Every trial design argument in rare disease eventually reaches randomization and blinding. For brain-computer interfaces, one of those two is not available, and the reason is worth understanding because it is not a limitation anyone can engineer around.

A speaker at the device session of the FDA and Duke-Margolis rare disease statistics workshop on September 29 said he did not believe participants can be blinded to BCI status in any effective or ethical way. Not because of the disease, and not because of anything particular to medical devices. Because of the direction the effect travels.

In a drug trial, and in most device trials, the intervention acts on the body. With an implanted BCI, the effect runs the other way, from the body out into the world. A participant either moves a cursor or does not.

“There is no placebo-induced telepathy. A computer will not click by itself through placebo.”

Device session speaker, on why BCI trials cannot be blinded

What Breaks When Blinding Is Impossible

The consequences run further than the blind itself. Without it there is no equipoise, which complicates randomized designs rather than merely inconveniencing them. The speaker was candid that he did not have a solution and invited the room to offer one.

It also contaminates a whole class of endpoints. Quality of life and other patient-reported measures carry response bias when the participant knows their assignment, so the certainty of evidence those endpoints generate is lower than the same instruments would produce in a blinded drug trial. For a technology whose entire value proposition is subjective restoration of function, that is an awkward place to be.

What remains available, in his framing: control of confounders becomes exceptionally important, since randomization cannot do that work. Temporary crossover is possible, assessing a participant in BCI-on and BCI-off conditions. And decentralized assessment may matter more here than in most trials, because a communication device should be measured in somebody’s everyday environment rather than a curated lab, where context of use drives whether the thing actually works.

Speed and Accuracy Are Not Clinical Benefit

The second methodological problem is that the field’s technical literature has optimized for metrics that are not the endpoint a regulator needs.

Decoding speed and accuracy are necessary. They are not communication, and they are not clinical benefit. The challenge industry faces is translating a given speed and accuracy into a demonstrable benefit, which is a different claim requiring different evidence.

A device developer presenting later made the same point from the other side. Accuracy is necessary but it is not communication. Whether a device lets someone convey something to another person, in a way that person understands, depends on context. A fast exchange with friends, where much is carried by shared context, makes different demands than a consultation with a clinician where specific information has to get across intact. Her argument was that understanding those contexts is part of defining the endpoint, not a detail to sort out afterwards.

The timing question sits underneath all of it. In a disease where individuals progress at different rates, when you assess is itself a design decision with consequences.

The Indication Question the Field Has Not Settled

The most unresolved issue raised was not statistical in the narrow sense. It was what these devices are actually indicated for.

A BCI addresses a functional deficit. It does not address underlying pathology. On the speaker’s own framing it is not a treatment in the conventional sense at all. And the functional deficit it addresses, loss of motor control or loss of speech, is common across a long list of unrelated conditions.

That creates genuine interest in an etiology-agnostic indication, one defined by the deficit rather than the disease. The statistical considerations for that are, by his account, still being worked through by the field.

And it inverts the usual rare disease arithmetic in a way worth noticing. ALS is a rare disease. In BCI terms it is a relatively prevalent one. The conditions underneath it, uncommon ataxias among them, involve numbers so small that, as he put it, a simple subgroup analysis will not cover it.

What an Early Feasibility Study Is Actually For

Paradromics presented the design of Connect-One, its early feasibility study of the Connexus BCI, and the useful part was less the device than what the company says it is using the study to learn.

Connexus places a cortical module carrying more than 400 electrodes roughly 1.5 millimeters into the cortical surface, building on the Utah array used in academic settings for two decades. A lead tunnels to an implanted internal transceiver, which links optically to an external transceiver that powers the system and carries neural data out to downstream compute, where decoders turn intended speech into synthesized audio or on-screen text. The surgical sequence was deliberately kept familiar to neurosurgeons who place deep brain stimulation systems: craniotomy, tunneled lead, subcutaneous pocket. Array placement is targeted using resting-state and task-based fMRI to locate speech areas.

Connect-One is primarily a safety study. It also carries two FDA breakthrough device designations, one for decoding imagined language and speech, the other for interacting with digital devices.

But the company framed the study’s real work as endpoint development for the pivotal trial that follows, and that framing is the transferable lesson.

What the feasibility study is being used to establish
1
Whether a decoder can be built at all
A training loop of copy tasks with known target words, expanding vocabulary and phonemes, iterating toward free text, with the data labeled because the intended utterance is known.
2
Within-participant variability
Intracortical BCIs generate unusual volumes of data per participant. That depth lets the company characterize variability inside one person before trying to characterize it across a population.
3
How much repetition a decoder needs
And whether that differs by word or by phoneme.
4
The right assessment timing
Tested in feasibility rather than assumed at pivotal.
5
Whether an instrument needs modifying
Feasibility is the stage to take an existing questionnaire and adapt it, then stand behind the modified endpoint later.
6
What counts as a responder
Rolling technical measures and patient-reported outcomes up into a participant-level definition of success.

Her three closing points were to use feasibility to test endpoints rather than only safety, to exploit the within-participant depth the technology uniquely provides, and at pivotal to pre-specify, understand what replication you are capable of, and be honest about the uncertainty.

Thirty Patients, and Everything After

On sample size, the device session was blunt. A pivotal study in this space may be small, perhaps around 30 participants, particularly where several early feasibility studies have already generated evidence.

A study that size will surface high-frequency safety events, and with long enough follow-up it will show the trajectory and stability of the device across the population. It will not surface rare events or rare device deficiencies. That is not a flaw to be designed away at thirty patients; it is arithmetic.

Which pushes the burden into the post-market world, and the speaker was explicit that it should: follow participants not for twelve months or two years but for as long as possible, and use registries to keep track of people once a study closes. For a permanently implanted device in a very small population, the evidence generation does not end at approval. It barely starts there.

The workshop produced no recommendations and was not a federal advisory committee. For sponsors in neurotechnology, its value is a public airing of the questions FDA reviewers are weighing before any of it reaches guidance.

References

  1. FDA, RISE Workshop 4: RISE to the Challenge: Statistical Considerations for Rare Disease Clinical Investigations, September 29, 2026. Device session: statistical considerations in clinical investigation of BCI devices for ALS.
  2. Connect-One: Early Feasibility Study of the Connexus Brain-Computer Interface, ClinicalTrials.gov, NCT07357428.
  3. Paradromics and the University of Michigan complete the first Connexus BCI implantation in the Connect-One study.
  4. Paradromics reports real-time speech with the Connexus BCI, September 2026.
  5. RISE Workshop #4 pre-read (PDF), Duke-Margolis and the FDA Rare Disease Innovation Hub.
  6. Federal Register notice establishing docket FDA-2026-N-10165, September 15, 2026.
  7. Full workshop recording (video). Quotations are taken from the recording and reported by role rather than by name.
  8. Part one: what the agency said about randomization and control arms.
Website | + posts

Moe Alsumidaie, MBA, MSF, is founder and Chief Editor of Vanguard Publications, which publishes Clinical Trial Vanguard, Pharma Vanguard and BullScope, and Head of Research at CliniBiz. He has two decades in clinical trial operations and data science, with earlier roles at Genentech, Abbott Vascular and Stanford University Medical Center, and is a guest lecturer in clinical trial sciences at Rutgers University.