FRONTIER SIGNAL | DEEP DIVE | BIOTECH & MEDTECH | 7 MIN READ | WEDNESDAY, 28 OCTOBER 2026

A Columbia team asked patients who could not move or speak to imagine opening and closing a hand. About one in three changed the sound they were hearing. The result may matter less for communication than for prognosis.

In an intensive care unit, a patient with a serious brain injury lies still. Eyes may open. Nothing responds to a spoken command. Doctors, and families, are asked to decide what happens next.

For years, researchers have suspected that some of these patients are more aware than they look. Their brains follow commands that their bodies cannot carry out. A new study in the journal Brain goes a step further: it lets such patients act on the world, through a tone that changes pitch and speed when they try.

The test used ordinary scalp electrodes of the kind already in ICUs. The only unusual part was custom software. That is why this is a medtech story as much as a neuroscience one.

13 OF 41 UNRESPONSIVE PATIENTS CONTROLLED IT · 31% VS 4% RECOVERED CONSCIOUSNESS BY DISCHARGE · STANDARD SCALP EEG

Brain, published 22 October 2026. Single-centre cohort: 41 unresponsive patients, 22 conscious patients, 8 healthy volunteers. Outcome figures are associations, not proof the test changes recovery. Authors’ reported figures, taken from the journal abstract and Columbia’s release; we could not open the full text.

Key takeaways

  • A closed loop, not just a scan. Patients tried to imagine or attempt hand opening and closing. A brain-computer interface read their EEG in real time and sped the tone up or slowed it down.

  • About one in three responded. 13 of 41 clinically unresponsive patients (32%) controlled the interface. All healthy volunteers did, and 9 of 22 conscious patients (41%).

  • Control tracked outcomes. Among controllers, 31% had recovered consciousness at hospital discharge, against 4% of non-controllers. At 12 months, 46% were functionally independent, against 7%.

  • It is not conversation. Nobody answered a question. The interface is one-way, and the authors call it early evidence that communication might be possible.

  • No product yet. We found no company, device clearance or regulatory filing attached to the study.

Here is what happened

  • The paper. “Clinically unresponsive patients with acute brain injury activate a brain-computer interface” appeared in Brain (Oxford University Press) on 22 October 2026 (DOI 10.1093/brain/awag210). Jan Claassen, a Columbia neurologist who directs the neurocritical care unit, is the senior and corresponding author. Qi Shen is first author.

  • The design. A prospective study at one centre of 41 unresponsive critically ill patients, 22 conscious critically ill patients and 8 healthy volunteers. Testing began a few days after hospital admission.

  • The funders. NIH, Paris Brain Institute America, the Canadian Institutes of Health Research and others. Columbia’s release reports no competing interests.

  • The coverage. Columbia and EurekAlert on 22 October, News-Medical on 23 October. We found no independent expert reaction.

How it works

  • Listen to the brain. EEG electrodes on the scalp record electrical activity. Trying to open and close a hand produces a recognisable pattern even when no muscle moves. An EEG cap that moves robotic fingers rests on the same idea.

  • Close the loop. Software detects the pattern and changes a tone, so the patient hears their own effort take effect. A recorded voice encourages them and congratulates them on reaching the target.

  • Slower, not absent. Unresponsive patients who controlled the tone took longer to respond, around 50 seconds, against 15 to 17 seconds in conscious patients and volunteers.

  • Compared with earlier work. A 2024 New England Journal of Medicine study of 353 patients found that about a quarter of those with no bedside command-following still showed sustained brain responses on EEG or fMRI, but typically months after injury. This new test runs on routine bedside equipment, in acute patients, and gives feedback.

THE SAME IDEA, WITHOUT THE SURGERY

Columbia reads the brain through the scalp to help patients who cannot move. Meta took the non-invasive route to a different goal: decoding the sentence a person is typing from brain scans. That one is an unreviewed preprint that needs a room-sized scanner, so it is a neighbour, not a competitor.

Why this matters

  • Hidden awareness is common enough to matter. Columbia’s earlier work found hidden consciousness in about 15% of unresponsive brain-injured patients using EEG and AI.

  • A better prognostic signal. The authors describe BCI control as a strong predictor. If it holds up, clinicians and families would have another piece of evidence beside bedside examination.

