Does response to vagus nerve stimulation for drug‐resistant epilepsy differ in patients with and without Lennox–Gastaut syndrome?

Abstract Introduction Literature on outcomes of patients with Lennox–Gastaut syndrome (LGS) receiving adjunctive vagus nerve stimulation (VNS) lacks information on seizure types and the time course of therapeutic effects. We have therefore performed what is to our knowledge the largest and most in‐depth analysis of the effectiveness of VNS in LGS patients paying special attention to the impact of VNS Therapy on individual seizure types. Methods The VNS Therapy Outcomes Registry includes over 7000 patients. A propensity score matching method was employed to match patients with LGS to non‐LGS patients with drug‐resistant epilepsy (DRE). Overall seizure frequencies were assessed prior to implantation and at 3‐, 6‐, 12‐, 18‐, and 24‐month follow‐ups to derive the main study outcomes: response rates and time to first response. Results A total of 564 LGS patients with sufficient data were identified in the registry and matched 2:1 to 1128 non‐LGS patients. Responder rates at 24 months were 57.5% in the LGS group and 61.5% in the non‐LGS group. Median seizure frequency reduction at 24 months was 64.3% versus 66.7% in the LGS versus non‐LGS group, respectively. In both groups, VNS was most effective at reducing focal aware seizures, “other” seizures, generalized‐onset non‐motor seizures, and drop attacks with relative reduction rates for these seizure types at 24 months exceeding 90% in both groups. Time‐to‐first response did not differ between the groups; however, there was a significantly higher proportion of patients who regressed from bilateral tonic–clonic (BTC) seizure response in the LGS group versus the non‐LGS group at 24 months: 22.4% versus 6.7%; p = .015. Conclusions Although limited by its retrospective design, the study shows that the effectiveness of VNS is comparable in DRE patients with and without LGS; however, LGS patients may be more prone to fluctuating control of BTCs.


INTRODUCTION
Lennox-Gastaut syndrome (LGS) is a severe epileptic encephalopathy of childhood-onset and poses a therapeutic challenge to physicians . By combining and balancing multiple therapeutic options to treat seizures, the goal is to prevent negative effects on comorbidities with the best possible behavioral and neurodevelopmental outcomes (Strzelczyk & Schubert-Bast, 2021). These treatments can be pharmacological or non-pharmacological and must address multiple seizure types in a highly drug-resistant population.
Vagus nerve stimulation (VNS) was approved as an adjunctive therapy for adults and children of all ages with drug-resistant seizures in 1994 in Europe, and for adults and children over 12 years of age with drug-resistant focal-onset seizures in 1997 in the United States (US).
The age limitation in the US was later reduced to children over 4 years of age.
VNS Therapy involves intermittent electrical stimulation of the left cervical vagus nerve. This stimulation induces action potentials traveling predominantly afferently to the brain where they modulate the metabolism and excitability of structures that compose what is being increasingly referred to as the vagal afferent network (Hachem et al., 2018). VNS Therapy is generally considered when patients with drug-resistant epilepsy (DRE) are not candidates for resective surgery, either for etiological reasons, patient preference, or caregiver preference. Typically, with adjunctive VNS Therapy, approximately 60% of DRE patients experience 50% or more seizure frequency reduction, whereas 40%-50% of patients with DRE experience a reduction in seizure severity or duration. Improvements in mood and certain domains of cognition have also been shown to be associated with VNS Therapy (Orosz et al., 2014;Spindler et al., 2019).

Numerous investigations of VNS Therapy in patients with
LGS have been published in the past 25 years; many of them report similar response rates for LGS patients and heterogeneous DRE populations . However, the majority of the studies do not report the effects of VNS Therapy on individual seizure types.
One study investigating the effects of VNS and callosotomy in LGS patients found both treatments to be effective in controlling atypical absences and bilateral tonic-clonic (BTC) seizures; VNS to be less effective than callosotomy in reducing drop seizures, and more effective than callosotomy in reducing myoclonic seizures; and both treatments to be ineffective in controlling tonic seizures. However, this was the experience of 44 patients at a single center, and these results have yet to be reproduced in larger cohorts (Cukiert et al., 2013). Our recently published analysis in which demographics and clinical characteristics of DRE patients with and without LGS who later underwent VNS implantation were compared found that prior to VNS, LGS patients were taking more anti-seizure medications (ASMs) with poorer seizure control and had more than twice the seizure burden than non-LGS patients with mainly BTCs contributing to this difference . Approximately 11% of LGS patients had undergone prior epilepsy surgery compared to 19% of non-LGS patients.
Considering these differences in disease burden, the question arises whether there are also differences in VNS response among the groups.
We therefore performed what is to our knowledge the largest and most in-depth analysis of the effectiveness of VNS in LGS patients paying special attention to the impact of VNS Therapy on individual seizure types and on the time course of response.

