Responsive stimulation of the thalamus for idiopathic generalized epilepsy: Results of the randomized controlled NAUTILUS trial through 18 months

Abstract

Objective

This study was undertaken to evaluate the safety and effectiveness of responsive thalamic stimulation as adjunctive therapy for drug‐resistant idiopathic generalized epilepsy (IGE) with generalized tonic–clonic seizures (GTCSs).

Methods

NAUTILUS is a prospective, multicenter, single‐blind, randomized sham‐controlled pivotal trial. Patients were ≥12 years of age with drug‐resistant IGE and ≥2 GTCSs over a 3‐month baseline. Bilateral depth leads were targeted to the centromedian thalamus. One month later, patients were randomized to Active (responsive stimulation, n = 44) or Sham (no stimulation, n = 43). The effectiveness evaluation period (EEP) began 3 months postimplant through 1 year. After a second GTCS in the EEP, patients transitioned to open‐label active stimulation. The primary safety endpoint was the serious adverse device‐related event (SADE) rate at 84 days postimplant. The primary effectiveness endpoint was time‐to‐second‐GTCS during the EEP. Additional endpoints included median percent change in days with any generalized seizure, GTCS frequency, and responder rate (RR).

Results

Eighty‐seven patients were implanted across 23 US centers. The SADE rate was significantly below the performance goal (6.9%, p < .0001), with no adverse effects on cognition, mood, or sleep. The prespecified primary effectiveness endpoint was not significant. However, a post hoc mixed‐effects model considering all EEP days demonstrated greater GTCS reduction in the originally randomized Active group (61%) compared to patients originally randomized to Sham (49%, p = .030). Eighteen‐month outcomes included 76.8% median GTCS reduction, 62.5% RR, 40% GTCS‐free at that timepoint, and 77.8% median reduction in days with any generalized seizure. More than 90% of patients and 86% of physicians reported improvement on Global Impression of Change scales.

Significance

NAUTILUS is the first randomized controlled neuromodulation trial in IGE. Responsive thalamic stimulation provided a clinically meaningful and durable reduction in seizures with an acceptable safety profile, offering a much‐needed option for drug‐resistant IGE.

Key points

• NAUTILUS is the first randomized controlled trial of neuromodulation therapy for drug‐resistant idiopathic generalized epilepsy, assessing safety and effectiveness of responsive stimulation of the thalamus.

• The primary safety endpoint was met, with an SADE rate of 6.9%, well below the 25% performance goal.

• The primary effectiveness endpoint, time to second GTCS, was not significant; a post hoc analysis considering data through 12 months showed a 23% greater reduction in GTCSs in patients originally randomized to Active as compared to Sham stimulation (p = .030), in whom there was a delayed start to stimulation.

• At 18 months, the median reduction in GTCSs was 76.8%, with a 62.5% responder rate; 40% of patients had no GTCS at that timepoint.

• There were no adverse effects on cognition, mood, or sleep observed through 18 months of follow‐up.

1. INTRODUCTION

Idiopathic generalized epilepsy (IGE) accounts for approximately 15%–20% of all epilepsies. ¹ IGE presents with absence, myoclonic, and generalized tonic–clonic seizures (GTCSs), alone or in combination; GTCSs confer the greatest risk of injury and sudden unexpected death in epilepsy (SUDEP). ² Approximately 20%–30% of patients with IGE are drug‐resistant. ³ Unlike patients with drug‐resistant focal epilepsy, those with drug‐resistant IGE have limited treatment options; resective/disconnective surgery and ablation are not options, because seizures arise from generalized networks, and other neuromodulation therapies have not been specifically evaluated or US Food and Drug Administration (FDA) approved.

The centromedian (CM) nucleus of the thalamus is a target for neuromodulation in generalized epilepsies, as it is a critical relay in thalamocortical circuits that underlie the generation and propagation of the spike–wave discharges characteristic of IGE. Early open‐label studies of CM nucleus stimulation demonstrated reductions in generalized seizures. ⁴ Small case series of responsive stimulation in the CM nucleus described a 75%–99% seizure reduction over 2 years of follow‐up, with decreased seizure duration, severity, or both, as well as quality‐of‐life improvements. ⁵ , ⁶ , ⁷ Responsive stimulation offers theoretical advantages over continuous or duty cycle stimulation paradigms, including patient‐specific detection algorithms, real‐time intervention, reduced total stimulation exposure, and the ability to review numbers and locations of detection and samples of intracranial electroencephalography (iEEG).

