Newer anticonvulsants in the treatment of anxiety disorders
Abstract
This is a protocol for a Cochrane Review (Intervention). The objectives are as follows:
1) To use evidence from RCTs in providing an estimate of the overall effects of the newer anticonvulsants in improving treatment response and reducing symptom severity in the treatment of anxiety disorders.
2) To determine whether particular anticonvulsants are more effective and tolerable than others in the treatment of anxiety disorders.
3) To determine whether the particular anxiety disorder treated predicts the effectiveness of anticonvulsants in terms of efficacy and tolerability.
4) To identify which factors (clinical, methodological) predict response to pharmacotherapy.
Background
Description of the condition
Anxiety disorders are associated with a high degree of morbidity and impaired quality of life. Characterised by a state of chronic physiological hyperarousal to fear‐inducing contexts, this group of disorders includes generalized anxiety disorder (GAD), obsessive compulsive disorder (OCD), panic disorder (PD), post‐traumatic stress disorder (PTSD), social anxiety disorder (SAD), and specific phobia (SP). Anxiety disorders are the most prevalent class of psychiatric disorders in the United States, with a nationally representative survey finding that over a quarter of adults have been diagnosed with an anxiety disorder once in their lifetime (Kessler 2005). They also tend to have an earlier age of onset than most other psychiatric disorders. The chronic course of anxiety disorders is reflected in their direct and indirect costs, which were estimated to be as high as 40 to 50 billion dollars in 1990 in the United States alone (Greenberg 1999). A significant proportion of this burden is carried by non‐psychiatric medical services.
Description of the intervention
Psychotherapy and pharmacotherapy have both demonstrated effectiveness in treating anxiety disorders. Cognitive behavioural therapy and the selective serotonin reuptake inhibitors (SSRIs) in particular are recommended by clinical guidelines as first‐line agents in treating this class of disorders (Baldwin 2005). Nevertheless, estimates of the number of patients who do not benefit from standard treatments range from 20 to 40% in PD (Bandelow 2004) to 40 to 60% in OCD (Pallanti 2002). It is therefore important that the potential of alternative medications for treating anxiety disorders be investigated. Anxiolytic effects in animal studies (Lamberty 2003; Mirza 2005), as well as in initial reports of elevated mood and reduced anxiety amongst epileptic patients following treatment with the newer anticonvulsants (Smith 1993) appear to warrant further investigation of the effectiveness of this class of medication in treating anxiety disorders. Evidence that the newer anticonvulsants possess a more beneficial side effect profile than their predecessors (Besag 2001) and are involved in fewer drug interactions (French 2004; LaRoche 2004) provides further support for the possible clinical utility of these agents.
A brief description of each of the newer anticonvulsants, as well as evidence of their anxiolytic efficacy and the side effects associated with their use, is presented below.
Felbamate
Felbamate is a broad spectrum anticonvulsant, and was the first of the new generation of anticonvulsants to receive FDA approval for the treatment of epilepsy (LaRoche 2004). The anxiolytic properties of this medication have received little investigation. An increased risk of aplastic anemia and hepatoxiticy has discouraged the use of felbamate as a first‐line agent in the treatment of epilepsy (Borowicz 2004).
Gabapentin
Gabapentin is a gaba aminobutyric acid (GABA) analogue. It appears to possess a broad mechanism of action, with more than 80% of its prescription for off‐label purposes, such as treating neuropathic pain, migraine headache, spasticity and bipolar disorder (McLean 2001). Although a recent review found little evidence of the efficacy of gabapentin as either a mood‐stabiliser or anti‐epileptic agent (LaRoche 2004), it resulted in significant reduction in the severity of symptoms in the treatment of refractory comorbid anxiety disorders in four patients (Pollack 1998).
