Vagus nerve stimulation for partial seizures
Abstract
Background
Vagus nerve stimulation (VNS) is a neuromodulatory treatment that is used as an adjunctive therapy for treating people with medically refractory epilepsy. VNS consists of chronic intermittent electrical stimulation of the vagus nerve, delivered by a programmable pulse generator. The majority of people given a diagnosis of epilepsy have a good prognosis, and their seizures will be controlled by treatment with a single antiepileptic drug (AED), but up to 20%‐30% of patients will develop drug‐resistant epilepsy, often requiring treatment with combinations of AEDs. The aim of this systematic review was to overview the current evidence for the efficacy and tolerability of vagus nerve stimulation when used as an adjunctive treatment for people with drug‐resistant partial epilepsy. This is an updated version of a Cochrane review published in Issue 7, 2010.
Objectives
To determine:
(1) The effects on seizures of VNS compared to controls e.g. high‐level stimulation compared to low‐level stimulation (presumed sub‐therapeutic dose); and
(2) The adverse effect profile of VNS compared to controls e.g. high‐level stimulation compared to low‐level stimulation.
Search methods
We searched the Cochrane Epilepsy Group's Specialised Register (23 February 2015), the Cochrane Central Register of Controlled Trials (CENTRAL) (The Cochrane Library 23 February 2015), MEDLINE (1946 to 23 February 2015), SCOPUS (1823 to 23 February 2015), ClinicalTrials.gov (23 February 2015) and ICTRP (23 February 2015). No language restrictions were imposed.
Selection criteria
The following study designs were eligible for inclusion: randomised, double‐blind, parallel or crossover studies, controlled trials of VNS as add‐on treatment comparing high and low stimulation paradigms (including three different stimulation paradigms ‐ duty cycle: rapid, mid and slow) and VNS stimulation versus no stimulation or a different intervention. Eligible participants were adults or children with drug‐resistant partial seizures not eligible for surgery or who failed surgery.
Data collection and analysis
Two review authors independently selected trials for inclusion and extracted data. The following outcomes were assessed: (a) 50% or greater reduction in total seizure frequency; (b) treatment withdrawal (any reason); (c) adverse effects; (d) quality of life; (e) cognition; (f) mood. Primary analyses were intention‐to‐treat. Sensitivity best and worst case analyses were also undertaken to account for missing outcome data. Pooled Risk Ratios (RR) with 95% confidence intervals (95% Cl) were estimated for the primary outcomes of seizure frequency and treatment withdrawal. For adverse effects, pooled RRs and 99% CI's were calculated.
Main results
Five trials recruited a total of 439 participants and between them compared different types of VNS stimulation therapy. Baseline phase ranged from 4 to 12 weeks and double‐blind treatment phases from 12 to 20 weeks in the five trials. Overall, two studies were rated as having a low risk of bias and three had an unclear risk of bias due to lack of reported information around study design. Effective blinding of studies of VNS is difficult due to the frequency of stimulation‐related side effects such as voice alteration; this may limit the validity of the observed treatment effects. Four trials compared high frequency stimulation to low frequency stimulation and were included in quantitative syntheses (meta‐analyses).The overall risk ratio (95% CI) for 50% or greater reduction in seizure frequency across all studies was 1.73 (1.13 to 2.64) showing that high frequency VNS was over one and a half times more effective than low frequency VNS. For this outcome, we rated the evidence as being moderate in quality due to incomplete outcome data in one included study; however results did not vary substantially and remained statistically significant for both the best and worst case scenarios. The risk ratio (RR) for treatment withdrawal was 2.56 (0.51 to 12.71), however evidence for this outcome was rated as low quality due to imprecision of the result and incomplete outcome data in one included study. The RR of adverse effects were as follows: (a) voice alteration and hoarseness 2.17 (99% CI 1.49 to 3.17); (b) cough 1.09 (99% CI 0.74 to 1.62); (c) dyspnea 2.45 (99% CI 1.07 to 5.60); (d) pain 1.01 (99% CI 0.60 to 1.68); (e) paresthesia 0.78 (99% CI 0.39 to 1.53); (f) nausea 0.89 (99% CI 0.42 to 1.90); (g) headache 0.90 (99% CI 0.48 to 1.69); evidence of adverse effects was rated as moderate to low quality due to imprecision of the result and/or incomplete outcome data in one included study. No important heterogeneity between studies was found for any of the outcomes.
Authors' conclusions
VNS for partial seizures appears to be an effective and well tolerated treatment in 439 included participants from five trials. Results of the overall efficacy analysis show that VNS stimulation using the high stimulation paradigm was significantly better than low stimulation in reducing frequency of seizures. Results for the outcome "withdrawal of allocated treatment" suggest that VNS is well tolerated as withdrawals were rare. No significant difference was found in withdrawal rates between the high and low stimulation groups, however limited information was available from the evidence included in this review so important differences between high and low stimulation cannot be excluded . Adverse effects associated with implantation and stimulation were primarily hoarseness, cough, dyspnea, pain, paresthesia, nausea and headache, with hoarseness and dyspnea more likely to occur on high stimulation than low stimulation. However, the evidence on these outcomes is limited and of moderate to low quality. Further high quality research is needed to fully evaluate the efficacy and tolerability of VNS for drug resistant partial seizures.
PICOs
Plain language summary
Vagus nerve stimulation for partial seizures
Epilepsy is a disorder where unexpected electrical discharges from the brain cause seizures. Most seizures can be controlled by a single antiepileptic drug but sometimes seizures do not respond to drugs. Some people require more than one antiepileptic medication to control their seizures, especially if these originate from one area of the brain (partial epilepsy), instead of involving the whole brain.
The vagus nerve stimulator (VNS) is a device that is useful as an additional treatment for epilepsy that does not respond well to drugs, and only affects one part of the brain. The device is connected to the vagus nerve in the neck, and sends mild electrical impulses to it. This is particularly important for treating people whose epilepsy does not respond well to drugs and who are not eligible for epilepsy surgery, or in whom surgery was not successful in reducing frequency of seizures. The aim of this systematic review was to look at the current evidence on how effective vagus nerve stimulation is in reducing frequency of epileptic seizures and any side effects associate with using the device.
