Abstract
Objective
Use of neuromodulation strategies targeting thalamic nuclei, including deep brain stimulation (DBS) and responsive neurostimulation (RNS), for treatment of pediatric drug‐resistant epilepsy (DRE) is increasing, despite limited evidence for efficacy and safety. We present the initial results from the Comparative Multicenter Evaluation of Thalamic Neuromodulation for Treatment‐Resistant Epilepsy in Children consortium, which was created to study thalamic modulation in children with DRE.
Methods
We performed a retrospective cohort study of children who underwent thalamic DBS or RNS for DRE between January 2015 and December 2024 across 12 centers. Primary outcomes were percent seizure reduction, Engel class, and responder rates (>50% seizure reduction) at 3, 6, and 12 months postoperatively. Secondary outcomes included antiseizure medications and surgical complications. Differences were quantified using ordinal logistic regression, generalized estimating equations with center clustering, and mixed‐effects models. Stratified analyses were performed by thalamic target (centromedian vs. anterior nucleus) and epilepsy diagnosis (Lennox–Gastaut syndrome, generalized, focal, and multifocal).
Results
Among the 221 included patients (111 DBS, 110 RNS) included, DBS and RNS achieved comparable Engel scores and 12‐month responder rates, but different temporal trajectories were observed. There was a significant device × time interaction (odds ratio = 1.10/month, 95% confidence interval = 1.01–1.20, p = .04), with RNS response rates increasing from 26.9% at 3 months to 55.6% at 12 months (paired p < .001) but with DBS response rates reaching 41.7% at 3 months and remaining comparatively stable over time. DBS achieved significantly higher 12‐month seizure freedom (9.5% vs. 1.2%, p = .03). Neither thalamic target selection nor diagnosis predicted seizure outcome. Safety profiles were equivalent.
Significance
This multicenter analysis provides pediatric‐specific data to inform surgical indications, device selection, and preoperative counseling. DBS and RNS were effective for treatment of pediatric DRE, with equivalent 12‐month seizure outcomes but different temporal trajectories. DBS provided earlier seizure control and higher rates of overall seizure freedom.
Keywords: deep brain stimulation, drug‐resistant epilepsy, neuromodulation, pediatric, responsive neurostimulation
Key points.
DBS and RNS achieved equivalent 12‐month seizure responder rates (50.0% vs. 56.6%) in 221 children across 12 centers.
DBS provided earlier seizure reduction that plateaued, whereas RNS showed progressive improvement.
No subgroup difference survived FDR correction; device selection should be individualized.
1. INTRODUCTION
Epilepsy is the most common serious neurological disorder of childhood, with approximately 25% of individuals progressing to drug‐resistant epilepsy (DRE), defined as failure of at least two appropriately selected antiseizure medications. 1 , 2 , 3 For children with DRE, resective surgery remains the gold standard because it may be curative; however, it is often not feasible in children with eloquent cortex involvement or multifocal/generalized onset. 4 , 5 Consequently, neuromodulation strategies, including deep brain stimulation (DBS) and responsive neurostimulation (RNS), have emerged as important therapeutic alternatives, presenting evolving treatment options for pediatric patients with DRE. Thalamic neuromodulation is increasingly used for generalized and multifocal seizure disorders, although data remain limited regarding the safety and efficacy of this approach in children. 6
The thalamus is a critical integrative hub that mediates reciprocal corticosubcortical connections and facilitates both seizure propagation and termination. 7 , 8 The anterior nucleus of the thalamus (ANT) and centromedian nucleus (CM) have become principal targets for neuromodulation in epilepsy. 9 The ANT modulates limbic and temporal lobe seizure networks, and the efficacy of ANT neurostimulation for DRE is supported by the landmark SANTE trial, which demonstrated significant seizure reduction with open‐loop DBS in adults with focal epilepsy. 10 , 11 Stimulation of the CM, an intralaminar nucleus with broad cortical and subcortical projections, has shown particular promise for generalized epilepsy syndromes, as demonstrated by the ESTEL trial for Lennox–Gastaut syndrome (LGS). 9 , 12 DBS delivers continuous, open‐loop stimulation to modulate network excitability, whereas RNS employs a closed‐loop system that is thought to detect aberrant electrographic activity and deliver targeted stimulation in real time. 13 , 14 Although randomized trials have focused on adult populations, their results have supported off‐label applications in children with DRE, 15 , 16 and a single partially randomized trial showed that the addition of DBS to optimally managed patients with vagus nerve stimulation results in better seizure control in pediatric populations. 17
Although several systematic reviews have supported the safety and feasibility of thalamic neuromodulation in children, the existing literature comprises small, retrospective cohorts with heterogeneous targets, devices, and outcome measures. 15 , 16 , 18 There is a lack of published data comparing DBS and RNS for pediatric thalamic neuromodulation, and no multicenter dataset has been assembled to evaluate these therapies across diverse institutional practices and patient populations. 19 As a result, surgeon experience, institutional practice patterns, and patient preference determine device selection rather than comparative evidence. To address this gap, we present a multicenter comparison of thalamic DBS and RNS for pediatric DRE from the Comparative Evaluation of Neuromodulation for Treatment‐Resistant Epilepsy (CENTRE) consortium. Our aim was to provide a systematic multicenter comparison of thalamic DBS and RNS in pediatric DRE, characterizing seizure outcomes, temporal trajectories of response, and adverse events across the two modalities, to inform shared decision‐making and establish equipoise for prospective comparative trials.
2. MATERIALS AND METHODS
2.1. Study design and population
This multicenter retrospective cohort study included patients from 12 pediatric epilepsy centers across North America. Specific inclusion criteria were age younger than 21 years at the time of device implantation, a diagnosis of DRE or refractory focal epilepsy as defined by International League Against Epilepsy criteria, 1 and surgical thalamic DBS or RNS device implantation for epilepsy between January 2015 and December 2024. Any patients who had a lead explanted during the follow‐up period were excluded. Institutional review board approval was obtained independently at each participating site with a waiver of consent. This paper was prepared in accordance with the STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) guidelines. 20
2.2. Data collection
Clinical data from electronic health records were collected retrospectively using the ADAPT (A Database of Pediatric Thalamic Neuromodulation) REDCap database, which was managed at the University of Utah and had an established data use agreement at each participating site. As a retrospective cohort study, target and device selection were at the discretion of multidisciplinary teams at each individual center. Variables included patient demographics (age at epilepsy onset, age at implantation, sex), epilepsy characteristics (diagnosis, seizure semiology, 21 preoperative electroencephalographic findings, number of preoperative antiseizure medications), surgical history (prior intracranial monitoring, prior epilepsy surgery), and surgical details (device type, thalamic target, implantation technique, laterality, number of leads). Epilepsy diagnoses were classified as LGS/LGS‐like, focal, multifocal, or other generalized epilepsy. Thalamic targets included the CM, ANT, and pulvinar. Seizure outcomes and antiseizure medication changes were recorded at 3, 6, and 12 months after implantation. Safety data, including intraoperative complications, hospital length of stay, and unplanned return to the operating room, were collected for all patients.