Lamotrigine
Lamotrigine is a phenyltriazine compound that is structurally unrelated to other antiepileptic drugs. Lamotrigine has demonstrated efficacy in randomised controlled trials (RCTs) of mood disorders (see Boylan 2002 for a review). One small placebo‐controlled RCT of PTSD found improvement on the reexperiencing and avoidance/numbing symptoms of this disorder following the administration of lamotrigine (Hertzberg 1999a). The most serious adverse effect of treatment with lamotrigine is the development of potentially life threatening rashes, including Stevens‐Johnson syndrome and toxic epidermal necrolysis, in approximately 1 in 1000 adult patients. These rashes have a mortality rate of 5 to 15% (Ghislain 2002).
Levetiracetam
Levetiracetam is an acetylcholine agonist with an unknown mechanism of action. Although an unpublished open‐label trial found levetiracetam to be effective in treating generalised SAD (Van Ameringen 2004a), this finding was not confirmed in a small placebo‐controlled RCT (Zhang 2005a). Levetiracetam resulted in significant improvement in the naturalistic augmentation of medication for 13 out of 23 patients with treatment‐resistant PTSD (Kinrys 2006). Levetiracetam is considered one of the 'nootropic' drugs, with possible enhancing effects on cognition.
Oxcarbazepine
Oxcarbazepine is a 10‐Keto analogue of the older anticonvulsant carbamazepine, with fewer side‐effects resulting from its failure to metabolise into the 10,11 epoxide. Berigan 2002a presented evidence of one patient diagnosed with PTSD with improved functioning following administration of oxcarbazepine.
Pregabalin
Pregabalin is a structural analogue to gabapentin, and is currently undergoing development as an anxiolytic for the treatment of GAD, PD and SAD (Feltner 2003b). There is evidence from RCTs that pregabalin is effective in treating both GAD (Feltner 2003b; Pohl 2005; Rickels 2005a) and social anxiety disorder (Pande 2004). In a pooled analysis of three randomised controlled trials of the efficacy of pregabalin in treating epilepsy, dizziness and somnolence emerged as the most frequently occurring adverse effects (Bialer 2002).
Topiramate
Treatment with the GABA agonist topiramate, resulted in lower levels of anxiety in a case series of PTSD (Berlant 2001), and reduced symptom severity in an open‐label trial of chronic PTSD amongst civilians (Berlant 2004). Topiramate has also demonstrated some efficacy in the open‐label treatment of generalised GAD (Van Ameringen 2004b). A retrospective case study found that treatment with this agent is significantly more likely to lead to psychosis than either gabapentin or lamotrigine (Crawford 1998). This is consistent with the development of psychosis in a patient with OCD following treatment with 200 mg/b.i.d. of topiramate (Kober 2005). The induction of panic attacks in the treatment of bipolar disorder (Damsa 2006), and the development of OCD following initiation of topiramate in treating epilepsy have also been reported (Ozkara 2005). In a review of the effect of antiepileptic drugs on cognitive function amongst epileptic patients, topiramate appeared to be the most troublesome of the newer antiepileptic drugs, at least in the initial treatment phase and if given in high doses (Brunbech 2002).
Tiagabine
Tiagabine is a selective GABA reuptake inhibitor, and was observed in an open label study to be effective across a range of anxiety disorders (Crane 2003). In a case series study tiagabine reduced symptom severity on the HAM‐A for patients diagnosed with panic disorder with and without agoraphobia (Zwanzger 2001). Similar observations were made in an open‐label study of tiagabine in treating PD in seven women (Taylor 2003). Provisional evidence of the efficacy of tiagabine in treating GAD, PTSD and SP have also been documented in open‐label studies (Van Ameringen 2004a). Tiagabine demonstrated effectiveness over a period of eight weeks in the open‐label augmentation of pharmacotherapy for anxiety symptoms (Schwartz 2005).
Vigabatrin
Vigabatran is an inhibitor of GABA transaminase, and its administration has been linked to the occurrence of visual field defects in a third of patients treated (Kalviainen 1999). There is a lack of clinical evidence on the efficacy of vigabatran in treating anxiety disorders.
Zonisamide
Zonisamide is a sulfonamide derivative which is chemically distinct from other anticonvulsants (Leppik 2004). While the efficacy of zonisamide in treating bipolar disorder has been evaluated in open‐label (Anand 2005) and case study trials (Berigan 2000), there is little clinical evidence of its efficacy in treating anxiety disorders.