Overall the five multi‐centre randomised controlled trials (RCTs) recruited a total of 439 participants and between them compared different types of VNS therapy. Overall there were three randomised controlled trials which compared high frequency stimulation to low frequency stimulation in participants aged 12‐60 years and another trial examined high frequency stimulation versus low frequency stimulation in children. Additionally, one trial examined three different stimulation frequencies.
The review of these trials found that vagus nerve stimulation is effective, when used with one or more antiepileptic drugs, to reduce the number of seizures for people whose epilepsy does not respond to drugs alone. Common side effects were voice alteration and hoarseness, pain, shortness of breath, cough, feeling sickly, tingling sensation, headache or infection at the site of the operation, with shortness of breath, voice alteration and hoarseness more common in people receiving high frequency stimulation compared to people receiving low level stimulation.
Out of the five included studies, two studies were rated individually as being of high quality and the other three studies were rated as being of unclear quality due to lack of reported information in the study paper about the methods of study design. The evidence for the effectiveness and side effects of VNS therapy was limited and imprecise from the small number of studies included in this review, so was rated as being of moderate to low quality. Further large, high quality studies are required to provide more information about the effectiveness and side effects of VNS therapy.
Authors' conclusions
Summary of findings
| High versus low stimulation for partial seizures | ||||||
| Patient or population: patients with partial seizures | ||||||
| Outcomes | Illustrative comparative risks* (95% CI) For adverse effects (99% CI) | Relative effect For adverse effects (99% CI) | No of Participants | Quality of the evidence | Comments | |
| Assumed risk | Corresponding risk | |||||
| Low stimulation (control) | High stimulation | |||||
| 50% reduction in seizure frequency (responders) | 144 per 1000 | 249 per 1000 | RR 1.73 | 373 | ⊕⊕⊕⊝ | RR>1 indicates outcome is more likely on High Stimulation |
| Withdrawals | 10 per 1000 | 26 per 1000 | RR 2.56 | 375 | ⊕⊕⊝⊝ | RR>1 indicates outcome is more likely on High Stimulation |
| Voice Alteration or Hoarseness | 251 per 1000 | 545 per 1000 | RR 2.17 | 330 | ⊕⊕⊕⊝ | RR>1 indicates outcome is more likely on High Stimulation |
| Cough | 291 per 1000 | 317 per 1000 | RR 1.09 | 334 | ⊕⊕⊕⊝ | RR>1 indicates outcome is more likely on High Stimulation |
| Dyspnea | 74 per 1000 | 181 per 1000 | RR 2.45 | 312 | ⊕⊕⊝⊝ | RR>1 indicates outcome is more likely on High Stimulation |
| Pain | 239 per 1000 | 241 per 1000 | RR 1.01 | 312 | ⊕⊕⊕⊝ | RR>1 indicates outcome is more likely on High Stimulation |
| Paresthesias | 172 per 1000 | 134 per 1000 | RR 0.78 | 312 | ⊕⊕⊕⊝ | RR>1 indicates outcome is more likely on High Stimulation |
| *The basis for the assumed risk (e.g. the median control group risk across studies) is provided in footnotes.4 The corresponding risk (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). | ||||||
| GRADE Working Group grades of evidence | ||||||
| 1 One study (VNS Study Group 1995) that contributed to this outcome was judged to be at high risk of bias, as it had incomplete outcome data, which could not be analysed by an intention to treat approach (see Characteristics of included studies for further information). 2Wide, imprecise confidence interval of the pooled effect estimate due to low withdrawal rates in the included studies. 3Wide, imprecise confidence interval of the pooled effect estimate due to low event rates in the included studies 4 Assumed Risk: The event rate in the low stimulation group multiplied by 1000. The event rate is the proportion of the total in which the event occurred. | ||||||
Background
This review is an update of a previously published review in The Cochrane Database of Systematic Reviews (The Cochrane Library, Issue 7, 2010).
Description of the condition
Epilepsy is a condition characterized by a tendency for recurrent seizures unprovoked by any known proximate insult. Epileptiform discharges involve either a localized area of the brain resulting in a partial seizure, or the entire brain resulting in a generalized seizure. The prevalence of epilepsy is estimated to be five to eight per 1000 population in developed countries, and in adults the most common type is partial epilepsy (Hauser 1975; Forsgren 2005). The majority of people given a diagnosis of epilepsy have a good prognosis, and their seizures will be controlled by treatment with a single antiepileptic drug (AED). However, 20% (reported in population‐based studies) to 30% (reported in clinical (non‐population‐based) series) will develop drug‐resistant epilepsy (Cockerell 1995; Kwan 2000), often requiring treatment with combinations of antiepileptic drugs (AEDs). These patients tend to have frequent, disabling seizures that limit their ability to work and participate in activities. Many of these patients also suffer from the chronic effects of long‐term, high‐dose AED polytherapy, while anxiety and depressive disorders are common in patients with epilepsy.The development of effective new therapies for the treatment of refractory seizures is therefore of considerable importance.
Description of the intervention
Vagus nerve stimulation (VNS) is a neuromodulatory treatment that is used as an adjunctive therapy for people with medically refractory epilepsy who are not eligible for epilepsy surgery or in whom surgery has failed. In this procedure, a pacemaker‐like device ‐ the Neuro‐cybernetic Prosthesis (NCP) ‐ is implanted under the skin of the chest. The stimulating electrodes of the NCP carry electrical signals from the generator to the left vagus nerve. By programming the device, the frequency, intensity, and duration of stimulation can be varied (the stimulation paradigm). In the initial trials, the vagus nerve was stimulated for 30 seconds, every five minutes (Sackeim 2001). During each 30‐second stimulation, the device delivered 500 microsecond pulses at 30 Hz frequency. For each individual, the intensity of the current was set at the highest that was tolerable, or to low intensity stimulation, depending on the allocated treatment group. Also, in an attempt to further abort seizures, patients could activate the device by placing a magnet over it when a seizure had occurred, or was about to occur. Participants enrolled into the initial randomised controlled trials of VNS had drug‐resistant partial epilepsy, and experienced a 24% to 28% median reduction in seizure frequency over a three month treatment period (Selway 1987).
How the intervention might work
Left VNS is a promising, relatively new treatment for epilepsy. In 1997, VNS was approved in the United States as an adjunctive treatment for medically refractory partial‐onset seizures in adults and adolescents. For some patients with partial‐onset seizures, the adverse effects of antiepileptic drugs (AEDs) are intolerable; for others, no single AED or combination of anticonvulsant agents is effective. Cerebral resective surgery is an alternative to pharmacotherapy in some cases, but many patients with partial‐onset seizures are not optimal candidates for intracranial surgery (Schachter 1998).
