Non‐invasive brain stimulation techniques for chronic pain

Summary of findings for the main comparison.

Repetitive transcranial magnetic stimulation (rTMS) compared with sham for chronic pain
Intervention: active rTMS Comparison: sham rTMS
Outcomes: pain (VAS or NRS)
Comparison	No of participants (studies)	Effect size (SMD, 95% CIs)	Relative effect (average % improvement (reduction) in pain (95% CIs) in relation to post‐treatment score from sham group)* *statistically significant outcomes with low heterogeneity only	Quality of the evidence (GRADE)
Pain: short‐term follow‐up Subgroup analysis: low‐frequency rTMS	81 (6)	Ineffective 0.15 (‐0.01 to 0.31) P = 0.07		⊕⊕⊝⊝ low
Pain: short‐term follow‐up subgroup analysis: high‐frequency rTMS	447 (20)	Effective ‐0.27 (‐0.35 to ‐0.20) P < 0.01		⊕⊕⊝⊝ low
Pain: short‐term follow‐up Subgroup analysis: motor cortex studies only, low‐frequency studies excluded, single‐dose studies	233 (12)	Effective ‐0.39 (‐0.51 to ‐0.27) P < 0.01	12% (8% to 15%)	⊕⊕⊝⊝ low
Pain: short‐term follow‐up Subgroup analysis: motor cortex studies only, low‐frequency studies excluded, multiple‐dose studies	157 (5)	Ineffective ‐0.07 (‐0.41 to 0.26) P = 0.68		⊕⊝⊝⊝ very low
Pain: short‐term follow‐up Subgroup analysis: prefrontal cortex studies only	68 (5)	Ineffective ‐0.47 (‐1.48 to 0.11) P = 0.36		⊕⊝⊝⊝ very low
Pain: medium‐term follow‐up rTMS all studies	184 (8)	Ineffective ‐0.18 (‐0.43 to 0.06) P = 0.15		⊕⊝⊝⊝ very low
Pain: long‐term follow‐up rTMS all studies	59 (3)	Ineffective ‐0.12 (‐0.46 to 0.21) P = 0.47		⊕⊕⊝⊝ low
CES compared with sham for chronic pain
Intervention: active CES Comparison: sham CES
Outcomes: pain (VAS or NRS)
Pain: short‐term follow‐up CES all studies	270 (5)	Ineffective ‐0.24 (‐0.48 to 0.01) P = 0.06		⊕⊕⊝⊝ low
tDCS compared with sham for chronic pain
Intervention: active tDCS Comparison: sham tDCS
Outcomes: pain (VAS or NRS)
Pain: short‐term follow‐up tDCS all studies	183 (10)	Ineffective ‐0.18 (‐0.56 to 0.09) P = 0.19		⊕⊝⊝⊝ very low
Pain: short‐term follow‐up Subgroup analysis: motor cortex studies only (single and multiple‐dose studies)	172 (10)	Ineffective ‐0.23 (‐0.48 to 0.01) P = 0.06		⊕⊕⊝⊝ low
Pain: short‐term follow‐up Subgroup analysis: motor cortex studies only (multiple‐dose studies only)	119 (7)	Ineffective ‐0.35 (‐0.79 to 0.09) P = 0.12		⊕⊝⊝⊝ very low
Pain: medium‐term follow‐up tDCS	77 (4)	Ineffective ‐0.20 (‐0.63 to 0.24) P = 0.37		⊕⊕⊝⊝ low
RINCE compared with sham for chronic pain
Intervention: active RINCE Comparison: sham RINCE
Outcomes: pain (VAS or NRS)
Pain: short‐term follow‐up tDCS all studies	91 (1)	Effective ‐1.41 (‐2.48 to ‐0.34) P = 0.01		⊕⊝⊝⊝ very low
GRADE Working Group grades of evidence High quality: Further research is very unlikely to change our confidence in the estimate of effect. Moderate quality: Further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate. Low quality: Further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate. Very low quality: We are very uncertain about the estimate. CES: cranial electrotherapy stimulation; CI: confidence interval; NRS: numerical rating scale; RINCE: reduced impedance non‐invasive cortical electrostimulation; rTMS: repetitive transcranial magnetic stimulation; tDCS: transcranial direct current stimulation; VAS: visual analogue scale
For full details of the GRADE judgements for each comparison see Appendix 6.

Background

This is an updated version of the original Cochrane review published in 2010, Issue 9, on non‐invasive brain stimulation techniques for chronic pain (O'Connell 2010).

Description of the condition

Chronic pain is a common problem. When defined as pain of greater than three months duration, prevalence studies indicate that up to half the adult population suffer from chronic pain, and 10% to 20% experience clinically significant chronic pain (Smith 2008). In Europe, 19% of adults experience chronic pain of moderate to severe intensity with serious negative implications for their social and working lives and many of these receive inadequate pain management (Breivik 2006). Chronic pain is a heterogenous phenomenon that results from a wide variety of pathologies including chronic somatic tissue injury such as arthritis, peripheral nerve injury and central nervous system injury, as well as a range of chronic pain syndromes such as fibromyalgia. It is likely that different mechanisms of pain production underpin these different causes of chronic pain (Ossipov 2006).

Description of the intervention

Brain stimulation techniques have been used to address a variety of pathological pain conditions including fibromyalgia, chronic post‐stroke pain and complex regional pain syndrome (Cruccu 2007; Fregni 2007; Gilula 2007), and clinical studies of both invasive and non‐invasive techniques have produced preliminary data showing reductions in pain (Cruccu 2007; Fregni 2007; Lefaucheur 2008b). Various types of brain stimulation, both invasive and non‐invasive, are currently in clinical use for the treatment of chronic pain (Cruccu 2007). Non‐invasive stimulation techniques require no surgical procedure and are therefore easier and safer to apply than invasive procedures.

Repetitive transcranial magnetic stimulation (rTMS) involves stimulation of the cerebral cortex (the outer layer of the brain) by a stimulating coil applied to the scalp. Electric currents are induced in the neurons (brain cells) directly using rapidly changing magnetic fields (Fregni 2007). Trains of these stimuli are applied to the target region of the cortex to induce alterations in brain activity both locally and in remote brain regions (Leo 2007). A recent meta‐analysis suggested that rTMS may be more effective in the treatment of neuropathic pain conditions (pain arising as a result of damage to the nervous system, as in diabetes, traumatic nerve injury, stroke, multiple sclerosis, epilepsy, spinal cord injury and cancer) with a central compared to a peripheral nervous system origin (Leung 2009).

