12 months, 6 of 13 patients with (46%) and 2 of 28 without BCI control (7%) recovered functional independence (OR 9.9; 95% CI 1.8-81.6). None of the patients that were unable to control the BCI and were CMD-negative recovered. Patients with BCI control had shorter times to recovery of functional independence by 12 months than those who did not (hazards ratio 6.0; 95% CI 1.2-30.6). All respondents of the BCI survey (n = 37) expressed strong support for the potential future use of BCI devices. These results suggest that BCI control is a strong predictor of outcomes following acute brain injury and provide initial evidence that a bedside, EEG-based BCI system might permit bidirectional communication in critically ill unresponsive patients.