The mechanism of action of VNS is not fully understood, but can be reasonably assumed to involve brainstem nuclei. The nucleus of the solitary tract, the main terminus for vagal afferents, has direct or indirect projections to the locus coeruleus, raphe nuclei, reticular formation, and other brainstem nuclei. These nuclei have been shown to influence cerebral seizure susceptibility, hence vagal modulation of one or more of these nuclei could plausibly represent the mechanism for seizure suppression (Krahl 2012). In this context, the immunomodulatory function of the vagus nerve is of particular interest. Afferent signals can activate the so‐called cholinergic anti‐inflammatory pathway upon inflammation. Through this pathway, efferent vagus nerve fibres inhibit the release of pro‐inflammatory cytokines and in this way reduce inflammation. In recent years, inflammation has been strongly implicated in the development of seizures and epilepsy and therefore the activation of the anti‐inflammatory pathway by VNS could decrease the inflammatory response and thereby explain its clinical effects. In addition to anticonvulsive effects, VNS might have positive effects on behavior, mood and cognition (Vonck 2014).
Why it is important to do this review
In this review we summarise evidence from randomised controlled trials where the efficacy and tolerability of VNS for people with drug‐resistant partial epilepsy have been investigated in order to aid clinical decision making when considering VNS treatment within this population.The aim was to evaluate the effects of VNS on seizure frequency and its tolerability and safety when used as an add‐on treatment for people with medically refractory epilepsy.
Objectives
To determine:
(1) The effects on seizures of VNS compared to controls e.g. high‐level stimulation compared to low‐level stimulation (presumed sub‐therapeutic dose); and
(2) The adverse effect profile of VNS compared to controls e.g. high‐level stimulation compared to low‐level stimulation.
Methods
Criteria for considering studies for this review
Types of studies
Trials had to meet all of the following criteria:
-
Randomised controlled trials;
-
Double blind trials;
-
Placebo controlled, active control (low stimulation) or other intervention control group; and
-
Parallel group or crossover studies.
Types of participants
Individuals of any age with partial epilepsy (i.e. experiencing simple partial, complex partial or secondarily generalized tonic‐clonic seizures) who have failed to respond to at least one AED, who were not eligible for surgery or in whom surgery had previously failed.
Types of interventions
-
VNS using high intensity (therapeutic) versus low intensity (presumed sub‐therapeutic) stimulation;
-
VNS stimulation versus different stimulation of VNS;
-
VNS stimulation versus no stimulation;
-
VNS stimulation versus different intervention.
Types of outcome measures
Primary outcomes
50% or greater reduction in seizure frequency
The primary outcome is the proportion of participants with a 50% or greater reduction in seizure frequency in the treatment period compared to the pre‐randomisation baseline period.
Secondary outcomes
Treatment withdrawal
The proportion of people having their allocated VNS paradigm stopped or altered during the course of the trial for whatever reason was used as a measure of "global effectiveness." Treatment is likely to be withdrawn due to adverse effects, lack of efficacy or a combination of both, and this is an outcome to which the individual makes a direct contribution.
Adverse effects
We reported the incidence of adverse events in all VNS implanted patients and according to randomised group. We chose to investigate the following side effects which are most common and important:
-
Infection at implantation site.
-
Haemorrhage at implantation site.
-
Voice alteration or hoarseness.
-
Pain.
-
Dyspnea.
-
Cough.
-
Ataxia.
-
Dizziness.
-
Paresthesias.
-
Fatigue.
-
Nausea.
-
Somnolence.
-
Headache.
In addition, we reported the five most common adverse effects (if different from those stated above).
Quality of life
The difference between intervention and control group(s) means on quality of life measures used in the individual studies.
Cognition
The difference between intervention and control group(s) means on cognitive assessments used in the individual studies.
Mood
The difference between intervention and control group(s) means on mood assessments used in the individual studies.
Search methods for identification of studies
Electronic searches
We searched the Cochrane Epilepsy Group's Specialized Register (23/02/2015) see Appendix 1. In addition we searched the following databases:
(a) The Cochrane Central Register of Controlled Trials (CENTRAL, The Cochrane Library, 23 February 2015) using the strategy outlined in Appendix 2;
(b) MEDLINE (Ovid) (1946 to 23/02/2015) using the strategy outlined in Appendix 3;
(c) SCOPUS (1823 to 23/02/2015) see Appendix 4;
(d) ClinicalTrials.gov 23/02/2015, see Appendix 5; and
(e) WHO International Clinical Trials Registry Platform ICTRP 23/02/2015 see Appendix 6.
No language restrictions were imposed.
Searching other resources
We reviewed reference lists of included studies to search for additional reports of relevant studies.
Data collection and analysis
Selection of studies
For the update, two review authors (MP and AR) independently assessed trials for inclusion. Any disagreements were resolved by discussion with a third author (JW). Three reviewers (MP, AR and JW) extracted data and assessed the risk of bias; disagreements were again resolved by discussion.
Data extraction and management
The following data were extracted for each trial using a data extraction form:
(1) Methodological/trial design:
(a) Method of randomisation.
(b) Method of allocation concealment.
(c) Method of double blinding.
(d) Whether any participants had been excluded from reported analyses.
(e) Duration of baseline period.
(e) Duration of treatment period.
(f ) Frequency of VNS tested.
(g) Information on Sponsorship/Funding.
(2) Participant/demographic information:
(a) Total number of participants allocated to each treatment group;
(b) Age/sex;
(c) Number with partial/generalized epilepsy;
(d) Seizure types;
(e) Seizure frequency during the baseline period;
(f) Number of background drugs.
(3) Outcomes
We recorded the number of participants experiencing each outcome (see Types of outcome measures) per randomised group and contacted authors of trials were for any missing information.
Assessment of risk of bias in included studies
MP and AR independently made an assessment of the risk of bias for each trial using the Cochrane 'Risk of bias' table as described in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011). Any disagreements were discussed and resolved. We rated all included studies as having a(n) low, high or unclear risk of bias on six domains applicable to randomised controlled trials: randomisation method, allocation concealment, blinding methods, incomplete outcome data, selective outcome reporting, and other sources of bias. We created summary of findings tables, employing the GRADE approach for assessing quality of evidence.