  • A path to communication. A patient who can steer a tone might one day answer yes or no. That is a hope, not a result.

  • The bar for care. Brain injury decisions, including withdrawal of life support, are made early. Better evidence about awareness can change them. That is why validation matters more than novelty.

Who captures the value

No company owns this result. The study names no spin-out, licence or device partner. The likeliest commercial route is software on existing bedside EEG hardware, not a new device. That is inference, and a regulatory path has not been described.

Ceribell (NASDAQ: CBLL) is the listed pure play on rapid ICU EEG. Second-quarter 2026 revenue was $28.1 million, up 33%, with 92% gross margin and a net loss of $19.3 million. Full-year revenue guidance was raised to $114 million to $117 million. Nothing links Ceribell to this study. It is the nearest listed company to the setting where such software would run.

Precision Neuroscience (private) raised a $250 million Series D on 24 September 2026, bringing its total funding to $430 million, and has FDA 510(k) clearance for a temporary cortical interface from March 2025. That is an implanted device, a different approach from Columbia’s scalp EEG. It shows investor appetite for brain-computer interfaces, not a competitor to this test. A different corner of brain technology, a server rack built on living human neurons, is further still from the bedside.

Other EEG and monitoring makers, among them large medical-device groups, would be natural partners. We did not verify figures for them and do not name them as beneficiaries.

The honest catch

  • Most patients did not respond. 28 of 41 did not control the interface. Failing the test is not proof of absent awareness, so false negatives are possible. The abstract says not every controller was identified by a separate test for hidden consciousness.

  • Outcomes are associations. A single centre, a small sample, and no sign in what we read that severity or treatment decisions were fully adjusted for. Patients who do better may simply be less severely injured. For how to read an early, small-cohort result, see our look at a six-patient gene-editing trial.

  • Missing details. The electrode count, per-patient accuracy, training time, injury types and trial registration were not in the sources we could read.

  • It does not change care yet. Columbia’s release says tools for identifying hidden consciousness are not yet available outside research.

  • Source limits. We read the journal abstract page, the Columbia release and press coverage. The full text is paywalled and the abstract figures were relayed through a page summariser, so check them against the paper before quoting.

The one question every reader should ask

Does a patient’s response to this test improve decisions, or only add a number to a hard conversation?

A prognostic signal is useful only if it is reliable, available at the bedside and used with care. Without trials that change practice and show better outcomes, it remains a research finding.

OUR VIEW

This is a strong, careful clinical study with a humane purpose. The most solid claim is the link between BCI control and outcomes, and the most exciting is the least proven: communication.

For readers watching medtech, the signal is that brain-computer interfaces are not only implants. A software layer on existing EEG is the cheaper, faster route to the ICU. Whether anyone builds a business on it depends on validation, regulation and reimbursement, none of which exist yet.

What to watch next

  • Replication. Do other hospitals reproduce the 13-of-41 result?

  • A two-way test. Can patients answer questions reliably?

  • Trials that change care. Does using the test alter treatment and improve outcomes?

  • A commercial partner. A licence, spin-out or EEG-maker partnership.

  • Regulation. A cleared pathway for software that reads awareness.

Quick questions

What did the study find?

About one in three clinically unresponsive ICU patients (13 of 41) could change a tone by imagining or attempting hand movements, as read by a standard EEG and custom software.

Does this mean these patients can communicate?

No. The interface is one-way. The authors describe it as early evidence that communication may become possible.

Why does it matter for recovery?

Patients who controlled the interface were more likely to have recovered consciousness at discharge (31% versus 4%) and to be functionally independent at 12 months (46% versus 7%). These are associations in a small single-centre study.

Is there a product or company behind it?

Not that we found. The study names no company, device or regulatory filing. Listed and private neurotechnology firms are not tied to it.

How is it different from earlier tests?

It uses routine bedside EEG in acute patients and gives the patient live feedback. Earlier research mostly detected responses without feedback, often later in recovery or with fMRI.

READ MORE: BIOTECH SIGNALS

Subscribe free, the frontier in plain English →

Sources

Frontier Signal explains frontier technology in plain English. Figures are from the journal abstract, the Columbia release and company filings; the full paper was not reviewed. This is general information, not medical or investment advice.

Keep Reading

View more
caret-right