How the intervention might work
The mechanisms through which anticonvulsants exert their anxiolytic effects are unknown. It is speculated that the inhibition of patterns of spontaneous neuronal firing, or kindling, may account for both the anticonvulsants' ability to reduce seizures and the psychotropic properties that many of them appear to possess (Post 2003). The stimulation of the inhibitory gaba aminobutyric acid (GABA), and the inhibition of its antagonist, guatamine, have been recognised as primary candidates for the neurochemical pathways mediating kindling (Ashton 2003). Such a model is consistent with evidence for the anxyliotic effects of endogenous GABA stimulation by means of the administration of vigabatrin (Zwanzger 2001b) and tiagabine (Zwanzger 2003) on experimentally induced panic. It has also formed the theoretical underpinning of open‐label PTSD trials (Berigan 2002b; Berlant 2001) in which some success in reducing the frequency of re‐experiencing/intrusive symptoms was reported. Finally, the side effects associated with anticonvulsants are consistent with activation of the GABA and guatamine pathways (Ketter 1999).
Why it is important to do this review
A systematic review of the efficacy of anticonvulsants in treating anxiety disorders would help establish the utility of these medications for patients who do not respond to first‐line interventions, as well as provide insight into the status of clinical factors, such as the presence of comorbid depression in predicting outcome. A systematic assessment would also help address the differential efficacy and tolerability of particular anticonvulsants for the treatment of specific anxiety disorders. This review aims to systematically assess the efficacy and tolerability of the newer anticonvulsants in treating anxiety disorders.
Objectives
1) To use evidence from RCTs in providing an estimate of the overall effects of the newer anticonvulsants in improving treatment response and reducing symptom severity in the treatment of anxiety disorders.
2) To determine whether particular anticonvulsants are more effective and tolerable than others in the treatment of anxiety disorders.
3) To determine whether the particular anxiety disorder treated predicts the effectiveness of anticonvulsants in terms of efficacy and tolerability.
4) To identify which factors (clinical, methodological) predict response to pharmacotherapy.
Methods
Criteria for considering studies for this review
Types of studies
All randomised and controlled trials (RCTs) of newer anticonvulsants in the treatment of anxiety disorders. Publication is not necessarily related to study quality and indeed publication may imply certain biases (Easterbrook 1991; Dickersin 1992; Scherer 1994), so unpublished abstracts and reports will also be considered. Trial reports in all languages will be included.
Types of participants
All participants diagnosed as possessing one or more anxiety disorders, including generalised anxiety disorder (GAD), obsessive‐compulsive disorder (OCD), panic disorder (PD), post‐traumatic stress disorder (PTSD), social phobia (SP) and specific phobia (SP), according to the Diagnostic and Statistical Manual (DSM III) (APA 1980) or DSM‐IV (APA 1994), or the International Classification of Diseases (ICD‐9 (WHO 1975) or ICD‐10 (WHO 1992)). No restrictions will be applied in terms of age, gender or race.
Types of interventions
RCTs of anticonvulsants developed from the last half of the 1980s will be included. The control may be either a placebo, an alternative medication, or a no treatment group. Trials of the following medications will be considered for inclusion: felbamate, gabapentin, lamatrogine, levetiracetam, oxcarbazepine, pregabalin, tiagabine, topiramate, zonisamide. Data from multi‐centre trials will also be included.
Types of outcome measures
Studies will be included if they provide post‐treatment medication and placebo treatment response on categorical outcome measures, or symptom severity means/standard deviations on dimensional outcome measures. Established scales that have undergone minor modifications, as determined by the review authors, will be included in the review.
Primary outcomes
1) Treatment response (responders versus non‐responders) will be determined from the Clinical Global Impressions scale ‐ Improvement item (CGI‐I) (or closely related measure), a widely used global outcome measure (Guy 1976). Responders are defined on the CGI‐I as those with a score of 1="very much" or 2="much" improved. The reduction of symptom severity will be determined from a variety of validated continuous outcome measures, such as the Liebowitz Social Anxiety Scale (LSAS) (Liebowitz 1987) and the Clinician Administered PTSD Scale (CAPS) (Blake 1990).