Transcranial direct current stimulation (tDCS) and cranial electrotherapy stimulation (CES) involve the safe and painless application of low‐intensity (commonly ≤ 2 mA) electrical current to the cerebral cortex of the brain (Fregni 2007; Gilula 2007; Hargrove 2012). tDCS has been developed as a clinical tool for the modulation of brain activity in recent years and uses relatively large electrodes that are applied to the scalp over the targeted brain area to deliver a weak constant current (Lefaucheur 2008a). Recent clinical studies have concluded that tDCS was more effective than sham stimulation at reducing pain in both fibromyalgia and spinal cord injury related pain (Fregni 2006a; Fregni 2006b). CES was initially developed in the USSR as a treatment for anxiety and depression in the 1950s and its use later spread to Europe and the USA where it began to be considered and used as a treatment for pain (Kirsch 2000). The electrical current in CES is commonly pulsed and is applied via clip electrodes that are attached to the patient's earlobes. A Cochrane Review of non‐invasive treatments for headaches identified limited evidence that CES is superior to placebo in reducing pain intensity after six to 10 weeks of treatment (Bronfort 2004). Reduced impedance non‐invasive cortical electrostimulation (RINCE) similarly applies an electrical current via scalp electrodes but utilises specific stimulation frequencies which are hypothesised to reduce electrical impedance from the tissues of the skin and skull, allowing deeper cortical penetration and modulation of lower‐frequency cortical activity (Hargrove 2012).

How the intervention might work

Brain stimulation techniques primarily seek to modulate activity in brain regions by directly altering the level of brain activity. The aim of brain stimulation in the management of pain is to reduce pain by altering activity in the areas of the brain that are involved in pain processing.

Both tDCS and rTMS have been shown to modulate brain activity specific to the site of application and the stimulation parameters. As a general rule, low‐frequency rTMS (≤ 1 Hz) results in lowered cortical excitability at the site of stimulation, whereas high‐frequency stimulation (≥ 5 Hz) results in raised cortical excitability (Lefaucheur 2008a; Pascual‐Leone 1999). Similarly, anodal tDCS, wherein the anode electrode is placed over the cortical target, results in a raised level of excitability at the target, whereas cathodal stimulation decreases local cortical excitability (Nitsche 2008). It is suggested that the observed alterations in cortical excitability (readiness for activity) following rTMS and tDCS that last beyond the time of stimulation are the result of long‐term synaptic changes (Lefaucheur 2008a). Modulation of activity in brain networks is also proposed as the mechanism of action of CES and RINCE therapy and it is suggested that the therapeutic effects are primarily achieved by direct action upon the hypothalamus, limbic system and/or the reticular activating system (Gilula 2007).

Imaging studies in humans suggest that motor cortex stimulation may reduce pain by modulating activity in networks of brain areas involved in pain processing, such as the thalamus, and by facilitating descending pain inhibitory mechanisms (Garcia‐Larrea 1997; Garcia‐Larrea 1999; Peyron 2007).

Sham credibility issues for non‐invasive brain stimulation studies

An issue regarding the credibility of sham conditions specifically for rTMS studies is whether the sham condition that is employed controls for the auditory (clicking sounds of various frequencies) and sensory stimulation that occurs during active stimulation (Lisanby 2001; Loo 2000). Various types of sham have been proposed including angling the coil away from the scalp (thus preserving the auditory cues but not the sensation of stimulation), using coils that mimic the auditory cues combined with gentle scalp electrical stimulation to mask the sensation and simple inert coils that reproduce neither the sound nor the sensation of active stimulation. Failure to control for such cues may impact negatively on patient blinding, particularly in cross‐over design studies. Lisanby 2001 and Loo 2000 suggest that an ideal sham condition for rTMS should:

not stimulate the cortex;
be the same as active stimulation in visual terms and in terms of its position on the scalp; and
not differ from active stimulation in terms of the acoustic and afferent sensory sensations that it elicits.

Strategies have been developed to try to meet these criteria (Borckardt 2008; Rossi 2007; Sommer 2006). There is evidence that simply angling the coil away from the scalp at an angle of less than 90° may still result in brain stimulation and not be truly inert (Lisanby 2001). This strategy is also easily detected by the recipient of stimulation. In these ways this type of sham might obscure or exaggerate a real clinical effect of active stimulation.

In studies of tDCS the sham condition commonly involves the delivery of a short initial period (30 seconds to one minute) of identical stimulation to the active condition, at which point the stimulation is ceased without the participant's knowledge. There is evidence that this achieves effective blinding of tDCS at stimulation intensities of 1 mA in naive participants (Ambrus 2012; Gandiga 2006), but at a stimulation intensity of 2 mA tDCS both participant and assessor blinding has been shown to be inadequate, since participants can distinguish the active condition more than would be expected by chance and a proportion of those receiving active stimulation develop a temporary but visible redness over the electrode sites (O'Connell 2012). At 1.5 mA there are detectable differences in the experience of tDCS that might compromise blinding (Kessler 2013), though a formal investigation of the adequacy of blinding at this intensity has not been published to date.

Why it is important to do this review

This approach to pain treatment is relatively novel. It is important to assess the existing literature robustly to ascertain the current level of supporting evidence and to inform future research and potential clinical use. Recent reviews have addressed this area and concluded that non‐invasive brain stimulation can exert a significant effect on chronic pain, but they have restricted their findings to specific cortical regions, types of painful condition or types of stimulation and did not carry out a thorough assessment of study quality or risk of bias (Lefaucheur 2008b; Leung 2009; Lima 2008).

Objectives

To review all randomised and quasi‐randomised studies of non‐invasive cortical stimulation techniques in the treatment of chronic pain. The key aims of the review were:

to critically evaluate the efficacy of non‐invasive cortical stimulation techniques compared to sham controls for chronic pain; and
to critically evaluate the influence of altered treatment parameters (i.e. stimulation method, parameters, dosage, site) on the efficacy of non‐invasive cortical stimulation for chronic pain.

Methods

Criteria for considering studies for this review

Types of studies

Randomised controlled trials (RCTs) and quasi‐randomised trials (e.g. by order of entry or date of birth) that utilise a sham control group were included. We included parallel and cross‐over study designs. We included studies regardless of language.

Types of participants

We included studies involving male or female participants over the age of 18 years with any chronic pain syndrome (with a duration of more than three months). It was not anticipated that any studies are likely to exist in a younger population. Migraine and other headache studies were not included due to the episodic nature of these conditions.