The RNS System (NeuroPace) is a responsive neurostimulator ⁸ that is FDA approved for focal epilepsy. It continuously monitors iEEG activity, detects abnormal patterns, and delivers targeted electrical stimulation in response (closed loop). The RNS System has demonstrated a safe and effective treatment profile in drug‐resistant focal epilepsy, ⁹ , ¹⁰ with median seizure reductions approaching 75% at 9 years of follow‐up in controlled clinical trials ¹¹ and 82% at 3 years of follow‐up in an open‐label FDA‐required postapproval study. ¹²

NAUTILUS is a randomized sham‐controlled trial designed to evaluate the safety and effectiveness of responsive stimulation targeting the CM nuclei in adults and adolescents with drug‐resistant IGE. This report provides safety and effectiveness results for the first 18 months of this pivotal trial.

2. MATERIALS AND METHODS

A prospective, multicenter, single‐blind, randomized, sham‐controlled pivotal trial was conducted at 23 epilepsy centers across the United States (NAUTILUS trial, ClinicalTrials.gov: NCT05147571). The study was approved by institutional review boards at all sites, and patients provided written informed consent. The trial was conducted in accordance with the Declaration of Helsinki and Good Clinical Practice guidelines.

2.1. Participants

Key inclusion criteria were as follows: age ≥ 12 years, confirmed IGE diagnosis by history and EEG, ¹³ drug‐resistant epilepsy defined as failure of ≥2 appropriate antiseizure medications (ASMs), GTCSs with or without absence or myoclonic seizures, and ≥2 GTCSs during the 3‐month baseline period. Key exclusion criteria included the following: focal epilepsy or EEG findings suggestive of focal onset; nonepileptic seizures that could not be distinguished from epileptic seizures, active psychosis, major depression, or suicidal ideation; and concurrent therapy with another active stimulation device for epilepsy.

2.2. Device placement and programming

Bilateral depth leads targeted the CM nucleus. The neurostimulator continuously monitored iEEG activity, detected abnormal electrographic patterns using patient‐specific algorithms, and delivered brief pulses of electrical stimulation in response to detected events.

2.3. Seizure diaries

Participants used an electronic seizure diary at enrollment, recording the type and number of seizures (GTCS, absence, or myoclonic) each day, seizure duration, postictal recovery time compared to baseline (shorter, usual, longer), and whether seizure rescue medications were taken.

2.4. Clinical trial design

The trial design is presented in Figure 1A. A time‐to‐event design was selected to minimize the time before patients randomized to a no‐treatment arm could access a potentially effective therapy. To meet implant eligibility, patients were required to have ≥2 GTCSs over a 2‐month retrospective and 1‐month prospective baseline. An RNS neurostimulator and one depth lead targeted to the left and right CM nuclei were implanted. Following a 1‐month postimplant stabilization period, patients were randomized to Active (stimulation enabled) or Sham (stimulation disabled; 1 month = 28 days). Randomization was balanced for study site, baseline seizure frequency, and prior vagus nerve stimulation (VNS) use.

FIGURE 1.

Clinical trial design and Consolidated Standards of Reporting Trials (CONSORT) participant flow diagram. (A) NAUTILUS study timeline showing baseline period, implant, randomization, effectiveness evaluation period (EEP), and open‐label follow‐up. Sham patients transitioned to Active stimulation after experiencing a second generalized tonic–clonic seizure (or other transition event) during the EEP while remaining blinded to original randomization assignment. (B) Patient disposition from enrollment through 18‐month follow‐up. ITT, intent‐to‐treat; LTFU, lost to follow‐up; Mo, months; Opt, Optimization; PP, per‐protocol.

Following randomization, the Active group underwent a 2‐month stimulation optimization period, during which detection and stimulation parameters were adjusted. Sham patients had detection parameters adjusted as needed and underwent simulated stimulation programming to maintain blinding. Patients entered the 9‐month effectiveness evaluation period (EEP) 3 months after implant.

The transition from blinded Active or Sham stimulation to open label active stimulation occurred when the patient's second GTCS occurred within the EEP, when the patient reached 1 year without a second GTCS, or when the 60th patient had a second GTCS, whichever occurred first (transition event). Patients were not informed of randomization assignment until their final study visit at 24 months (see Figure 1A).

2.5. Endpoints

The primary safety endpoint was the rate of serious adverse device‐related events (SADEs) at 84 days postimplant using a one‐sided exact binomial test. Secondary safety endpoints included annualized SADE rate, serious adverse events (SAEs) of particular relevance (implant site infection, intracranial hemorrhage, suicidality, cognitive deterioration), and SUDEP rate.

The primary effectiveness endpoint was time to second GTCS during the EEP. The secondary effectiveness endpoints included percent change in days with any type of generalized seizure (GTCS, absence, or myoclonic) during the EEP compared to baseline.

Additional prespecified outcomes during stimulation treatment included median percent change in GTCS, in days on which GTCSs, myoclonic seizures, or absence seizures occurred, and GTCS 50% and 100% responder rates.

Quality of life (Patient‐Weighted Quality of Life in Epilepsy Inventory‐31/Quality of Life in Epilepsy Inventory for Adolescents‐48), ¹⁴ neurocognitive function (NIH Toolbox Cognition Battery), ¹⁵ depression (Beck Depression Inventory‐II), sleep quality (Pittsburgh Sleep Quality Index), and patient and physician global impressions of change were assessed at regular intervals throughout the study.

