Neuromodulation for Drug-Resistant Epilepsy

Evidence status. VNS, anterior-nucleus DBS, and RNS have different FDA-labeled populations. Their evidence supports seizure reduction rather than guaranteed seizure freedom; resective or ablative candidacy must be assessed first at a comprehensive epilepsy center. All three U.S. labels use the pre-2017 term partial onset; read it as focal onset throughout.

1.Drug Resistance and the Surgical Gateway

Drug-resistant epilepsy has a formal definition: failure of adequate trials of two tolerated, appropriately chosen and used antiseizure medication schedules, as monotherapy or in combination, to achieve sustained seizure freedom. Reaching that threshold should trigger referral to a comprehensive epilepsy center while medication optimization continues, because the probability that a third or fourth drug produces freedom is low, and the years lost to repeated trials are years of seizure-related risk, including SUDEP.

At the center, the presurgical evaluation asks whether there is a single seizure focus that can be removed or ablated. Laser interstitial thermal therapy is an option for selected mesial temporal cases. Selection balances seizure-freedom probability, anatomy, language and memory risk, invasiveness, and patient preference; it is not universally preferred over open resection and does not eliminate cognitive risk. When the answer is yes, resection or ablation is usually the most effective option and neuromodulation is not the first move. Neuromodulation becomes the answer when that evaluation returns one of a few recurring patterns:

• The seizures cannot be localized to a single focus, or arise from multiple independent foci.
• The focus is bilateral, classically bilateral independent mesial temporal onset.
• The onset zone overlaps eloquent cortex (motor, language, primary sensory) that cannot be resected without unacceptable deficit.
• Resection has failed, or the patient is not a candidate for or declines open resection.

2.Vagus Nerve Stimulation

Vagus nerve stimulation (VNS) is the oldest of the three and the only one with no intracranial component. A helical electrode is placed around the left cervical vagus nerve and connected to a pulse generator in the chest, delivering scheduled, intermittent stimulation; modern generators can add cardiac-triggered stimulation when a programmed heart-rate change is detected. This is a physiologic trigger, not a specific seizure detector, because not every seizure produces tachycardia and not every tachycardia is ictal. Patients are also given a magnet they can swipe over the generator to deliver an extra train during an aura or a seizure. The responsive idea in VNS started there, long before cardiac-based detection. Because nothing enters the cranium, VNS carries the lowest procedural risk profile of the three and is broadly used. In the United States, the epilepsy label is adjunctive therapy for refractory partial-onset seizures in patients age 4 years and older; generalized-epilepsy use is evidence-supported but off-label.

VNS is palliative and its effect is cumulative. Roughly half of patients reach a 50% or greater reduction in seizure frequency with continued use over time. That is the figure most clinicians quote, though published long-term series span a considerably wider range on either side of it. Complete seizure freedom is uncommon. Its characteristic side effects come from vagal stimulation itself and occur during the on-cycle: voice alteration or hoarseness, cough, throat sensation, paresthesias, and shortness of breath, which are generally tolerable and often diminish over time. Separately, and easy to miss on a page of stimulation side effects: VNS can cause or worsen sleep apnea, obstructive more often than central. Ask about it before implantation and again after titration. For the patient who is not a candidate for a focus-directed device, or who wants to avoid intracranial surgery altogether, VNS is the natural choice.

3.Deep Brain Stimulation of the Anterior Nucleus

DBS of the anterior nucleus of the thalamus (ANT) treats epilepsy as a network disease. The anterior nucleus is a node in the limbic circuit of Papez, and bilateral leads placed there are thought to interrupt the propagation of seizures rather than to act on a single cortical focus. Stimulation is open-loop, delivered on a fixed cycle, and the device does not detect seizures.

The pivotal SANTE trial established the approach: in the last month of the three-month blinded phase, bilateral ANT stimulation produced a 40.4% unadjusted median reduction in seizure frequency against 14.5% in controls, and (the defining feature of thalamic DBS) the benefit grew substantially with time, reaching roughly 75% median reduction at seven years of open-label follow-up with responder rates that continued to climb. Two caveats belong with that curve. The first is that SANTE is no longer the only randomized evidence: the FRANCE trial (Chabardes, Epilepsia 2026) randomized 61 patients to ANT-DBS or best medical treatment and did not demonstrate superiority, with a numerically greater median reduction in severe seizure frequency at 12 months (44% versus 6%; p = 0.09). This was an open-label trial in patients who had previously failed VNS; the between-group primary comparison, rather than within-group improvement, determines the superiority conclusion. The second is that the long-term SANTE figures come from an open-label extension, a point taken up below. The U.S. label covers patients age 18 or older with refractory partial-onset seizures, with or without secondary generalization, after failure of at least three antiseizure medications. Safety and effectiveness were demonstrated in patients averaging at least six seizures per month over the preceding three months, with no seizure-free interval exceeding 30 days; less frequent seizures were not evaluated. Clinically, ANT-DBS is often considered for focal network epilepsy not reducible to one or two resectable or directly stimulable foci; temporal and frontal onsets were prominent in the evidence base. Reported adverse effects include self-reported memory difficulty and depressive symptoms, each reported by a minority of participants during the blinded phase and by a larger fraction cumulatively across years of follow-up. Name both explicitly in counseling, because the lead sits inside a memory circuit. The honest qualifier is that formal neuropsychological testing in the SANTE cohort did not show group-level cognitive decline, so these are subjective complaints in a minority rather than a demonstrated group deficit. The usual implantation risks of intracranial leads apply on top.