Types of interventions

We included studies investigating the therapeutic use of non‐invasive forms of brain stimulation (tDCS, rTMS CES or RINCE). We did not include studies of electroconvulsive therapy (ECT) as its mechanism of action (the artificial induction of an epileptic seizure (Stevens 1996)) differs substantially from the other forms of brain stimulation. Invasive forms of brain stimulation involving the use of electrodes implanted within the brain and indirect forms of stimulation, such as caloric vestibular stimulation and occipital nerve stimulation, were also not included. In order to meet our second objective of considering the influence of varying stimulation parameters, we included studies regardless of the number of stimulation sessions delivered, including single‐dose studies.

Types of outcome measures

Primary outcomes

The primary outcome measure was change in self reported pain using validated measures of pain intensity such as visual analogue scales (VAS), verbal rating scales (VRS) or numerical rating scales (NRS).

Secondary outcomes

Secondary outcomes that we extracted when available included self reported disability data, quality of life measures and the incidence/nature of adverse events.

Search methods for identification of studies

Electronic searches

For the OVID MEDLINE search, we ran the subject search with the Cochrane Highly Sensitive Search Strategy for identifying randomised trials in MEDLINE: sensitivity maximising version (2008 revision) as referenced in Chapter 6 and detailed in box 6.4c of the Cochrane Handbook for Systematic Reviews of Interventions Version 5.0.1 (Higgins 2011). We have slightly adapted this filter to include the term 'sham' in the title or abstract. The search strategies are presented in Appendix 1 and included a combination of controlled vocabulary (MeSH) and free‐text terms. We based all database searches on this strategy but appropriately revised them to suit each database.

Electronic databases

We ran the original search for the review in November 2009 and searched all databases from their inception. To identify studies for inclusion in this update we searched the following electronic databases from 2009 to July 2013 to identify additional published articles:

the Cochrane Central Register of Controlled Trials (CENTRAL) (The Cochrane Library 2013, Issue 6);
OVID MEDLINE & MEDLINE in Process to 23 July 2013;
OVID EMBASE to 2013 week 29;
PsycINFO to July week 3 2013;
CINAHL to July 2013;
LILACS to January 2013;

For full details of the search parameters including dates for this update see Appendix 1; Appendix 2; Appendix 3.

Searching other resources

Reference lists

We searched reference lists of all eligible trials, key textbooks and previous systematic reviews to identify additional relevant articles.

Unpublished data

We searched the National Research Register (NRR) Archive, Health Services Research Projects in Progress (HSRProj), Current Controlled Trials register (incorporating the meta‐register of controlled trials and the International Standard Randomised Controlled Trial Number (ISRCTN)) to January 2013 to identify research in progress and unpublished research.

Language

The search attempted to identify all relevant studies irrespective of language. We assessed non‐English papers and, if necessary, translated with the assistance of a native speaker.

We sent a final list of included articles to two experts in the field of therapeutic brain stimulation with a request that they review the list for possible omissions.

Data collection and analysis

Selection of studies

Two review authors (NOC and BW) independently checked the search results and included eligible studies. Initially two review authors (NOC and BW) read the titles or abstracts (or both) of identified studies. Where it was clear from the study title or abstract that the study was not relevant or did not meet the selection criteria we excluded it. If it was unclear then we assessed the full paper, as well as all studies that appeared to meet the selection criteria. Disagreement was resolved through discussion between the two review authors. Where resolution was not achieved a third review author (LDS) considered the paper(s) in question.

Data extraction and management

Two review authors (NOC and BW) extracted data independently using a standardised form that was piloted by both authors independently on three randomised controlled trials of transcutaneous electrical nerve stimulation prior to the searches. We resolved discrepancies by consensus. The form included the following.

'Risk of bias' assessment results.
Country of origin.
Study design.
Study population ‐ condition; pain type; duration of symptoms; age range; gender split; prior management.
Sample size ‐ active and control groups.
Intervention ‐ stimulation site, parameters and dosage (including number and duration of trains of stimuli and number of pulses for rTMS studies).
Type of sham.
Credibility of sham (for rTMS studies ‐ see below).
Outcomes ‐ mean post‐intervention pain scores for the active and sham treatment groups at all follow‐up points.
Results ‐ short, intermediate and long‐term follow‐up.
Adverse effects.
Conflict of interest disclosure.

Assessment of risk of bias in included studies

We assessed risk of bias using the Cochrane 'Risk of bias' assessment tool outlined in Chapter 8 of the Cochrane Handbook for Systematic Reviews of Interventions Version 5.0.1 (Higgins 2011).

The criteria assessed for parallel study designs (using low/high/unclear judgements) were: adequate sequence generation; adequate allocation concealment; adequate blinding of assessors; adequate blinding of participants; adequate assessment of incomplete outcome data; whether free of suggestion of selective outcome reporting; and whether free of other bias.

The criteria assessed for cross‐over study designs (using low/high/unclear judgements) were: adequate sequence generation; whether data were clearly free from carry‐over effects; adequate blinding of assessors; adequate blinding of participants; whether free of the suggestion of selective outcome reporting; and whether free of other bias.

For this update, in compliance with new author guidelines from the Cochrane Pain, Palliative and Supportive Care review group and the recommendations of Moore 2010 we added two criteria, 'study size' and 'study duration', to our 'Risk of bias' assessment using the thresholds for judgement suggested by Moore 2010:

Size (we rated studies with fewer than 50 participants per arm as being at high risk of bias, those with between 50 and 199 participants per arm at unclear risk of bias, and 200 or more participants per arm at low risk of bias).

Duration (we rated studies with follow‐up of less than two weeks as being at high risk of bias, two to seven weeks at unclear risk of bias and eight weeks or longer at low risk of bias).

Two review authors (NOC and BW) independently checked risk of bias. Disagreement between review authors was resolved through discussion between the two review authors. Where resolution was not achieved a third review author (LDS) considered the paper(s) in question.

Assessment of sham credibility

We rated the type of sham used in studies of rTMS for credibility: as optimal (the sham controls for the auditory and sensory characteristics of stimulation and is visually indistinguishable from real stimulation (Lisanby 2001; Loo 2000)) and sub‐optimal (fails to account for either the auditory and sensory characteristics of stimulation, or is visually distinguishable from the active stimulation, or fails on more than one of these criteria). We made a judgement of 'unclear' where studies did not adequately describe the sham condition.

In light of empirical evidence that tDCS may be inadequately blinded at intensities of 2 mA (O'Connell 2012), and of detectable differences in the experience of tDCS at 1.5 mA (Kessler 2013), for this update we assessed studies that used these stimulation intensities to be at unclear risk of bias for participant and assessor blinding. We chose 'unclear' instead of 'high' risk of bias as the available evidence demonstrates the potential for inadequate blinding rather than providing clear evidence that individual studies were effectively unblinded. We applied this rule to all newly identified studies and retrospectively to studies identified in the previous version of this review.

