Summary of findings
Description of the condition
Epilepsy is a common neurological disorder affecting 0.5% to 1% of the population (Forsgren 2005). More than 30% of all epilepsy patients suffer from uncontrolled seizures or have unacceptable medication-related side effects (Kwan and Brodie 2000). Alternative treatment options are available for patients with refractory seizures. Addition of newly developed antiepileptic drugs to the treatment regime may result in freedom from seizures in this population group. However, the chance of becoming seizure-free with this strategy is limited and estimated to be around 6% when compared to placebo (Beyenburg 2009). Surgery for epilepsy leads to long-term freedom from seizures in approximately 58% of suitable surgery candidates (Engel 2003). For the remainder, few options are left and neurostimulation may provide an alternative treatment (Engel 2003).
Description of the intervention
Both extracranial (vagus nerve stimulation) and intracranial (deep brain stimulation (DBS) and cortical (neocortex and cerebellar cortex) stimulation) neurostimulation have been used as treatments for epilepsy (Boon 2007a). Intracranial stimulation is the direct application of an electrical current to central nervous system structures by means of implanted (DBS) or subdural (cortical stimulation) electrodes connected to an implantable pulse generator.
How the intervention might work
The precise mechanism of action of DBS still needs to be elucidated. Several mechanisms of action have been proposed. By continuous application of current via the electrodes, the targeted brain structures may be (functionally) inhibited. This is done in a reversible manner since the stimulation can be stopped at any time. The effect of the inhibition depends on the targeted structures, thus depending on the location of the implanted electrodes in the brain. Stimulation of electrodes placed in the epileptic onset region (for example the hippocampus) may lead to 'local' inhibition of the hyperexcitable region and to seizure suppression. Stimulation of electrodes placed in key structures responsible for seizure propagation (for example the thalamus) may additionally lead to suppression of seizure spread, based on the connections between the area of stimulation and other parts of the central nervous system. This may provide a likely hypothesis when crucial structures in the epileptogenic networks are involved (Boon 2007a).
Why it is important to do this review
For both deep brain and cortical stimulation, several uncontrolled and unblinded trials with discongruent results and high risk of bias exist. Randomized controlled trials have been performed but not systematically reviewed. Until now, no clear descriptions of the outcomes and side effects have been available. The aim of this systematic review is to give an overview of the current evidence for the use of DBS and cortical stimulation as treatments for refractory epilepsy.
To assess the efficacy, safety and tolerability of deep brain and cortical stimulation for refractory epilepsy based on randomized controlled trials.
Criteria for considering studies for this review
Types of studies
Randomized controlled trials (RCTs) investigating deep brain or cortical stimulation in patients with refractory epilepsy were selected. Blinded as well as unblinded studies were considered for inclusion in this review.
Types of participants
Patients with refractory epilepsy with partial or generalized seizures, or both. Partial seizures are found in a localization-related form of epilepsy in which seizure semiology or findings from investigations disclose a localized origin of the seizures. With generalized seizures the first clinical changes indicate involvement of both hemispheres (ILAE classification). Patients are considered to be refractory if they suffer from uncontrolled seizures despite adequate treatment with at least two first-line antiepileptic drugs (either as monotherapy or in combination) that are appropriate for the epileptic syndrome, or they experience unacceptable medication-related side effects. In adults, at least two years of treatment is recommended before drug-resistant epilepsy can be diagnosed (Kwan 2010; Kwan and Brodie 2009).
Both patients with normal and abnormal magnetic resonance imaging (MRI) were included. Patients who had undergone other treatments besides antiepileptic drugs (for example resective surgery or vagus nerve stimulation) were also included.
Types of interventions
Deep brain stimulation (DBS) (in different intracranial regions) or cortical (neocortex or cerebellar cortex) stimulation. Both treatments could have been compared to a control patient group: 1) receiving sham stimulation, 2) undergoing resective surgery, or 3) being further treated with antiepileptic drugs, depending on the study protocol.
Types of outcome measures
(1) Seizure freedom: the proportion of participants that was free of seizures (complete absence of seizures, comparable with Engel classification class I (Jehi 2008)) during the randomized period, i.e. the phase of the trial during which, according to treatment allocation, one group of patients received the intracranial neurostimulation treatment and the other group the control treatment (in contrast to open-label follow-up periods of the same trials during which (nearly) all patients received the neurostimulation treatment under investigation in an unblinded manner, without any control group). For RCTs with longer randomized phases, subanalyses per three-month epochs were performed (e.g. months one to three, months four to six).
(2) Responder rate: proportion of patients with at least a 50% seizure frequency reduction, compared to the baseline period, throughout the randomized period.
(1) Seizure frequency reduction: percentage reduction in seizure frequency during the randomized phase of the trial compared to baseline. When the needed data were not presented in the respective article, they were calculated (if raw data were present) or the authors were contacted. When necessary to avoid treatment effects > 100%, we directly compared 'on' to 'off' stimulation periods instead of referring to baseline seizure frequency (as for Van Buren 1978, see also Appendix 1).
(2) Adverse events: adverse events occurring throughout the randomized period including surgery-related and device-related adverse events.
(3) Neuropsychological testing: results of neuropsychological testing during or at the end of the randomized period.
(4) Quality of life: results of questionnaires concerning quality of life that were completed during or at the end of the randomized period.
Search methods for identification of studies
We searched the following electronic databases, without any language restrictions:
(1) PubMed (6 August 2013), using the search strategy outlined in Appendix 2;
(2) the Cochrane Epilepsy Group Specialized Register (31 August 2013), which was searched by Alison Beamond and Graham Chan using the search strategy outlined in Appendix 2; and
(3) the Cochrane Central Register of Controlled Trials (CENTRAL) (The Cochrane Library 2013, Issue 7), using the search strategy outlined in Appendix 2.
Searching other resources
We reviewed the reference lists of retrieved studies to search for additional reports of relevant studies.
We contacted authors of relevant trials identified by our search, other researchers in the field, and manufacturers of the devices to identify unpublished or ongoing studies, or studies published in non-English journals.
Data collection and analysis
Selection of studies
Four review authors (Mathieu Sprengers (MS), Kristl Vonck (KV), Evelien Carrette (EC) and Paul Boon (PB)) independently assessed the identified trials for inclusion. Any disagreements were solved by discussion and by involving another review author (Anthony Marson (AM)).
Data extraction and management
Relevant data were extracted into a prespecified data extraction form by two review authors (MS and KV). If additional data were needed, the investigators of the studies were contacted. Disagreements were solved by discussion.
The following data were extracted.
(1) Methodological and trial design:
(a) method of randomization and sequence generation;
(b) method of allocation concealment;
(c) blinding methods (patient, physician, outcome assessor);
(d) information about sponsoring;
(e) whether any participants had been excluded from reported analyses;
(f) duration of period between implantation and start of the treatment period;
(g) duration of treatment period and, in the case of a cross-over design, washout period;
(h) antiepileptic drug (AED) policy.
(2) Participants and demographic information:
(a) number of participants allocated to each treatment group;
(b) age and sex;
(c) information about type of epilepsy and seizures types;
(d) duration of epilepsy;
(e) additional information if applicable and available (intellectual capacities, neuroimaging results).
(a) stimulation target;
(b) output voltage and current;
(c) stimulation frequency;
(d) pulse width;
(e) continuous, intermittent or responsive ('closed-loop') stimulation.
(a) seizure freedom;
(b) responder rate;
(c) seizure frequency reduction;
(d) adverse events;
(e) neuropsychological outcome;
(f) quality of life.
Assessment of risk of bias in included studies
The methodological quality of the studies was independently evaluated by two review authors (MS and KV) according to the guidelines in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011).
- The risk of bias was assessed for each individual study using the Cochrane Collaboration's tool for assessing risk of bias.
- Randomization: only RCTs were included in this review. Studies with inadequate methods of allocation concealment were planned to be excluded.
- Blinding of participants, personnel and outcome assessors: double-blind studies were preferred but single-blind and even unblinded (comparison to resective surgery or antiepileptic drugs) studies were also eligible for inclusion in the review.
- Incomplete outcome data: this was evaluated separately for each study. Studies where losses to follow-up differed significantly between the treatment and control groups were planned to be excluded.
- Selective reporting: this was evaluated separately for each study (selective outcome reporting) and, furthermore, if sufficient studies were identified we planned to explore if there was any evidence of publication bias using funnel plots.
Several studies have reported results that may be consistent with an outlasting effect after intracranial stimulation (Andrade 2006; Lim 2007; McLachlan 2010; Velasco 2007). Such an effect could mask or reduce any treatment effect if seizure frequency in the control group is evaluated after previous stimulation without an adequate washout period. As there is no general consensus concerning this outlasting effect, we judged the risk of bias in such studies as 'uncertain', whereas studies without prior stimulation or with an adequate washout period were classified as 'at low risk of bias'.
Finally, we also made judgements if antiepileptic drugs were changed during the trial as this could also influence observed treatment effects.
Measures of treatment effect
We planned to express results of categorical outcomes as relative risks (RR) with 95% confidence intervals (CI). However, to combine results from parallel group (unpaired data) and cross-over trials (paired data) we used the method described by Curtin 2002, Elbourne 2002 and Stedman 2011. This method makes use of maximum likelihood estimate odds ratios (OR) (Mantel-Haenszel ORs) for parallel trials and marginal Becker-Balagtas ORs (Becker 1993) for cross-over trials. Treatment effects of continuous outcomes were expressed as mean differences with 95% CIs.
Unit of analysis issues
Results from cross-over trials were analysed and incorporated in the meta-analysis as paired data, using the approach proposed by Curtin 2002.
Dealing with missing data
Where data for our chosen outcomes were not provided in trial reports, the original investigators were contacted and further data were requested. If raw data were available, missing outcomes were calculated, if possible (for example seizure frequency reduction).
Assessment of heterogeneity
Clinical heterogeneity was assessed by comparing the clinical and trial characteristics and a judgement was made as to whether significant clinical heterogeneity was present. Statistical inconsistency was assessed by visual inspection of the forest plots and by using the I² statistic and the Chi² test (Q test).