Measures of treatment effect
We analysed the primary outcome of seizure reduction as a binary outcome and presented it as a risk ratio. We also analysed secondary outcomes including adverse effects and treatment withdrawal as binary outcomes and presented risk ratios. We planned to analyse quality of life and cognition as continuous outcomes, presented using the standardised mean difference, but this was not possible due to limited information from single studies for cognition outcomes and heterogenous measurement scales for quality of life outcomes. Therefore these outcomes are discussed narratively.
Unit of analysis issues
We did not encounter any unit of analysis issues, as we did not find any crossover studies; we analysed all outcomes as risk ratios as planned or discussed them narratively.
Dealing with missing data
We sought missing data by contacting the study authors. We carried out intention‐to‐treat (ITT), best case and worst case analysis on the primary outcome to account for any missing data (see Data synthesis). All analyses are presented in the main report.
Assessment of heterogeneity
We assessed clinical heterogeneity by comparing the distribution of important individual participant factors among trials (for example age, seizure type, duration of epilepsy, number of AEDs taken at the time of randomisation) and trial factors (for example randomisation concealment, blinding, losses to follow‐up). We examined statistical heterogeneity using a Chi2 test and I2 statistic; providing no significant heterogeneity was present (P > 0.10), we employed a fixed‐effect model. In the event heterogeneity was found (> 50%), a random‐effects model analysis was planned.
Assessment of reporting biases
We requested protocols from study authors for all included studies to enable a comparison of outcomes of interest. If outcome reporting bias was suspected for any included study, we planned to further investigate using the ORBIT matrix system (Kirkham 2010). Examination of asymmetry funnel plots was planned to establish publication bias, but such an assessment was not possible due to the small number of studies included in the review.
Data synthesis
We employed a fixed‐effect model meta‐analysis to synthesise the data. Comparisons we expected to carry out included:
-
intervention group versus controls on seizure reduction;
-
intervention group versus controls on treatment withdrawal;
-
intervention group versus controls on adverse effects;
-
intervention group versus controls on quality of life;
-
intervention group versus controls on cognition;
-
intervention group versus controls on mood.
Each comparison was to be stratified by type of control group, that is level of stimulation, and study characteristics to ensure the appropriate combination of study data. Our preferred estimator for all binary outcomes was the Mantel‐Haenzsel risk ratio (RR). For the outcomes 50% or greater reduction in seizure frequency and treatment withdrawal, we used 95% confidence intervals (Cls). For individual adverse effects we used 99% Cls to make an allowance for multiple testing. Our analyses included all participants in the treatment groups to which they had been allocated following transplantation. For the efficacy outcome (50% or greater reduction in seizure frequency) we undertook three analyses:
(1) Primary (ITT) analysis
Participants not completing follow up or with inadequate seizure data were assumed non‐responders. To test the effect of this assumption, we undertook the following sensitivity analyses. Analysis by ITT was done where this was reported by the included studies.
(2) Worst case analysis
Participants not completing follow up or with inadequate seizure data were assumed to be non‐responders in the high‐level stimulation group, and responders in the low‐level stimulation group.
(3) Best case analysis
Participants not completing follow up or with inadequate seizure data were assumed responders in the high‐level stimulation group, and non‐responders in the low‐level stimulation group.
Subgroup analysis and investigation of heterogeneity
We performed a subgroup analysis for adverse effects. We intended to investigate heterogeneity using sensitivity analysis if deemed appropriate.
Sensitivity analysis
We also intended to carry out sensitivity analysis if peculiarities were found between study quality, characteristics of participants, interventions and outcomes.
Results
Description of studies
Results of the search
The search (carried out 19 September 2013) revealed 749 records identified from the databases outlined in Electronic searches. We screened the 486 records that remained after duplicates were removed for inclusion in the review. We excluded 438 at this point leaving 48 full‐text articles to be assessed for eligibility. Following this, we excluded 23 (see Figure 1 and Characteristics of excluded studies for reasons of exclusion). We included a total of five studies in the review, four of which were included in meta‐analyses (Handforth 1998; Klinkenberg 2012; Michael 1993; VNS Study Group 1995). DeGiorgio 2005 was not included in the meta‐analysis of high versus low stimulation, because this trial compared three different duty cycles paradigms; results of this trial are described narratively. We identified five ongoing studies and six abstracts for studies awaiting classification; authors of these studies were contacted for more information, providing their contact details were available.
Study flow diagram (reflecting results of the search carried out on 19 September 2013).
A pre‐publication search carried out on 23 February 2015 identified 67 results; two studies have been identified as potentially relevant (Klinkenberg 2014 and NCT02089243 2014). These will be addressed in the next update.
Included studies
Overall, five randomised controlled trials which recruited a total of 439 participants were included in this review (DeGiorgio 2005; Handforth 1998; Klinkenberg 2012; Michael 1993; VNS Study Group 1995). Trial characteristics are summarised below. For further information on each trial please see Characteristics of included studies.
Three trials compared high stimulation to low stimulation in participants aged 12‐60 years (Handforth 1998, Michael 1993, VNS Study Group 1995) and another trial examined high stimulation versus low stimulation in children (Klinkenberg 2012). One trial examined three different stimulation paradigms (DeGiorgio 2005). In the majority of the trials, participants were eligible to take part in the double‐blind phase if they were found to experience a minimum of six partial seizures per month and were drug‐resistant.
One multi‐centre (USA) parallel trial (DeGiorgio 2005) randomised 64 subjects > 12 years of age to one of three treatment arms, corresponding to rapid, medium, and slow duty‐cycles: group A, seven seconds on and 18 seconds off (n = 19); group B, 30 seconds on and 30 seconds off (n = 19); group C, 30 seconds on and three minutes off (n = 23). The baseline period was four weeks in duration with a treatment period of three months.
Another multi‐centre (USA) parallel trial (Handforth 1998) included 198 subjects aged 13‐60 years and had two treatment arms: intervention, high stimulation (n = 95) and active control group, low stimulation (n = 103). This trial had a baseline period of 12‐16 weeks and a treatment period of 16 weeks. Dodrill 2000 and Amar 1998 are linked to this study.