2.6. Statistical analysis

The intent‐to‐treat (ITT) population included all implanted patients. The per‐protocol (PP) population excluded patients who did not meet eligibility or protocol requirements. The prespecified primary safety and effectiveness analyses used the ITT population; additional analyses were conducted in the PP population.

The prespecified primary effectiveness analysis used a Cox proportional hazards model stratified by baseline GTCS frequency (≤2 vs. >2 per month) and prior VNS, which considered one time‐to‐event measurement for each patient. A post hoc alternative primary effectiveness analysis used a mixed‐effects model with negative binomial variance structure. ¹⁶ This model used monthly seizure count data from randomization (month 1; Active group receives stimulation) through 12 months postimplant, with fixed effects for time and treatment group and a random effect for subject, employing a difference‐of‐differences approach. The post hoc analysis considered 1249 monthly observations across 82 patients, providing substantially greater power to detect treatment effects.

Additional effectiveness endpoints used Wilcoxon signed‐rank tests with Bonferroni correction for within‐group comparisons of change from baseline. Neuropsychological outcomes were analyzed using 90% reliable change index (RCI) for individual‐level change and paired group contrasts for group‐level analyses.

2.7. Exploratory analyses

2.7.1. ASM changes

ASM changes were identified from concomitant medication records for all PP patients. ASMs were to be held constant over the first year. Relevant ASM changes were addition or discontinuation of an ASM or a dose adjustment of >25%. At each cumulative timepoint (3–18 months), patients were classified as having no change, addition/increase only, discontinuation/decrease only, or both. Median percent change in GTCS from baseline was compared across categories of ASM changes using the Kruskal–Wallis test.

2.7.2. Rescue medication use

Intermittent use of benzodiazepines as rescue therapy was evaluated in PP patients. The proportion of days with GTCSs on which rescue benzodiazepines were used was modeled using binomial generalized estimating equations (GEEs) with an exchangeable correlation structure, logit link, and robust standard errors (n = 80). Rescue diary entries were classified using standardized mapping of free‐text diary entries, with ambiguous entries conservatively assumed to be benzodiazepines. Time was divided into 84‐day bins using the implant date for baseline bins and the stimulation date for poststimulation bins. Three models were fit to the data: pooled baseline versus poststimulation, a poststimulation linear trend, and bin‐by‐bin contrasts against pooled baseline with Benjamini–Hochberg false discovery rate (FDR) correction.

2.7.3. Seizure duration and postictal recovery

Patient‐reported ordinal outcomes (shorter, usual, or longer) were compared between preimplant baseline and poststimulation periods grouped into 3‐month (84 day) bins. A cumulative link mixed model was fitted with time bin as a fixed effect and a random subject intercept. Odds ratios (ORs) < 1 indicated a shift toward improvement. Bonferroni correction was applied across six poststimulation comparisons (p < .0083). The analysis included 1679 seizures from 80 subjects.

2.7.4. Injury event rates

Injury‐due‐to‐seizure event rates were compared pre‐ and posttreatment using a within‐subject pairwise design across all 87 ITT patients, with each patient serving as their own control. Two Poisson regression approaches were employed: a fixed‐effects model with sandwich standard errors and a GEE model with exchangeable correlation structure.

3. RESULTS

3.1. Patient disposition and baseline characteristics

Of 100 enrolled patients, 87 were implanted and randomized (44 Active, 43 Sham). Five patients were excluded from the PP population. Eighty‐four patients completed 12 months of follow‐up, and 80 completed 18 months (Figure 1B). As of 18 months of follow‐up, two deaths had occurred: one definite SUDEP event (stimulation enabled) and one from a plane crash.

Baseline demographic and clinical characteristics for the ITT population are summarized in Table 1. Mean age at enrollment was 30.6 years (SD = 11.3, range = 14–66), with mean epilepsy duration of 18.2 years (SD = 12.0). Patients were treated with a mean of 3.7 ASMs (SD = 1.3), and 33.3% had prior VNS therapy. Most patients had multiple seizure types; 42.5% had GTCSs with absence and myoclonic seizures, 27.6% had GTCSs and absence seizures, 17.2% had GTCSs and myoclonic seizures, and 12.6% had only GTCSs. There was a substantial range in the rate of baseline GTCSs, from .3 to as high as 24 GTCSs per month. The median and mean for baseline seizure rates were equivalent between Active and Sham randomization groups (Figure 2).

TABLE 1.

Baseline demographics and clinical characteristics.