Study design
This is a post-market registry-based prospective cohort study of patients diagnosed with DRE and treated with VNS Therapy adjunctive to ASMs, which uses propensity score matching (PSM) to compare outcomes of two patient groups within the registry. The design and analysis of this study were conducted in accordance with the Strengthening the Reporting of Observational Studies in Epidemiology and the Reporting of Studies Conducted using Observational Routinely-Collected Health Data statements (Benchimol et al., 2015;Von Elm et al., 2007). This retrospective analysis was approved by the ethical committee of the Friedrich Schiller University Hospital (2022-2575).

Setting and participants
The VNS Therapy Patient Outcome Registry includes patients receiving VNS Therapy as an adjunctive treatment for DRE. The registry was established in 1999 by the device manufacturer Cyberonics, Inc. (now LivaNova PLC) after the approval of adjunctive VNS Therapy in the US for DRE. The goal of the registry was to systematically monitor treatment outcomes in implanted patients.

Data sources and variables
The registry data were prospectively and voluntarily provided by 1285 prescribing physicians from 978 centers of which 911 were in the US and Canada and 67 in other countries across the world. The physicians or their designated clinical staff completed standard case report forms based on a patient's medical history or current visit and voluntarily submitted the forms to the registry for data entry. Previously, investigators have authenticated the integrity of the systems for collecting and processing registry data using an independent auditing agency (Amar et al., 2004).
All study data were de-identified prior to analysis. Individual, de-identified data were only used to construct aggregate statistics, including age-standardization. Only aggregate data were retained and presented.
Access to data was restricted to a minimum number of study investigators and accessed via encrypted security codes without further distribution prior to de-identification.
The database was queried to extract seizure outcomes reported at baseline and up to 24 months after VNS implantation, as well as safety data of patients with DRE with or without LGS.

Bias
The PSM homogeneous population was identified to control for selection bias. A propensity score method was used to match patients in the disease populations, that is, LGS patients with DRE and non-LGS patients with DRE (Rosenbaum & Rubin, 1983). An ordinal logistic regression model was run to regress the disease population variable on age at implant, age at diagnosis, and sex. Patients were paired based on an optimal matching with a 1:2 ratio assuming that the number of LGS patients in the registry is at least half of the number of non-LGS patients.

Sample size
This is an enumerative study, and the sample size is not based on a statistical power calculation. The eligibility of all patients in the VNS Therapy Patient Outcome Registry was evaluated for inclusion in the analysis.

Statistical methods
Demographics and de-identified patient characteristics were analyzed the patient to fall such as generalized onset atonic or BTC. Analysis of the seizure type "aura" was discontinued owing to a low sample size delivering unreliable data.
Analyses by seizure type were performed in patients having a seizure count of >0 for a specific seizure type at baseline.
Subjects with a new seizure type onset were not counted to evaluate changes from baseline. At each post-baseline visit, a continuity correction was applied to 0 seizure counts becoming 0.5. However, the number of patients for whom a seizure type was reported at a followup visit that was not reported during the baseline period was assessed per seizure type.
The primary endpoint in this study was the seizure response rate at each available follow-up assessment in LGS patients with DRE versus matched non-LGS patients with DRE included in the PSM population (hereafter referred to as the LGS group and the non-LGS group, respectively). Responder status was defined as a reduction in seizure frequency from the baseline of 50% or more. A regressed status was defined as a subject who was in responder status at the previous follow-up visit but had less than 50% seizure frequency reduction compared to baseline at a later follow-up visit.
At each time point, the proportions of responders and of regressed subjects, together with the corresponding 95% confidence intervals (95% CI), were derived from the Clopper-Pearson method.
Independently, for each follow-up, a two-proportion z-test was applied to compare the responder rates between the LGS group and the non-LGS group. As this is a non-confirmatory study, no approaches were considered to correct for multiplicity.
Missing data on the primary endpoint were imputed according to missing at random assumption and taking into consideration specific baseline characteristics, that is, age at implant, age at diagnosis, sex, average seizure count at baseline, and seizure type at baseline. Twohundred imputation datasets were generated and combined for the inference using the Little and Rubin framework (Little & Rubin, 1987).
Distribution of average count of seizures per month and change from baseline in seizure counts were summarized descriptively by LGS and non-LGS populations. Due to the non-normal distribution of the seizure differences, the median of changes from baseline (i.e., calculated as the median of all differences between the seizure count at the follow-up visit and at baseline) was provided together with the 95% CI derived by the bootstrap method (Puth et al., 2015). The Wilcoxon test was used to compare the medians of the two groups.
Generalized Estimating Equations (GEE) models assuming (i) a binomial distribution with a logit link function for the responder rate outcome and (ii) a negative binomial distribution with a log-linear link function for the post-implant average seizures counts were run overall and by seizure type to evaluate the effect of the disease populations (LGS vs. non-LGS) (Liang & Zeger, 1986). The models were adjusted for the effect of the follow-up. The exchangeable correlation was assumed within-patients.
Kaplan-Meier plots and estimates for time-to-first response were provided for each disease population. The starting date of the survival functions was from the study day 0 (implant date). Time was in months until termination, and it was censored at the date of discontinuation or analysis cutoff date.
As our baseline analysis indicated a higher proportion of patients with developmental delay and mental retardation in the LGS group, safety endpoints were chosen that are less biased by the impairment of patients to report them (e.g., hoarseness in a nonverbal patient). Therefore, safety was evaluated based on reported hospitalizations and all-cause mortality obtained from the total patient years of exposure during the registry period.