A recent multi‐centre (Holland) parallel trial (Klinkenberg 2012) investigated children only and consisted of two treatment arms including high stimulation (n = 21) and low stimulation (n = 20). The baseline period was 12 weeks in duration followed by a treatment period of 20 weeks. After the blinded phase all participants underwent a non‐controlled follow‐up, in which they received high stimulation (add‐on phase). Aalbers 2012 is linked to this study.
One multi‐centre (USA and Europe) parallel trial (Michael 1993) had a pre‐randomisation period of 12 weeks and a treatment period of 14 weeks, where 22 adults were randomised to one of two treatment arms: high, therapeutic (n = 10) or low, sub‐therapeutic (n = 12). All patients completed the acute phase of the study and entered the extension phase.
A further multi‐centre (USA, Sweden and Germany) parallel trial (VNS Study Group 1995) randomised 114 patients to one of two treatment arms: high‐level stimulation (n = 54) and low‐level stimulation (n = 60). This trial had a baseline period of 12 weeks and a treatment period of 14 weeks. Patients exiting the study were offered indefinite extension treatment in an open trial. Ben‐Menachem 1994, Ben‐Menachem 1995, Elger 2000, Ramsay 1994, Holder 1992 and Lotvall 1994, are linked to this study.
Excluded studies
We excluded 23 studies for the following reasons: 18 studies were not randomised trials; two studies did not study an eligible population; three studies had been terminated (no results available). For further information on each trial please see Characteristics of excluded studies.
Risk of bias in included studies
See Figure 2 and Figure 3 for a summary of the risk of bias in each included study. Each study was allocated an overall rating for risk of bias: low, high, or unclear. See below for specific domain ratings.
Risk of bias graph: review authors' judgements about each risk of bias item presented as percentages across all included studies.
Risk of bias summary: review authors' judgements about each risk of bias item for each included study.
Allocation
In three trials (Handforth 1998; Klinkenberg 2012; VNS Study Group 1995) we rated the method by which allocation was concealed as having a low risk of bias. Two trials did not provide clear methods and were rated as unclear on this domain (DeGiorgio 2005;Michael 1993). As for the domain of sequence generation, we rated four studies (DeGiorgio 2005; Handforth 1998; Klinkenberg 2012, VNS Study Group 1995) as having a low risk of bias due to using a computer‐generated randomisation schedule or random number tables/random permuted blocks. We rated one study (Michael 1993) as unclear due to a lack of details on the methods used.
In the DeGiorgio 2005 study, randomisation occurred in blocks of six (two for each group), with a unique predetermined randomisation schedule for each site. In the Handforth 1998 trial, randomisation schedule was generated by a statistical consultant. A consultant organisation served as central randomiser using this schedule. Randomisation for each study site was in groups of four participants. In order to randomise an individual, the unblinded programmer at that site telephoned the randomiser to obtain the randomised treatment assignment. The blinded interviewer at each site had no access to the randomiser, nor to the treatment assignment list, which was kept in a locked place by the site programmer. In Klinkenberg 2012 participants were allocated to a treatment condition by one trial nurse using a computer program. Finally, in the VNS Study Group 1995 report, randomisation tables were developed by an independent statistician. Assignments were made by a look‐up table in the software used to program the generator. Participants were randomised in blocks of four at each site. The investigators and sponsor were unaware of the assignments prior to randomisation.
Blinding
In all of the studies blinding was achieved by using identical implants within the different groups. We judged blinding of participants as unclear in two papers (DeGiorgio 2005; Michael 1993), as no details of the method of blinding were provided. We rated the other three studies (Handforth 1998;Klinkenberg 2012;VNS Study Group 1995) as having a low risk of bias on this particular domain, because to assure blinding, at each treatment‐phase visit the device was temporarily turned off while the participant was assessed by the blinded interviewer. An important issue in blinded trials on VNS is the difficulty of effectively blinding the patients given the frequency of stimulation‐related side effects such as voice alteration. This could limit the validity of the observed treatment effects.
Incomplete outcome data
We rated three studies (Handforth 1998; Klinkenberg 2012; Michael 1993) as having a low risk of bias for incomplete outcome data as low amounts of missing data were reported and either intention to treat analysis was employed or there were no concerns of missing data having an effect on the overall outcome estimate.We also rated the DeGiorgio 2005 study as unclear risk of bias as three participants out of 64 exited early from the study and an intention to treat analysis was not employed, however it is unclear if this approach as influenced the results of the study. We rated the VNS Study Group 1995 as having a high risk of bias because in this trial 57 patients were randomized to low stimulation, however three patients allocated to high stimulation had their stimulator programmed for low stimulation in error. These patients were analysed in the low stimulation group rather than the high stimulation group they were randomised to. This is not an intention to treat analysis.
Selective reporting
None of the protocols for the included studies were available, therefore we could not compare a priori methods and outcomes to the published reports. Because of this, we rated all but one study (Klinkenberg 2012) as having an unclear risk of bias. It should be noted, however, that based on the information contained in the publications, there was no suspicion of reporting bias. In Klinkenberg 2012, the secondary outcome of IQ which was described in the methods section was not reported in the results, thus we assessed this study as having a high risk of reporting bias..
Other potential sources of bias
All studies were sponsored by Cyberonics, Inc, Webster (TX), the manufacturers of the device, and were therefore rated as having an unclear risk of bias on this domain.
Effects of interventions
See: Summary of findings for the main comparison High versus low stimulation for partial seizures
1) High versus Low Vagus Nerve Stimulation
50% or greater reduction in seizure frequency
Data from 4 studies contributed to this outcome.
(a) Intention‐to‐treat analysis:
Results of a Chi2 test showed no significant heterogeneity between trial for a response to VNS (Chi2 = 3.67, df = 3, P = 0.30, I2=18%). The overall risk ratio (RR) for a response to high stimulation compared to low stimulation using the fixed‐effect model is 1.73 (95% Cl 1.13 to 2.64, P = 0.01), showing that patients receiving high stimulation are more likely to show a 50% or greater reduction in seizure frequency (See Analysis 1.1.).
(b) Best and worst case scenarios:
No significant heterogeneity was found for these outcomes (worst case: Chi2 = 4.94, df = 3, P = 0.18; best case: Chi2 = 2.13, df = 3, P = 0.55); I2= 0% and 39% respectively. The overall worst case RR (95%) for a response to VNS is 1.61 (95% CI 1.07 to 2.43, P = 0.02) and best case is 1.91 (95% CI 1.27 to 2.89, P = 0.002). For all three analyses, results suggest a statistically significant treatment effect for high stimulation (See Analysis 1.2 and Analysis 1.3)..