If neither clinical nor statistical heterogeneity were found, results were pooled using a fixed-effect model. We planned to use the Mantel-Haenszel method for dichotomous outcomes and the inverse variance method for continuous outcomes. However, to combine data from parallel and cross-over trials we had to use the generic inverse variance method. This approach also allowed incorporation of treatment effects estimated by regression and other models. As none of the cross-over trials evaluated the effect of stimulation on quality of life, we used the inverse variance method for this continuous outcome. Although quality of life was evaluated using the QOLIE-89 or QOLIE-31 (abbreviated version of QOLIE-89) questionnaires in different trials, we chose the mean difference (MD) approach instead of the standardized mean difference (SMD) approach. Firstly, both questionnaires have the same range and very similar means, standard deviations and minimally clinically important change values in the same population (Cramer 1998; Devinsky 1995; Wiebe 2002). Second, we thought the MD approach would introduce less error then the SMD approach, which attributes differences in standard deviations entirely to differences in measurement scales and ignores real differences in variability among study populations. Finally, unlike the SMD approach, the MD approach allows us to combine final values and change scores. In view of the difficulty in combining neuropsychological data from various studies, we summarized the data for this outcome only qualitatively in the text. The same was true for adverse events, due to their diverse nature.
Subgroup analysis and investigation of heterogeneity
As stimulation of different intracranial structures may not be equally effective, and lead to different adverse events, results were not pooled across different targets but were presented per individual target for reasons of clinical heterogeneity.
If sufficient studies were found, we planned to assess the effect of study quality on the outcome. Because we initially planned to express results of categorical outcomes as RR instead of OR, we also performed a sensitivity analysis using RR as described by Zou 2007. Furthermore, if different strategies could be followed we planned to analyse their consequences in a sensitivity analysis.
Description of studies
Results of the search
See Figure 1 for a flow-diagrammatic summary of the search results. Eighty-eight articles were identified as potentially eligible for inclusion in this review. Sixty-three articles were excluded as they did not meet the eligibility criteria: 53 were not RCTs, eight assessed intracranial stimulation for other purposes (or in another population) than refractory epilepsy, and in two articles the efficacy of another intervention (transcranial direct current stimulation) was evaluated.
|Figure 1. Study flow diagram.|
Four records described three recent studies. One is still recruiting patients (Boon 2007b: hippocampal stimulation), two others have been preliminarily terminated (Chabardes 2005: subthalamic nucleus stimulation; Wiebe 2008: hippocampal stimulation). When the results of the latter are not published, the authors will be contacted and asked to provide their partial results.
Four records mentioned an RCT evaluating the efficacy and safety of DBS of the mammillary bodies and mammillothalamic tracts (van Rijckevorsel 2004). However, up to now the results have not been published. The authors were contacted but have not provided data yet. Further efforts to acquire these data will be undertaken by the first update of this review. Another record is a recently published congress abstract of a single-blind within-subject control study of centromedian thalamic DBS (Valentin 2012). Upon a more detailed full-text article publication, eligibility for inclusion in this review will be assessed.
Sixteen articles describing 10 studies fulfilled the criteria for inclusion in this review. As the results of one of these studies (Velasco 2000) were only presented in a graph (no exact figures), only nine studies were included in the quantitative synthesis (meta-analysis).
Description of the included studies
Nine out of 10 included studies evaluated the safety and efficacy of open-loop (scheduled) stimulation, the remaining studies concerned closed-loop (responsive) stimulation. Stimulation of the ictal onset zone (including the hippocampus (three studies) and the trial about responsive stimulation) as well as of more remote network structures has been studied. The latter included the cerebellar cortex (three studies) and the anterior (one study) and centromedian (two studies) thalamic nucleus.
1. Anterior thalamic nucleus stimulation
Fisher 2010, also known as the SANTE trial, is a parallel group RCT evaluating the efficacy and safety of bilateral anterior thalamic nucleus DBS in 109 patients (age 18 to 65 years) with refractory partial-onset epilepsy (mean duration of epilepsy: 22.3 years, median baseline seizure frequency: 19.5 per month). After one month of postoperative recovery, patients entered a three-month blinded randomized phase during which half of the participants received stimulation and half did not. This was followed by a nine-month open-label period during which all patients received stimulation in an unblinded way and stimulation parameters could be programmed on an individual basis but antiepileptic drugs (AED) were still kept constant. From the 13th month on, AEDs could vary freely ('long-term follow-up'). All outcomes considered for this review were examined.
2. Centromedian thalamic nucleus stimulation
1. Fisher 1992 was a cross-over randomized trial in seven patients (age 16 to 41 years) who were found to be poor candidates for epilepsy surgery, two of them having (multi)focal epilepsy and five generalized epilepsy (2/5 had Lennox-Gestaut syndrome). The patients had been suffering from epilepsy for 14 to 29 years and had a mean monthly baseline seizure frequency of 23.4 seizures. Patients were randomized one to two months postoperatively to first receive either bilateral centromedian thalamic nucleus (two hours per day) or sham stimulation. The two treatment blocks lasted three months with a three-month washout phase between them. After this nine-month randomized and blinded period, all patients were stimulated during the long-term open-label follow-up period. All outcomes considered for this review were studied and reported except for quality of life.
2. Velasco 2000 was a cross-over randomized trial in 13 patients (age 4 to 31 years) with refractory epilepsy for 4 to 33 years (8 with Lennox-Gestaut syndrome and 5 with localization-related epilepsy) and a median baseline seizure frequency of 119 seizures per month. After six to nine months of stimulation in all participants, patients entered a six-month randomized double-blind cross-over protocol. In half of the patients the stimulator was turned off for three months, between month six and nine, the other half underwent the same manoeuvre nine to 12 months postoperatively. Between month 13 and 15 stimulation was restarted in all patients in an unblinded manner. Two of the original 15 patients were explanted before initiation of the randomized double-blind period due to skin erosions. Seizure frequency during the blinded three-month period without stimulation was presented in a graph and compared to the preceding three months (with stimulation). As these three months only coincided with the three-month stimulation 'on' period of the double-blind protocol in half of patients, and furthermore no exact figures were provided, this study could not be included in the meta-analysis but only in the qualitative synthesis.
3. Cerebellar stimulation
1. Van Buren 1978 reported their results of cerebellar stimulation (superior surface of the cerebellum parallel to and about 1 cm from either side of the midline) in five patients (age 18 to 34 years) with refractory epilepsy for 8 to 23 years, with a mean baseline seizure frequency of 5.1 seizures per day. Presumably four had (multi)focal epilepsy and one had generalized epilepsy. Stimulation was initiated as soon as preoperative seizure frequency had resumed after electrode implantation. Over the ensuing 15 to 21 months, patients were hospitalized three or four times for four to six weeks. During these admissions, seizure frequency was evaluated with and without stimulation. This was performed in a blinded as well as an unblinded way. For this review, only the double-blind data were considered (in total 26 days 'on' and 26 days 'off'). As four out of five patients' seizure frequency increased during the trial (with as well as without stimulation), we decided to directly compare seizure frequency during the stimulation 'on' and 'off' periods to avoid treatment effects with > 100% reductions in seizure frequency (see Appendix 1). The analysis expressing treatment effects with regard to baseline seizure frequency was performed as a sensitivity analysis.
2. Wright 1984 was a cross-over randomized trial in 12 patients (age 20 to 38 years) who had had epilepsy for 10 to 32 years. Five patients had only generalized seizures, one only partial seizures, four partial and generalized seizures, and in two patients seizures were difficult to classify (complex partial seizures versus complex absences). The type of epilepsy was not reported. The six-month randomized phase started several months after electrode implantation, after the patient had returned to his preoperative seizure frequency, and consisted of three two-month periods: continuous, contingent (that is patients received only stimulation when the 'seizure button' was depressed (during an aura or seizure) and for two minutes after it was released) and sham stimulation of the upper surface of the cerebellum (electrodes ± 2 cm parasagittally from the midline). As there was no baseline period, the sham stimulation period seizure frequency (mean: 62 seizures per month) served as reference data for the meta-analysis. Apart from quality of life, all outcomes considered for this review were evaluated.
3. Velasco 2005 studied the efficacy and safety of bilateral stimulation of the superomedial surface of the cerebellum in five patients (age 16 to 35 years) with generalized (n = 3) or (multi)focal frontal lobe epilepsy (n = 2) for 11 to 27 years (mean baseline seizure frequency: 14.1 seizures per month). All patients had generalized tonic-clonic seizures and 4/5 had tonic seizures. The three-month parallel-group randomized phase was initiated one month after electrode implantation and was followed by unblinded stimulation in all patients for 21 months. Seizure frequency and adverse events were evaluated.
4. Hippocampal stimulation
1. Tellez-Zenteno 2006 was a multiple cross-over RCT in four patients (age 24 to 37 years) with refractory left medial temporal lobe epilepsy with mesial temporal sclerosis on MRI whose risk of postoperative memory deficits prevented resective surgery. Duration of epilepsy ranged from 16 to 24 years and the mean monthly baseline seizure frequency was between two and four in three participants and 25 in another. Left hippocampal stimulation was compared to sham stimulation in three two-month treatment pairs, each containing one month with and one month without stimulation. All outcomes considered for this review were studied. With regards to quality of life, see Appendix 3.
2. Velasco 2007 reported their results of uni- or bilateral hippocampal stimulation (according to seizure focus) in nine patients (age 14 to 43 years) with intractable temporal lobe epilepsy for 3 to 37 years (mean baseline seizure frequency: 37.9 seizures per month) who were poor surgery candidates. Five had a normal MRI and four had hippocampal sclerosis. Seizure frequency and adverse events were assessed in a double-blind manner during the first postoperative month during which half of the participants received stimulation and half did not. After this randomized one-month period stimulation was turned 'on' in all patients (follow-up: 18 to 84 months).
