A drug trial can give some patients a sugar pill without telling them. A brain implant that lets a paralysed person with amyotrophic lateral sclerosis (ALS) click a mouse or speak through a computer offers no such cover: the clicks either happen or they do not. “There’s no placebo-induced telepathy,” Adam Fry, clinical science manager at Synchron, told an FDA workshop on 29 September. “A computer won’t click by itself through placebo.”
That problem, and what it means for the trials that could bring speech and computer-control implants to market, filled a session of RISE to the Challenge, a workshop on the statistics of rare disease trials. The FDA’s Rare Disease Innovation Hub and the Duke-Margolis Institute for Health Policy co-convened it at the National Press Club in Washington. The background paper from the FDA’s device centre uses brain-computer interfaces for ALS as its case example, and in the session an FDA official, an independent statistician and two implant makers worked through how a pivotal trial, the larger study that supports an approval decision, can be designed when few patients are available. The FDA says these workshops deal with issues common to many rare diseases and do not cover individual products under review. The recording was posted in October, and the comment period on the FDA’s questions closed on 13 November.
The questions the FDA asked
On 15 September the FDA published its questions in the Federal Register and opened a public docket. Three of its four questions on medical devices note that the device paper uses brain-computer interfaces for ALS as its example. They ask:
- for new Class III devices, the highest-risk category, which needs premarket approval, what matters when evidence collection for a small population shifts from before approval to after it, and which tools, such as registries, electronic health records or post-approval studies, would best support timely data collection after approval;
- when a single-arm study, in which every participant gets the device, measured against a performance goal or an external control can provide reasonable assurance of safety and effectiveness, and what has to be specified in advance;
- what would make an outside data source suitable for future pivotal studies in small populations, and who should build that infrastructure.
The fourth asks which prior information, such as feasibility studies, natural history data, international experience and earlier device generations, is credible enough to build on, and whether Bayesian methods, which formally combine earlier evidence with new trial data, can speed up reliable evidence.
What the FDA official said
Daniel Caños directs the Office of Clinical Evidence and Analysis in the FDA’s Center for Devices and Radiological Health, which houses its device biostatisticians. From 2016 to 2019 he led the evidence development division in Medicare’s coverage group, which handled national coverage decisions that tie payment to further evidence collection.
When randomisation is not feasible, Caños said, a single-arm study measured against a performance goal, a success rate set in advance that the device must beat, is an approach FDA guidance recognises, and the agency recognises that device trials can face challenges with blinding and equipoise, the genuine uncertainty about which arm is better that justifies randomising. A credible goal can come from natural history data or published results for the best available care, should reflect the patients the device is meant for, and needs a primary endpoint, the main outcome the trial is judged on, fixed in advance.
Outside data can be brought in through Bayesian statistics under a 2010 FDA guidance, he said, which recommends that prior assumptions be based on data, prespecified and agreed with the FDA before the study starts. “There’s no free lunch in statistics,” he said.
He then walked through an illustrative example from the FDA’s 2019 guidance on uncertainty in benefit-risk decisions. A device treats a disease with 10,000 new cases a year, too common for the FDA’s humanitarian exemption for very small populations, and must beat a 60 per cent success rate, with 66 per cent of patients observed to succeed. At the conventional one-sided significance level of 2.5 per cent, the accepted risk of wrongly concluding the device works, the sponsor would need 274 patients. At 10 per cent it would need 128, with a modest post-market study as a condition of approval. If the FDA accepted a 20 per cent level, 65 patients would be enough, provided the sponsor collected relatively robust data after approval, for example through a registry that tracks patients after treatment.
For devices meant for small populations, where serious unmet need, constrained enrolment and a lack of alternatives coincide, “greater pre-market uncertainty may be acceptable,” Caños said. In return the agency looks for post-market data that is actually collected on time, labelling and public summaries that describe it, and accountability. That includes advisory committee review and, where post-market data raise questions about whether the device still meets the standard, options up to withdrawal, sales restrictions or a narrower indication, the use a device is approved for.
The statistician’s caution
Janet Wittes, principal of Wittes LLC, ran the biostatistics research branch at the National Heart, Lung, and Blood Institute in the 1980s and later founded the consultancy Statistics Collaborative. Her starting point: “A trial should be a trial.” It should be rigorous, she said, and if a trial can be randomised it should be, though randomisation “is necessary but not sufficient.” She set out the legal line between the two kinds of product: drugs need substantial evidence that they work, which she likened to proof beyond reasonable doubt in a criminal trial, while Class III devices need reasonable assurance of safety and effectiveness, closer to the civil standard. A device whose benefit is plain, like a walking cane, may not need a control group, she said, and with devices “the benefit is often apparent, but we need to figure out how to evaluate safety.”
Synchron: trials that cannot be blinded
Fry presented online after a storm cancelled his flight. “I don’t think it is possible to blind participants to a BCI status, at least not in an effective or ethical way,” he said. The implant’s effect runs from the body to the outside world, so a sham device produces nothing. That leaves no equipoise between arms, and self-reported measures such as quality of life carry bias when people know whether their device works. He pointed to alternatives: careful control of confounding factors, temporary crossover comparing the implant switched on and off, and assessing participants in their everyday surroundings.