Treatment withdrawal
Data from four studies contributed to this outcome. Five participants withdrew from high level stimulation and three participants withdrew from low level stimulation in Handforth 1998 and Klinkenberg 2012 combined, while no participant withdrew from either stimulation paradigm in Michael 1993 and VNS Study Group 1995. A Chi2 test revealed no significant statistical heterogeneity (Chi2 = 0.11, df = 1, P = 0.74). The overall risk ratio (RR) for withdrawal for any reason is 2.56 (95% CI 0.51 to 12.71). We found no significant difference in withdrawal between high and low level stimulation P = 0.2 (See Analysis 1.4).
Adverse effects
The risk ratios (RR) of adverse effects were as follows:
(a) Voice alteration and hoarseness: 2.17 (99% CI 1.49 to 3.17, P = 0.23, I2 = 32%) from three studies recruiting 334 participants; 159 to high level stimulation and 175 to low level stimulation (Handforth 1998; Michael 1993; VNS Study Group 1995). 87 participants reported this adverse effect in the high level stimulation group and 44 in the low level stimulation group (see Analysis 1.5).
(b) Cough: 1.09 (99% CI 0.74 to 1.62, P = 0.54, I2 = 0%) from three studies recruiting 334 participants; 159 to high level stimulation and 175 to low level stimulation (Handforth 1998; Michael 1993; VNS Study Group 1995); 51 participants reported this adverse effect in the high level stimulation group and 51 in the low level stimulation group (see Analysis 1.6).
(c) Dyspnea: 2.45 (99% CI 1.07 to 5.60, P = 0.77, I2 = 0%) from two studies recruiting 312 participants; 149 to high level stimulation and 163 to low level stimulation (Handforth 1998; VNS Study Group 1995); 27 participants reported this adverse effect in the high level stimulation group and 12 in the low level stimulation group (see Analysis 1.7).
(d) Pain: 1.01 (99% CI 0.60 to 1.68, P = 0.57, I2 = 0%) from two studies recruiting 312 participants; 149 to high level stimulation and 163 to low level stimulation (Handforth 1998; VNS Study Group 1995); 36 participants reported this adverse effect in the high level stimulation group and 39 in the low level stimulation group (see Analysis 1.8).
(e) Paresthesia 0.78 (99% CI 0.39 to 1.53, P = 0.36, I2 = 0%) from two studies recruiting 312 participants; 149 to high level stimulation and 163 to low level stimulation (Handforth 1998; VNS Study Group 1995); 20 participants reported this adverse effect in the high level stimulation group and 28 in the low level stimulation group (see Analysis 1.9).
(f) Nausea: 0.89 (99% CI 0.42 to 1.90, P = 0.10, I2 = 64%) from two studies recruiting 312 participants; 149 to high level stimulation and 163 to low level stimulation (Handforth 1998; Michael 1993); 17 participants reported this adverse effect in the high level stimulation group and 21 in the low level stimulation group (see Analysis 1.10).
(g) Headache: 0.90 (99% CI 0.48 to 1.69, P = 0.16, I2 = 50%) from two studies recruiting 220 participants; 105 to high level stimulation and 115 to low level stimulation (Handforth 1998; VNS Study Group 1995); 24 participants reported this adverse effect in the high level stimulation group and 29 in the low level stimulation group (see Analysis 1.11)
One out of 198 participants in Handforth 1998 and two out of 41 participants in Klinkenberg 2012, experienced an infection after implantation and were not randomised. Overall, this shows that voice alteration and hoarseness, and dyspnea are significant adverse effects of high stimulation. Haemorrhage at implantation site, ataxia, dizziness, fatigue and somnolence were not reported in any included studies.
Quality of life (QOL)
Data from two studies are reported narratively; the Dodrill 2000 paper as part of Handforth 1998 and the Holder 1992 paper as part of VNS Study Group 1995 contributed to this outcome.
On the Quality of Life in Epilepsy‐31 (QOLIE‐31), there was no statistically significant differences of interaction effects between high versus low VNS frequency. On the Short Form (36) Health Survey(SF‐36), there was a statistically significant difference between groups on the domains of Physical Function and Social Function, but no difference for all other domains. There were no statistically significant differences between groups on the Medical Outcomes Study (MOS) and Health‐Related Hardiness Scale (H‐RHS). On the Washington Psychosocial Seizure Inventory (WPSI), only financial status was reported to be significantly different between groups (Handforth 1998).
In VNS Study Group 1995, Quality of Life was evaluated according to 'patient,' 'investigator' and 'companion' Global Rating Scales; an evaluation of the patient's condition at each follow up visit compared to baseline measured on a Visual Analogue Scale. Quality of life was significantly improved compared to baseline according to all three Global Rating Scales (patient, investigator and companion) in both the high stimulation and low stimulation groups compared to the low stimulation group (p<0.001 for both groups for all three scales), but no significant difference between stimulation groups was observed in Quality of Life.
In addition, results from these two studies showed that the patients who had at least 50% seizure reduction exhibited signs of slight improvement in QOL variables compared to those patients who did not demonstrate this degree of seizure reduction. Overall, a small number of favourable QOL effects were associated with levels of VNS stimulation that are now typically used clinically.
Cognition
Data from one study (Dodrill 2000 paper as part of Handforth 1998) contributed to this outcome. No statistically significant differences in interaction effects were reported between high versus low frequency VNS on all four measures used: Wonderlic Personnel Test, Digit Cancellation, Stroop Test, Symbol Digit Modalities.
Mood
Data from one study (Elger 2000 paper as part of VNS Study Group 1995) contributed to this outcome. Mood was measured according to the Montgomery–Åsberg Depression Rating Scale (MADRS) at baseline 3 months and 6 months. A MADRS Total score of between 10 and 20 (maximum score 20) indicated mild depressive mood disorder. At baseline, three out of five patients in the low stimulation group and four out of six patients in the high stimulation group had MADRS scores greater than 10, at three months, two out of five patients in the low stimulation group and two out of six patients in the high stimulation group had MADRS scores greater than 10 and at six months, one out of five patients in the low stimulation group and one out of six patients in the high stimulation group had MADRS scores greater than 10. Overall, four out of five patients in the low stimulation group and five out of six patients in the high stimulation group showed decreases in MADRS scores over the study, but no statistically difference in was present between high and low stimulation groups (Mann‐Whitney's test; P < 0.10).