4.Responsive Neurostimulation

Responsive neurostimulation (RNS) is the closed-loop device. A cranially implanted generator is connected to one or two leads (depth, strip, or both) placed directly at the seizure focus or foci. The system continuously monitors local electrocorticography and delivers stimulation when clinician-programmed detection features are met. Detections are not seizures, and the gap is much wider than that phrase suggests: most patients receive hundreds to thousands of brief stimulations a day, vastly more than any plausible seizure count, so the overwhelming majority of stimulation is delivered in the interictal state. This observation is consistent with network modulation contributing to benefit, but stimulation counts alone do not establish the mechanism. It is a focus-targeted, responsive therapy rather than a perfect seizure-prediction system.

The U.S. label covers adults with partial-onset seizures localized to no more than two foci, refractory to at least two antiseizure medications, with frequent disabling seizures. The supporting population averaged at least three disabling seizures monthly over the preceding three months, with no month containing fewer than two; patients with less frequent seizures were not evaluated. Within that population, RNS is especially useful for one or two well-localized foci that cannot be resected. The two situations that recur most often are bilateral independent mesial temporal lobe epilepsy and a focus sitting in eloquent cortex. As with the other two devices, the benefit is progressive: the pivotal trial showed a 37.9% reduction on stimulation against 17.3% on sham during the blinded phase, and nine-year prospective follow-up demonstrated a 75% median reduction with a 73% responder rate and roughly a third of patients achieving very high (90% or greater) seizure reduction. Seizure-free stretches do occur, but they are usually intervals rather than a settled state. A distinctive secondary benefit is the chronic ambulatory electrocorticography the device records, which gives the clinician a long-term window into interictal burden, lateralization, and multiday and circadian rhythms. Read it correctly: detection counts depend on programmed features and can include interictal activity or artifact. Long episodes may be a useful electrographic-seizure proxy only after review of stored ECoGs validates that relationship for the patient and settings; and the published outcome figures still rest on patient seizure diaries rather than on device counts. In practice this recording redirects management, most strikingly when a patient implanted for presumed bilateral mesial temporal epilepsy proves strongly lateralized over months of recording and becomes a resection candidate after all. Principal risks are those of implanted intracranial hardware: hemorrhage and device-site infection.

5.A Fourth Target, Off-Label

Three devices carry an epilepsy label; a fourth target is used off-label often enough that a trainee should be able to name it. Bilateral stimulation of the centromedian nucleus of the thalamus is considered at experienced centers for Lennox-Gastaut syndrome and symptomatic generalized epilepsy: precisely the population that fits none of the three labeled devices well, and the population a strict reading of the decision path would send to VNS by default. The evidence is largely case series plus one small randomized trial. ESTEL, an Australian trial of centromedian DBS in 19 patients with a Lennox-Gastaut phenotype, missed its blinded primary endpoint: 50% of stimulated participants reached a 50% or greater reduction in diary-recorded seizures against 22% of controls, a difference that was not statistically significant, though the electrographic seizure outcome did favor stimulation. Offer centromedian DBS as a reasonable off-label option at a center that does it, not as a proven one.

Responsive stimulation delivered through thalamic leads is likewise off-label. The randomized NAUTILUS trial in idiopathic generalized epilepsy did not meet its prespecified primary effectiveness endpoint. Its 18-month open-label follow-up reported a 76.8% median reduction in generalized tonic-clonic seizures from baseline; that within-patient comparison does not establish blinded superiority. NeuroPace reported on July 28, 2026 that FDA considered the indication-expansion supplement not approvable in its current form and requested additional clinical evidence. This indication remains investigational in the United States as of September 2026.

6.Matching Device to Patient

Localization provides the organizing framework, but devices do not sort perfectly. Network topology, age and label, prior resection, comorbid mood or memory disease, MRI needs, battery and follow-up burden, patient preference, and center expertise can outweigh a simple focal-versus-diffuse rule. Two practical points follow. The devices are not mutually exclusive, and selected centers use combined VNS and intracranial stimulation after individualized assessment; evidence for combinations is mainly observational, and compatibility, interactions, and follow-up burden require review. And because the mechanisms differ, failure of one modality does not preclude benefit from another. A patient who did poorly on VNS is not thereby a poor candidate for RNS or thalamic DBS.

Read the table as a decision path rather than a feature list. Start by confirming that resection is genuinely off the table. Then: if there are one or two discrete foci you simply cannot remove (bilateral hippocampi, or a focus in language or motor cortex), RNS lets you put therapy exactly where the seizures start and supplies chronic recordings that can refine the localization hypothesis. If the epilepsy is focal but will not reduce to one or two stimulable points, ANT-DBS treats the network node through which those seizures propagate. And if localization fails entirely, the epilepsy is generalized, or the patient wants to avoid intracranial hardware (including many children), VNS is the least invasive effective option. And for generalized epilepsy with drop attacks, do not forget corpus callosotomy: more invasive than VNS, but it acts sooner, and it is not mutually exclusive with a device.

7.What to Tell the Patient

Counseling for any of the three rests on the same three points. First, the goal is fewer and less severe seizures, not a cure; seizure freedom is possible but uncommon. Second, the benefit builds over time, so the first months are not the verdict; patience is part of the therapy. Third, each device carries its own risk signature, from the on-cycle vocal effects of VNS to the intracranial implantation risks of the two brain devices, and these should be named specifically. Set against the alternative (continued uncontrolled seizures with their attendant injury, cognitive, psychosocial, and mortality risks), meaningful palliation is a substantial gain, and framing it honestly is what makes it durable.