3. McLachlan 2010 was another study evaluating hippocampal stimulation as a treatment for medically intractable epilepsy in two patients (age 45 to 54 years) with independent bitemporal originating seizures for 15 to 29 years (with 32 and 16 seizures per month, respectively). MRI was normal in one and showed bilateral hippocampal sclerosis in the other patient. A three-month postoperative baseline period was followed by a cross-over protocol which contained three months of bilateral hippocampal stimulation followed by a three-month washout period and three months of sham stimulation (control). All outcomes considered for this review were evaluated except for quality of life.
5. Closed-loop ictal onset zone stimulation
Morrell 2011, also known as the Neuropace study, was a parallel-group RCT in 191 patients (age 18 to 66 years) with intractable partial onset seizures for 2 to 57 years with one (45%) or two (55%) seizure foci. The mean daily baseline seizure frequency was 1.2. After a 12-week baseline period, one or two recording and stimulating depth or subdural cortical strip leads, or both, were surgically placed in the brain according to the seizure focus or foci. A four-week postoperative stabilization period (neurostimulator programmed to sense and record the electrocorticogram; all patients) and a four-week stimulation optimization period (optimization of stimulation parameters; only patients randomized to treatment group) preceded the 12-week blinded evaluation period (BEP) during which, in half of the participants, the seizure focus was stimulated in response to epileptiform electrographic events. This was followed by an open-label evaluation period with stimulation 'on' in all patients. All outcomes considered for this review were evaluated in this trial.
Risk of bias in included studies
Detailed assessments of each risk of bias item for each included study can be found in the risk of bias tables in the section 'Characteristics of included studies'. A summary of the review authors' judgements is shown in Figure 2.
|Figure 2. Risk of bias summary: review authors' judgements about each risk of bias item for each included study.|
Methods for random sequence generation and treatment allocation concealment (selection bias) were often poorly described in the published articles. After personal communication with the authors, however, these were found to be adequate in all trials for which such additional information could be obtained. As some authors could not be contacted or provide any further explanation, there remained some uncertainty about two trials (Tellez-Zenteno 2006; Wright 1984).
All 10 trials were reported to be double-blind RCTs. However, only for eight out of 10 trials the blinding of patients, personnel and outcome assessors was assessed as adequate. Some uncertainty remained with regards to Van Buren 1978. For this RCT (which contained both double-blind and unblinded evaluation periods, see above) it was not reported whether neuropsychological testing was performed during the blinded or unblinded evaluation period and if the sealed notes containing the treatment code for the double-blind evaluation period were double-opaque and by whom they were handled (for more details: see Characteristics of included studies). Finally, although the double-blinding procedure in Velasco 2000 seemed adequate, the authors compared seizure frequency between stimulation 'off' periods (blinded) and the three-month periods preceding these. Only in about 50% of participants these latter periods coincided with blinded stimulation 'on' periods. For the other half, these three months corresponded to unblinded stimulation 'on' periods, which could have resulted in performance or detection bias (the seizure frequency during blinded stimulation 'on' periods could not be obtained from the authors).
Morrell 2011 was the sole study where patients were asked at the end of the BEP if they knew or could guess if they had received 'real' or sham stimulation. This was of particular importance in this trial as stimulation parameters were determined individually after randomization and only in patients allocated to the stimulation group (for more details: see Characteristics of included studies).
Incomplete outcome data
Risk of bias arising from incomplete outcome data was assessed as high for Fisher 1992 only. In this study one of the two patients who improved noticeably with stimulation experienced a marked seizure frequency increase in the washout period and, therefore, was dropped from the blinded protocol whereafter stimulation was successfully reinstalled. As there were only seven patients (two responders) this one patient represented a significant proportion, especially when taking into consideration the reason for dropout and the fact that a paired analysis of outcome data did not allow inclusion of this patient in the (default) meta-analysis.
Evidence suggesting selective reporting was present for a number of trials. Statistical analysis included only a subgroup of patients in Fisher 1992 (only patients with generalized tonic-clonic seizures, not prespecified in the 'Methods' section) or a subset of available data in McLachlan 2010 (median monthly seizure frequency instead of total number of seizures). As raw data were published in the original articles or provided upon our request, this had no influence on the review.
Fisher 2010 did not report on or mention all available outcome measures in the published paper (for example seizure-free days and seizure-free intervals) but only reported that 'changes in additional outcome measures did not show significant differences'. Again, this had no direct consequences for this review as these outcome variables were not taken into consideration.
In various trials results were incompletely reported, however without strong evidence of selective reporting.
- Neuropsychological testing results were often only reported to be non-significant (Fisher 1992; Wright 1984) or were incompletely published (Tellez-Zenteno 2006). However, as: 1) neuropsychological testing yields too abundant data for publication in a journal article (and therefore not entirely reporting them does not necessarily reflect study quality), and 2) we did not attempt to incorporate these results into a meta-analysis but rather described them in a qualitative way; we think this is of less concern for this review.
- Finally, as not all exact figures with regards to adverse events, neuropsychological outcome and quality of life could be reported in Morrell 2011 (too much data), the authors provided us with these data.
Outlasting effect after prior stimulation
Four trials with a parallel-group design (Fisher 2010; Morrell 2011; Velasco 2005; Velasco 2007) and two cross-over trials with a three-month washout period (Fisher 1992; McLachlan 2010) were judged as being at low risk of bias. Two cross-over trials (Tellez-Zenteno 2006; Wright 1984) did not contain any washout period, which could mask or reduce any treatment effect if stimulation had an outlasting effect. This was even more true for Van Buren 1978 and Velasco 2000, two cross-over trials for which the randomized evaluation took place only after 6 to 21 months of stimulation, without any washout period.
Antiepileptic drug (AED) policy
The antiepileptic drug regimen was kept unchanged in all trials but Tellez-Zenteno 2006, in which it was changed in three out of four patients during the trial. Morrell 2011 allowed benzodiazepines for seizure clusters or prolonged seizures but it was unlikely this significantly influenced the reported results.
Effects of interventions
See: Summary of findings for the main comparison Anterior thalamic nucleus stimulation; Summary of findings 2 Centromedian thalamic nucleus stimulation; Summary of findings 3 Cerebellar stimulation; Summary of findings 4 Hippocampal stimulation; Summary of findings 5 Responsive ictal onset zone stimulation
|Figure 3. Forest plot of comparison: 1 Stimulation versus sham stimulation, outcome: 1.1 Seizure freedom.|
|Figure 4. Forest plot of comparison: 1 Stimulation versus sham stimulation, outcome: 1.2 Responder rate.|
1. Anterior thalamic nucleus stimulation
a. Seizure freedom
During the three-month blinded randomized phase of Fisher 2010 1/55 patients in the control group was seizure-free versus 0/54 in the stimulated group (OR 0.33; 95% CI 0.01 to 8.36).
b. Responder rate
Responder rate was not significantly different in the stimulated (29.6%) compared to the control (25.9%) group (OR 1.20; 95% CI 0.52 to 2.80).
c. Seizure frequency reduction
Over the entire blinded randomized period anterior thalamic nucleus stimulation resulted in a significantly (-17.4%; 95% CI -31.2 to -1.0) higher seizure frequency reduction compared to sham stimulation. The authors reported a trend for increasing differences in median monthly seizure frequency reduction over time between the groups (stimulation versus control: month 1: -33.9% versus -25.3%, month 2: -42.1% versus -28.7% and month 3: -40.4% versus -14.5%; the adjusted treatment effects being -10% (P = 0.37), -11% (P = 0.34) and -29% (P = 0.002) respectively).
d. Adverse events
During the BEP, two self-reported adverse events occurred significantly more frequently in the stimulated group compared to the control group: depression (14.8% versus 1.8%; P = 0.02, Fisher's Exact Test) and subjective memory impairment (13.0 versus 1.8%; P = 0.03). On the contrary, there were significantly fewer epilepsy-related injuries (7.4% versus 25.5%; P = 0.01). Differences for other adverse events were not statistically significant and included: confusional state (7.4% versus 0.0%; P = 0.06), anxiety (9.3% versus 1.8%; P = 0.11), paraesthesia (9.3% versus 3.6%; P = 0.27), new or worse partial seizures with secondary generalization (9.3% versus 5.5%; P = 0.48) and new or worse simple (5.6% versus 1.8%; P = 0.36) or complex (9.3% versus 7.3%; P=0.74) partial seizures. One patient experienced 210 complex partial seizures in the three days after turning on the stimulator (baseline seizure frequency of 19 seizures per month), resolving with reprogramming of the stimulator.
Over the entire study period, five asymptomatic haemorrhage events were reported (four after the initial implant procedure, one following a seizure and a fall and remote from the lead tract). All were asymptomatic. Fourteen participants (12.7%) developed implant site infections, either in the stimulator pocket (7.3%), the lead extension tract (5.5%) or at the site of the burr hole (1.8%). There were no parenchymal brain infections. In nine patients this eventually led to (temporary) hardware removal. Leads initially implanted outside the target structure had to be replaced in 8.2% of participants. Implant site pain was reported by 10.9% of participants during the first year of the trial. Five participants experienced status epilepticus during the trial, two of them with stimulation 'on': one during month two of the blinded phase (complex partial status) and one when the stimulator was turned 'on' after the blinded phase (complex partial status, resolving within five days after switching stimulation 'off'). Five participants died during the course of the trial but none of the deaths were judged as device-related. Mortality causes were: SUDEP (n = 2 + 1 before device implantation), and drowning and suicide (probably in relation to recent life events). The SUDEP rate during stimulation (2 SUDEPs over 325 patient-years with stimulation = 6.2 per 1000 patient-years) fell within the range reported in comparable refractory epilepsy populations (2.2 to 10 per 1000 patient-years) (Tellez-Zenteno 2005; Tomson 2008).
e. Neuropsychological outcome
Although self-reported depression and subjective memory impairment occurred significantly more frequently in the stimulated group (see above), changes in neuropsychological test scores for cognition and mood were very similar in the treatment and control groups and were not significantly different. The evaluated items can be found in Characteristics of included studies. Looking at the individual patients, worsening (> 1 standard deviation change (SD)) of Profile of Mood States Depression subscale (POMS-D) was present in 3/8 stimulated participants with self-reported depression. None of the seven patients with subjective memory impairment showed worsening (> 1 SD) of verbal or visual memory scores.
f. Quality of life
Changes from baseline in overall QOLIE-31 scores were comparable for the treatment (+ 2.5) and control (+ 2.8) group. The MD in change score (-0.30) was neither statistically (95% CI -3.50 to 2.90) nor clinically significant (positive is better, improvements of 5 to 11.7 have been defined in the literature (Borghs 2012; Cramer 2004; Wiebe 2002) as being clinically meaningful).