2.3. Outcome measures
The primary outcome measures were (1) seizure reduction category, measured on a four‐level ordinal scale: <50% reduction, 50%–75% reduction, >75%–99% reduction, and seizure freedom; (2) Engel classification; and (3) responder rates at 3, 6, and 12 months after implantation. Seizure reduction categories were determined by the treating clinician at each follow‐up visit based on caregiver‐reported seizure frequency relative to the preimplantation baseline, consistent with outcome reporting in prior neuromodulation trials. 10 , 12 Responder rate was defined as the proportion of patients achieving at least a 50% reduction in seizures to index clinically meaningful responses. 22 Seizure freedom (0 seizures since previous time point), change in antiseizure medication burden, and the occurrence of surgical complications, including intraoperative adverse events and return to the operating room, were assessed as secondary outcomes.
2.4. Statistical analysis
Descriptive statistics were used to summarize baseline demographic and clinical characteristics. Continuous variables were compared between device groups using the Mann–Whitney U‐test and reported as mean ± SD or median (interquartile range), as appropriate. Categorical variables were reported as frequency (percentage) and compared using the chi‐squared test or Fisher exact test (n < 5), depending on the sample size. The primary analysis compared seizure‐reduction categories between DBS and RNS at each postimplantation time point using an ordinal logistic regression (proportional odds model), with device type as the primary predictor. The proportional odds assumption was verified using the Brant test and was satisfied across all time points and model specifications. Adjusted models included thalamic target, epilepsy diagnosis, age at modulation, number of antiseizure medications, baseline seizure frequency, prior vagus nerve stimulation, prior resective surgery, prior intracranial electroencephalographic (EEG) monitoring, baseline focal motor clonic seizures, and treating site as covariates.
To account for within‐center correlations arising from the multicenter design, generalized estimating equations (GEE) with an exchangeable correlation structure were used for the binary responder outcome, with center as the clustering variable. 10 , 23 The exchangeable structure was selected based on the assumption that outcomes within a given center are equally correlated regardless of the temporal ordering of implantations. Mixed‐effects logistic models with random intercept for center were also fitted, with both approaches yielding concordant results, and GEE reported as the primary clustered analysis. The longitudinal trajectory of treatment response was evaluated using a repeated‐measures GEE model with observations at 3, 6, and 12 months, with patients as the clustering unit and an exchangeable correlation structure. A device × time interaction term was included to test whether the temporal trajectory of response differed between DBS and RNS. Within‐device temporal changes were assessed using McNemar test for the binary responder outcome and the Wilcoxon signed‐rank test for ordinal seizure reduction category in patients with data available at both 3 and 12 months.
Prespecified subgroup analyses were performed stratified by thalamic target (CM, ANT) and epilepsy diagnosis (LGS/LGS‐like, focal, multifocal, or other generalized). An exploratory subgroup analysis stratified by seizure onset location on preoperative EEG (focal, multifocal, or generalized) was performed in a sensitivity manner. A device × target interaction term was tested in the ordinal logistic model to evaluate whether device effects varied by thalamic target. All subgroup analyses were considered exploratory. Benjamini–Hochberg false discovery rate (FDR) correction was applied within each subgroup family (thalamic target or epilepsy diagnosis) to account for multiple comparisons. Both nominal and FDR‐corrected p‐values are reported for subgroup analyses. Complete‐case analysis was used at each time point, with the pattern and extent of missing data reported. Notably, differential missingness was observed at the 3‐month time point, with a higher proportion of missing data in the DBS group compared with the RNS group. Differential missingness may reflect expected differences in postoperative follow‐up cadence between devices, with RNS patients typically returning earlier for initial device interrogation. As a sensitivity analysis for missingness, multiple imputation by chained equations (m = 20, proportional‐odds methods for the ordinal seizure reduction outcome) was performed, conditioning on baseline covariates and outcomes at adjacent time points. All tests were two‐sided with a significance threshold of p < .05. Analyses were performed using Python 3.12 and R 4.3.
3. RESULTS
3.1. Patient demographics, baseline characteristics, and safety
A total of 221 patients (111 DBS, 110 RNS) from 12 centers met the inclusion criteria. The mean age at device implantation was 14.7 ± 5.7 years (range = 2.3–20.6) in the DBS group and 15.3 ± 4.2 years (range = 3.2–20.7) in the RNS group (p = .68; Table 1). There were 45 (40.5%) female patients in the DBS cohort and 45 (40.9%) in the RNS cohort (p > .99). Epilepsy diagnosis distribution was comparable between groups (p = .16), with LGS/LGS‐like representing the largest category (35.7%), followed by other generalized epilepsy (21.7%), multifocal (19.5%), and focal (20.4%). There were significantly more focal motor clonic seizures in the DBS group (52.3% vs. 30.0%, p < .001), whereas the rates of other seizure types were similar between the two groups.
TABLE 1.
Baseline characteristics of included patients.