Two independent review authors (NOC and BW) performed rating of sham credibility. We resolved disagreement between review authors through consensus. Where resolution was not achieved a third review author (LDS) considered the paper(s) in question. Where sham credibility was assessed as unclear or sub‐optimal we made a judgement of 'unclear' for the criterion 'adequate blinding of participants' in the 'Risk of bias' assessment.

Measures of treatment effect

We used standardised mean difference (SMD) to express the size of treatment effect on pain intensity measured with a VAS or NRS. In order to aid interpretation of the pooled effect size we back‐transformed the SMD to a 0 to 100 mm VAS format on the basis of the mean standard deviation from trials using 0 to 100 mm VAS. We considered the likely clinical importance of the pooled effect size using the criteria proposed in the IMMPACT consensus statement (Dworkin 2008). Specifically, we judged a decrease in pain of < 15% as no important change, ≥ 15% as a minimally important change, ≥ 30% as a moderately important change and ≥ 50% as a substantially important change.

Unit of analysis issues

We entered cross‐over trials into a meta‐analysis where it was clear that these data were free of carry‐over effects. We combined the results of cross‐over studies with parallel studies using the generic inverse‐variance method as suggested in the Cochrane Handbook for Systematic Reviews of Interventions, section 16.4.6.2 (Higgins 2011). We imputed the post‐treatment between‐condition correlation coefficient from an included cross‐over study that presented individual patient data and used this to calculate the standard error of the standardised mean difference (SE (SMD)). Where data from the same cross‐over trials were entered more than once into the same meta‐analysis we corrected the number of participants by dividing by the number times data from that trial were entered in the meta‐analysis. We calculated the SMD(SE) for parallel studies in RevMan. For each study we entered the SMD (SE) into the meta‐analysis using the generic inverse‐variance method.

Dealing with missing data

Where insufficient data were presented in the study report to enter a study into the meta‐analysis, we contacted the study authors to request access to the missing data.

Data synthesis

We performed pooling of results where adequate data supported this using RevMan 5 software (version 5.2) (RevMan 2012), with a random‐effects model. Where an analysis included parallel and cross‐over trials we used the generic inverse variance method (see Unit of analysis issues). We conducted separate meta‐analyses for different forms of stimulation intervention (i.e. rTMS, tDCS, CES and RINCE) and for short‐term (0 to < 1 week post‐intervention), mid‐term (≥ 1 to 6 weeks post‐intervention) and long‐term (≥ 6 weeks post‐intervention) outcomes where adequate data were identified.

Where more than one data point was available for short‐term outcomes, we used the first post‐stimulation measure, and where multiple treatments were given we took the first outcome at the end of the treatment period. For medium‐term outcomes where more than one data point was available, we used the measure that fell closest to the mid‐point of this time period. We excluded studies from the meta‐analysis that we rated at high risk of bias on any criteria, excluding the criteria 'study size' and 'study duration'.

For this update we utilised the GRADE approach to assessing the quality of a body of evidence (Guyatt 2008). To ensure consistency of GRADE judgements we applied the following criteria to each domain equally for all key comparisons of the primary outcome:

Limitations of studies: downgrade once if less than 75% of included studies are at low risk of bias across all 'Risk of bias' criteria.
Inconsistency: downgrade once if heterogeneity is statistically significant and the I² value is more than 40%.
Indirectness: downgrade once if more than 50% of the participants were outside the target group.
Imprecision: downgrade once if fewer than 400 participants for continuous data and fewer than 300 events for dichotomous data (Guyatt 2011).
Publication bias: downgrade where there is direct evidence of publication bias.

While we had planned to use GRADE in our initial protocol we introduced these criteria specifically for this update.

Subgroup analysis and investigation of heterogeneity

We assessed heterogeneity using the Chi² test to investigate its statistical significance and the I² statistic to estimate the amount. Where significant heterogeneity (P < 0.1) was present we explored subgroup analysis. Pre‐planned comparisons included site of stimulation, frequency of TMS stimulation (low ≤ 1 Hz, high ≥ 5 Hz), multiple versus single‐dose studies and the type of painful condition (central neuropathic versus peripheral neuropathic versus non‐neuropathic pain versus facial pain (for each stimulation type). Central neuropathic pain included pain due to identifiable pathology of the central nervous system (e.g. stroke, spinal cord injury), peripheral neuropathic pain included injury to the nerve root or peripheral nerves, facial pain included trigeminal neuralgia and other idiopathic chronic facial pains, and non‐neuropathic pain included all chronic pain conditions without a clear neuropathic cause (e.g. chronic low back pain, fibromyalgia, complex regional pain syndrome type I).

Sensitivity analysis

When sufficient data were available, we conducted sensitivity analyses on the following study factors: risk of bias, sham credibility (for rTMS studies) and cross‐over versus parallel‐group designs.

Results

Description of studies

See: Characteristics of included studies; Characteristics of excluded studies.

Results of the search

Published data

In our original review the search strategy identified 1148 citations, including 305 duplicates. See Appendix 4 and Appendix 5 for full details of the search results from the original review. Screening of the 843 unique citations by title and abstract identified 39 as potentially eligible for the review. Three studies were identified from handsearching of the reference lists of included studies of which two were not retrievable in abstract or full manuscript form. The level of agreement between review authors, calculated using the kappa statistic for study eligibility based on title and abstract alone, was 0.77. We identified three more papers that were not picked up from the search strategy. We also deemed these to be potentially eligible for the review. One of the experts contacted to review the search results for possible omissions identified one additional study. The full‐text screening of the 44 citations identified 33 eligible studies (19 of rTMS, 422 participants randomised; six of tDCS, 124 participants randomised; eight of CES, 391 participants randomised) (André‐Obadia 2006; André‐Obadia 2008; Boggio 2009; Borckardt 2009; Capel 2003; Carretero 2009; Cork 2004; Defrin 2007; Fenton 2009; Fregni 2005; Fregni 2006a; Fregni 2006b; Gabis 2003; Gabis 2009; Hirayama 2006; Irlbacher 2006; Kang 2009; Katsnelson 2004; Khedr 2005; Lefaucheur 2001a; Lefaucheur 2001b; Lefaucheur 2004; Lefaucheur 2006; Lefaucheur 2008; Lichtbroun 2001; Mori 2010; Passard 2007; Pleger 2004; Rollnik 2002; Saitoh 2007; Tan 2000; Tan 2006; Valle 2009). The kappa level of agreement between authors for eligibility from full‐text screening was 0.87.