Characteristic	ITT, N = 87	PP, N = 82
Age at enrollment, years, mean ± SD [range]	30.6 ± 11.3 [14–66]	30.5 ± 10.6 [15–60]
Age at epilepsy onset, years, mean ± SD	12.4 ± 6.1	12.4 ± 6.1
Duration of epilepsy, years, mean ± SD	18.2 ± 12.0	18.1 ± 11.6
Female sex, % (n)	54.0% (47)	54.9% (45)
Race: White, % (n)	80.5% (70)	79.3% (65)
Ethnicity: not Hispanic/Latino, % (n)	85.1% (74)	84.1% (69)
ASMs at enrollment, mean ± SD	3.7 ± 1.3	3.7 ± 1.3
Prior VNS therapy, % (n)	33.3% (29)	31.7% (26)
Monthly baseline rate of GTCSs, mean ± SD	2.7 ± 3.6	2.7 ± 3.7
Monthly baseline rate of any generalized seizure, mean ± SD	9.2 ± 8.1	9.0 ± 8.1
Seizure types at baseline, % (n)
GTCS only	12.6% (11)	—
GTCS + absence	27.6% (24)	—
GTCS + myoclonic	17.2% (15)	—
GTCS + absence + myoclonic	42.5% (37)	—

Abbreviations: ASM, antiseizure medication; GTCS, generalized tonic–clonic seizure; ITT, intent‐to‐treat; PP, per‐protocol; VNS, vagus nerve stimulation.

FIGURE 2.

Baseline generalized tonic–clonic seizure (GTCS) distribution. Number of GTCSs during baseline is provided in summary table with median, mean, first and third quartiles, and minimum and maximum. The maximum number of seizures was higher in the Sham group but, as shown in the figure, the distribution of baseline GTCSs was similar.

3.2. Safety

The primary safety endpoint was met; the SADE rate at 84 days postimplant was 6.9% (6/87 patients, 95% confidence interval [CI] = 2.9%–14.5%), with the upper bound of the confidence interval well below the 25% performance goal (p < .0001).

Over the 18‐month follow‐up, there were a total of nine SADEs reported across seven patients. Events were implant site erosion and headache in one patient; implant site infection in two patients (both underwent explant); implant site fibrosis in one patient; chronic encephalopathy in one patient; transient intracranial hemorrhage near one lead in one patient; and two transient procedural complications in one patient (pulmonary edema and acute encephalopathy). All SADEs resolved, except one headache and one encephalopathy, which were ongoing at the time of study withdrawal at 125 days (this patient was excluded from PP analyses due to nonepileptic seizure events).

SAEs of particular relevance as defined in the protocol were implant site infection (n = 2, described above), intracranial hemorrhage (n = 2, one described above as implant site hemorrhage and one subdural hematoma due to a seizure), suicidality (n = 1, patient had a prior history of suicidality and comorbid depression and anxiety), and cognitive deterioration (patient with chronic encephalopathy described above). ¹⁷ There were no SAEs attributed to acute or chronic stimulation.

One definite SUDEP event occurred. To provide a confident estimate of the SUDEP rate, patient data were combined across all prospective RNS System clinical trials (N = 693 patients, 3502.0 implant years). The estimated SUDEP rate of 2.9 (lower confidence limit = 1.5, upper confidence limit = 5.3) per 1000 patient‐years is lower than expected rates in surgery candidates (9.3/1000) and ASM trial placebo arms (6.1/1000). ¹⁸ , ¹⁹

3.2.1. Neurocognitive, mood, and sleep outcomes

NIH Toolbox Cognition Battery measure (Picture Vocabulary, Flanker Inhibitory Control, Picture Sequence Memory) data were available for 56 paired datasets at 12 months and remained stable from baseline. Individual‐level analysis using 90% RCI thresholds showed no systematic decline.

Mean mood scores (Beck Depression Inventory‐II) remained within the “none to minimal” depression range of 0–13 (baseline mean = 10.8 [n = 84], 18‐month mean = 12.2 [n = 66], p = .41). Sleep quality was impaired at baseline; sleep quality assessments (Pittsburgh Sleep Quality Index) indicated there was no adverse impact of RNS System treatment, with a trend toward improvement (p = .46).

3.3. Effectiveness

The prespecified primary effectiveness time‐to‐event analysis with the Cox proportional hazards model did not demonstrate a statistically significant difference between groups. Median time to second GTCS was 76.0 days in the Active group (range = 39.0–127.0) versus 67.0 days in Sham (range = 40.0–84.0), with a hazard ratio of .895 (95% CI = .53–1.50, p = .675). Given the observed data, the prespecified model was not well suited to capture the time‐varying treatment response, as it relied on a single data point per patient —time to second GTCS—without incorporating subsequent observations across the 9‐month EEP. The time‐to‐event design did not provide sufficient time to discriminate the treatment response, particularly in patients with high seizure frequencies at baseline.