| Variable | DBS, n = 111 | RNS, n = 110 | p |
|---|---|---|---|
| Age at implant, years (range) | 14.7 ± 5.7 (2.3–20.6) | 15.3 ± 4.2 (3.2–20.7) | .68 |
| Age at epilepsy onset, months | 46.4 ± 50.9 | 49.3 ± 49.5 | .47 |
| Female sex | 45 (40.5%) | 45 (40.9%) | >.99 |
| Number of preop ASMs | 2.6 ± .6 | 2.6 ± .6 | .76 |
| Diagnosis | .16 | ||
| LGS/LGS‐like | 39 (35.8%) | 40 (37.7%) | – |
| Focal | 17 (15.6%) | 28 (26.4%) | – |
| Multifocal | 25 (25.7%) | 18 (17.0%) | – |
| Other generalized | 28 (25.7%) | 20 (18.9%) | – |
| Seizure types | – | ||
| Myoclonic | 32 (28.8%) | 26 (23.6%) | .43 |
| Epileptic spasms | 29 (26.1%) | 21 (19.1%) | .24 |
| Atonic | 45 (40.5%) | 40 (36.4%) | .57 |
| Generalized tonic | 57 (51.4%) | 50 (45.5%) | .43 |
| Focal motor clonic | 58 (52.3%) | 33 (30.0%) | .001 |
| Atypical absence | 43 (38.7%) | 41 (37.3%) | .89 |
| Thalamic target | – | ||
| CM | 75 (67.6%) | 77 (70.0%) | .77 |
| ANT | 34 (30.6%) | 17 (15.5%) | .01 |
| Pulvinar | 0 (.0%) | 10 (9.1%) | <.001 |
| Prior VNS | 56 (50.5%) | 29 (26.4%) | <.001 |
| Prior resective surgery | 11 (9.9%) | 23 (20.9%) | .03 |
| Previous intracranial monitoring | 30 (27.0%) | 46 (41.8%) | .02 |
| Implantation technique | (n = 89) | (n = 107) | <.001 |
| Robotic | 27 (30.3%) | 100 (93.5%) | – |
| Frameless | 14 (15.7%) | 3 (2.8%) | – |
| Frame‐based | 48 (53.9%) | 4 (3.7%) | – |
| Days to stimulation initiation, median [IQR] | 26 [13–41] | 35 [22–46] | .01 |
| Intraoperative complications | 0 (.0%) | 2 (1.8%) | .25 |
| Hemorrhage | 0 (.0%) | 2 (1.8%) | – |
| Return to operating room | 3 (2.7%) | 2 (1.8%) | > .99 |
| Infection‐related | 3 (2.7%) | 1 (.9%) | – |
| Hemorrhage | 0 (.0%) | 1 (.9%) | – |
| Hospital LOS, days, median [IQR] | 2 [1–4] | 2 [1–3] | .55 |
Note: Continuous values are reported as mean ± SD and categorical values as n (%) unless otherwise indicated. Boldface font indicates statistical significance at p < .05.
Abbreviations: ANT, anterior nucleus; ASM, antiseizure medication; CM, centromedian nucleus; DBS, deep brain stimulation; IQR, interquartile range; LGS, Lennox–Gastaut syndrome; LOS, length of stay; RNS, responsive neurostimulation; VNS, vagus nerve stimulation.
The CM nucleus was the most common thalamic target in both groups (DBS 67.6% and RNS 70.0%), but the ANT was targeted more frequently in the DBS group (30.6% vs. 15.5%, p = .01). The pulvinar was targeted exclusively in RNS patients (9.1%, p < .001). Prior vagus nerve stimulation was significantly more common in the DBS group (50.5% vs. 26.4%, p < .001), whereas prior resective epilepsy surgery was more common in the RNS group (9.9% vs. 20.9%, p = .03). RNS patients were more likely to have undergone prior intracranial monitoring (27.0% vs. 41.8%, p = .02), and RNS devices were more likely to be implanted robotically (30.3% vs. 93.5%, p < .001). The median time from device implantation to stimulation initiation was shorter for DBS than for RNS (26 days [interquartile range (IQR) = 13–41] vs. 35 days [IQR 22–46], p = .01). Both device types demonstrated favorable safety profiles. Intraoperative complications did not occur in any DBS patients but occurred in two RNS patients (1.8%, both hemorrhage, p = .25). Return to the operating room was infrequent and comparable between DBS (2.7%) and RNS (1.8%, p > .99). Four of five unplanned returns to the operating room (80%) were infection‐related, and the remaining RNS return was for hemorrhage. Two patients (both DBS) underwent device explantation during follow‐up, both for surgical site infection. The median hospital length of stay was 2 days in both groups (DBS IQR = 1–4 and RNS IQR = 1–3, p = .55).
3.2. Overall seizure outcomes
Availability of outcome data varied by time point (Table S1). At 3 months, data were available for 165 patients (72/111 DBS [64.9%] and 93/110 RNS [84.5%]), reflecting differential missingness in the DBS group. Data were available for 165 patients at 6 months (82/111 DBS [73.9%] and 83/110 RNS [75.5%]) and for 165 patients at 12 months (84/111 DBS [75.7%] and 81/110 RNS [73.6%]).
At 12 months, responder rates were similar between devices (DBS 50.0% vs. RNS 55.6%, p = .39). Engel class was comparable between groups (DBS median 3 [IQR = 3–4] vs. RNS 3 [IQR = 3–4], p = .69; Table 2). The numbers of patients achieving <50%, 50%–75%, >75%–99%, or 100% seizure reduction with DBS and RNS are shown in Table S2. Ordinal logistic regression demonstrated no significant difference in seizure reduction category at 12 months (odds ratio [OR] = .71, 95% confidence interval [CI] = .35–2.43, p = .49; Table 3).
TABLE 2.
Seizure outcomes after DBS or RNS device implantation.
| Variable | DBS | RNS | p |
|---|---|---|---|
| 3 months | |||
| Responder rate a | 30/72 (41.7%) | 25/93 (26.9%) | .046 |
| Seizure freedom | 4/72 (5.6%) | 1/93 (1.1%) | .17 |
| Engel class b | 4 [3–4] (n = 63) | 4 [3–4] (n = 93) | .13 |
| 6 months | |||
| Responder rate a | 34/82 (41.5%) | 38/83 (45.8%) | .58 |
| Seizure freedom | 4/82 (4.9%) | 0/83 (.0%) | .06 |
| Engel class b | 3 [3–4] (n = 51) | 3 [3–4] (n = 83) | .29 |
| 12 months | |||
| Responder rate a | 42/84 (50.0%) | 45/81 (55.6%) | .53 |
| Seizure freedom | 8/84 (9.5%) | 1/81 (1.2%) | .03 |
| Engel class b | 3 [3–4] (n = 54) | 3 [3–4] (n = 72) | .69 |
| Within‐patient change (3 to 12 months) | |||
| Responder → responder | 18/59 (30.5%) | 18/71 (25.4%) | – |
| Nonresponder → responder | 14/59 (23.7%) | 24/71 (33.8%) | – |
| Responder → nonresponder | 7/59 (11.9%) | 1/71 (1.4%) | – |
| Nonresponder → nonresponder | 20/51 (33.9%) | 28/71 (39.4%) | – |
| McNemar p | .13 | <.001 | – |
| Wilcoxon signed‐rank p | .15 | <.001 | – |
Note: Continuous values are reported as mean ± SD and categorical values as n (%) unless otherwise indicated. Boldface font indicates statistical significance at p < .05.
Abbreviations: DBS, deep brain stimulation; RNS, responsive neurostimulation.
Responder rate is number of patients achieving ≥50% seizure reduction.
I, free of disabling seizures; II, rare disabling seizures; III, worthwhile improvement; IV, no worthwhile improvement.
