← Back to Blog

The Clinical Trial Model Wasn't Built for a Therapy That Is Different in Every Patient

6 min read

The randomized controlled trial is the closest thing medicine has to a universal proof standard. Give one group the treatment, give another group a placebo or the current standard of care, and measure the difference. It works because it assumes the intervention itself stays roughly fixed while the population varies.

That assumption quietly breaks down for BCI, and almost nobody discusses it outside of the regulatory teams who lose sleep over it.

A BCI decoder is not really one fixed intervention. It's retrained, recalibrated, and adapted for each individual patient, often continuously, because the whole premise of a working system is that it learns the specific relationship between one person's neural signals and their intent. Two patients in the same trial arm, implanted with identical hardware, can end up running meaningfully different models within a few weeks. That's not a flaw in the trial. It's the product working as designed. But it means the classic assumption behind a randomized trial — that the treatment is constant and only the patients vary — doesn't really hold.

Identical implants feeding two patients whose decoder models diverge as each system personalizes
Identical hardware. Different models within weeks. That's the product working as designed.

There's a second problem sitting on top of that one. Blinding is central to a well-run trial, and a sham arm is how you rule out placebo effects. Sham surgery — opening a patient's skull and doing nothing — is an ethically fraught proposition even in conditions where it has occasionally been used, and it becomes even harder to justify for a population that is often severely disabled and has real, scarce windows in which surgery is medically feasible at all. So most BCI trials today end up small, open-label, and single-arm almost out of necessity rather than choice, which is a weaker evidentiary standard than what regulators typically expect from a novel implantable therapy.

This isn't a new problem in medicine, even if it's a new one for this category. Gene therapies and certain cell therapies ran into a similar wall years ago, because a treatment manufactured individually for each patient doesn't fit neatly into a framework built around a single, fixed drug tested across a large population. Regulators responded, slowly and imperfectly, with frameworks built around single-patient evidence, adaptive trial designs, and structured collection of real-world outcomes over time rather than one pivotal study that has to carry the entire evidentiary weight on its own.

BCI is heading toward the same fork in the road, and the companies that get there first with a workable framework will likely move faster than the ones trying to force their technology into a trial design built for a different kind of medicine entirely.

Where this could be wrong

If decoding approaches converge toward more standardized, foundation-model-style systems that behave similarly across patients rather than being deeply personalized to each one, the core premise here weakens considerably, and classic trial designs may work perfectly well after all. Regulators have also already shown more flexibility than this argument implies, through breakthrough device programs and a growing openness to real-world evidence, so some of this mismatch may resolve through existing tools faster than a new framework would need to be built from scratch. It's possible the field simply grows into the regulatory apparatus that already exists rather than needing a new one.

What this means for founders

Worth building a clinical and regulatory strategy that assumes personalization is a feature to be proven, not a complication to be minimized. That means engaging regulators early on adaptive and single-patient trial frameworks rather than defaulting to a standard pivotal trial structure and hoping it fits, building infrastructure from day one for continuous, longitudinal outcome collection rather than treating data gathering as something that ends once the trial does, and defining success in terms that make sense for an individual patient's functional outcome, not only in terms of statistical significance pooled across a group that was never really receiving the same treatment to begin with. The companies that treat this as a design problem early will likely spend less time stuck arguing with regulators later.

WorldChangers is a BCI startup accelerator — at most 10 early-stage companies per batch. If you're building at the brain-machine boundary, applications for Batch 01 are open.

Building at the brain–computer interface?

Apply to Batch 01