A post hoc mixed‐effects model with negative binomial variance structure appropriately accounted for the time‐varying treatment effect. Over the first postimplant year, patients randomized to Active stimulation received an average of 296.2 days of stimulation, compared to an average of 148.6 days for patients randomized to Sham, who had a delayed transition to stimulation treatment (illustrated in Figure 3). Patients randomized to Active treatment achieved a 61% reduction in GTCS from baseline (95% CI = 54%–67%) compared to 49% in the patients originally randomized to Sham (95% CI = 40%–57%), representing a 23% additional reduction attributable to a longer duration of responsive stimulation (95% CI = 2.5%–40%). This treatment effect estimate was significant (p = .03) and was robust across sensitivity analyses including subgroup covariates (gender, prior VNS, age, seizure type, baseline frequency) and site as a random effect.

FIGURE 3.

Days on which individual patients entered the stimulation active group. Modified primary analysis leverages difference in number of days of stimulation across randomization group. Treatment group receives stimulation at 1 month postimplant. Sham group has a delayed and staggered entry into the stimulation‐on condition.

All additional effectiveness endpoints through 18 months showed statistically significant reductions as compared to baseline (Figure 4A–E). Median percent GTCS reduction was 68.9% at 3 months, maintained through 18 months with a 76.8% median reduction (all p < .001 from baseline at every 3‐month interval; Figure 4E). At 18 months, the GTCS 50% responder rate (RR) was 62.5% and 40% of patients were GTCS‐free at that timepoint.

FIGURE 4.

Median percent reduction during stimulation treatment. Median percent change from baseline in (A) days with any generalized seizure, (B) days with generalized tonic–clonic seizures (GTCSs), (C) days with myoclonic seizures, (D) days with absence seizures, and (E) GTCSs is shown by 84‐day intervals from stimulation initiation. Bars represent interquartile range; lines represent median. All timepoints are significantly different at p < .001 from baseline (Wilcoxon signed‐rank test with Bonferroni correction). Per‐protocol population is shown. Note that the apparent dip in median percent reduction over days 253‐336 is not significantly different from other time bins and is most parsimoniously explained by the expected variance.

Significant and sustained reductions in days with any type of generalized seizure (GTCS, myoclonic, or absence) were observed during times that patients were receiving stimulation (all p < .001), ranging from 55.5% to 77.8% across 3‐month intervals (Figure 4A). For patients with baseline absence seizures (n = 45), the median percent reduction in days with absence seizures was 98.2% at 3 months, reaching 100% median reduction by 12 months, and sustained thereafter (Figure 4D). The median reduction in days with myoclonic seizures (38 patients had baseline myoclonic seizures) was 88.0% at 3 months, reaching 100% at 15 and 18 months (Figure 4C).

3.4. Quality of life and global impressions of change

Quality of life did not decline in any domain and improved in some. The most consistent improvements were in seizure‐related domains; Seizure Worry improved significantly at 6, 12, and 18 months (p = .004, .015, and .032, respectively), and Daily Activities improved significantly at 6 and 18 months (p = .028). At 18 months, >90% of patients reported improvement on the Patient Global Impression of Change, and 80% of physicians reported improvement on the Clinical Global Impression of Improvement.

3.5. Additional analyses

3.5.1. ASM changes

These analyses evaluated the number of patients who had clinically meaningful changes in ASMs during the EEP, when medications were to remain stable, and over 12–18 months. The majority of patients had no clinically meaningful ASM change. At 12 months 79% (64/81) and at 18 months 59% (45/76) remained on the same ASMs as baseline. The median percent change in the rate of GTCSs did not differ significantly between patients who had no change, addition/increase, or discontinuation/decrease in ASMs at 12 months (Kruskal–Wallis p = .739) or 18 months (p = .134).

3.5.2. Rescue medication use

The odds of using a benzodiazepine as a rescue medication for a GTCS were 44% lower during responsive stimulation treatment compared to preimplant baseline (OR = .56, 95% CI = .42–.74, p < .001). Bin‐by‐bin contrasts showed significant reductions at months 3–6 (OR = .57) and months 6–9 (OR = .53) after FDR correction, with later bins showing a similar direction of effect but not reaching statistical significance.

3.5.3. Seizure duration and postictal recovery

Postictal recovery time improved across the first 15 months poststimulation (ORs = .38–.54, all p < .001), indicating patients were approximately twice as likely to report shorter recovery times relative to baseline. Five of six time bins achieved significance after Bonferroni correction. Seizure duration showed modest improvement, with two of six time bins reaching corrected significance (months 6–9: OR = .57, p = .004; months 12–15: OR = .47, p < .001), although all bins up to 15 months had ORs < 1. Neither outcome showed significant improvement at months 15–18, likely reflecting reduced power (n = 79 seizures from 24 subjects).

3.5.4. Injury event rates

Two Poisson regression models estimated a reduction in injury event rates posttreatment, although neither reached statistical significance. The fixed‐effects model yielded a rate ratio of .70 (95% CI = .42–1.14, p = .153), corresponding to an estimated 30% reduction, and the GEE model yielded a rate ratio of .66 (95% CI = .34–1.27, p = .214), corresponding to an estimated 34% reduction. The direction and magnitude of the effect were consistent across both modeling approaches, although the CIs were wide, suggesting the study was not powered to detect this effect size.