TABLE 3.
Multivariable regression analyses.
| Predictor | OR (95% CI) | p |
|---|---|---|
| Seizure category at 12 months, n = 165 a | ||
| RNS vs. DBS | .64 (.23–1.74) | .38 |
| ANT vs. CM target | .42 (.15–1.17) | .10 |
| Focal vs. LGS epilepsy | 1.76 (.60–5.22) | .30 |
| Multifocal vs. LGS epilepsy | 1.72 (.59–5.01) | .32 |
| Other generalized vs. LGS epilepsy | 1.09 (.42–2.81) | .86 |
| Responder status across 3, 6, and 12 months b | ||
| RNS vs. DBS at 6 months | .56 (.26–1.24) | .15 |
| Time, per month | 1.06 (.99–1.13) | .09 |
| Device × time | 1.10 (1.01–1.20) | .04 |
| ANT vs. CM target | .43 (.19–.97) | .04 |
| CM target subgroup, device effect by time point c | ||
| 3 months, n = 113 | .43 (.19–.92) | .03 |
| 6 months, n = 113 | 1.07 (.52–2.18) | .86 |
| 12 months, n = 116 | 1.21 (.61–2.41) | .58 |
Note: Boldface font indicates statistical significance at p < .05.
Abbreviations: ANT, anterior nucleus of the thalamus; CI, confidence interval; CM, centromedian nucleus; DBS, deep brain stimulation; LGS, Lennox–Gastaut syndrome; OR, odds ratio; RNS, responsive neurostimulation.
Cross‐sectional ordinal logistic regression, proportional odds.
Longitudinal generalized estimating equations logistic regression, exchangeable correlation, clustered by patient. Time is centered at 6 months.
Ordinal logistic regression for RNS vs. DBS. OR > 1 favors RNS; OR < 1 favors DBS.
3.3. Temporal trajectories
Despite equivalent 12‐month outcomes, DBS and RNS exhibited significantly different temporal trajectories of treatment response (Figure 1A). DBS demonstrated proportionally earlier seizure reduction, with a 3‐month responder rate of 41.7% compared with 26.9% for RNS (Table 2). Although the DBS responder rate remained relatively stable across follow‐up (Table 2; 41.7% to 50.0%, paired McNemar p = .13, Wilcoxon signed‐rank p = .15), RNS was slower to achieve a responder status, with responder rates increasing from 26.9% at 3 months to 55.6% at 12 months. Paired analysis among 73 RNS patients with data at both time points confirmed the trajectory; 26 patients gained responder status, whereas only one responder became a nonresponder (McNemar p < .001), with a significant shift in response category (Wilcoxon p < .001; Table 2). The longitudinal GEE model with a device × time interaction confirmed the rate of improvement differed significantly between devices (OR = 1.10 per month, 95% CI = 1.01–1.20, p = .04; Table S3). These findings were supported by multiple imputation under the missing at random method (pooled OR = 1.15, 95% CI = 1.05–1.25, p = .002; Table S4). Seizure freedom at 12 months (0 seizures since 6‐month follow‐up appointment) was achieved in 9.5% of DBS patients compared with 1.2% of RNS patients (p = .03; Figure 1B). In the adjusted longitudinal model, ANT targeting was associated with lower odds of responder status compared with CM targeting (OR = .43, 95% CI = .19–.97, p = .04; Table 3).
FIGURE 1.
Overall seizure outcomes after treatment with deep brain stimulation (DBS) or responsive neurostimulation (RNS). (A) Responder rate (patients achieving ≥50% seizure reduction) over time. (B) Seizure freedom rates over time. * indicates statistical significance.
3.4. Outcomes by thalamic target
In the CM subgroup (75 DBS, 77 RNS; with outcome data available for 40–59 DBS and 56–63 RNS patients across time points), responder rates were 46.0% for DBS and 25.4% for RNS at 3 months (p = .02), with rates converging by 12 months (47.5% vs. 56.9%, p = .31; Table S5). Seizure freedom was 8.5% at 12 months after CM‐targeted DBS and 1.7% after CM‐targeted RNS (p = .21). The device × target interaction was not significant (OR = 1.12, 95% CI = .25–4.93, p = .88; Table S3). In the ANT subgroup (34 DBS, 17 RNS; with outcome data available for 22–24 DBS and 11–16 RNS patients across time points), responder rates were equivalent between DBS and RNS across all time points (3 months 31.8% vs. 31.2%, 6 months 34.8% vs. 35.7%, 12 months 54.2% vs. 54.5%, all p > .99), although interpretation is limited by the small number of ANT‐targeted RNS patients with outcome data (Table S4). The pulvinar was targeted exclusively with RNS (n = 10, from three centers), predominantly for multifocal epilepsy. Given the small sample size, outcomes are reported descriptively; responder rates were 25.0% at 3 months (2/8), 66.7% at 6 months (4/6), and 40.0% at 12 months (2/5). No patients who underwent pulvinar‐targeted RNS achieved seizure freedom. No target subgroup comparison remained significant after FDR correction (Table S6).
3.5. Outcomes by epilepsy diagnosis
Exploratory subgroup analyses of responder rates stratified by epilepsy diagnosis revealed no significant difference between devices for LGS/LGS‐like (DBS 53.1% vs. RNS 50.0%, p > .99), multifocal epilepsy (DBS 52.9% vs. RNS 53.8%, p > .99), focal epilepsy (DBS 64.3% vs. RNS 55.6%, p = .72), or other generalized epilepsy (DBS 33.3% vs. 68.4%, p = .06) at 12 months (Figure 2, Table S6). Among patients with other generalized epilepsy, RNS demonstrated a nominally higher 12‐month responder rate than DBS; however, this did not survive correction for multiple comparisons (Benjamini–Hochberg q = .22).
FIGURE 2.
Responder rate by epilepsy diagnosis and device. Patients achieving ≥50% seizure reduction at 12 months are shown. DBS, deep brain stimulation; LGS, Lennox–Gastaut syndrome; RNS, responsive neurostimulation.
3.6. Outcomes by seizure onset location
In an exploratory analysis prompted by the distinct connectivity profiles of the anterior and centromedian thalamic nuclei, we stratified patients by predominant seizure onset location on preoperative EEG (focal, multifocal, or generalized) and restricted the analysis to 150 patients with a pure category designation. No statistically significant onset × cohort interactions were detected. Relative to focal onset, the RNS versus DBS effect did not differ significantly for multifocal onset (OR = 1.52, 95% CI = .26–8.99, p = .65) or generalized onset (OR = 2.73, 95% CI = .44–17.05, p = .28; Table S7).