Secondary outcomes
2) The response of comorbid symptoms will be measured by :
(a) Depression scales, such as the Beck Depression Inventory (BDI) (Beck 1961), the Hamilton Depression Scale (HAM‐D) (Hamilton 1959), and the Montgomery‐Asberg Depression Rating Scale (MADRS) (Montgomery 1979).
(b) Anxiety scales, such as the Hamilton Anxiety Scale (HAM‐A) (Hamilton 1960).
3) Quality of life measures.
4) Measures of functional disability, such as the Sheehan Disability Scale (SDS) which includes subscales to assess work, social and family related impairment (Sheehan 1996).
Data from these scales will be included as indicators of medication effectiveness.
5) The total proportion of participants who withdrew from the RCTs due to treatment emergent adverse events will be included in the analysis as a surrogate measure of medication acceptability, in the absence of other more direct indicators of acceptability. In addition, the most common drug‐related adverse events for both the included and excluded studies (defined as those occurring in at least 10% of the participants given medication), as well as significant differences in the rate of occurrence of drug‐related adverse events between medication and control groups will be described as part of the narrative review.
Search methods for identification of studies
Electronic Searches
1) The Cochrane Collaboration Depression, Anxiety and Neurosis Controlled Trials Register (CCDANCTR‐Studies) will be searched with the following search strategy:
CCDANCTR‐Studies
Diagnosis = Anxiety or Anxious or Phobi* or Panic or Obsess* or compulsi* or Post‐Traumatic
and
Intervention = felbamate or gabapentin or lamatrogine or levetiracetam or oxcarbazepine or pregabalin or tiagabine or topiramate or zonisamide
CCDANCTR‐References
Keyword = Anxiety or Anxious or Phobi* or Panic or Obsess* or compulsi* or Post‐Traumatic
and
Free‐Text = felbamate or gabapentin or lamatrogine or levetiracetam or oxcarbazepine or pregabalin or tiagabine or topiramate or zonisamide
2) The Cochrane Central Register of Controlled Trials (CENTRAL) (The Cochrane Library Issue 2 2006) will be searched.
3) Additional searches with be carried out on MEDLINE via PubMed (January 1985 to December 2005), and through PsycINFO (1983 to 2005, Part B).
The complete MEDLINE search query, as derived from the search strategy developed by Robinson 2002, is provided below:
(randomized controlled trial [pt] OR controlled clinical trial [pt] OR randomized controlled trials [mh] OR random allocation [mh] OR double‐blind method [mh] OR single‐blind method [mh] OR clinical trial [pt] OR clinical trials [mh] OR ("clinical trial" [tw]) OR ((singl* [tw] OR doubl* [tw] OR trebl* [tw] OR tripl* [tw]) AND (mask* [tw] OR blind* [tw])) OR ("latin square" [tw]) OR placebos [mh] OR placebo* [tw] OR random* [tw] OR research design [mh:noexp] OR comparative study [mh] OR evaluation studies [mh] OR follow‐up studies [mh] OR prospective studies [mh] OR cross‐over studies [mh] OR control* [tw] OR prospectiv* [tw] OR volunteer* [tw]) NOT (animal [mh] NOT human [mh]) AND (anxiety disorder [mh:noexp] OR "anxiety disorder" [tw] OR phobic disorders [mh:noexp] OR obsessive‐compulsive disorder [mh] OR "obsessive‐compulsive" [tw] OR stress disorders, post‐traumatic [mh:noexp] OR "post‐traumatic" [tw]) AND (anti‐convulsants [tw] OR anticonvulsants [tw] OR anticonvulsants [mh] OR anti‐epileptics [tw] OR antiepileptics [tw] OR felbamate [tw] OR gabapentin [tw] OR lamotrigine [tw] OR levetiracetam [tw] OR oxcarbazepine [tw] OR pregabalin [tw] OR topiramate [tw] OR tiagabine [tw] OR zonisamide [tw])
The PsycINFO search strategy will include the following search query: ("randomisation" OR "randomization") OR "controlled" AND ("anticonvulsants" OR "antiepileptics").
