Deep brain stimulation occupies an unusual position in medicine: it is an established therapy whose mechanism remains contested. Electrodes implanted in deep grey matter deliver continuous or patterned current, and patients with Parkinson's disease, dystonia, essential tremor or Huntington's disease can gain years of better motor control, fewer fluctuations and improved quality of life. The physiological account of that benefit has shifted repeatedly. Early reasoning held that high-frequency stimulation silences overactive nuclei by mimicking a lesion; later work showed that activity continues during stimulation and that the dominant effect is disruption of pathological oscillatory synchrony; more recent formulations describe network-level changes in which stimulation alters information flow across cortico-basal ganglia-thalamo-cortical loops. Each account explains part of the clinical picture, none explains all of it, and the disagreement is not academic, since mechanism determines which target is chosen and how current is programmed.
Indications have expanded faster than the mechanistic understanding. Approved uses cover Parkinson's disease, dystonia, essential tremor, obsessive-compulsive disorder and epilepsy, while investigation has moved into depression, Alzheimer's disease, Tourette syndrome and other conditions. Expansion brings a recurring difficulty, because the circuits that generate symptoms are distributed and patient-specific. A target that works in one phenotype may fail in another carrying the same diagnostic label, which is why response rates in psychiatric indications have been less consistent than in movement disorders, and why trials in those conditions need careful phenotypic selection, sham control and extended follow-up. Huntington's disease illustrates the same tension: the disorder is progressive and degenerative, so stimulation can relieve specific symptoms without altering the course of the disease. Large cohorts have made response reasonably predictable in movement disorders and much less so in psychiatry, and that asymmetry is a mechanism question rather than an accident of trial design.
Safety and tolerability define the practical ceiling. Surgical risks include infection, intracranial haemorrhage and electrode displacement, and reported series document both infections and haemorrhage in patients undergoing implantation, with rates varying by technique and by patient factors. Stimulation produces its own adverse effects, and these are target-dependent. Stimulation-induced dyskinesia in Parkinson's disease can reflect the interaction of current with levodopa, so programming and medication must be adjusted together. Speech disturbance tends to appear at high amplitudes, wide pulse widths or high frequencies, particularly when current spreads to the internal capsule, and subthalamic stimulation has been associated with reduced vocal volume and dysphonia more often than pallidal stimulation. Cognitive change follows a similar pattern, with greater declines in executive function and verbal fluency reported for subthalamic than for pallidal targets, while post-stimulation dementia is uncommon and usually attributed to disease progression. Most of these effects are adjustable rather than permanent, which is the advantage of a reversible therapy and also the reason programming demands expertise and time.
Forward progress is technical. Electrodes with directional leads allow current to be shaped rather than simply delivered, and adaptive stimulation, in which parameters respond to sensed biomarkers such as oscillatory activity, promises to replace fixed settings with closed-loop control. Both refinements address the same weakness: therapy is currently tuned by clinical observation at intervals, whereas the underlying pathology fluctuates from hour to hour and from state to state, so patients spend much of the day outside the range their settings were chosen for. Deeper mechanistic work is expected to accompany these devices, since recording from implanted electrodes during everyday behaviour offers a window on human circuit dynamics that no animal model provides. What the field still lacks is a principled way to decide, before surgery, which patient will respond to which target, and that gap separates a powerful technique from a predictable treatment. Closing it will require the same kind of stratified evidence that drug development demands, rather than the accumulated experience of specialist centres.
DOI:10.1007/s11684-026-1231-1