3.7. Model concordance
All modeling approaches (ordinal logistic regression with site as a fixed effect, GEE clustered by center, and mixed‐effects logistic regression with random intercept for site) yielded concordant conclusions, with all point estimates near 1 and no statistically significant 12‐month differences between devices (Table 3, Figure 3, Table S3).
FIGURE 3.
Forest plot examining device effect across models and subgroups. Odds ratios (ORs; 95% confidence interval [CI]) for deep brain stimulation (DBS) versus responsive neurostimulation (RNS) are shown across models and subgroups. ANT, anterior nucleus of the thalamus; CM, centromedian nucleus; GEE, generalized estimating equations; LGS, Lennox–Gastaut syndrome; mo, month.
4. DISCUSSION
This study from the CENTRE consortium provides a large multicenter comparison of DBS and RNS for pediatric DRE and refractory focal epilepsy, encompassing 221 patients across 12 centers. Within this cohort, three principal findings emerged. First, both devices demonstrated favorable safety profiles, with similarly low rates of intraoperative complications, return to the operating room, and short hospital length of stay. Second, DBS and RNS achieved similar overall seizure reduction at 12 months, with responder rates of 50.0% and 56.6%, respectively, but exhibited distinct temporal trajectories; DBS provided earlier seizure reduction that plateaued over time, whereas RNS showed slower reduction rates between 3 and 12 months (device × time interaction, p = .04). Third, DBS was associated with a higher rate of seizure freedom at 12 months (9.5% vs. 1.2%, p = .03). Taken together, these findings support the safety and efficacy of RNS and DBS thalamic neuromodulation for DRE in this patient population, while highlighting temporal and seizure freedom differences that may inform device selection and family counseling.
4.1. Temporal divergence
The underlying mechanisms of DBS and RNS may explain the divergent temporal trajectories we observed between these devices. DBS uses open‐loop stimulation to modulate thalamic relay circuits. 14 This seemingly instantaneous “on” state may explain the earlier onset of seizure reduction observed in our cohort. Although DBS programming parameters, including frequency, pulse width, and voltage, are iteratively adjusted over the initial weeks to months after implantation, this optimization period is considerably shorter than the RNS titration cycle, which involves both detection signal refinement and stimulation parameter adjustment, allowing it to operate as a closed‐loop system that relies on detecting abnormal electrocorticographic activity and requires titration of target doses. 13 , 14 Consistent with this mechanistic framework, we observed that RNS activation occurred a median of 9 days later than DBS activation (35 vs. 26 days, p = .01), reflecting the additional time required for detection algorithm optimization. However, this 9‐day offset is modest relative to the 12‐month follow‐up window and may not fully account for the sustained divergence in trajectory. Continued improvement in postoperative seizure outcomes with RNS has previously been reported by Nair et al., 24 who described median seizure reduction rates increasing from 44% to 75% over 9 years of follow‐up in adults with focal epilepsy treated with cortical RNS, with responder rates continuing to improve beyond 5 years. Chronic neurostimulation with RNS (and likely also DBS) may induce corticothalamocortical plasticity, progressively remodeling the epileptic network. 25 , 26 Repeated interruption of seizure propagation is also thought to produce an antikindling effect, which gradually raises the seizure threshold over time. 27 The relative plateau in the DBS response rate may indicate that continuous stimulation achieves maximal network modulation earlier after device implantation. Whether DBS or RNS outcomes continue to improve beyond 12 months in children, as adult data suggest, 24 is the basis of ongoing study in the CENTRE consortium.
4.2. Seizure freedom
In comparison with RNS, DBS was associated with greater seizure freedom at 12 months; however, in view of the relatively small subset of patients who achieved seizure freedom (eight DBS vs. one RNS patient), this finding should be considered hypothesis‐generating rather than definitive. Our 12‐month DBS seizure freedom rate (9.5%) is consistent with that observed in the systematic review by Yan et al. (12.5%), 15 is slightly lower than that seen in the SANTE long‐term data (16% at 5 years), 11 and exceeds the pooled estimates from the recent 2025 meta‐analysis by Dhaliwal et al. (ANT 3.57%, CM 1.32%). 28 Although these comparisons are limited by differences in follow‐up duration, outcome definitions, and population characteristics, several mechanistic considerations may explain the differential seizure freedom rates between devices. The higher rate of seizure freedom with DBS may reflect fundamental differences between the neuromodulation techniques. Continuous thalamic stimulation produces sustained disruption of corticothalamic relay circuits, potentially achieving a threshold of network inhibition sufficient to fully suppress seizure generation. Bernstein et al. 29 recently described the long‐term effects of continuous ANT stimulation, demonstrating that seizure freedom may be maintained for years in generalized epilepsy, perhaps suggesting that tonic modulation of thalamic nodes may produce a durable suppressive effect on seizure networks. Further, as Samanta et al. 30 discussed in their review of neuromodulation for developmental and epileptic encephalopathies, CM‐targeted DBS exerts broad modulatory effects across multiple seizure types by continuously engaging thalamocortical circuits. In contrast, responsive event‐triggered stimulation may not replicate this effect in the interictal period. Whether longer follow‐up narrows this gap remains an open question.
4.3. Subgroup analyses
Subgroup analyses by thalamic target and epilepsy diagnosis, although exploratory, yielded several findings that may direct future device selection and investigation. Within the CM subgroup, which comprised the majority of patients, no statistically significant difference in responder rate was observed between devices at any time point, although a numerically higher early response was noted for DBS at 3 months (46.0% vs. 25.4%, p = .02). This descriptive pattern converged by 12 months, and the device × target interaction was not significant (p = .88), confirming that the overall temporal trajectory finding was likely not driven by differences in target distribution between devices. In the adjusted longitudinal model, ANT targeting was associated with lower odds of responder status compared with CM targeting; however, crude responder rates were similar across targets, and this adjusted signal was most likely driven by within‐site differences likely reflecting residual confounding by indication. Yet, this possible early CM‐targeted DBS effect is consistent with the established efficacy of CM stimulation for generalized seizure types and with short‐term improvements reported in the ESTEL trial. 12 , 31 This pattern reproduces the overall trajectory finding at the target level, indicating that the divergent temporal profiles are likely not caused by differences in target distribution between devices. In the ANT subgroup, device outcomes were equivalent across all time points, although the small number of ANT‐targeted RNS patients (n = 11–16 with outcome data) limits the reliability of this comparison. Nevertheless, this finding aligns with the wider body of research. Talbot et al. 32 reported no significant differences in seizure reduction between ANT‐targeted DBS and RNS in a recent meta‐analysis. Similarly, in a meta‐analysis of pediatric thalamic neuromodulation involving 88 patients, Sharma et al. 33 found no significant association between target or modality and treatment response.