3.6. RNS System programming

The neurostimulator was programmed to detect low‐frequency activity (>2 Hz) with a high‐frequency cutoff of <12 Hz to avoid detecting and stimulating sleep spindle activity. An example of detection is provided in Figure 5. Bipolar stimulation was delivered via the electrodes exhibiting the most prominent epileptiform activity. Initial stimulation parameters consisted of high‐frequency bursts (143 Hz, 5000‐ms duration) at charge densities of 1.5 μC/cm². If paresthesia was reported, stimulation parameters were adjusted by the treating physician. Recommended adjustments included reducing stimulation frequency to ≤20 Hz or decreasing charge density. Parameters were subsequently refined at intervals of 3 months or more based on clinical response and electrographic findings. The median total daily stimulation time per day was 61.5 min (average = 71.3 min).

FIGURE 5.

Intracranial electroencephalographic (iEEG) patterns consistent with generalized seizures detected in the centromedian nucleus bilaterally. (A) A discharge of sharply contoured rhythmic delta with overlying beta is detected (B1B2) and is followed by a stimulation burst (Tx). A redetection of higher amplitude delta (A1A2) is followed by another stimulation burst, after which the iEEG returns to baseline. The period during which the signal is blanked is the time when stimulation is delivered. (B) The same event with an expanded time window. The patient did not experience a seizure. LCM is the left centromedian nucleus of the thalamus with sensing and stimulating electrode contacts 1–2 and 3–4. The same convention is used for the right centromedian nucleus (RCM) electrodes 1–2 and 3–4. ECOG, electrocorticogram.

4. DISCUSSION

The NAUTILUS study represents the first randomized controlled trial of neuromodulation therapy for idiopathic generalized epilepsy and establishes responsive stimulation of the thalamus as a viable treatment option for patients with drug‐resistant IGE. The study met its primary safety endpoint, with an SADE rate well below the prespecified performance goal. Although the primary effectiveness endpoint, time to second GTCS, did not reveal a significant difference between Active and Sham, a modified analysis that considered each data point according to original randomization group showed that Active treatment conferred a significant benefit compared to Sham (p = .030). Sham patients transitioned to open‐label active stimulation only after their transition event and thus received less stimulation overall during the EEP. Prespecified additional endpoints showed statistically significant, clinically meaningful, and durable reductions in the number of GTCSs, the number of days with GTCSs, and days with any type of generalized seizure (all p < .001).

NAUTILUS was the first neuromodulation trial to use a time‐to event design. The lack of significance in the prespecified primary effectiveness endpoint, in retrospect, may reflect methodological limitations of the Cox proportional hazards model, which assumes constant hazard ratios over time, an assumption that is violated when treatment effects increase with duration of therapy. This phenomenon is well documented in focal epilepsy, where RNS System efficacy continues to improve through at least 9 years of follow‐up. ¹¹ , ²⁰ , ²¹ The post hoc mixed‐effects analysis, which appropriately models repeated‐measures data and time‐varying effects, utilized 1249 monthly observations across 82 PP subjects, compared to 82 time‐to‐event measurements in the prespecified Cox proportional hazard analysis, providing substantially greater statistical power to detect treatment effects. This mixed‐effects analysis demonstrated a statistically significant treatment benefit with Active stimulation, providing an additional reduction in GTCSs compared to patients originally randomized to Sham, for whom the start of stimulation was delayed.

The magnitude and durability of seizure reduction observed in these patients with drug‐resistant IGE is clinically meaningful. Median GTCS reduction of 76.8% at 18 months, a 62.5% RR, and 40% seizure freedom rate for GTCSs at that timepoint are notable for these patients with more than 18 years of IGE despite multiple ASM trials. The efficacy for absence and myoclonic seizures, with median reductions approaching or reaching 100%, suggests that responsive stimulation of the thalamus may be effective for all IGE seizures.

In addition to substantial reductions in GTCSs and other IGE seizures, patients demonstrated robust and sustained improvement in postictal recovery time, along with more modest improvements in seizure duration, outcomes that directly reflect the lived burden of epilepsy and are unlikely to be captured by seizure counts alone. The significant reduction in benzodiazepine rescue medication use on GTCS days further suggests that seizures occurring after stimulation treatment may be less severe or less likely to require acute pharmacologic intervention. Two Poisson regression models estimated a 30%–34% reduction in rates of seizure‐related injury with treatment; although not statistically significant, the direction and magnitude were consistent across approaches, suggesting the study was underpowered to detect this effect size. Importantly, these improvements appear related to responsive stimulation therapy; similar benefits occurred in patients who had stable ASM regimens and in the minority of patients who had changes in ASMs.