4) Ongoing trials will be located using the metaRegister of Controlled Trials database (mRCT) (http://www.controlled‐trials.com), as well as the National Institute of Health's Computer Retrieval of Information on Scientific Projects (CRISP) service (1972‐2005). The search terms '"anticonvulsants OR antiepileptics" will be entered into the search interface for these databases.
Reference Lists
The bibliographies of all identified trials will be scanned for additional studies.
Correspondence
1) Published and unpublished trials will also be obtained from key researchers, as identified by the frequency with which they are cited in the bibliographies of RCTs and open‐label studies.
2) Pharmaceutical companies will also be contacted. They will be identified through the source of funding cited in published RCTs, as well as the companies with which the authors are affiliated. Additional pharmaceutical companies who are actively involved in this area will be identified through the innovation.org website (http://www.innovation.org) of the Pharmaceutical Research and Manufacturers of America foundation (PhRMA).
Data collection and analysis
Selection of studies
RCTs identified from the search will be independently assessed for inclusion by two raters, based on information included in the abstract and/or method section of the trial report. The raters will also independently collate the data listed under "Data extraction and management" from RCTs which they both regard as satisfying the inclusion criteria specified in the "criteria for considering studies" section. Studies for which additional information is required in order to determine their suitability for inclusion in the review will be listed in the "studies awaiting assessment" table in the Review Manager (RevMan) software, pending the availability of this information. Any disagreements in the independent trial assessment and data collation procedures will be resolved by discussion.
Data extraction and management
Spreadsheet forms will be designed for the purpose of recording descriptive information, summary statistics of the outcome measures, the quality scale ratings, and associated commentary. Once this data has been entered, it will be exported to the Review Manager (RevMan) software, which will be used to conduct the meta‐analysis. Where information is missing, the review authors will contact investigators by email in an attempt to obtain this information.
The following information will be collated from each trial:
(a) Description of the trials, including the primary researcher, the year of publication, and the source of funding.
(b) Characteristics of the interventions, including the number of participants randomised to the treatment and control groups, the number of total drop‐outs per group as well as the number that dropped out due to adverse effects, the dose of medication and the period over which it was administered, and the name of the particular anticonvulsant tested.
(c) Characteristics of trial methodology, including the diagnostic (e.g. DSM‐IV (APA 1994)) and exclusionary criteria employed, the screening instrument used (e.g. the Structured Clinical Interview for DSM‐IV (SCID) (Spitzer 1996)) for both the primary and comorbid diagnoses, the presence of comorbid major depressive disorder (MDD), the use of a placebo run‐in, whether a minimal severity criterion was employed, the number of centres involved, and the trial's total score on the CCDAN Quality Rating Scale (CCDAN‐QRS) (Moncrieff 2001).
(d) Characteristics of participants, including gender distribution and mean and range of ages, mean length of time since diagnosis with the anxiety disorder, whether they have been treated with the medication in the past (treatment naivety), the number of participants in the sample with MDD, and the baseline severity of the anxiety disorder, as assessed by the trial's primary outcome measure or another commonly employed scale.
(e) Outcome measures employed (primary and secondary), and summary continuous (means and standard deviations) and dichotomous (number of responders) data. Additional information will be included, such as whether data reflected the intent‐to‐treat (ITT) with last observation carried forward (LOCF) or completer/observed cases (OC) sample, and the minimal period required for inclusion of participants in the LOCF analyses. Other methods of estimating the outcome for participants who dropped out of the study, such as the mixed effects (ME) model, will also be recorded.
(f) Quality assessment, including the number of randomised participants who were not included in the analysis (lost to follow‐up (LTF)), whether blinding occurred for the assessor/s, patients, or those who administered medication, as well as whether the allocation of medication was randomised and the allocation sequence was concealed (the methods used in implementing these respective bias reduction measures will also be documented).