Among the exploratory subgroup analyses, a numerically higher RNS responder rate was observed in other generalized epilepsy (68.4% vs. 33.3%, p = .06, FDR‐corrected q = .22), which did not reach statistical significance. Prior single‐center comparisons have generally found no difference in neuromodulation effectiveness between focal and generalized epilepsy or between open‐loop and closed‐loop paradigms. 34 Although the present study was underpowered to detect subgroup differences, this relationship should be examined further in larger prospective cohorts. The hypothesis aligns with emerging evidence that CM‐targeted RNS may be particularly effective for generalized and multifocal epilepsy syndromes. Nanda et al. 35 reported an average seizure reduction of 82.6% and an 18% Engel class I rate with CM‐targeted RNS in patients with idiopathic generalized and rapidly generalizing epilepsy, and pediatric‐specific series from Bercu et al. 36 and Speakes et al. 37 have reported responder rates of 83% and 71%, respectively, in children with generalized DRE treated with CM‐targeted RNS. However, in a recent review, Hsu et al. 38 noted that no modality has demonstrated clear superiority, and our subgroup findings should be considered hypothesis‐generating. Our exploratory analysis, stratified by EEG‐defined seizure onset location, likewise did not detect significant onset × cohort interactions, although cell sizes were modest and CIs were wide. Whether anterior versus centromedial thalamic targeting confers differential benefit by seizure onset location merits dedicated prospective evaluation. The pulvinar, targeted exclusively with RNS in our cohort (n = 10), demonstrated variable responder rates (25%–67% across time points) without achieving seizure freedom, resulting, most likely, from the small sample size, which precluded meaningful comparison. Emerging interest in pulvinar stimulation for posterior quadrant epilepsy 39 warrants further dedicated study.
4.4. Device selection: What is the answer?
Device selection across the CENTRE consortium was influenced by a range of factors beyond individual patient characteristics, including local device availability, surgeon training and familiarity, programming infrastructure, and payer coverage. Although we adjusted for treating site as a fixed effect, we cannot fully rule out time‐varying site‐level effects or device × site interactions that are not captured by mean site differences. However, it is plausible that DBS may be advantageous when rapid response is clinically important, such as in the setting of a high seizure burden, in status‐prone phenotypes, or when minimizing programming visits is a priority. RNS, with its closed‐loop responsive paradigm and on‐device electrocorticography, may offer particular value when chronic electrocorticographic data are clinically useful or when localized seizure onset is well defined but unresectable. Yet, because subgroup analysis did not survive FDR correction, prospective data are needed before more specific device recommendations can be made.
4.5. Limitations
Our study was retrospective and nonrandomized, and factors such as surgeon preference, institutional experience, and unmeasured patient characteristics may have affected device choice. Seizure outcomes were assessed using clinician‐determined reduction categories based on caregiver report rather than prospective seizure diaries, introducing potential recall and classification bias. Differential missingness was observed at 3 months (DBS 35.1% vs. RNS 15.5%), which may bias early time point comparisons. This asymmetry resolved by 12 months, and sensitivity analysis using multiple imputation under the missing at random method supported the primary device × time interaction. However, severely informative missingness that depends on unobserved outcome values cannot be fully ruled out. Follow‐up was limited to 12 months, which may have been insufficient to capture the full trajectory of either device, given that the adult data demonstrate continued improvement over years to decades. 24 Patients who had their device explanted before enrollment were explicitly excluded, and this group warrants further evaluation. We were unable to account for differences in surgical targeting, programming strategies, stimulation parameters, or detection algorithm optimization across the 12 participating centers, all of which may influence outcomes independently of device type. Stimulation‐related and behavioral side effects, including mood changes, were not systematically captured in the retrospective dataset and may be underreported. Prospective registries with standardized outcome capture, including patient‐reported and neuropsychological measures, will be necessary to address these limitations and establish definitive comparative guidance.
4.6. Future directions
These findings underscore the need for prospective, comparative studies of thalamic neuromodulation in pediatric patients with DRE. The CENTRE consortium is currently collecting 2‐year follow‐up data from the ADAPT database to further characterize seizure responses between DBS and RNS. Several other critical clinical questions remain. Prospective registries with standardized outcome measures and seizure diaries would address the limitations inherent in retrospective caregiver‐reported data. We will also continue to collect patient data through the ADAPT database to achieve more detailed subgroup analyses. Finally, integrating electrophysiologic biomarkers into prospective trial designs may enable early prediction of treatment response and individualized device selection, moving the field beyond empiric choice toward precision neuromodulation.
5. CONCLUSIONS
This large multicenter study from the CENTRE consortium indicated that thalamic DBS and RNS may offer similar seizure reduction in children with DRE, although their effects appear to unfold differently over time. DBS tended to produce more rapid benefits and a higher rate of seizure freedom, whereas RNS showed notable improvements gradually over the first year. These results support choosing the device based on individual clinical needs and epilepsy features, and they emphasize the importance of prospective multicenter registries with standardized outcomes and long‐term follow‐up to inform treatment for pediatric patients with DRE.
AUTHOR CONTRIBUTIONS
Samuel A. Tenhoeve: Data curation; formal analysis; investigation; methodology; writing—original draft; writing—review and editing. Nebras M. Warsi: Formal analysis; investigation; methodology; writing—original draft; writing—review and editing. Allison M. Ludwick: Data curation; investigation; project administration; writing—review and editing. Adam C. Glaser: Data curation; investigation; writing—review and editing. Marcella R. Ruppert‐Gomez: Data curation; investigation; writing—review and editing. Scellig S. Stone: Data curation; investigation; writing—review and editing. Megan Votoupal: Data curation; investigation; writing—review and editing. Jeffrey S. Raskin: Data curation; investigation; writing—review and editing. Shanique Doyle: Data curation; investigation; writing—review and editing. Benjamin Kennedy: Data curation; investigation; writing—review and editing. Deborah Grenard: Data curation; investigation; writing—review and editing. Dominic Nistal: Data curation; investigation; writing—review and editing. Hannah E. Goldstein: Data curation; investigation; writing—review and editing. Jason S. Hauptman: Data curation; investigation; writing—review and editing. Gavin Yoon: Data curation; investigation; writing—review and editing. Kurtis I. Auguste: Data curation; investigation; writing—review and editing. David D. Gonda: Data curation; investigation; writing—review and editing. Alon Kashanian: Data curation; investigation; writing—review and editing. Shaun D. Rodgers: Data curation; investigation; writing—review and editing. Janette Coble: Data curation; investigation; writing—review and editing. Jarod L. Roland: Data curation; investigation; writing—review and editing. Nunthasiri Wittayanakorn: Data curation; investigation; writing—review and editing. Chima O. Oluigbo: Data curation; investigation; writing—review and editing. Somnath Das: Data curation; investigation; writing—review and editing. Jasmine A. Thum: Data curation; investigation; writing—review and editing. Carolina Gorodetsky: Data curation; investigation; writing—review and editing. MyLoi Huynh: Data curation; investigation; writing—review and editing. George M. Ibrahim: Data curation; investigation; writing—review and editing. Robert J. Bollo: Conceptualization; formal analysis; investigation; methodology; supervision; writing—original draft; writing—review and editing.