The safety profile of responsive stimulation of the thalamus in IGE is consistent with experience in focal epilepsy. The types and rates of adverse events, primarily procedure‐related in the early postoperative period, are expected for any implanted neurostimulation device. The SUDEP rate of 2.9/1000 patient‐years is lower than published rates for drug‐resistant epilepsy populations, consistent with potential protective effects of seizure reduction. ¹⁸ , ¹⁹

Critically, assessments of cognition, mood, or sleep showed no negative effect of stimulation therapy. There were no adverse behavioral outcomes associated with responsive stimulation of the thalamus at any timepoint, including in the most clinically relevant domains of depression and cognitive dysfunction. There was no effect of thalamic stimulation on sleep quality, which is an important consideration, because poor sleep quality is a well‐described comorbidity of IGE. ²² These observations speak to the safety of closed loop, responsive stimulation compared to tonic, programmed stimulation of deep brain nuclei.

The CM nucleus of the thalamus serves as a strategic target for neuromodulation in generalized epilepsies. As a key relay in thalamocortical networks, the CM nucleus participates in the generation and propagation of the bilateral synchronous spike–wave discharges that characterize IGE. Responsive stimulation may act through multiple mechanisms: acute desynchronization of epileptiform activity during detected events and longer term network plasticity effects that reduce seizure susceptibility over time. NAUTILUS results support a robust and sustained response to thalamic stimulation; future analyses at the conclusion of the study (24‐months postimplant) will evaluate whether there are continued improvements as a result of network level modulation over time.

Several limitations warrant consideration. The time to second seizure design was selected because it did not require patients to remain in a prolonged nontreatment arm if there were frequent GTCSs. This design, although attractive to patients and physicians, and ethically appropriate to protect patients from the considerable risks of GTCSs, limited the duration of the blinded comparison and may have diminished the ability to distinguish the effect of responsive stimulation. Many patients, particularly those with a higher baseline seizure frequency, had their second GTCS and exited the blinded period before the stimulation therapy was likely to have an effect. An implant effect, such as has been described in other neuromodulation studies, ⁹ , ²⁰ would impact both treatment groups and may have obscured the effect of stimulation. The duration of the implant effect has not been characterized for this indication. The relatively small number of patients limits subgroup analyses, although treatment effects were consistent across examined covariates. Finally, the current analysis is limited to 18 months; longer term follow‐up at 24 months will be important to further describe longer term safety and the durability of effects on seizure frequency.

In conclusion, the NAUTILUS trial demonstrates that RNS System responsive stimulation of the thalamus is safe and effective for drug‐resistant IGE. The therapy provides clinically meaningful and durable seizure reduction for GTCSs and days with other generalized seizure types, with an acceptable safety profile, and no observed adverse effects on sleep, cognition, or mood. These results establish thalamic responsive neurostimulation as a therapeutic option for patients with drug‐resistant IGE, addressing a significant unmet need in epilepsy care.

AUTHOR CONTRIBUTIONS

Utku Uysal: Investigation (equal); review and editing (equal). Patrick Landazuri: Investigation (equal); review and editing (equal). David E. Burdette: Investigation (equal); review and editing (equal). Sanjay Patra: Investigation (equal); review and editing (equal); supervision (equal). Angela N. Crudele: Investigation (equal); review and editing (equal). Dario J Englot: Investigation (equal); review and editing (equal). Jay R. Gavvala: Investigation (equal); review and editing (equal). Nitin Tandon: Investigation (equal). Sandipan Pati: Investigation (equal); review and editing (equal). Lesley C. Kaye: Investigation (equal); review and editing (equal). Steven Ojemann: Investigation (equal); review and editing (equal). Daniel Barnett: Investigation (equal); review and editing (equal). Joseph Neimat: Investigation (equal); review and editing (equal). Adriana E. Palade: Investigation (equal); review and editing (equal). Shervin Rahimpour: Investigation (equal); review and editing (equal). Amir A. Arain: Investigation (equal); review and editing (equal). Robert Mark Richardson: Investigation (equal); review and editing (equal). Sydney S. Cash: Investigation (equal); review and editing (equal). Vicenta Salanova: Investigation (equal); review and editing (equal). Alexandra Urban: Investigation (equal); review and editing (equal). William P. Welch: Investigation (equal); review and editing (equal). Jorge A. Gonzalez‐Martinez: Investigation (equal). Jamie Joseph Van Gompel: Investigation (equal); review and editing (equal). Keith Starnes: Investigation (equal); review and editing (equal). David Spencer: Investigation (equal); review and editing (equal). Lia D. Ernst: Investigation (equal); review and editing (equal). Ushtar Amin: Investigation (equal); review and editing (equal). Angelica Rivera‐Cruz: Investigation (equal); review and editing (equal). Patricia Dugan: Investigation (equal); review and editing (equal). Marytery Fajardo: Investigation (equal); review and editing (equal). Matt Lallas: Investigation (equal); review and editing (equal). Barbara C. Jobst: Investigation (equal); review and editing (equal). Nicole Odom: Investigation (equal); review and editing (equal). Jarod L. Roland: Investigation (equal); review and editing (equal). Jon T. Willie: Investigation (equal); review and editing (equal). Sameer A. Sheth: Investigation (equal); review and editing (equal). Alica M. Goldman: Investigation (equal); review and editing (equal). Christopher T. Skidmore: Investigation (equal); review and editing (equal). Chengyuan Wu: Investigation (equal); review and editing (equal). Cornelia Drees: Investigation (equal); review and editing (equal). Jonathon Parker: Investigation (equal); review and editing (equal). Taneeta Mindy Ganguly: Investigation (equal); review and editing (equal). Jerzy Szaflarski: Investigation (equal); review and editing (equal). Saadi Ghatan: Supervision (equal); reviewing and editing (equal). Lise Johnson participated in the analysis of the data and drafting the manuscript for intellectual content. Jacob Norman participated in analysis of the data. Brett Wingeier participated in the analysis of the data. Cairn G. Seale participated in the interpretation of the data and drafting the manuscript for intellectual content. Martha J. Morrell participated in the design and conceptualization of the study, analysis and interpretation of the data, and drafting the manuscript for intellectual content.