In this update we conducted a full search in February 2013 and updated the search of the main databases on 12 June 2013 and again on 24 July 2013. We included a further 23 completed studies with 773 participants randomised (range of n = 3 to 105, see Figure 1 for a flow chart of the search process). Of these, 11 studies (324 participants randomised) investigated rTMS (Ahmed 2011; André‐Obadia 2011; Avery 2013, Fregni 2011; Hosomi 2013; Jensen 2013; Lee 2012; Mhalla 2011; Picarelli 2010; Short 2011; Tzabazis 2013), eight studies (177 participants randomised) investigated tDCS (Antal 2010; Jensen 2013; Mendonca 2011; Portilla 2013; Riberto 2011; Soler 2010; Villamar 2013; Wrigley 2014), three studies (181 participants randomised) investigated CES (Rintala 2010; Tan 2011; Taylor 2013), and one study investigated a novel form of stimulation (reduced impedance non‐invasive cortical electrostimulation (RINCE)) that did not fit neatly into any of the three broad categories (Hargrove 2012, 91 participants)). Overall this updated review included 56 studies (1710 participants randomised), with 30 trials of rTMS (746 participants randomised), 14 trials of tDCS (301 participants randomised), 11 studies of CES (572 participants randomised) and one study of RINCE stimulation (91 participants randomised).

Figure 1

Study flow diagram for updated search.

We identified an additional 11 conference abstracts that were not related to full published studies (Acler 2012; Albu 2011; Ansari 2013; Fricova 2009; Fricova 2011; Klirova 2010; Klirova 2011; Knotkova 2011; Pellaprat 2012; Schneider 2012; Yaĝci 2013). We contacted the authors of these abstracts to try to ascertain whether they were unique studies or duplicates and to acquire full study reports. Where we were unable to obtain this information we placed these records in Studies awaiting classification. For two of these abstracts the authors confirmed that they referred to studies that are either in the analysis/write‐up stage or under review for publication, and as such were unavailable for this review update (Knotkova 2011; Schneider 2012). For the remaining abstracts identified in this update our attempts to contact the authors were not successful (Acler 2012; Albu 2011; Ansari 2013 Fricova 2009; Fricova 2011; Klirova 2010; Klirova 2011; Pellaprat 2012; Yaĝci 2013). We sent requests by email where possible in February 2013, with a follow‐up email in April and June 2013, for those identified in the first search of this update, and in June 2013 for those identified by the second round of searching.

Unpublished data

In our original review the search strategy identified 5920 registered studies. Screening of the studies by the register records identified 23 studies that might potentially produce relevant data. Of these, seven were duplicated across trials registers, leaving 16 unique registered studies. We contacted the contact author for each of these studies by post or email with a request for any relevant data that might inform the review. No data were available from any of these studies for inclusion in this review.

In this update our search of the trials registers identified 599 records from which 11 relevant ongoing trials were identified. In addition to the two ongoing studies remaining from the last update (NCT00947622; NCT00815932); this makes a total of 13 ongoing studies identified. We contacted the contact author for each of these studies by post or email with a request for any relevant data that might inform the review. No data were available from any of these studies for inclusion in this review. We sent initial request emails for this update in April, and where no response was received also in May and in June 2013. Unpublished data and a full study report was provided for one study of rTMS identified from the trials registers search of the last update of this review (reference was Wajdik 2009, now Avery 2013).

Included studies

See Characteristics of included studies.

Country of origin and language of publication

All but one of the studies (Irlbacher 2006, written in German) were written in English. Studies were undertaken in Brazil, Egypt, Europe (France, Germany, Italy, Spain and the UK), Israel, Japan, Russia, South Korea and the USA. Most studies were based in a laboratory or outpatient pain clinic setting.

Type of stimulation, application and use

In total 30 studies investigated rTMS (Ahmed 2011; André‐Obadia 2006; André‐Obadia 2008; André‐Obadia 2011; Avery 2013; Borckardt 2009; Carretero 2009; Defrin 2007; Fregni 2005; Fregni 2011; Hirayama 2006; Hosomi 2013; Irlbacher 2006; Kang 2009; Khedr 2005; Lee 2012; Lefaucheur 2001a; Lefaucheur 2001b; Lefaucheur 2004; Lefaucheur 2006; Lefaucheur 2008; Mhalla 2011; Onesti 2013; Passard 2007; Picarelli 2010; Pleger 2004; Rollnik 2002; Saitoh 2007; Short 2011; Tzabazis 2013). Eleven studies investigated CES (Capel 2003; Cork 2004; Gabis 2003; Gabis 2009; Katsnelson 2004; Lichtbroun 2001; Rintala 2010; Tan 2000; Tan 2006; Tan 2011; Taylor 2013), 14 studies investigated tDCS (Antal 2010; Boggio 2009; Fenton 2009; Fregni 2006a; Fregni 2006b; Jensen 2013; Mendonca 2011; Mori 2010; Portilla 2013; Riberto 2011; Soler 2010; Valle 2009; Villamar 2013; Wrigley 2014), and one study investigated RINCE stimulation (Hargrove 2012). We had not been aware of RINCE therapy until it was identified in this search update. While it bears similarities with CES the author of the included trial suggested that due to the specific unique stimulation parameters that differ from conventional forms of CES, it represents a novel form of cortical stimulation (Hargrove 2012).

Study designs

There were a mixture of parallel and cross‐over study designs. For rTMS there were 12 parallel studies (Ahmed 2011; Avery 2013; Carretero 2009; Defrin 2007; Fregni 2011; Khedr 2005; Lee 2012; Mhalla 2011; Passard 2007; Picarelli 2010; Short 2011; Tzabazis 2013), and 18 cross‐over studies (André‐Obadia 2006; André‐Obadia 2008; André‐Obadia 2011, Borckardt 2009; Fregni 2005; Hirayama 2006; Hosomi 2013; Irlbacher 2006; Kang 2009; Lefaucheur 2001a; Lefaucheur 2001b; Lefaucheur 2004; Lefaucheur 2006; Lefaucheur 2008; Onesti 2013; Pleger 2004; Rollnik 2002; Saitoh 2007). For CES there were eight parallel studies (Gabis 2003; Gabis 2009; Katsnelson 2004; Lichtbroun 2001; Rintala 2010; Tan 2006; Tan 2011; Taylor 2013), and three cross‐over studies (Capel 2003; Cork 2004; Tan 2000), of which we considered two as parallel studies, with only the opening phase of the study considered in this review because subsequent phases were unblinded (Capel 2003; Cork 2004). For tDCS there were seven parallel studies (Fregni 2006a; Fregni 2006b; Mendonca 2011; Mori 2010; Riberto 2011; Soler 2010; Valle 2009), and seven cross‐over studies (Antal 2010; Boggio 2009; Fenton 2009; Jensen 2013; Portilla 2013; Villamar 2013; Wrigley 2014), of which we considered one as a parallel study with only the opening phase of the study considered in this review due to excessive attrition after the first phase (Antal 2010).