2. Centromedian thalamic nucleus stimulation
a. Seizure freedom
None of the patients in the Fisher 1992 trial (two hours of intermittent stimulation per day) achieved seizure freedom, neither with nor without stimulation (OR 1.00; 95% CI 0.11 to 9.39).
Although one patient was completely seizure-free at the maximum open-label follow-up (minimum follow-up of one year, mean 41.2 months), Velasco 2000 (24 hours of intermittent stimulation per day) did not report on differences in seizure freedom between stimulation 'on' versus 'off' periods in the double-blind protocol performed between month six and month 12 of the trial. However, as mean seizure frequency reductions were very similar in both groups, major differences in seizure freedom seem unlikely.
b. Responder rate
Statistically significant differences in responder rate, favouring either the stimulation or the control group, could not be demonstrated by Fisher 1992 (OR 1.00; 95% CI 0.27 to 3.69). Two patients did experience ≥ 50% seizure frequency reductions with stimulation 'on' compared to baseline, but one of them had a similar reduction without stimulation and the other could not be included in a paired analysis as he was dropped from the blinded protocol due to a seizure frequency increase during the washout period (see also 'Sensitivity analyses').
Eleven out of 13 patients showed ≥ 50% seizure reductions at maximum follow-up in Velasco 2000, but again the authors did not report on differences in responder rates between stimulation 'on' versus 'off' periods. As for seizure freedom, however, important differences in responder rate were improbable as mean seizure frequency reductions were comparable for stimulation 'on' and 'off' periods.
c. Seizure frequency reduction
Paired analysis (thus excluding one patient) revealed a non-significant 7.1% seizure frequency increase during stimulation 'on' compared to stimulation 'off' periods in Fisher 1992 (95% CI -44.1 to 58.2). Successive months of stimulation were not associated with a clear trend for increasing efficacy over time during the three-month stimulation 'on' period.
Velasco 2000 found very similar and statistically not significantly different reductions in seizure frequency during stimulation 'off' periods in the double-blind phase of the trial and the three-month period preceding it (with stimulation 'on'). Graphs showed approximately a mean 75% reduction in total seizure frequency during stimulation 'on' as well as stimulation 'off' periods (P = 0.23).
Some open-label trials have reported that complex partial seizures may be less prone to centromedian thalamic nucleus stimulation (Velasco 1993; Velasco 1995). Excluding patients with only complex partial seizures (n = 1) in a subgroup analysis of Fisher 1992 showed a non-significant -8.9% MD in seizure frequency reduction (95% CI -79.0 to 61.3%). Although, compared to baseline seizure frequency, reductions in generalized tonic-clonic seizures and atypical absences in Velasco 2000 were more pronounced than those found for complex partial seizures, very similar reductions in seizure frequency were found for any seizure type during stimulation 'on' and 'off' periods and statistically significant differences could not be demonstrated (P values being 0.27, 0.29 and 0.72 respectively).
d. Adverse events
Stimulation-related side effects did not occur in Fisher 1992 or Velasco 2000. Fisher 1992 explicitly reported that no single patient had new seizures or worsening of seizures after initiation of stimulation.
However, various patients in both trials experienced some device- or procedure-related adverse events. One patient in Fisher 1992 required repair of the connection to the pulse generator on one side because no stimulation effect was evident at any intensity, either behaviorally or by electroencephalogram (EEG) monitoring. A postimplantation computed tomography (CT) scan in another patient revealed an asymptomatic and minimal haemorrhage in the vicinity of one depth electrode. Skin erosion forced explantation in three patients of the Velasco 2000 trial, including two children (five and six years old) whose stimulators had to be removed before the double-blind protocol took place. Young children seemed particularly vulnerable to skin erosions because of the size of the hardware, which is designed for an adult population.
e. Neuropsychological outcome
Multivariate analysis with repeated measures showed no significant differences in any of the neuropsychological tests between baseline and stimulation 'on' and 'off' periods in Fisher 1992. The cognitive assessment battery can be found in Characteristics of included studies.
f. Quality of life
None of the two studies evaluated the impact of centromedian thalamic stimulation on quality of life.
3. Cerebellar stimulation
a. Seizure freedom
Regardless of stimulation status, seizure freedom could not be achieved in any of the trials evaluating cerebellar stimulation (OR 0.96; 95% CI 0.22 to 4.12).
b. Responder rate
Cerebellar stimulation did not result in a statistically significantly higher responder rate compared to sham stimulation (OR 2.43; 95% CI 0.46 to 12.84). In the treatment groups, there were 1/5 (Van Buren 1978), 1/9 (Wright 1984) and 2/3 (Velasco 2005) responders, whereas sham stimulation was associated with a ≥ 50% reduction in seizure frequency in 1/5, 0/9 and 0/2 patients, respectively.
There were no responders with contingent stimulation in Wright 1984 (OR 1.00; 95% CI 0.12 to 8.64).
c. Seizure frequency reduction
The pooled mean treatment effect was a -12.4% change in seizure frequency in favour of cerebellar stimulation but this effect did not reach statistical significance (95% CI -35.3 to 10.6). Only Velasco 2005 reported enough details to evaluate a possible trend for increasing efficacy over successive months of stimulation. Although the treatment effect was most pronounced in the third month of stimulation (month 1: -54% versus -29%, month 2: -31% versus -14%, month 3: -82% versus -14%), the small number of patients and the observed variability make it premature to draw any conclusions on this issue. Finally, Van Buren 1978 stated that no slow trends toward improvement could be noticed.
Contingent stimulation was not associated with changes in seizure frequency in Wright 1984 (treatment effect +0.9%; 95% CI -23.2 to 24.9%).
d. Adverse events
Stimulation-related side effects were not reported in any of the trials. Psychiatric evaluation after completion of the Wright 1984 trial did not detect adverse psychiatric sequelae as a result of the stimulation trial.
In contrast, device- or procedure-related adverse events were not uncommon. Electrode migration necessitating repeated surgery occurred in 3/12 and 3/5 patients in Wright 1984 and Velasco 2005 respectively. An electrode lead causing pain needed to be repositioned in one patient and a receiver pocket that had burst open had to be resutured in another (Wright 1984). Leakage of cerebrospinal fluid into the subcutaneous apparatus tracts required resuturing in 3/5 patients of Van Buren 1978, and Wright 1984 reported that most patients experienced temporary swelling over one or both receiver sites, presumably due to cerebrospinal fluid accumulation, but that this spontaneously resolved. A subcutaneous seroma had to be drained in one of Velasco's patients. Wound infections could be settled with antibiotics in two patients but required total hardware removal in one patient (Velasco 2005; Wright 1984). Finally, repeated surgery was performed in another two patients due to a defective receiver and abdominal wound erosion (Wright 1984). Taken all together, in every trial about half of the patients required repeated surgery (3/5 in Van Buren 1978, 6/12 in Wright 1984 and 3/5 in Velasco 2005).
e. Neuropsychological outcome
Each patient in Wright 1984 was assessed by a clinical psychologist in every phase of the trial but 'psychometry' could not reveal any major change in any of the patients. More details were provided by Van Buren 1978. Consistent changes in full scale intelligence or memory quotients could not be detected, nor were there any significant changes in subtests (performance and oral intelligence quotient). Comparing 'on' to 'off' stimulation, the test scores of the four individuals they evaluated showed very similar results in two participants, a moderate increase in one patient, and a moderate decrease in another.
f. Quality of life
None of the trials on cerebellar stimulation formally evaluated impact on quality of life. However, Wright 1984 reported that all his patients but one felt better for cerebellar stimulation, thought it had helped them, and wished to continue it after completion of the trial. However, only five patients chose one phase of the trial as being different from the others: two singled out the continuous, one the contingent, and two others the no-stimulation phase. Moreover, only one patient's subjective impression agreed with the authors' assessment and in this patient the no-stimulation period was his best. Finally, one patient reported a reduction of episodes of incontinence with contingent but not continuous stimulation, which beneficially affected his social possibilities.
4. Hippocampal stimulation
a. Seizure freedom
No single patient was seizure-free for the duration of the RCT they had been included in (OR 1.03; 95% CI 0.21 to 5.15).
b. Responder rate
Hippocampal stimulation was not associated with significantly higher responder rates compared to sham stimulation (OR 1.20; 95% CI 0.36 to 4.01). There were no responders in McLachlan 2010, 1/4 patient experienced a ≥ 50% reduction in seizure frequency with as well as without stimulation in Tellez-Zenteno 2006, and Velasco 2007 reported 1/4 responder in the treatment group compared to 0/5 in the control group.
c. Seizure frequency reduction
Hippocampal stimulation significantly reduced seizure frequency with a pooled mean treatment effect of -28.1% (95% CI -34.1 to -22.2). None of the authors provided enough data to allow evaluation for trends of increasing efficacy over time.
d. Adverse events
No adverse events occurred in relation to stimulation and there were no early surgical complications in any of the trials (McLachlan 2010; Tellez-Zenteno 2006; Velasco 2007). However, skin erosion and local infection 24 months after implantation required explantation in 3/9 patients in Velasco 2007.
e. Neuropsychological outcome
Neuropsychological testing in Tellez-Zenteno 2006 could not reveal significant differences between baseline, 'on' and 'off' periods in any of the formal or subjective measures (see Characteristics of included studies for the different tests they performed). Moreover, reported mean scores were exactly or nearly the same for the 'on' and 'off' periods. Of particular interest was a patient who previously had a right temporal lobectomy and whose memory scores were not influenced by left hippocampal stimulation. The Center for Epidemiologic Studies Depression (CES-D) scale could not demonstrate meaningful changes in mood states during baseline (19), 'on' (20) and 'off' (18) stimulation periods.