Demographics and clinical characteristics
The
In the LGS group, the median reduction of total seizures was 48.6%, 52.1%, 66.7%, 56.4%, and 64.3% at 3, 6, 12, 18, and 24 months, respectively, compared with 45.5%, 50%, 60%, 61.4%, and 66.7% in the non-LGS group. The Wilcoxon test comparing the medians of the two groups did not find significant differences in total seizure frequency reduction at any of the follow-up time points.
The prevalence of individual seizure types showed a different distribution between the two groups. BTCs were the most prevalent seizure type in the LGS group affecting more than half of patients at baseline, whereas BTCs only occurred in a third of non-LGS patients, in whom FIAs were the most common seizure type at baseline. When comparing the baseline visit to the 24-month follow-up visit, no significant changes in seizure type prevalence were observed in the non-LGS group, whereas FAs were completely eradicated, and the prevalence of "other" seizures were reduced by 15% in the LGS group ( Figure S1).
The Wilcoxon test comparing the medians of both groups did not identify any statistically significant differences in the relative reduction of any individual seizure type at 24 months. In both groups, VNS was most effective at reducing FAs, "other" seizures, GONMs, and "drop attacks" with relative reduction rates for these seizure types at 24 months exceeding 90% in both groups. The median relative reduction for FIAs, BTC, and FBTC at 24 months was 80%, 71.7%, and 70% in the LGS group and 70.2%, 71.4%, and 87.5% in the non-LGS group, respectively ( Figure 2).
As relative changes in seizure frequency of individual seizure types can only be assessed in patients who experienced the seizure type at baseline, it is important to also analyze how many patients reported an emergence of an individual seizure type that they did not report at baseline.
Cumulative rates of patients who experienced an emergence of an individual seizure type that was not reported at baseline are provided in Figure 3. BTCs, FIAs, and drop attacks were the most frequently emerging seizure types in the LGS group with cumulative rates of 12.2%, 11.4%, and 8.6% at 24 months, respectively. In the non-LGS group FIAs, FBTCs, and BTCs were the most frequently emerging seizure types with cumulative rates of 16.5%, 9.5%, and 8% at 24 months, respectively.
The highest responder rates by seizure type were for drop attacks (78% and 90% in LGS DRE and non-LGS groups, respectively) and for FBTCs (74.1% and 73.2% in LGS and non-LGS groups, respectively) (Figure 4). No differences in time-to-first response were found between the LGS group (6, 3-20 months) and the non-LGS (6, 3-18 months).

Safety findings
Within the 24 months of follow-up, there were 4 deaths in the LGS group and 5 deaths in the non-LGS group corresponding to F I G U R E 1 Response rate based on all seizure types: Forest plot for response rates based on all seizures for both groups. CI, confidence intervals; LGS, Lennox-Gastaut syndrome.
F I G U R E 2 Reduction of seizures by seizure type: (A) relative change in seizure frequency by seizure type within Lennox-Gastaut syndrome (LGS) drug-resistant epilepsy (DRE) patients experiencing that seizure type at baseline; (B) relative change in seizure frequency by seizure type within non-LGS DRE patients experiencing that seizure type at baseline. BTC, bilateral tonic-clonic; FA, focal aware; FBTC, focal-to-BTC; FIA, focal impaired awareness; GONM, generalized onset non-motor.
all-cause mortality rates of 8.4 per 1000 patient years in the LGS group and 5.47 per 1000 patient years in the non-LGS group. Hospitalizations could not be analyzed due to low data quality: Data coverage was less than 5% and decreased with every follow-up visit leading to an absence of an evaluable amount of hospitalization data.

DISCUSSION
To our knowledge, this analysis represents the largest and most in- quency reduction and responder rates were 55%-60% at 24 months. This is consistent with our recent meta-analysis of adjunctive VNS Therapy in patients with LGS that found a responder rate of 54% across studies . Furthermore, VNS was most effective in reducing the frequency of the same seizure types ("other" seizures, GONMs, and drop attacks) and did so to a similar degree in both populations.

Limitations
The

CONCLUSION
VNS is a safe and effective adjunctive treatment for patients with DRE with or without LGS. The present analysis suggests that LGS patients may be more prone to fluctuating control of BTCs; however, it is unclear whether this is a result of VNS efficacy or reflects the known evolution of seizure types over time in LGS. Major differences in response to VNS between the LGS and non-LGS groups were not identified. VNS was associated with a significant reduction of the frequency of BTCs and drop attacks by more than 70% and 90%, respectively, in both LGS and non-LGS groups.

ACKNOWLEDGMENTS
The authors extend their sincere gratitude to all the patients, their families and caregivers, and the clinical staff who contributed to this study.

DATA AVAILABILITY STATEMENT
The data that support the findings of this study are available upon request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.