A pivotal study in these rare diseases might be small, perhaps around 30 people, he said, particularly if several early feasibility studies, the small first studies in people, have already produced evidence. On endpoints, “speed and accuracy does not equal clinical benefit,” he said. The industry’s task is to show what a given speed and accuracy mean for a patient, with the added question of when to measure in a disease where people decline at different rates.
The unsettled question he raised was the indication. The technology “addresses a functional deficit. It does not address the underlying pathology,” he said. Loss of movement or speech occurs across many diseases, so there is interest in an indication defined by the deficit, whatever its cause. ALS, rare in general, may count as a relatively common condition in the brain-computer interface field, he said, while rarer causes such as uncommon ataxias involve numbers too small for a simple subgroup analysis.
In November, Synchron said it had paused enrolment in trials of its updated Stentrode over a performance problem with the implant’s lead, and that no device-related safety issues have been found.
Paradromics: the feasibility study as a rehearsal
Marie Fahey, Paradromics’ director of clinical affairs, described Connect-One, the early feasibility study of the company’s Connexus implant. Its primary endpoint is safety; its secondary endpoints are speech decoding performance, clinical outcome assessments and patient-reported outcomes, and participants are supported across a six-year study. Connexus holds two Breakthrough Device designations, Fahey said, one for decoding imagined language and speech and one for controlling digital devices. In October Paradromics announced that the FDA had approved a protocol change letting people living anywhere in the US be considered, removing a requirement to live within a four-hour drive of one of the three sites, at UC Davis Health, Mass General Brigham and University of Michigan Health.
Fahey said the company is using the study to design the endpoints of its pivotal trial. Implanted devices generate unusually dense data within each participant. That data can show how much a measure varies between attempts and sessions, how many repetitions give a stable estimate, which assessment window is meaningful and reliable, and what level and consistency of performance counts as success for one person. She ranked the measures by how much they reveal about clinical benefit: decoder accuracy, then speed, reliability, whether the intended information reached another person, and finally whether the device meets real communication needs. Accuracy alone does not make communication, she said; a fast conversation with friends makes different demands from a visit to a doctor.
A small study can characterise frequent device and procedure events, communication performance, variability and durability over repeated follow-up, Fahey said. It cannot settle rare adverse event rates, uncommon device failures or very long-term durability, so she argued for following participants for as long as possible, through registries where needed. Her conclusion: “Small participant numbers do not have to mean weak clinical evidence,” provided the endpoint is clinically interpretable, the analysis is prespecified, results replicate across participants and the uncertainty is stated openly.
Signal: Why it matters
The pivotal trial sets the cost, timing and labelled use of the first FDA-approved implants. In the FDA’s illustrative example, which comes from nonbinding 2019 guidance and does not involve a brain implant, the same device could be judged on 274 patients or on 65 if the FDA decides more uncertainty before approval is acceptable and the sponsor commits to registry follow-up after it. For companies recruiting from small groups of severely paralysed patients, how the FDA applies that principle could shape how long it takes to reach the market and how much evidence spending continues after launch.
The indication question goes to market size. A label defined by lost speech or movement, whatever the cause, would cover more than ALS. Connect-One’s stated intended use is already written around the deficit, for people with impaired communication from severe loss of voluntary motor control, and Paradromics’ study page names ALS and brainstem stroke as examples. Endpoints also matter for coverage: Fry’s work at Synchron focuses on designing trials and outcome measures that support payer approval as well as regulatory approval.
Not disclosed
What performance goal, sample size or significance level the FDA would accept for a speech or computer-control implant, and whether it would accept an indication defined by the deficit.
Forward Radar: What to watch
What the FDA does with the comments on docket FDA-2026-N-10165, and whether it updates its 2021 guidance on implanted brain-computer interfaces.
Which company sets out the first pivotal trial design for an implant, and what indication it seeks.
When Synchron restarts enrolment, and what Connect-One reports next.
Sources
Primary:
- Duke-Margolis Institute for Health Policy: RISE to the Challenge, 29 September 2026, with agenda, speaker biographies, slides and recording
- Duke-Margolis: workshop agenda (PDF)
- Duke-Margolis: speaker biographies (PDF)
- Duke-Margolis: workshop slide deck, including Paradromics’ case study (PDF)
- Duke-Margolis: workshop recording, YouTube
- FDA: RISE Workshop 4, RISE to the Challenge
- Federal Register: Statistical considerations for the design of rare disease clinical investigations, docket FDA-2026-N-10165, 15 September 2026
- FDA: Consideration of uncertainty in making benefit-risk determinations in medical device premarket approvals, De Novo classifications and humanitarian device exemptions, August 2019 (PDF)
- FDA: Guidance for the use of Bayesian statistics in medical device clinical trials, 2010
- FDA: Implanted brain-computer interface devices for patients with paralysis or amputation, guidance, May 2021
- Paradromics: Connect-One clinical study page
- Paradromics via Business Wire: FDA approval to expand Connect-One recruitment nationwide, 26 October 2026
Coverage:
Cross-reference to prior InsideBCI coverage:
- 13 November 2026: Synchron pauses enrolment in its Stentrode trials over a problem with the implant’s lead
- 16 September 2026: Paradromics’s first Connexus participant used a brain implant to speak her own words
- 26 August 2026: Paradromics wins FDA approval to connect its Connexus BCI to personal laptops, tablets and phones inside Connect-One