McLachlan 2010 assessed the objective and subjective memory of their two patients during baseline, 'on', washout and 'off' periods. They found no changes in one participant and contradictory results in the other. This latter patient reported improved subjective memory during the stimulation 'on' period (baseline second, 'off' third to - sixth and 'on' 12th to 13th percentile (pc), higher was better) but formal testing pointed towards worsening of verbal (baseline first, 'off' 14th and 'on' second pc) as well as visuospatial (baseline 21st, 'off' 42nd and 'on' first pc) memory.
f. Quality of life
Only Tellez-Zenteno 2006 evaluated the impact of hippocampal DBS on quality of life. Repeated (once per month) testing in three patients could not demonstrate statistically significant differences between QOLIE-89 scores during baseline (57), 'on' (55) and 'off' (60) periods (treatment effect -5.0; 95% CI -53.3 to 43.3), which was obviously not surprising given the small number of patients. This five-point difference was clinically of borderline significance (positive was better, improvements of 5 to 11.7 have been defined in the literature (Borghs 2012; Cramer 2004; Wiebe 2002) as being clinically meaningful).
5. Closed-loop ictal onset zone stimulation
a. Seizure freedom
There were no statistically significant differences in seizures freedom during the three-month BEP of Morrell 2011, with 2/97 and 0/94 patients being seizure-free in the treatment and control group, respectively (OR 4.95; 95% CI 0.23 to 104.44).
b. Responder rate
With 28.9% of participants experiencing ≥ 50% reductions in seizure frequency in the treatment group compared to 26.6% in the group receiving sham stimulation, stimulation status did not significantly influence responder rates (OR 1.12; 95% CI 0.59 to 2.11).
c. Seizure frequency reduction
Closed-loop stimulation of the ictal onset zone significantly reduced seizure frequency, the treatment effect being -24.9% (95% CI -40.1% to -6.0%). A trend for increasing efficacy over time could be observed during the three-month BEP, with statistically significant reductions in seizure frequency from the second month of stimulation on (treatment versus control group: month 1: -34.2% versus -25.2% (P = 0.28), month 2: -38.1% versus -17.2% (P = 0.016) and month 3: -41.5% versus -9.4% (P = 0.008)).
d. Adverse events
There were no significant differences between the treatment and sham groups in the percentages of patients with mild or serious adverse events (overall or for any type). In fact, with the exception of increased complex partial seizures (treatment versus sham: n = 2 versus n = 2), headache (n = 3 versus n = 1) and incision site infection (n = 2 versus n = 0), each individual type of device-related (definite or uncertain) adverse event occurred in no more than one participant in the treatment group. Two participants had device-related serious adverse events: one patient of the treatment group and another of the control group had one and three events related to a change in seizures respectively.
Intracranial haemorrhage occurred in nine participants (4.7%). The majority of these (7/9) were considered as being serious, but none of the patients had permanent neurologic sequelae. Six of the nine events were postoperative: three epidural haematomas, two intraparenchymal haemorrhages and one subdural haematoma. The other three events were subdural haematomas attributed to seizure-related head trauma. Implant or incision site soft tissue infections occurred in 5.2% and about half of them urged explantation (2.1%). There were no parenchymal brain infections. The most frequently reported adverse events during the first year of the trial were related to the cranial implantation of the pulse generator and included implant site pain (15.7%), headache (10.5%), procedural headache (9.4%) and dysesthesia (6.3%). Six participants died over the entire 340 years of patient experience. Causes were: lymphoma (n = 1), suicide (history of depression, n = 1) and SUDEP (n = 4, 3 had stimulation enabled). The SUDEP rate (4 SUDEPs over 340 patient-years = 11.8 per 1000 patient-years) was slightly higher than those usually reported in refractory epilepsy patients (2.2 to 10 per 1000 patient-years) (Tellez-Zenteno 2005; Tomson 2008). However, the relatively limited number of patient-years (in other studies often > 2000 years) made it premature to draw firm conclusions on this issue. Nevertheless, close monitoring of the SUDEP rate is definitely needed.
e. Neuropsychological outcome
Neuropsychological assessment at the end of the BEP could not reveal any significant differences between the treatment and sham groups in any measure. In addition, there were no adverse changes in mood inventories at the end of the blinded phase of the trial. The neuropsychological and mood assessment batteries can be found in Characteristics of included studies. Self-reported depression occurred in one patient in each group and subjective memory impairment was reported by one participant belonging to the treatment group.
f. Quality of life
Changes from baseline in overall QOLIE-89 scores were comparable for the treatment (+2.04) and control (+2.18) groups. The MD in change score (-0.14) was neither statistically (95% CI -2.88 to 2.60) nor clinically significant (positive was better, improvements of 5 to 11.7 have been defined in the literature (Borghs 2012; Cramer 2004; Wiebe 2002) as being clinically meaningful). These conclusions applied to the overall as well as any subscale QOLIE-89 score.
Expressing treatment effects of dichotomous outcomes as relative risks (RR) instead of odds ratios (OR) did not change our conclusions. For seizure freedom, effect estimators were nearly identical however with slightly smaller CIs. With regards to the responder rate, effect estimators were discretely lower and CIs smaller when using RR.
Empty cells hindered calculation of odds or risk ratios. In these situations, it was customary to add +0.5 to each cell (Deeks 2011). Given the small number of included patients in most trials, we examined if adding +0.25 instead of +0.5 would change our conclusions. In general, this was not the case. Concerning seizure freedom, however, CIs were larger (for all targeted structures, for OR as well as RR) and the treatment effect seemed more pronounced (but with higher uncertainty) for closed-loop stimulation of the ictal onset zone (OR 8.91; 95% CI 0.14 to 560). With regards to the responder rate, treatment effect estimators and CIs were comparable (except perhaps for a higher degree of uncertainty for cerebellar stimulation).
Including only trials with a low risk of bias due to an outlasting effect after prior stimulation (and thus excluding three cross-over trials without washout periods) did not change our conclusions. For cerebellar stimulation only one trial remained (Velasco 2005); and for hippocampal stimulation the following pooled effect estimates were calculated: seizure freedom OR 1.06 (95% CI 0.12 to 9.62), responder rate OR 1.75 (95% CI 0.22 to 14.13) and seizure frequency reduction -28.5% (95% CI -34.6 to -22.4). Risks of other types of bias which could have directly influenced our conclusions were mainly present in the three cross-over trials.
As the two participants in McLachlan 2010 experienced very similar treatment effects, the standard error associated with the MD in seizure frequency in this study was the lowest (3.13) among all trials on hippocampal stimulation. In this way this very small cross-over study (n = 2) substantially influenced the pooled mean treatment effect. As its weight in the standard analysis appeared disproportionally high (94%), we performed a sensitivity analysis using 29.01 (the standard error of Velasco 2007) instead of 3.13 as the standard error for McLachlan 2010. This alternative analysis yielded a similar -28.2% treatment effect, however with a higher degree of uncertainty (95% CI -50.7 to -5.8). Excluding Tellez-Zenteno 2006 (a cross-over trial without washout period) in this latter analysis resulted in a -45.7% treatment effect for hippocampal stimulation (95% CI -85.9 to -5.5).
To avoid treatment effects > 100%, we directly compared 'on' and 'off' stimulation periods for Van Buren 1978 (see Appendix 1). However, taking baseline seizure frequency as the reference also for Van Buren 1978 (responder rate OR 2.40; 95% CI 0.21 to 26.82; seizure frequency reduction -123.5%; 95% CI -280.3 to 33.3) did not change our conclusion regarding the efficacy of cerebellar stimulation (responder rate OR 2.85; 95% CI 0.64 to 12.68; seizure frequency reduction -15.9%; 95% CI -40.3 to 8.5).
Finally, an unpaired analysis of Fisher 1992 ('best case scenario'), including the patient who seemed to benefit from stimulation but whose absence of stimulation 'off' data (see Characteristics of included studies) prevented inclusion in a paired analysis, could not demonstrate a significant responder rate increase (OR 2.00; 95% CI 0.13 to 29.81) or reduction in seizure frequency (-6.6%; 95% CI -93.7 to 80.5), even after exclusion of a patient with only complex partial seizures (OR 2.00; 95% CI 0.13 to 31.98; -20.7% 95% CI -101.6 to 60.2).
More than 30% of all epilepsy patients have pharmacologically refractory epilepsy (Kwan and Brodie 2000). Resective surgery is the first treatment of choice for these patients. However, most patients are not suitable surgical candidates, some are reluctant to undergo brain surgery, and many do not achieve long-term seizure freedom (de Tisi 2011; Engel 2003). Other treatment options include vagus nerve stimulation, following a specific diet (for example a ketogenic diet) and inclusion in trials with newly developed drugs. However, these options yield seizure freedom in only a small minority of patients. Invasive brain stimulation, including deep brain and cortical stimulation, may be an alternative treatment for these patients. Open-label trials have often shown promising but at the same time mixed results, and in addition are at high risk of bias. To increase our understanding of the efficacy and safety of invasive brain stimulation we performed a systematic review of the literature selecting only randomized controlled trials (RCTs).