Study participants

The included studies were published between 2000 and 2013. In rTMS studies sample sizes at the study outset ranged from four to 70 participants. In CES studies sample size ranged from 19 to 105 participants, in tDCS studies sample size ranged from three to 41 participants and the single RINCE study recruited 91 participants.

Studies included a variety of chronic pain conditions. Nine rTMS studies included participants with neuropathic pain of mixed origin; of these seven included a mix of central, peripheral and facial neuropathic pain patients (André‐Obadia 2006; André‐Obadia 2008; André‐Obadia 2011; Hirayama 2006; Hosomi 2013, Lefaucheur 2004; Lefaucheur 2008), two included a mix of central and peripheral neuropathic pain patients (Lefaucheur 2006; Saitoh 2007), of which one study included a patient with phantom limb pain (Saitoh 2007). One study included a mix of central neuropathic pain and phantom limb pain patients (Irlbacher 2006). One study included a mix of central and facial neuropathic pain patients (Lefaucheur 2001a), two rTMS studies included only central neuropathic pain patients (Defrin 2007; Kang 2009), one included only peripheral neuropathic pain patients (Borckardt 2009), and nine studies included non‐neuropathic chronic pain including fibromyalgia (Carretero 2009; Lee 2012; Mhalla 2011; Passard 2007; Short 2011; Tzabazis 2013), chronic widespread pain (Avery 2013), chronic pancreatitis pain (Fregni 2005; Fregni 2011), and complex regional pain syndrome type I (CRPSI) (Picarelli 2010; Pleger 2004). One study included only phantom limb pain (Ahmed 2011). Finally one study included a mix of peripheral neuropathic and non‐neuropathic chronic pain (Rollnik 2002), including one participant with phantom limb pain and one with osteomyelitis. The majority (17) of rTMS studies specified chronic pain that was refractory to current medical management (André‐Obadia 2006; André‐Obadia 2008, André‐Obadia 2011; Defrin 2007; Hirayama 2006; Hosomi 2013; Kang 2009; Khedr 2005; Lefaucheur 2001a; Lefaucheur 2001b; Lefaucheur 2004; Lefaucheur 2006; Lefaucheur 2008; Onesti 2013; Picarelli 2010; Rollnik 2002; Saitoh 2007). This inclusion criterion was varyingly described as intractable, resistant to medical intervention or drug management.

Of the studies investigating CES, one study included participants with pain related to osteoarthritis of the hip and knee (Katsnelson 2004), and two studied chronic back and neck pain (Gabis 2003; Gabis 2009). Of these, the later study also included participants with chronic headache but these data were not considered in this review. Three studies included participants with fibromyalgia (Cork 2004; Lichtbroun 2001; Taylor 2013), and three studies included participants with chronic pain following spinal cord injury (Capel 2003; Tan 2006; Tan 2011), although only one of these reports specified that the pain was neuropathic (Tan 2011). One study included participants with a mixture of "neuromuscular pain" excluding fibromyalgia of which back pain was reportedly the most prevalent complaint (Tan 2000), although further details were not reported. One study included participants with chronic pain related to Parkinson's disease (Rintala 2010).

Of the studies of tDCS one study included participants with a mixture of central, peripheral and facial neuropathic pain (Boggio 2009), one study included participants with neuropathic pain secondary to multiple sclerosis (Mori 2010), three included participants with central neuropathic pain following spinal cord injury (Fregni 2006a; Soler 2010; Wrigley 2014), one with neuropathic or non‐neuropathic pain following spinal cord injury (Jensen 2013), and six studies included non‐neuropathic pain, specifically chronic pelvic pain (Fenton 2009), and fibromyalgia (Fregni 2006b; Mendonca 2011; Riberto 2011; Villamar 2013), or a mixed group (Antal 2010). One study included participants with neuropathic pain following burn injury (Portilla 2013). Four studies of tDCS specified recruiting participants with pain that was refractory to medical management (Antal 2010; Boggio 2009; Fenton 2009; Fregni 2006a). The study relating to RINCE stimulation included participants with fibromyalgia (Hargrove 2012).

Most studies included both male and female participants except the studies of Fenton 2009 (chronic pelvic pain) and Fregni 2006b, Valle 2009, Riberto 2011 and Mhalla 2011; Lee 2012 (fibromyalgia), which recruited females only and Fregni 2006a (post‐spinal cord injury pain), which recruited only males. Two studies did not present data specifying the gender distribution of participants (Capel 2003; Katsnelson 2004).

Outcomes

Primary outcomes

All included studies assessed pain using self reported pain visual analogue or numerical rating scales. There was variation in the precise measure of pain (for example, current pain intensity, average pain intensity over 24 hours) and in the anchors used particularly for the upper limit of the scale (e.g. "worst pain imaginable", "unbearable pain", "most intense pain sensation"). Several studies did not specify the anchors used.

All studies assessed pain at the short‐term (< 1 week post‐treatment) follow‐up stage. Twenty‐three studies reported collecting medium‐term outcome data (≥ 1 to 6 weeks post‐treatment) (Ahmed 2011; André‐Obadia 2008; Antal 2010; Borckardt 2009; Carretero 2009; Defrin 2007; Fenton 2009; Fregni 2006a; Fregni 2006b; Fregni 2011; Gabis 2009; Kang 2009; Khedr 2005; Lee 2012; Lefaucheur 2001a; Mori 2010; Passard 2007; Picarelli 2010; Short 2011; Soler 2010; Tzabazis 2013; Valle 2009; Wrigley 2014). Only three studies collected controlled outcome data on long‐term (> 6 weeks post‐treatment) follow‐up (Avery 2013; Kang 2009; Passard 2007).