CONFLICT OF INTEREST STATEMENT
G.M.I. receives investigator‐initiated grant funding from LivaNova, is a consultant for Medtronic, Synergia, and LivaNova, and serves on the scientific advisory board of Synergia. No other authors have any financial disclosures. 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.
ETHICS APPROVAL
This study was approved by the institutional review board at each of the 12 participating centers.
CONSENT
The requirement for individual patient consent was waived by each site's institutional review board given the retrospective nature of the study and use of deidentified data.
Supporting information
Table S1. Seizure reduction data availability by time point and device.
Table S2. Full seizure reduction category distributions.
Table S3. Complete regression model outputs.
Table S4. Multiple imputation under missing at random–pooled generalized estimating equation responder model.
Table S5. Seizure outcomes for the thalamic target region subgroups.
Table S6. Subgroup analyses with multiple comparison correction.
Table S7. Seizure onset category × cohort interaction.
ACKNOWLEDGMENTS
We thank Kristin Kraus and Cortlynd Olsen for editorial assistance.
DATA AVAILABILITY STATEMENT
The data that support the findings of this study are available from the corresponding author upon reasonable request.
REFERENCES
- 1. Kwan P, Arzimanoglou A, Berg AT, Brodie MJ, Allen Hauser W, Mathern G, et al. Definition of drug resistant epilepsy: Consensus proposal by the ad hoc task force of the ILAE commission on therapeutic strategies. Epilepsia. 2010;51:1069–1077. [DOI] [PubMed] [Google Scholar]
- 2. Kwan P, Brodie MJ. Early identification of refractory epilepsy. N Engl J Med. 2000;342:314–319. [DOI] [PubMed] [Google Scholar]
- 3. Xue‐Ping W, Hai‐Jiao W, Li‐Na Z, et al. Risk factors for drug‐resistant epilepsy: A systematic review and meta‐analysis. Medicine (Baltimore). 2019;98:e16402. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4. Jobst BC, Cascino GD. Resective epilepsy surgery for drug‐resistant focal epilepsy: A review. JAMA. 2015;313:285–293. [DOI] [PubMed] [Google Scholar]
- 5. Cross JH, Reilly C, Gutierrez Delicado E, Smith ML, Malmgren K. Epilepsy surgery for children and adolescents: Evidence‐based but underused. Lancet Child Adolesc Health. 2022;6:484–494. [DOI] [PubMed] [Google Scholar]
- 6. Sisterson ND, Kokkinos V, Urban A, Li N, Richardson RM. Responsive neurostimulation of the thalamus improves seizure control in idiopathic generalised epilepsy: Initial case series. J Neurol Neurosurg Psychiatry. 2022;93:491–498. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7. Penfield W. Epileptic automatism and the centrencephalic integrating system. Res Publ Assoc Res Nerv Ment Dis. 1952;30:513–528. [PubMed] [Google Scholar]
- 8. Piper RJ, Richardson RM, Worrell G, Carmichael DW, Baldeweg T, Litt B, et al. Towards network‐guided neuromodulation for epilepsy. Brain. 2022;145:3347–3362. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9. Warsi NM, Yan H, Suresh H, Wong SM, Arski ON, Gorodetsky C, et al. The anterior and centromedian thalamus: Anatomy, function, and dysfunction in epilepsy. Epilepsy Res. 2022;182:106913. [DOI] [PubMed] [Google Scholar]
- 10. Fisher R, Salanova V, Witt T, Worth R, Henry T, Gross R, et al. Electrical stimulation of the anterior nucleus of thalamus for treatment of refractory epilepsy. Epilepsia. 2010;51:899–908. [DOI] [PubMed] [Google Scholar]
- 11. Salanova V, Witt T, Worth R, Henry TR, Gross RE, Nazzaro JM, et al. Long‐term efficacy and safety of thalamic stimulation for drug‐resistant partial epilepsy. Neurology. 2015;84:1017–1025. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12. Dalic LJ, Warren AEL, Bulluss KJ, Thevathasan W, Roten A, Churilov L, et al. DBS of thalamic centromedian nucleus for Lennox‐Gastaut syndrome (ESTEL trial). Ann Neurol. 2022;91:253–267. [DOI] [PubMed] [Google Scholar]
- 13. Morrell MJ. Responsive cortical stimulation for the treatment of medically intractable partial epilepsy. Neurology. 2011;77:1295–1304. [DOI] [PubMed] [Google Scholar]
- 14. Gouveia FV, Warsi NM, Suresh H, Matin R, Ibrahim GM. Neurostimulation treatments for epilepsy: Deep brain stimulation, responsive neurostimulation and vagus nerve stimulation. Neurotherapeutics. 2024;21:e00308. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15. Yan H, Toyota E, Anderson M, Abel TJ, Donner E, Kalia SK, et al. A systematic review of deep brain stimulation for the treatment of drug‐resistant epilepsy in childhood. J Neurosurg Pediatr. 2019;23:274–284. [DOI] [PubMed] [Google Scholar]
- 16. Khan M, Paktiawal J, Piper RJ, Chari A, Tisdall MM. Intracranial neuromodulation with deep brain stimulation and responsive neurostimulation in children with drug‐resistant epilepsy: A systematic review. J Neurosurg Pediatr. 2022;29:208–217. [DOI] [PubMed] [Google Scholar]
- 17. Suresh H, Mithani K, Warsi N, Ochi A, Otsubo H, Drake JM, et al. Add‐on deep brain stimulation versus continued Vagus nerve stimulation for childhood epilepsy (ADVANCE): A partially randomized patient preference trial. Ann Neurol. 2024;96:405–411. [DOI] [PubMed] [Google Scholar]
- 18. Mithani K, Niazi F, Suresh H, Alrumayyan Y, Rayco ER, Ochi A, et al. Deep brain stimulation of the centromedian nucleus for drug‐resistant epilepsy in children: Quality‐of‐life and functional outcomes from the CHILD‐DBS registry. Epilepsia. 2025;66:2225–2238. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19. Samanta D, Aungaroon G, Albert GW, Karakas C, Joshi CN, Singh RK, et al. Advancing thalamic neuromodulation in epilepsy: Bridging adult data to pediatric care. Epilepsy Res. 2024;205:107407. [DOI] [PubMed] [Google Scholar]