Summary of main results
We identified nine RCTs which met our eligibility criteria and could be included in the meta-analysis, including one trial on anterior thalamic nucleus DBS for (multi)focal epilepsy (n = 109), one trial on centromedian thalamic DBS for (multi)focal or generalized epilepsy (n = 7; 14 treatment periods due to cross-over design), three trials on cerebellar stimulation for (multi)focal or generalized epilepsy (n = 22; 39 treatment periods), three RCTs on hippocampal DBS for medial temporal lobe epilepsy (n = 15; 21 treatment periods) and one trial on responsive stimulation of the ictal onset zone (one or two epileptogenic regions) (n = 191). In addition, the results of one RCT on centromedian thalamic DBS for (multi)focal or generalized epilepsy (n = 13; 26 treatment periods) were qualitatively described as the unavailability of exact figures prevented inclusion in the meta-analysis. All trials compared stimulation to sham stimulation. For reasons of clinical heterogeneity, we did not combine results across different stimulated targets but pooled data, if applicable, per individual target.
Statistically significant effects on seizure freedom during the BEPs (one to three months) could not be demonstrated for any target. However, the small number of trials and patients cannot exclude the possibility of clinically meaningful improvements for any target. Nevertheless, it should be noticed that across all different trials only three patients were seizure-free for the duration of the BEP. Two of these belonged to the treatment group of the RCT evaluating closed-loop stimulation of the ictal onset zone (OR 4.95; 95% CI 0.23 to 104.4) and another to the sham group of the trial on anterior thalamic nucleus DBS (OR 0.33; 95% CI 0.01 to 8.35).
Besides seizure freedom, the 50% responder rate was our other primary outcome measure. Statistically significant effects on responder rates after one to three months of stimulation could not be observed for any target, but again the wide CIs cannot exclude clinically meaningful changes for either the stimulation or the control group. The fact that ORs were ≥ 1.00 in every single trial and > 1.00 for every target (except for centromedian thalamic DBS: OR 1.00; 95% CI 0.27 to 3.69) do not suggest equivalence. However, apart from cerebellar stimulation (OR 2.43; 95% CI 0.46 to 12.84), the pooled effect estimates seem of little clinical importance for anterior thalamic nucleus DBS (OR 1.20; 95% CI 0.52 to 2.80), hippocampal DBS (OR 1.20; 95% CI 0.36 to 4.01) and responsive ictal onset zone stimulation (OR 1.12; 95% CI 0.59 to 2.11).
Statistically significant seizure frequency reductions were demonstrated for anterior thalamic DBS (-17.4%; 95% CI -32.1 to -1.0), hippocampal DBS (-28.1%; 95% CI -34.1 to -22.2) and responsive ictal onset zone stimulation (-24.9%; 95% CI -40.1 to -6.0). When interpreting these results, one should keep in mind that these effect estimates may be rather conservative due to observed trends for increasing efficacy over time for anterior thalamic DBS (month 1: -10%, month 3: -29%) and responsive ictal onset zone stimulation (month 1: -9%, month 3: -32%) and a possible outlasting effect in the stimulation 'off' period in Tellez-Zenteno 2006, a (high-weighted) cross-over trial on hippocampal DBS without any washout period. Significant reductions could not be demonstrated for cerebellar (-12.4%; 95% CI -35.3 to 10.6%) or centromedian thalamic (+7.1%; 95% -44.1 to 58.2%; no effect in another cross-over trial (Velasco 2000), P = 0.23) stimulation, although the small number of patients and possible carryover effects in stimulation 'off' periods in Velasco 2000 (centromedian thalamic DBS), Van Buren 1978 and Wright 1984 (cerebellar stimulation) preclude more definitive judgements.
Only for anterior thalamic DBS there were statistically significant differences in stimulation-related adverse events. These included (treatment versus control group) depression (14.8% versus 1.8%; P = 0.02), subjective memory impairment (13.8% versus 1.8%; P = 0.03) and epilepsy-related injuries (7.4% versus 25.5%; P = 0.01). In addition, confusional state and anxiety were more frequent, and standard stimulation parameters could be inappropriate and increase seizure frequency in a small minority of patients. For the other targets, stimulation-related adverse events did not occur (centromedian thalamic DBS, cerebellar and hippocampal stimulation) or were very rare and not significantly more prevalent in the treatment group (responsive ictal onset zone stimulation). In general, however, the size of the included studies (in particular those on centromedian thalamic DBS, cerebellar and hippocampal stimulation) is too limited to make more conclusive statements, although responsive ictal onset zone stimulation seems to be well tolerated except perhaps for the SUDEP rate. The SUDEP rate was 2 per 325 (6.2 per 1000) patient-years with stimulation 'on' for anterior thalamic DBS and 4 per 340 (11.8 per 1000) patient-years for responsive ictal onset zone stimulation compared to 2.2 to 10 per 1000 patient-years as usually reported in refractory epilepsy patients (Tellez-Zenteno 2005; Tomson 2008). Although the limited number of patient-years prevent firm conclusions on this issue, close monitoring is certainly indicated for the latter.
The invasive nature of direct brain stimulation treatments resulted in various surgery- or device-related adverse events. In the two largest trials, asymptomatic intracranial haemorrhages were detected postoperatively in 3.1% to 3.7% of participants and implant or incision site infection occurred in 5.2% to 12.7% resulting in hardware removal in 2.1% to 8.2% (Fisher 2010; Morrell 2011). Inadequate stereotactic placement of electrodes needed repeated surgery in 8.2% of patients in Fisher 2010. Electrode migration seems of particular concern for cerebellar stimulation electrodes (n = 6/22). Other adverse events included skin erosions, defective hardware, leakage of cerebrospinal fluid, a lead causing pain and a subcutaneous seroma. Cranial implantation of the neurostimulator in Morrell 2011 was associated with implant site pain (16% in year one), headache (11%), procedural headache (9%) and dysesthesia (6%).
Statistically significant differences in formal neuropsychological testing results could not be demonstrated on the group level for any target. However, only for responsive ictal onset zone stimulation there is reasonable evidence for the absence of adverse neuropsychological sequelae. In contrast, the higher prevalence of depression and subjective memory impairment with anterior thalamic DBS (see above) and the low number of (neuropsychologically tested) participants in studies on centromedian thalamic DBS, cerebellar and hippocampal stimulation urge further research. In this respect, it should be mentioned that one (n = 1/6) patient receiving hippocampal stimulation showed objective worsening of memory scores although he reported a subjective memory improvement. In addition, results were often incompletely published and the content of the neuropsychological test battery was not clear for Wright 1984 (cerebellar stimulation).
Anterior thalamic nucleus DBS and responsive ictal onset zone stimulation do not significantly improve or worsen quality of life after three months of stimulation. With regards to the other targets, only one trial on hippocampal stimulation (n = 3) (Tellez-Zenteno 2006) has formally evaluated quality of life, while in Wright 1984 the patients' impressions on cerebellar stimulation were described. Although no clear and unambiguous impact on quality of life was found, data are too sparse to make any sensible conclusion.
Quality of the evidence
For a more detailed assessment of the quality of the evidence see Summary of findings for the main comparison; Summary of findings 2; Summary of findings 3; Summary of findings 4; Summary of findings 5.
Several factors affect the quality of currently available evidence. Of major importance is the limited number of trials, which in addition mostly have very small sample sizes. Although this holds true for every target, this is of particular concern for centromedian thalamic DBS, cerebellar and hippocampal stimulation. Moreover, neuropsychological testing and assessment of quality of life were only performed in a subset of trials. These limitations make it harder to demonstrate statistical significance of clinically meaningful differences or to exclude the possibility of such improvements when clinically non-meaningful differences are found.
In four cross-over RCTs on cerebellar (n = 2/3), centromedian thalamic (n = 1/2) and hippocampal (n = 1/3) DBS there was no washout period before outcome measures were evaluated during stimulation 'off' periods (Tellez-Zenteno 2006; Van Buren 1978; Velasco 2000; Wright 1984). As some or all patients had previously been stimulated and findings consistent with a carryover effect of invasive neurostimulation have been reported in the literature (Andrade 2006; Lim 2007; McLachlan 2010; Velasco 2007; Vonck 2013) this may mask or reduce possible beneficial or adverse effects of stimulation. In addition, changes in the antiepileptic drug (AED) regimen in 3/4 patients during the trial may further have influenced the results of Tellez-Zenteno 2006 (hippocampal stimulation). A sensitivity analysis excluding those four trials did not change our main conclusions, although this did result in more pronounced estimates of stimulation effects for cerebellar (responder rate OR 8.33; 95% CI 0.22 to 320.4; seizure frequency reduction -36.7%; 95% CI -95.5 to 21.1) and hippocampal stimulation (responder rate OR 1.75; 95% CI 0.22 to 14.1; if also larger standard error for McLachlan 2010 for seizure frequency reduction of -45.7%; 95% CI -85.9 to -5.5). Obviously, in the case of a clear absence of any effect (for example on seizure freedom) the possibility of an outlasting effect in these trials does not complicate interpretation of the results.
The quality of the evidence on centromedian thalamic DBS is very low. Two RCTs were identified in the literature. However, one trial (Velasco 2000) (n = 13) evaluated stimulation 'off' periods after six to nine months of stimulation without any washout period. The trial only studied two outcome measures (seizure frequency reduction and adverse events), compared blinded stimulation 'off' to the three months preceding it (instead of consistently comparing outcomes to blinded stimulation 'on' periods), and the non-reporting of exact figures prevented inclusion in the meta-analysis. In the second trial (Fisher 1992) seven patients received only two hours of stimulation per day and incomplete outcome data could have biased the results.
Risk of bias was present or unclear in various other trials. It was unclear if the neuropsychological outcome in Van Buren 1978 (cerebellar stimulation) was assessed during blinded or unblinded evaluation periods; methods for random sequence generation and allocation concealment were not well described in Tellez-Zenteno 2006 (hippocampal stimulation) and Wright 1984 (cerebellar cortical stimulation), and evidence of selective reporting was present in two other trials (Fisher 2010 for anterior thalamic DBS; McLachlan 2010 for hippocampal DBS), although we think the latter has not greatly affected the results of this review. Some trials also reported their results incompletely (mainly neuropsychological testing results) and without evidence for selective reporting (Fisher 1992 for centromedian thalamic DBS; Tellez-Zenteno 2006 for hippocampal DBS; Wright 1984 for cerebellar cortical stimulation).
As no more than three trials could be identified for each individual target, we were not able to assess the risk of publication bias.