2) Rapid versus Mild versus Slow duty cycle VNS
50% or greater reduction in seizure frequency
Data from just one study (DeGiorgio 2005) contributed to this comparison. The reduction in seizure frequency was 22% for rapid cycle (P = 0.0078), 26% for medium cycle (P = 0.0270), and 29% for slow cycle (P = 0.0004). For all three groups combined, the reduction in seizure frequency was 40%. Between group comparisons revealed no statistically significant differences in seizure frequency (Kruskal Wallis test, p value is not reported). The >50% responder rate was the same for all three groups (six participants in each group achieved 50% or greater reduction in seizure frequency).
Treatment withdrawal
Throughout the study period only three patients withdrew and the reasons for this were: one developed a device infection, one was lost to follow‐up, and one could not tolerate stimulation (rapid cycle). The randomisation assignments of the first two patients are unknown, presumably as this was deemed by the study authors to be irrelevant to the conclusion.
Adverse effects
The percentage of adverse effects in all three groups combined were as follows: post‐operative pain at the generator 21.3%, throat pain and pharyngitis 9.8%, cough 9.8%, voice alteration 4.9%, vocal cord paralysis 1.6%, abdominal pain and diarrhoea 1.6%. Cough and voice alteration were more common among rapid cycle stimulation (26%, versus 5% medium cycle and 9% for slow cycle). Other adverse effects were not sub‐divided into treatment groups.
Discussion
Summary of main results
Five randomised controlled trials which recruited 439 participants were included in this review. All trials were sponsored by Cyberonics, Inc., Webster, TX, USA. In four trials, participants were randomised to high or low VNS and in the remaining trial participants were randomised to duty cycles of rapid, mild or slow VNS.
Results of the overall efficacy analysis show that people receiving high VNS stimulation were 1.73 (95% Cl 1.13 to 2.64) times more likely to have reduced seizures compared to those receiving low stimulation. This effect did not vary substantially and remained statistically significant for both the best and worst case scenarios, accounting for missing outcome data in one study. Results from one study contributed to the comparison of three different duty cycles paradigms (rapid versus mild versus slow), showing no statistically significant differences in seizure frequency. All included studies were of short duration, so no conclusions can be drawn about long term efficacy of VNS.
Results for treatment withdrawal rates of high and low stimulation groups were similar; there was little difference between the numbers of people who dropped out of the high stimulation group (n = 5) compared to the low stimulation group (n = 2). However, due to imprecision of the pooled result (RR 2.56 [95% Cl 0.51 to 12.71]), important differences in withdrawal rates between the stimulation paradigms cannot be ruled out based on current evidence. The most common adverse events were voice alteration and hoarseness, cough, dyspnea, pain, paresthesias, nausea, and headache. For these adverse effects, voice alteration/hoarseness and dyspnea were over twice as likely to occur in people receiving high stimulation. However, there is some uncertainty and imprecision in reported differences in adverse events between groups; there are often wide confidence intervals, making it difficult to draw conclusions. Only three out of sixty‐four subjects exited early in the study comparing rapid versus mild versus slow stimulation (DeGiorgio 2005): one developed a device infection, requiring removal, one could not tolerate stimulation (high stimulation group), and one was lost to follow‐up. This total percentage withdrawal rate of 4.7% suggests that the treatment is well tolerated.
Results from two studies (Handforth 1998; VNS Study Group 1995)contributed to QOL outcome and showed that a small number of favourable QOL effects were associated with VNS stimulation, but no statistically significant differences between high and low level stimulation groups. Data from one study contributed to cognition outcome (Handforth 1998); no statistically significant differences were reported between high versus low frequency VNS on all four measures used. Data from one study contributed to mood outcome (VNS Study Group 1995); the majority of patients showed an improvement in mood on the MADRS scale compared to baseline but no statistically significant differences were reported between high versus low frequency VNS.
Overall completeness and applicability of evidence
Currently there are only five studies which look at VNS for partial seizures, with less than 500 participants in total. The addition of further evidence from future studies may change the results and conclusions of this review.
This review focuses on the use of VNS in drug‐resistant partial seizures. The results cannot be extrapolated to other patient groups like those with generalised epilepsy. The results of this review indicate that VNS is an effective add‐on treatment for drug‐resistant seizures, but we cannot state how VNS compares to other antiepileptic treatments. This is especially true for VNS because it was tested in an active control situation whereas antiepileptic drugs are tested against placebo. Head‐to‐head trials are needed to assess the relative efficacy and tolerability of antiepileptic treatments.
Quality of the evidence
Out of the five included studies, two studies (Handforth 1998; VNS Study Group 1995) were rated individually as having a low risk of bias and the other three studies were rated as having an unclear risk of bias (DeGiorgio 2005; Klinkenberg 2012; Michael 1993), due to lack of methodological detail concerning study design. The GRADE approach was employed to rate the level of evidence per outcome and this is presented in a summary of findings table (see summary of findings Table for the main comparison). For the main outcome of 50% reduction in seizure frequency, the quality of evidence was rated as moderate due to incomplete outcome data from one study contributing to the analysis (VNS Study Group 1995), however results of best case and worst case scenario analysis to account for missing outcome data showed similar results (see Effects of interventions). Tolerability outcomes (withdrawal and adverse effects) were judged as moderate to low quality due to the imprecision of pooled results and incomplete outcome data from one study contributing to the analysis (VNS Study Group 1995).
In three trials, adequate methods of allocation concealment and randomisation were described (Handforth 1998; Klinkenberg 2012; VNS Study Group 1995). In two papers, no details on the methods of blinding of participants and outcome assessors were reported (DeGiorgio 2005; Michael 1993). The issue of blinding in these studies is somewhat more complex than in typical randomised controlled trials of antiepileptic drugs where a placebo tablet can be administered. The different stimulation paradigms raise the possibility that investigators and participants may not have been adequately blinded to the treatment groups because individuals in the control group may been able to detect that the stimulations were very infrequent; this is likely to be an issue in all trials involving a VNS device.
All protocols for the trials were requested from authors, but no protocols were provided to assist in fully assessing the risk of selective outcome reporting, so this remains unclear in most included studies.