Secondary outcomes

We only considered secondary outcomes that distinctly measured self reported disability or quality of life for extraction and inclusion in the Characteristics of included studies table. Nine studies used measures of disability or pain interference (Avery 2013; Cork 2004; Kang 2009; Mhalla 2011; Passard 2007; Short 2011; Soler 2010; Tan 2000; Tan 2006), and 14 studies collected measures of quality of life (Avery 2013; Fregni 2006b; Lee 2012; Lichtbroun 2001; Mhalla 2011; Mori 2010; Passard 2007; Picarelli 2010; Riberto 2011; Short 2011; Tan 2011; Taylor 2013; Tzabazis 2013; Valle 2009).

Adverse event reporting

Seventeen studies did not report any information regarding adverse events (Ahmed 2011; André‐Obadia 2011; Borckardt 2009; Cork 2004; Defrin 2007; Gabis 2009; Jensen 2013; Kang 2009; Katsnelson 2004; Khedr 2005; Lefaucheur 2006; Lefaucheur 2008; Lichtbroun 2001; Pleger 2004; Riberto 2011; Tan 2000; Tan 2006).

Studies of rTMS

See Table 1 for a summary of stimulation characteristics utilised in rTMS studies.

Table 1. rTMS studies ‐ characteristics of stimulation

Study	Location of stimulation	Coil orientation	Frequency (Hz)	Intensity (% RMT)	Number of trains	Duration of trains	Inter‐train intervals (sec)	Number of pulses per session	Treatment sessions per group
Ahmed 2011	M1 stump region	45° angle from sagittal line	20	80	10	10 sec	50	2000	5, x 1 daily
André‐Obadia 2006	M1 contralateral to painful side	Posteroanterior	20, 1	90	20 Hz: 20 1Hz: 1	20 Hz: 4 sec 1 Hz: 26 min	20 Hz: 84	1600	1
André‐Obadia 2008	M1 contralateral to painful side	Posteroanterior Medial‐lateral	20	90	20	4 sec	84	1600	1
André‐Obadia 2011	M1 hand area, not clearly reported but likely contralateral to painful side	Not specified	20	90	20	4 sec	84	1600	1
Avery 2013	Left DLPFC	Not specified	10	120	75	4	26	3000	15
Short 2011	Left DLPFC	Para‐sagittal	10	120	80	5 sec	10 sec	4000	10, x 1 daily (working days) for 2 weeks
Borckardt 2009	Left PFC	Not specified	10	100	40	10 sec	20	4000	3 over a 5‐day period
Carretero 2009	Right DLPFC	Not specified	1	110	20	60 sec	45	1200	Up to 20 on consecutive working days
Defrin 2007	M1 midline	Not specified	5	115	500	10 sec	30	? 500*	10, x 1 daily
Fregni 2005	Left and right SII	Not specified	1	90	Not specified	Not specified	Not specified	1600	1
Fregni 2011	Right SII	Not specified	1	70% maximum stimulator output intensity (not RMT)	1	Not specified	Not specified	1600	10, x 1 daily (week days only)
Hirayama 2006	M1, S1, PMA, SMA	Not specified	5	90	10	10 sec	50	500	1
Hosomi 2013	M1 corresponding to painful region	Not specified	5	90	10	10 sec	50	500	10, x 1 daily (week days only)
Irlbacher 2006	M1 contralateral to painful side	Not specified	5, 1	95	Not specified	Not specified	Not specified	500	1
Kang 2009	Right M1	45º postero‐lateral	10	80	20	5 sec	55	1000	5, x 1 daily
Khedr 2005	M1 contralateral to painful side	Not specified	20	80	10	10 sec	50	2000	5, x 1 daily
Lee 2012	Right DLPFC (low‐frequency) Left M1 (high‐frequency)	Not specified	10, 1	10 Hz: 80 1 Hz: 110	10 Hz:25 1 Hz: 2	10 Hz: 8 sec 1 Hz: 800 sec	10 Hz: 10 1 Hz: 60	10 Hz: 2000 1 Hz: 1600	10, x 1 daily (week days only)
Lefaucheur 2001a	M1 contralateral to painful side	Not specified	10	80	20	5 sec	55	1000	1
Lefaucheur 2001b	M1 contralateral to painful side	Posteroanterior	10, 0.5	80	10 Hz: 20 0.5 Hz: 1	10 Hz: 5 sec 0.5 Hz: 20 min	10 Hz: 55	10 Hz: 1000 0.5 Hz: 600	1
Lefaucheur 2004	M1 contralateral to painful side	Posteroanterior	10	80	20	5 sec	55	1000	1
Lefaucheur 2006	M1 contralateral to painful side	Posteroanterior	10, 1	90	10 Hz: 20 1 Hz: 1	10 Hz: 6 sec 1 Hz: 20 min	10 Hz: 54	10 Hz: 1200 1 Hz: 1200	1
Lefaucheur 2008	M1 contralateral to painful side	Posteroanterior	10, 1	90	10 Hz: 20 1 Hz: 1	10 Hz: 6 sec 1 Hz: 20 min	10 Hz: 54	10 Hz: 1200 1 Hz: 1200	1
Mhalla 2011	Left M1	Posteroanterior	10	80	15	10 sec	50	1500	14, 5 x 1 daily (working days), then 3 x 1 weekly, then 3 x 1 fortnightly, then 3 x 1 monthly
Onesti 2013	M1 deep central sulcus	H‐coil	20	100	30	2.5 sec	30	1500	5, x 1 daily on consecutive days
Passard 2007	M1 contralateral to painful side	Posteroanterior	10	80	25	8 sec	52	2000	10, x 1 daily (working days)
Picarelli 2010	M1 contralateral to painful side	Posteroanterior	10	100	25	10 sec	60	2500	10, x 1 daily (working days)
Pleger 2004	M1 hand area	Not specified	10	110	10	1.2 sec	10	120	1
Rollnik 2002	M1 midline	Not specified	20	80	20	2 sec	Not specified	800	1
Saitoh 2007	M1 over motor representation of painful area	Not specified	10, 5, 1	90	10 Hz; 5 5 Hz: 10 1 Hz: 1	10 Hz: 10 sec 5 Hz: 10 sec 1 Hz: 500 sec	10 Hz: 50 5 Hz: 50	500	1
Tzabazis 2013	Targeted to ACC	4‐coil configuration	1 Hz (10 Hz data excluded as not randomised)	110	Not reported	Not reported	Not reported	1800	20, x 1 daily (working days)

ACC: anterior cingulate cortex; DLPFC: dorsolateral prefrontal cortex; M1: primary motor cortex; PFC: prefrontal cortex; PMA: pre‐motor area; RMT: resting motor threshold; dS1: primary somatosensory cortex; SII: secondary somatosensory cortex; SMA: supplementary motor area