Overall completeness and applicability of evidence
Currently available evidence is far from complete. The completeness and applicability of the evidence are highly dependent on its quality. All factors limiting the quality of the evidence at the same time limit, to a greater or lesser extent, the completeness and applicability of the evidence. In this review this is especially the case for the small number of trials and patients in which deep brain and cortical stimulation have been studied. Furthermore, only a subset of trials have evaluated the impact of stimulation on the neuropsychological outcome (seven out of 10 trials, with varying degree of extensiveness of testing) and on quality of life (only three to four out of 10 trials). More large and well-designed RCTs are definitely needed to demonstrate or exclude benefits and side effects of invasive brain stimulation therapies. This applies to every single target although there are important differences between the different targeted structures. Taken together, evidence is most complete for responsive ictal onset zone stimulation, followed by anterior thalamic DBS, hippocampal DBS, cerebellar cortical stimulation and finally centromedian thalamic DBS. In addition, several other targets have yielded promising results in open-label trials but have not been studied in blinded and randomized conditions (or the results have not been published yet), for example the subthalamic nucleus (Chabardes 2002; Wille 2011), the caudate nucleus (Chkhenkeli 2004) and the motor cortex (Elisevich 2006).
Trials on cerebellar and centromedian thalamic DBS included both patients with (multi)focal epilepsy and patients suffering from generalized epilepsy. In contrast, trials on anterior thalamic DBS, hippocampal DBS and responsive ictal onset zone stimulation recruited only (multi)focal, temporal lobe and focal (one or two epileptogenic regions) epilepsy patients, respectively. Although this makes sense for hippocampal DBS and responsive ictal onset zone stimulation, further studies are needed to determine if anterior thalamic DBS could also be useful for generalized epilepsy patients.
Only Velasco 2000 (centromedian thalamic DBS) recruited a substantial number of minors; 5/13 or 7/15 patients were between four and 15 years old. Authors reported that skin erosion may be of particular concern in children under eight years of age as a result of the relatively large size of the pulse generator and the leads, originally designed for an adult population. Of the other trials, Fisher 1992 (centromedian thalamic DBS), Velasco 2005 (cerebellar stimulation) and Velasco 2007 (hippocampal stimulation) each included one 14 to 16 year old adolescent, whereas in all other trials all patients were adult. Therefore, current evidence is basically limited to adult refractory epilepsy patients. Fisher 2010 (anterior thalamic DBS) only allowed adults with normal mental capacities (intelligence quotient (IQ) > 70). These are important restrictions which should be taken into consideration when evaluating the overall completeness and applicability of current evidence. Furthermore, evidence is limited to stimulation parameters or parameter strategies used in the respective trials and to the RNS® System (NeuroPace, Mountain View, CA) for responsive ictal onset zone stimulation.
Besides the low number of trials and patients, the limited duration of the BEPs (one to three-month stimulation 'on' periods) represents a second major gap in the available evidence. This seems of particular concern for invasive brain stimulation therapies as increasing efficacy over time has been reported during BEPs in some RCTs (Fisher 2010; Morrell 2011), during open-label follow-up after completion of RCTs (Fisher 2010; Morrell 2011; Velasco 2007) and in some small open-label trials (Franzini 2008; Khan 2009). Various RCTs have followed their patients for many months or years after the randomized and blinded phase had been finished and it may be relevant for the reader to cite the results they reported to illustrate the shortcomings of today's evidence. Fisher 2010 (anterior thalamic DBS) reported seizure freedom in 0% at the end of the BEP (n = 54), in 2.0% at the end of the ensuing nine month open-label period (stimulation parameters adjusted on an individual basis, AEDs unchanged) (n = 99) and in 4.5% after two years of follow-up (changes in the AED regimen were allowed) (n = 81). Responder rates were 30%, 43% and 54% respectively, with mean seizure frequency reductions of -40%, -41% and -56%. Fisher 1992 (centromedian thalamic DBS) observed a 50% seizure reduction in 3/7 patients (2/7 during the BEP) after an additional three to 13 months of open-label follow-up (24 hours of stimulation per day), the mean reduction in seizure frequency being -30% (-7% during the BEP). With regards to the same target, Velasco 2000 reported seizure freedom in 1/13 patients (7.7%), a 85% responder rate and a mean 72% seizure frequency reduction at maximum follow-up (12 to 94 months). Velasco 2005 (cerebellar stimulation) showed a 50% improvement in 2/3 patients during the BEP (mean seizure frequency reduction of 56%) and in 4/5 patients after 12 to 24 months follow-up (68% reduction). The most spectacular improvement was found in Velasco 2007 (hippocampal stimulation) who reported seizure freedom in 4/9 patients after 18 months follow-up (0/4 during the BEP), a 50% reduction in all nine patients (1/4 during the BEP) and a mean seizure frequency reduction of -85% (-30% during the BEP). Finally, three-month seizure freedom and 50% responder rate after two years of open-label follow-up (n = 102) in Morrell 2011 (responsive ictal onset zone stimulation) were 7.1% and 46% (mean seizure frequency reduction not reported) compared to 2.1% and 29% respectively during the BEP. Notwithstanding that these open-label data often show very favourable results, we would like to emphasize that at the same time these are at high risk of bias, including but not limited to placebo effects and improvements due to changes in AED or spontaneous evolution of the disease (see also below). Only RCTs with more extensive BEP can unequivocally determine whether and to what extent the efficacy of invasive brain stimulation treatments increases over time. Meanwhile, we reported for each individual study if and to what extent such an increasing efficacy over time was observed during the BEP.
Finally, no trials comparing invasive intracranial neurostimulation treatments to resective surgery or further treatment with AED ('best medical practice') have been published yet.
Potential biases in the review process
We chose to describe the risk of bias present in different trials rather than excluding all trials with some 'acceptable' risk of bias. Given the limited number of RCTs on deep brain and cortical stimulation published in the literature, we thought such an approach would be more useful to the reader than just concluding that more well-designed trials are needed. However, such an approach adds some risk of bias to the review process. This remark holds particularly true for the inclusion of four cross-over trials without any washout period and therefore at (unknown) risk of bias due to an outlasting effect after stimulation. We therefore performed a sensitivity analysis excluding these trials. Although this resulted in a slightly more favourable effect estimate, it did not change the review's main conclusions.
As empty cells hinder calculation of odds ratios (seizure freedom, responder rate), it is customary to add +0.5 to each cell if applicable (Deeks 2011). However, given the small number of patients included in most trials, this approach may have biased our results. A sensitivity analysis adding +0.25 instead of +0.5 did not change our main conclusions but did increase the degree of uncertainty around the effect estimates for seizure freedom.
For cerebellar and hippocampal stimulation, results of BEPs with different durations (one to three months) were pooled. As some reports have suggested increasing efficacy over time this may have lead to an overestimation compared to the one-month treatment effect and an underestimation compared to the three-month treatment effect. We therefore refer to the observed treatment effects as occurring after 'one to three months' of stimulation. In addition, we described in the text if and to what extent increasing efficacy over time was observed during the BEP of each individual trial. For future RCTs with longer BEPs we plan to pool results per three-month epoch (for example month one to three, month four to six).
Agreements and disagreements with other studies or reviews
Although various non-systematic reviews have been published the past years, to our knowledge this is the first systematic review on RCTs studying deep brain and cortical stimulation. The non-systematic reviews also discussed uncontrolled, often unblinded trials. In this regard, it is appealing that these trials have often yielded much more favourable results than RCTs. Besides the placebo effect, several other factors may account for this discrepancy. First of all, RCTs compare real stimulation to sham stimulation whereas in uncontrolled trials baseline seizure frequency is taken for the reference data. Accordingly, seizure frequency reductions due to (temporary) implantation effects (Fisher 2010; Hodaie 2002;Lim 2007; Morrell 2011) and microlesions resulting from electrode insertion (Boëx 2011; Katariwala 2001; Schulze-Bonhage 2010) contribute to the observed treatment effects in uncontrolled trials whereas they do not in RCTs. Second, uncontrolled trials have longer follow-up periods and increasing efficacy over time has been suggested (see above). However, one should realize that medication-induced and spontaneous improvements can be quite impressive on a group level (Neligan 2012; Selwa 2003) and therefore are likely to contribute to the more favourable results obtained in uncontrolled trials. Third, the cross-over design used in four RCTs may undervalue the efficacy of neurostimulation treatments, as discussed above. Finally, further improvements due to optimization of stimulation parameter settings have been reported (Boëx 2011; Vonck 2013; Wille 2011) and uncontrolled trials often use variable parameter settings whereas RCTs have a fixed stimulation protocol. In conclusion, it is likely that several factors overestimate the efficacy of invasive neurostimulation in uncontrolled trials whereas some others may contribute to an underestimation of its full potential in RCTs.
Vagus nerve stimulation is another type of invasive neurostimulation which nowadays has become routinely available in many epilepsy centres worldwide. Although the treatment effects reported in two large RCTs (-12.7% and -18.4%) (Handforth 1998; VNS Study Group 1995) were similar or slightly inferior to those of anterior thalamic DBS (-17.4%), hippocampal DBS (-28.1%) and closed-loop ictal onset zone stimulation (-24.9%), a Cochrane Review on vagus nerve stimulation (including only those two trials) did demonstrate a significant higher responder rate with vagus nerve stimulation using a high stimulation paradigm ('standard stimulation') compared to a low stimulation paradigm ('sham stimulation') (OR 1.93; 95% CI 1.1 to 3.4) (Privitera 2002). As outlined above, we did not find such a significant improvement for any intracranial target.
Implications for practice
Making general recommendations about the practical usefulness of intracranial neurostimulation treatments implies making trade-offs between potential benefits and harms, costs, healthcare resources and alternative treatments such as newly developed drugs, the ketogenic diet, vagus nerve stimulation and epilepsy surgery. We believe such a trade-off should be made on an individual patient basis, differing from country to country, and therefore goes beyond the scope of this review. In this section we will consequently only focus on available evidence on the benefits and harms of intracranial neurostimulation treatments.