Potential biases in the review process
Although all protocols were requested, the time frame in which the majority of the studies were conducted made retrieval of all of these difficult. This could lead to potential bias through omitted information to which we did not have access.
Agreements and disagreements with other studies or reviews
The studies included in this review were essentially active control trials, thus results may be difficult to compare against other meta‐analyses of antiepileptic drugs that were compared against placebo. The magnitude of the risk ratio for high VNS treatment compared to control would tend to be reduced by any anti‐seizure effect of the low stimulation. A higher risk ratio may have been found if high VNS treatment was compared against no stimulation, but the investigators decided that with a low level of stimulation that was thought to be ineffective, participants were less likely to guess which treatment they were receiving.
Another recent review on VNS (Morris 2013) supports the conclusions made in this review with regard to a positive outcome through use of VNS. This review also describes an association between VNS and mood in adult patients, and includes children‐specific analyses.
Study flow diagram (reflecting results of the search carried out on 19 September 2013).
Risk of bias graph: review authors' judgements about each risk of bias item presented as percentages across all included studies.
Risk of bias summary: review authors' judgements about each risk of bias item for each included study.
Comparison 1 High versus low stimulation, Outcome 1 50% responders.
Comparison 1 High versus low stimulation, Outcome 2 50% responders worst case.
Comparison 1 High versus low stimulation, Outcome 3 50% responders best case.
Comparison 1 High versus low stimulation, Outcome 4 Withdrawals.
Comparison 1 High versus low stimulation, Outcome 5 Voice alteration or hoarseness.
Comparison 1 High versus low stimulation, Outcome 6 Cough.
Comparison 1 High versus low stimulation, Outcome 7 Dyspnea.
Comparison 1 High versus low stimulation, Outcome 8 Pain.
Comparison 1 High versus low stimulation, Outcome 9 Paresthesias.
Comparison 1 High versus low stimulation, Outcome 10 Nausea.
Comparison 1 High versus low stimulation, Outcome 11 Headache.
| High versus low stimulation for partial seizures | ||||||
| Patient or population: patients with partial seizures | ||||||
| Outcomes | Illustrative comparative risks* (95% CI) For adverse effects (99% CI) | Relative effect For adverse effects (99% CI) | No of Participants | Quality of the evidence | Comments | |
| Assumed risk | Corresponding risk | |||||
| Low stimulation (control) | High stimulation | |||||
| 50% reduction in seizure frequency (responders) | 144 per 1000 | 249 per 1000 | RR 1.73 | 373 | ⊕⊕⊕⊝ | RR>1 indicates outcome is more likely on High Stimulation |
| Withdrawals | 10 per 1000 | 26 per 1000 | RR 2.56 | 375 | ⊕⊕⊝⊝ | RR>1 indicates outcome is more likely on High Stimulation |
| Voice Alteration or Hoarseness | 251 per 1000 | 545 per 1000 | RR 2.17 | 330 | ⊕⊕⊕⊝ | RR>1 indicates outcome is more likely on High Stimulation |
| Cough | 291 per 1000 | 317 per 1000 | RR 1.09 | 334 | ⊕⊕⊕⊝ | RR>1 indicates outcome is more likely on High Stimulation |
| Dyspnea | 74 per 1000 | 181 per 1000 | RR 2.45 | 312 | ⊕⊕⊝⊝ | RR>1 indicates outcome is more likely on High Stimulation |
| Pain | 239 per 1000 | 241 per 1000 | RR 1.01 | 312 | ⊕⊕⊕⊝ | RR>1 indicates outcome is more likely on High Stimulation |
| Paresthesias | 172 per 1000 | 134 per 1000 | RR 0.78 | 312 | ⊕⊕⊕⊝ | RR>1 indicates outcome is more likely on High Stimulation |
| *The basis for the assumed risk (e.g. the median control group risk across studies) is provided in footnotes.4 The corresponding risk (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). | ||||||
| GRADE Working Group grades of evidence | ||||||
| 1 One study (VNS Study Group 1995) that contributed to this outcome was judged to be at high risk of bias, as it had incomplete outcome data, which could not be analysed by an intention to treat approach (see Characteristics of included studies for further information). 2Wide, imprecise confidence interval of the pooled effect estimate due to low withdrawal rates in the included studies. 3Wide, imprecise confidence interval of the pooled effect estimate due to low event rates in the included studies 4 Assumed Risk: The event rate in the low stimulation group multiplied by 1000. The event rate is the proportion of the total in which the event occurred. | ||||||
| Outcome or subgroup title | No. of studies | No. of participants | Statistical method | Effect size |
| 1 50% responders Show forest plot | 4 | 373 | Risk Ratio (M‐H, Fixed, 95% CI) | 1.73 [1.13, 2.64] |
| 2 50% responders worst case Show forest plot | 4 | 373 | Risk Ratio (M‐H, Fixed, 95% CI) | 1.61 [1.07, 2.43] |
| 3 50% responders best case Show forest plot | 4 | 373 | Risk Ratio (M‐H, Fixed, 95% CI) | 1.91 [1.27, 2.89] |
| 4 Withdrawals Show forest plot | 4 | 375 | Risk Ratio (M‐H, Fixed, 95% CI) | 2.56 [0.51, 12.71] |
| 5 Voice alteration or hoarseness Show forest plot | 3 | 334 | Risk Ratio (M‐H, Fixed, 99% CI) | 2.17 [1.49, 3.17] |
| 6 Cough Show forest plot | 3 | 334 | Risk Ratio (M‐H, Fixed, 99% CI) | 1.09 [0.74, 1.62] |
| 7 Dyspnea Show forest plot | 2 | 312 | Risk Ratio (M‐H, Fixed, 99% CI) | 2.45 [1.07, 5.60] |
| 8 Pain Show forest plot | 2 | 312 | Risk Ratio (M‐H, Fixed, 99% CI) | 1.01 [0.60, 1.68] |
| 9 Paresthesias Show forest plot | 2 | 312 | Risk Ratio (M‐H, Fixed, 99% CI) | 0.78 [0.39, 1.53] |
| 10 Nausea Show forest plot | 2 | 220 | Risk Ratio (M‐H, Fixed, 99% CI) | 0.89 [0.42, 1.90] |
| 11 Headache Show forest plot | 2 | 312 | Risk Ratio (M‐H, Fixed, 99% CI) | 0.90 [0.48, 1.69] |