- 20. von Elm E, Altman DG, Egger M, Pocock SJ, Gøtzsche PC, Vandenbroucke JP. The strengthening the reporting of observational studies in epidemiology (STROBE) statement: Guidelines for reporting observational studies. J Clin Epidemiol. 2008;61:344–349. [DOI] [PubMed] [Google Scholar]
- 21. Beniczky S, Trinka E, Wirrell E, Abdulla F, al Baradie R, Alonso Vanegas M, et al. Updated classification of epileptic seizures: Position paper of the international league against epilepsy. Epilepsia. 2025;66:1804–1823. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22. van der Kop ML. The need for an individualized approach to what is considered a clinically significant reduction in seizure frequency: A patient's perspective. Epilepsia. 2023;64:1469–1471. [DOI] [PubMed] [Google Scholar]
- 23. Liang K‐y, Zeger SL. Longitudinal data analysis using generalized linear models. Biometrika. 1986;73:13–22. [Google Scholar]
- 24. Nair DR, Laxer KD, Weber PB, Murro AM, Park YD, Barkley GL, et al. Nine‐year prospective efficacy and safety of brain‐responsive neurostimulation for focal epilepsy. Neurology. 2020;95:e1244–e1256. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25. Anderson DN, Charlebois CM, Smith EH, Davis TS, Peters AY, Newman BJ, et al. Closed‐loop stimulation in periods with less epileptiform activity drives improved epilepsy outcomes. Brain. 2024;147:521–531. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26. Khambhati AN, Shafi A, Rao VR, Chang EF. Long‐term brain network reorganization predicts responsive neurostimulation outcomes for focal epilepsy. Sci Transl Med. 2021;13:eabf6588. [DOI] [PubMed] [Google Scholar]
- 27. Tass PA, Majtanik M. Long‐term anti‐kindling effects of desynchronizing brain stimulation: A theoretical study. Biol Cybern. 2006;94:58–66. [DOI] [PubMed] [Google Scholar]
- 28. Dhaliwal JK, Ruiz‐Perez M, Chari A, Piper RJ, Tisdall MM, Hart M. Deep brain stimulation for epilepsy: A systematic review and meta‐analysis of randomized and non‐randomized studies of thalamic targeting. Epilepsy Res. 2025;216:107607. [DOI] [PubMed] [Google Scholar]
- 29. Bernstein ZL, Raghu AL, Divanbeighi Zand AP, Kheder A, Gross RE. Long‐term seizure reduction in generalized epilepsy after anterior nucleus of the thalamus stimulation. Epilepsy Behav. 2025;167:110389. [DOI] [PubMed] [Google Scholar]
- 30. Samanta D, Haneef Z, Albert GW, Naik S, Reeders PC, Jain P, et al. Neuromodulation strategies in developmental and epileptic encephalopathies. Epilepsy Behav. 2024;160:110067. [DOI] [PubMed] [Google Scholar]
- 31. Velasco F, Velasco AL, Velasco M, Jiménez F, Carrillo‐Ruiz JD, Castro G. Deep brain stimulation for treatment of the epilepsies: The centromedian thalamic target. Acta Neurochir Suppl. 2007;97:337–342. [DOI] [PubMed] [Google Scholar]
- 32. Talbot NC, Bridges JR, Luther PM, Folse MS, Peco BV, Galardo C, et al. Comparison of deep brain stimulation of the hippocampus to thalamic targets and responsive neurostimulation for adult intractable epilepsy: A systematic review and meta‐analysis. World Neurosurg. 2025;194:123621. [DOI] [PubMed] [Google Scholar]
- 33. Sharma A, Parfyonov M, Tiefenbach J, Hogue O, Nero N, Jehi L, et al. Predictors of therapeutic response following thalamic neuromodulation for drug‐resistant pediatric epilepsy: A systematic review and individual patient data meta‐analysis. Epilepsia. 2024;65:542–555. [DOI] [PubMed] [Google Scholar]
- 34. Alcala‐Zermeno JL, Gregg NM, Starnes K, Mandrekar JN, van Gompel JJ, Miller K, et al. Invasive neuromodulation for epilepsy: Comparison of multiple approaches from a single center. Epilepsy Behav. 2022;137:108951. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35. Nanda P, Sisterson N, Walton A, Chu CJ, Cash SS, Moura LMVR, et al. Centromedian region thalamic responsive neurostimulation mitigates idiopathic generalized and multifocal epilepsy with focal to bilateral tonic‐clonic seizures. Epilepsia. 2024;65:2626–2640. [DOI] [PubMed] [Google Scholar]
- 36. Bercu MM, Zargar BS, Spykman KE, Heredia G, Mogilner AY, Hernandez AW, et al. Centromedian nucleus targeting in the pediatric population treated with thalamic responsive neurostimulation for drug‐resistant epilepsy. Epilepsia Open. 2026;11:322–331. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 37. Speakes ME, Reznik‐Schaefer K, Al‐Ramadhani R, Fernandez LD, Hect JL, Abel TJ, et al. Treatment of pediatric drug‐resistant generalized epilepsy with responsive neurostimulation of the centromedian nucleus of the thalamus: A case series of seven patients. Epilepsy Res. 2025;210:107516. [DOI] [PubMed] [Google Scholar]
- 38. Hsu J, Ali R, Air E, Englot DJ, Muh CR, Richardson RM, et al. Thalamic neuromodulation: New frontiers. Epilepsy Curr. 2026;22:15357597261416388. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 39. de Monaco BA, Santos AN, Cordeiro JG, et al. Pulvinar responsive neurostimulation for posterior quadrant epilepsy: An illustrative case with connectome projection. Neurosurg Rev. 2025;49:51. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Table S1. Seizure reduction data availability by time point and device.
Table S2. Full seizure reduction category distributions.
Table S3. Complete regression model outputs.
Table S4. Multiple imputation under missing at random–pooled generalized estimating equation responder model.
Table S5. Seizure outcomes for the thalamic target region subgroups.
Table S6. Subgroup analyses with multiple comparison correction.
Table S7. Seizure onset category × cohort interaction.
Data Availability Statement
The data that support the findings of this study are available from the corresponding author upon reasonable request.