Of all potential intracranial targets, only five have been studied in randomized and double-blind conditions so far. The main limitation is the number of trials, which in addition mostly have very small sample sizes and are of short duration. Nevertheless, high-quality evidence is available that three months of anterior thalamic nucleus DBS and responsive ictal onset zone stimulation can reduce seizure frequency in refractory (multi)focal epilepsy patients, whereas moderate-quality evidence shows the same for one to three months of hippocampal DBS in refractory temporal lobe epilepsy patients. However, compared to sham stimulation, the observed improvements were moderate (ranging between 17% and 28%) and there is no evidence for either a clinically or statistically significant impact on seizure freedom, responder rate or quality of life (although anterior thalamic DBS did reduce epilepsy-associated injuries). Given these rather moderate improvements, possible harms should be carefully considered. Anterior thalamic DBS and responsive ictal onset zone stimulation were in general safe and well-tolerated, but 1) anterior thalamic DBS was associated with statistically significant higher incidences of self-reported depression (no group-level changes in objective measures) and subjective memory impairment (no group-level changes in objective measures) besides statistically non-significant increases in anxiety, confusional state and seizure frequency in some patients; and 2) SUDEP rate should be closely monitored in future for responsive ictal onset zone stimulation. Hippocampal DBS seemed safe and relatively well-tolerated in 15 patients but these findings should be confirmed in more larger trials, with particular concern for memory impairment (found in 1/6 neuropsychologically tested patients). Besides stimulation-related side effects, the invasive nature of these treatments resulted in soft tissue infections and asymptomatic intracranial haemorrhages, but no permanent symptomatic sequelae resulting from electrode implantation were reported. Finally, when balancing benefits and risks of the aforementioned treatments one should keep in mind that many of the patients included in the trials on intracranial neurostimulation had previously turned out to be refractory to various other treatments (including AED, resective surgery and vagal nerve stimulation) and most of them probably had no other treatment options.
Besides the three targets mentioned in the previous paragraph, centromedian thalamic nucleus DBS and cerebellar cortical stimulation have been studied in RCTs but no significant effects were found in these small trials, which in addition suffered from various other limitations. In conclusion, there is insufficient evidence to accept or refuse their efficacy or tolerability. No trials comparing intracranial stimulation to 'best medical practice' or surgery have been published yet.
Finally, it is remarkable that non-randomized unblinded trials on intracranial neurostimulation treatments have often reported more favourable results. However, these trials probably overestimate the treatment effect attributable to stimulation. At the same time, some factors may have underestimated the true treatment effect in RCTs, such as the cross-over design, individually suboptimal stimulation parameter settings and the short duration of follow-up. These last statements, however, have not been studied in randomized and double-blind conditions and therefore remain speculative.
Implications for research
Given the limited number of RCTs identified in the literature, more randomized double-blind controlled clinical trials are required to provide evidence on the efficacy and safety of intracranial neurostimulation treatments for refractory epilepsy. These trials should preferably:
Additionally, there is a need for RCTs comparing intracranial neurostimulation treatments to 'best medical practice' (including vagal nerve stimulation); reported trends for increasing efficacy over time should be verified in randomized and if possible double-blind conditions (comparison to 'best medical treatment' could overcome ethical issues); and, finally, more efforts should be made to identify optimal stimulation parameter paradigms, which could be patient-specific.
We thank A Beamond and G Chan for searching the Cochrane Epilepsy Group Specialized Register and Dr M Miatton (Ghent University Hospital) for her valuable assistance in the interpretation of the neuropsychological data. Dr M Sprengers is supported by an “FWO-aspirant” grant (Research Foundation Flanders). Prof Dr K Vonck is supported by a BOF-ZAP grant from Ghent University Hospital. Prof Dr P Boon is supported by grants from FWO-Flanders, grants from BOF, and by the Clinical Epilepsy Grant from Ghent University Hospital.
Data and analyses
- Top of page
- Summary of findings [Explanations]
- Authors' conclusions
- Data and analyses
- Contributions of authors
- Declarations of interest
- Sources of support
- Differences between protocol and review
- Index terms
Appendix 1. Calculation of treatment effects in Van Buren 1978
We illustrate the way we calculated treatment effects for Van Buren 1978 taking patient 2 of their trial as an example. Van Buren 1978 reported 183% seizure frequency increase during the early double-blind stimulation ON period, a 125% increase during the late double-blind stimulation ON period, a 812% increase during the early double-blind stimulation OFF period and finally a 156% increase during the late double-blind stimulation OFF period. This can be formulated as 283%, 225%, 912% and 256% of baseline seizure frequency respectively. Comparing stimulation ON to stimulation OFF periods with regard to baseline seizure frequency would result in a 330% seizure reduction with stimulation ON [(283-912+225-256)% x ½]. As 4 out of 5 patients' seizure frequency increased during the trial (more accurate seizure detection? spontaneous evolution of their disease?), we decided to directly compare stimulation ON to stimulation OFF periods to avoid treatment effects > 100%. For patient 2, this results into 69% (1-[283/912]) and 12% (1-[225/256]) seizure frequency reductions during early and late double-blind evaluations respectively, or a mean 41% ([69+12)% x ½) reduction in seizure frequency across both periods. Responders during stimulation ON periods were defined as subjects experiencing a ≥ 50% seizure frequency reduction with regard to stimulation OFF periods (direct comparison), whereas the inverse definition was used to define responders during stimulation OFF periods.
Appendix 2. Search strategies
1. CENTRAL search strategy
#1 MeSH descriptor Epilepsy explode all trees
#2 MeSH descriptor Seizures explode all trees
#3 epilep* OR seizure* OR convulsion*
#4 (#1 OR #2 OR #3)
#5 MeSH descriptor Deep Brain Stimulation explode all trees
#7 (#5 OR #6)
#8 (#4 AND #7)
2. PubMed search strategy
Our search strategy is based on the Cochrane Highly Sensitive Search Strategy for identifying randomized trials in MEDLINE (sensitivity-maximizing version, 2008 revision; Pubmed format) (Lefebvre 2011).
#1 randomized controlled trial [pt]
#2 controlled clinical trial [pt]
#3 random* [tiab]
#4 placebo [tiab]
#5 sham [tiab]
#6 trial [tiab]
#7 groups [tiab]
#8 blind* [tiab]
#9 (#1 OR #2 OR #3 OR #4 OR #5 OR #6 OR #7 OR #8)
#10 animals [mh] NOT humans [mh]
#11 (#9 NOT #10)
#12 epilepsy [MeSH]
#13 seizures [MeSH]
#14 epileps* OR epilept*
#17 (#12 OR #13 OR #14 OR #15 OR #16)
#18 deep brain stimulation [MeSH]
#19 stimulat* OR stimuli* OR stimulu*
#20 (#18 OR #19)
#21 (#11 AND #17 AND #20)
3. Cochrane Epilepsy Group Specialized Register search strategy
#1 MeSH DESCRIPTOR Deep Brain Stimulation Explode All WITH AE CL CT EC ES HI IS MT MO NU PX ST SN TD UT VE
#2 (cort* OR brain OR thalam* OR hippocamp* OR cerebel* OR cerebr*) NEAR4 stimul*
#3 "transcranial magnetic stimulation" OR rTMS OR "vagus nerve stimulation" OR "vagal nerve stimulation"
#4 (#1 OR #2) NOT #3
Appendix 3. Quality of life in Tellez-Zenteno 2006
Tellez-Zenteno 2006 reported mean QOLIE-89 scores of 57 (standard deviation (SD) 47), 55 (SD 33) and 27 (SD 60) during baseline, stimulation ON and stimulated OFF periods. These scores are based on repeated testing (once per month) in 3 patients, resulting in 9 QOLIE-89 scores in total. Tellez-Zenteno 2006 also reported median QOLIE-89 scores (with corresponding interquartile ranges), being 57 (24 to 90), 64 (30 to 78) and 61 (39 to 80) respectively. Taking into account the total number of QOLIE-89 scores (only 9), the different effect estimators and their corresponding measures of variability, we assume that the authors switched figures for the QOLIE-89 score during the stimulation OFF period, the mean being 60 and 27 representing the standard deviation. Indeed, it is impossible to calculate a mean score of 27 when the median is 61 and the interquartile range (39 to 80), with only 9 measurements in total.
Contributions of authors
Mathieu Sprengers, Paul Boon, Evelien Carrette and Kristl Vonck co-operated in the literature search, data extraction, data analysis and in writing the review. Anthony Marson contributed in the case of disagreements.
Declarations of interest
Medtronic Inc has provided support in terms of free devices for a pilot study and an international multicentre randomized trial of hippocampal deep brain stimulation in epilepsy co-ordinated by Ghent University Hospital.
Sources of support
- Dr. M. Miatton, Belgium.Assistance in the interpretation of the neuropsychological data
- No sources of support supplied
Differences between protocol and review
The title of the review was changed from 'Deep brain and cerebellar stimulation for epilepsy' to 'Deep brain and cortical stimulation for epilepsy' as we thought neocortical stimulation also fits the scope of this review (which may be particularly relevant for future updates of the review).
The percentage seizure frequency reduction was added as an additional outcome measure. This was done in a prespecified way after one author involved in the writing of the protocol (Annelies Van Dycke) was replaced by another author (MS). The reason to do so was to allow a more precise estimation of the efficacy of the different invasive intracranial neurostimulation treatments.
We planned to express the treatment effect for dichotomous outcome measures by relative risk (RR). However, for reasons outlined in the 'Methods' section, we used odds ratios and performed a sensitivity analysis with RRs to evaluate any possible influence of this change.
Medical Subject Headings (MeSH)
Cerebral Cortex; Deep Brain Stimulation [instrumentation; *methods]; Electrodes, Implanted [adverse effects]; Epilepsy [*therapy]; Outcome Assessment (Health Care); Randomized Controlled Trials as Topic
MeSH check words
* Indicates the major publication for the study