Assessment of methodological quality of included studies
The quality of the trials will be assessed by two review authors using the CCDAN Quality Rating Scale (CCDAN‐QRS) (Moncrieff 2001). This 23 item scale assesses a range of features such as sample size, the duration of the intervention, inclusion and exclusion criteria, and whether or not the power of the trial to detect a treatment effect was calculated. In addition, data for other trial characteristics which have been recognised as potential sources of bias, such as the method used in generating the allocation sequence, the concealment of allocation, whether outcome assessment was blinded, and the number of participants lost to follow up, will be independently collated by 2 review authors as part of the general data collection process (see Data extraction and management section). This is regarded as necessary given doubts concerning the usefulness of an overall quality score from a scale composed of multiple items (Alderson 2003).
Measures of treatment effect
Categorical data
Relative risk of response to treatment will be calculated for the dichotomous outcome of interest (CGI‐I or related measure). Relative risk will be used instead of odds ratios, as odd ratios are less easily interpreted. Odds ratios also tend to overestimate the size of the treatment effect when interpreted as relative risks. This is especially the case when the occurrence of the outcome of interest is common (as anticipated in this review, with an expected response greater than 20%) (Deeks 2003).
The number needed to treat (NNT) will also be calculated for statistically significant measures of treatment response on the CGI‐I (or related measure). NNT is defined as the inverse of the absolute risk difference due to the medication intervention. In this review it will be used to indicate the number of patients who require treatment with medication, relative to a control, before a single additional patient in the medication group responds to treatment.
Continuous data
Weighted mean differences (WMD) will be calculated for continuous summary data derived from the same scale. In cases in which a range of scales are employed for each outcome, such as in the assessment of symptom severity on the LSAS and CAPS, as well as in the assessment of comorbid depression on the MADRS and HAM‐D, the standardised mean difference (SMD) will be determined. This method of analysis standardises the differences between the means of the treatment and control groups in terms of the variability observed in the trial.
Studies with multiple treatment groups
In trials comparing the efficacy of multiple anticonvulsants in treating anxiety disorders, placebo‐medication outcome comparisons will be restricted to one of the anticonvulsants. Data from only one anticonvulsant group will be used so as to avoid possible bias through multiple comparisons with the same placebo group. Separate head‐to‐head primary outcome comparisons of the respective anticonvulsants will be conducted for trials comparing multiple anticonvulsants.
In the case of data from trials employing multiple fixed doses of medication, the bias introduced through comparing the summary statistics for multiple groups against the same placebo control will be avoided by pooling the means and standard deviations across all of the treatment arms as a function of the number of participants in each arm (see Law 2003).
Cross‐over trials
Cross‐over trials will only be included in the calculation of the outcomes of interest when it is (a) possible to extract medication and placebo/comparator data from the first treatment period, or (b) when the inclusion of data from both treatment periods is justified through a wash‐out period of sufficient duration as to minimise the risk of carry‐over effects. The minimum wash‐out period required will be assessed on the basis of the plasma half‐life of the particular agent, as determined by consulting the Lundbeck Psychotropics website (http://www.psychotropics.dk/luinst.asp). For trials in which the washout period is regarded as adequate, data from both periods will only be included when it is possible to determine the standard error of the mean difference in response between groups (Elbourne 2002). The summary statistics required to derive the standard error of interest will be obtained from the trial report, or for trials for which this information is missing, will be imputed through averaging the relevant statistic from other included crossover trials with comparable control conditions.
Dealing with missing data
All analyses of dichotomous data will be intention‐to‐treat (ITT). The total number of participants randomised to the different comparison groups will be used as the denominator in comparisons of treatment response. Only data from trials which provide information on the original group size (prior to drop‐outs) will be included in the analyses of treatment response. Preference will be given to the inclusion of summary statistics for continuous outcome measures derived from mixed effects models, followed by last observation carried forward (LOCF) and observed cases (OC) summary statistics (in that order). This is in line with evidence that ME methods are more robust to bias than LOCF analyses (Verbeke 2000).
