Adaptive deep brain stimulation
Deep brain stimulation is almost thirty years old. What is new is not the electrode: it is that the system adjusts its own output according to what it detects, in real time. Evidence that this works better does not yet exist.
A fine electrode, implanted with millimetre precision in a deep brain nucleus, connected to a generator that delivers continuous electrical pulses. That is how deep brain stimulation has worked since it received its first regulatory approval in 1997, for tremor1. Today it has established indications in Parkinson's disease and refractory epilepsy, and two more — dystonia and obsessive-compulsive disorder — authorised through a special pathway that requires demonstrating probable benefit rather than proven efficacy1. That regulatory nuance is almost never reported, and yet it changes entirely how the word "approved" should be read.
The novelty of recent years fits in one adjective: adaptive. The same electrode that stimulates can listen. It records local electrical activity in the nucleus where it sits and, when that signal crosses a threshold, the system raises or lowers stimulation intensity on its own. Instead of a fixed dose programmed in the clinic, a dose that follows the patient's state hour by hour.
In February 2025 the US agency approved the first such system for Parkinson's disease2. It is worth reading the document rather than the headline: what was approved is an optional programming feature added to an already authorised device, with no change of indication. The signal it uses lies in the alpha-beta band, and the loop's response time is configurable between a fraction of a second and half an hour3. In other words: depending on how it is programmed, this can resemble a reflex or a trend adjustment.
What the trial showed, and what it did not
The study behind that approval set itself a modest goal and met it: that adaptive stimulation should be no worse than well-programmed conventional stimulation. It achieved that, with good tolerability, no serious device-related adverse events and somewhat lower energy consumption4. What it did not set out to demonstrate — and therefore did not demonstrate — is that it is better. The agency itself recorded that the definition of the success criterion was revised after the fact, which precludes confirmatory statistical inference3.
The rest of the field's randomised evidence is at a scale that surprises anyone arriving from the headlines: the only blinded trial comparing adaptive against optimised conventional stimulation has four patients5. It also used a different biomarker from the one the approved system employs. A 2026 study addressing what classical stimulation never resolved well — gait disturbance — decodes ongoing locomotion to adapt parameters to context; it too is a feasibility study, again with four participants, and its own authors present it as a blueprint for the next generation of therapies, not as a therapy6.
The honest conversation with the patient
There is a figure no discussion of closed-loop stimulation should obscure, and it is the most useful of all in the consulting room. Five-year follow-up of the largest controlled trial of subthalamic stimulation shows that motor improvement, which at one year is around half the baseline score, falls to somewhat over a third by the fifth year; activities of daily living erode further still7. What does hold over time is control of dyskinesia and reduction of medication.
None of this makes stimulation a poor indication: it remains one of the most transformative treatments in functional neurosurgery, and in selected patients with early motor complications it clearly improves quality of life8. But it defines the promise precisely. The device treats symptoms; it does not halt the disease. The window of greatest benefit is the first year. And there is a price: in that same trial, serious adverse events were more frequent in the operated group.
What happens in the operating room
Two convictions deeply held in our specialty do not survive scrutiny of the pooled data.
The first is that the patient must be awake. A systematic review of thirty-one studies and more than two thousand five hundred patients finds no difference in motor outcome between awake surgery and image-guided surgery under general anaesthesia; nor according to the use of microelectrode recording9. The choice depends on patient preference, the team's experience and available resources — which, in a country where those resources are unevenly distributed, is not a minor detail.
The second is that we are more precise than we are. The real mean error between planned and achieved target, across more than six thousand trajectories, is close to two millimetres10. Robotic assistance reduces it independently, by somewhat under a millimetre. And intraoperative electrophysiological recording, contrary to what is usually assumed, was associated with a larger error, not a smaller one.
What to take away
That a brain implant should respond to what it detects is not in itself a 2025 novelty: a neurostimulator that detects epileptic activity and delivers stimulation in response has been approved since 201311. What is genuinely new is applying that principle to Parkinson's disease and modulating continuously rather than firing on an event.
The direction is right, and it is probably the definitive direction: devices that not only act on the brain but listen to it. But between "it is reasonable to think this will be better" and "it is demonstrated that this is better" there is a distance that has not been closed — and saying so is not scepticism. It is the only way that, when the evidence arrives, it can be believed.