CONFLICT OF INTEREST STATEMENT

P.L. has served as a consultant and advisory board member for NeuroPace. D.E.B. has received support from NeuroPace and has served as a paid consultant for NeuroPace. A.N.C. has received research funding from NeuroPace and Rapport Pharmaceuticals. D.E. has received consulting fees from NeuroPace. S.R. has received speaker fees from NeuroPace. N.T. is a co‐founder and equity owner of BrainDynamics and has stock options with Nervonik. S.O. has received speaking honoraria from Insightec, has served as director of fellowship supported by educational grants from Medtronic and Boston Scientific, and has consulted for Alpha Omega Engineering and Aspen Neurosciences. R.M.R. has received speaking and consulting fees from NeuroPace. A.U. has served as a consultant/speaker for NeuroPace. J.A.G.M has consulted for Neurologic. J.J.V.G. is a named inventor for intellectual property licensed to Cadence Neuroscience, has licensed intellectual property developed at Mayo Clinic to NeuroOne, and owns stock in NeuroOne. D.S. has served as chair of a DSMB for NeuroPace, and site PI for clinical trials sponsored by Neurona and NeuroPace, but did not receive any personal income from these sources. L.D.E. served as a consultant for Jazz Pharmaceuticals on one occasion in 2025. M.L. has served as a consultant or been part of speakers bureaus for Jazz, Neurelis, UCB, and Pyros. P.D. receives consultant fees from NeuroPace. S.A.S. reports consulting or advisory relationships with Boston Scientific, Zimmer Biomet, Abbott, Koh Young Technology, and NeuroPace and is cofounder of Motif Neurotech. C.T.S. has received research support from NeuroPace and Rapport Therapeutics. C.W. has received consulting fees from BrainLab, NeuroOne, and Renishaw and research support from DIXI Medical. T.M.G. has received research funding from NeuroPace. J.S. reports current and recent funding (past 24 months; longer if directly related) from NIH, NSF, DoD, Shor Foundation for Epilepsy Research, UCB Biosciences, NeuroPace, LivaNova, Ceribell, UniQure, State of AL; has recently served on consulting/advisory boards for PureTech Health, Biopharmaceutical Research Company, LivaNova, UCB Biosciences, AdCel Pharma, and iFovea; is an editorial board member for Epilepsy & Behavior, Epilepsy & Behavior Reports, Journal of Epileptology (associate editor), Journal of Medical Science, and Folia Medica Copernicana; and has served on the Alabama State Medical Cannabis Study Commission and on the Alabama Medical Cannabis Commission (unpaid). S.G. has served as a consultant to NeuroPace. L.J. is an employee of NeuroPace and has equity ownership/stock options with NeuroPace. J.No. is an employee of NeuroPace and has equity ownership/stock options with NeuroPace. C.G.S. is an employee of NeuroPace and has equity ownership/stock options with NeuroPace. B.W. is an employee of NeuroPace and has equity ownership/stock options with NeuroPace. M.J.M. is an employee of NeuroPace and has equity ownership/stock options with NeuroPace. None of the other authors has any conflict of interest to disclose. We confirm that we have read the Journal's position on issues involved in ethical publication and affirm that this report is consistent with those guidelines.

SOCIAL MEDIA AND ARTICLE PROMOTION

First RCT of responsive thalamic neuromodulation in drug‐resistant idiopathic generalized epilepsy reports safety with 77% reduction in tonic–clonic seizures.

DATA AVAILABILITY STATEMENT

The data that support the findings of this study are available from NeuroPace, Inc. Restrictions apply to the availability of these data, which were used under license for this study.