Assessment of heterogeneity
Heterogeneity of treatment response and symptom severity will be assessed visually from the forest plot of relative risk. This will help determine whether the differences between the results of trials were greater than would be expected by chance alone. Heterogeneity will also be assessed by means of the chi‐square test of heterogeneity. If the chi‐square test has a p‐value of less than 0.10, this will be interpreted as evidence of heterogeneity, given the low power of the chi‐square statistic when the number of trials is small (Deeks 2003).
In addition, the I‐square heterogeneity statistic reported by RevMan will be used to quantify the inconsistency of the trial results within each subgroup analysis (Higgins 2003). Differences on continuous measures in medication efficacy between these subgroups will be assessed by means of Deeks' stratified test of heterogeneity (Deeks 2001). This method subtracts the sum of the chi‐square statistics available for each of the subgroups in the study from the chi‐statistic available for all the trials, to provide a measure (Qb) of heterogeneity between groups. Differences in treatment response on the CGI‐I will be determined by whether the confidence intervals for the effect sizes of the subgroups overlap. This method was chosen in preference to the stratified test, due to inaccuracies in the calculation in RevMan of the chi‐square statistic for dichotomous measures (Deeks 2003).
Assessment of reporting bias
Publication is not necessarily related to study quality and indeed publication may imply certain biases (Easterbrook 1991; Dickersin 1992; Scherer 1994). Small‐sample effects (including publication bias) will be determined by visual inspection of a funnel plot of treatment response. A more objective measure of bias for treatment response and reduction of symptom severity will be obtained by regressing the standardised treatment effects for the RCTs on the inverse of the standard error of their treatment estimate (Egger 1997). Although detection of publication bias is compromised when a review includes only a few small trials (Egger 1998), detecting bias using this method provides relatively strong evidence for the selective publication of trial reports. This procedure will be performed using the meta package of the R statistical language (R 2005).
Data synthesis
Categorical and continuous treatment effects will be obtained from a random effects model (the random effects model includes both within‐study sampling error and between‐studies variation in determining the precision of the confidence interval around the overall effect size, whereas the fixed effects model takes only within‐study variation into account). The outcomes will be expressed in terms of an average effect size for each subgroup, as well as by means of 95% confidence intervals.
Subgroup analyses and investigation of heterogeneity
Subgroup analyses (Thompson 1994) will be undertaken in order to assess the degree to which methodological differences between trials might have systematically influenced differences observed in the primary treatment outcomes.
The trials are to be grouped according to the following methodological sources of heterogeneity (number of trials permitting):
(a) The involvement of participants from a single centre or multiple centres. Single centre trials are more likely to be associated with lower sample size but less variability in clinician ratings.
(b) Whether or not trials were industry funded. In general, published trials which are sponsored by pharmaceutical companies appear more likely to report positive findings than trials which are not supported by for‐profit companies (Als‐Nielsen 2003; Baker 2003).
(c) Whether a placebo or alternative medication was used as a control group.
In addition, the following criteria will be used to assess the extent of clinical sources of heterogeneity:
(a) Whether or not the sample included patients diagnosed with major depression. Such an analysis might assist in determining the extent to which the efficacy of an anticonvulsant in treating anxiety disorders is independent of its ability to reduce symptoms of depression.
In recognition of the possibility of differential effects for different anticonvulsants, all of the comparisons will be stratified by the anticonvulsant agent employed. In addition, comparisons will also be stratified by the specific anxiety disorder (GAD, OCD, PD, PTSD, SAD, or SP) targeted.
Sensitivity analysis
Sensitivity analyses determine the robustness of the reviewer's conclusion to methodological assumptions made in conducting the meta‐analysis. A sensitivity analysis will be conducted to determine whether treatment response varies as a function of the use of treatment response versus non‐response as an outcome statistic. This comparison may be necessary in the light of evidence that treatment response may result in less consistent outcome statistics than non‐response (Deeks 2002) when the control group event rate is higher than 50%. This sensitivity analysis will accordingly only be performed if the majority of trials report a control group event rate higher than 50%.