Stimulation location

The parameters for rTMS application varied significantly between studies including by site of stimulation, stimulation parameters and the number of stimulation sessions. The majority of rTMS studies targeted the primary motor cortex (M1) (Ahmed 2011; André‐Obadia 2006; André‐Obadia 2008; André‐Obadia 2011; Defrin 2007; Hirayama 2006; Hosomi 2013; Irlbacher 2006; Kang 2009; Khedr 2005; Lee 2012, Lefaucheur 2001a; Lefaucheur 2001b; Lefaucheur 2004; Lefaucheur 2006; Lefaucheur 2008; Mhalla 2011; Onesti 2013; Passard 2007; Picarelli 2010; Pleger 2004; Rollnik 2002; Saitoh 2007). Of these, one study specified stimulation of the right hemisphere (Kang 2009), one study specified the left hemisphere (Mhalla 2011), and two studies specified stimulation over the midline (Defrin 2007; Pleger 2004). One study used a novel H‐coil to stimulate the motor cortex of the leg representation situated deep in the central sulcus (Onesti 2013), and the remainder stimulated over the contralateral cortex to the side of dominant pain. One of these studies also investigated stimulation of the supplementary motor area (SMA), pre‐motor area (PMA) and primary somatosensory cortex (S1) (Hirayama 2006). Two studies stimulated the dorsolateral pre‐frontal cortex (DLPFC), with two studies stimulating the left hemisphere (Borckardt 2009; Short 2011), and two studies the right (Carretero 2009; Lee 2012). One study investigated stimulation of the left and right secondary somatosensory cortex (SII) as separate treatment conditions (Fregni 2005), and another investigated stimulation to the right SII area (Fregni 2011). One study used a four‐coil configuration to target the anterior cingulate cortex (Tzabazis 2013).

Stimulation parameters

Frequency

Eleven studies investigated low‐frequency (< 5 Hz) rTMS (André‐Obadia 2006; Carretero 2009; Fregni 2005; Fregni 2011; Irlbacher 2006; Lee 2012; Lefaucheur 2001b; Lefaucheur 2006; Lefaucheur 2008; Saitoh 2007; Tzabazis 2013). Of these, one study used a frequency of 0.5 Hz in one treatment condition (Lefaucheur 2001b), and the rest used a frequency of 1 Hz. Twenty‐seven studies investigated high‐frequency (≥ 5 Hz) rTMS (Ahmed 2011; André‐Obadia 2006; André‐Obadia 2008; André‐Obadia 2011; Avery 2013; Borckardt 2009; Defrin 2007; Fregni 2005; Hirayama 2006; Hosomi 2013; Irlbacher 2006; Kang 2009; Khedr 2005; Lee 2012; Lefaucheur 2001a; Lefaucheur 2001b; Lefaucheur 2004; Lefaucheur 2006; Lefaucheur 2008; Mhalla 2011; Onesti 2013; Passard 2007; Picarelli 2010; Pleger 2004; Rollnik 2002; Saitoh 2007; Short 2011). While the study by Tzabazis 2013 did apply high‐frequency stimulation to some participants, the allocation of the high‐frequency groups was not randomised in that study (confirmed through correspondence with authors) and so those data will not be considered further in this review as they do not meet our inclusion criteria.

Other parameters

We observed wide variation between studies for various stimulation parameters. The overall number of rTMS pulses delivered varied from 120 to 4000. The study by Defrin 2007 reported a total number of pulses of 500 although the reported stimulation parameters of 500 trains, delivered at a frequency of 5 Hz for 10 seconds would imply 25,000 pulses. Eight studies specified a posteroanterior or parasagittal orientation of the stimulating coil (André‐Obadia 2006; Lefaucheur 2001b; Lefaucheur 2004; Lefaucheur 2006; Lefaucheur 2008; Passard 2007; Picarelli 2010; Short 2011), two studies specified a coil orientation 45º to the midline (Ahmed 2011; Kang 2009), one study compared a posteroanterior coil orientation with a medial‐lateral coil orientation (André‐Obadia 2008), one used an H‐coil (Onesti 2013), one used a four‐coil configuration (Tzabazis 2013), and the remaining studies did not specify the orientation of the coil. Within studies that reported the information, the duration and number of trains and the inter‐train intervals varied. Two studies did not report this information (Fregni 2005; Fregni 2011).

Type of sham

rTMS studies employed a variety of sham controls. In 11 studies the stimulating coil was angled away from the scalp to prevent significant cortical stimulation (Ahmed 2011; André‐Obadia 2006; André‐Obadia 2008; Carretero 2009; Hirayama 2006; Kang 2009; Khedr 2005; Lee 2012; Pleger 2004; Rollnik 2002; Saitoh 2007), of which two studies also simultaneously electrically stimulated the skin of the scalp in both the active and sham stimulation conditions in order to mask the sensations elicited by active rTMS and thus preserve participants' blinding (Hirayama 2006; Saitoh 2007). The remaining studies utilised sham coils. Of these, eight studies specified that the sham coil made similar or identical sounds to those elicited during active stimulation (André‐Obadia 2011; Borckardt 2009; Defrin 2007; Irlbacher 2006; Mhalla 2011; Passard 2007; Picarelli 2010; Tzabazis 2013), and five specified that the sham coil made similar sounds, looked the same and elicited similar scalp sensations as the real coil (Avery 2013; Fregni 2011; Hosomi 2013; Onesti 2013; Short 2011). Six studies did not specify whether the sham coil controlled for the auditory characteristics of active stimulation (Fregni 2005; Lefaucheur 2001a; Lefaucheur 2001b; Lefaucheur 2004; Lefaucheur 2006; Lefaucheur 2008).

Studies of CES

See Table 2 for a summary of stimulation characteristics utilised in CES studies.

Table 2. CES studies ‐ characteristics of stimulation

Study	Electrode placement	Frequency (Hz)	Pulse width (msec)	Waveform shape	Intensity	Duration (min)	Treatment sessions per group
Capel 2003	Ear clip electrodes	10	2	Not specified	12 μA	53	x 2 daily for 4 days
Cork 2004	Ear clip electrodes	0.5	Not specified	Modified square wave biphasic	100 μA	60	? daily for 3 weeks
Gabis 2003	Mastoid processes and forehead	77	3.3	Biphasic asymmetric	≤ 4 mA	30	x 1 daily for 8 days
Gabis 2009	Mastoid processes and forehead	77	3.3	Biphasic asymmetric	≤ 4 mA	30	x 1 daily for 8 days
Katsnelson 2004	Mastoid processes and forehead	Not specified	Not specified	2 conditions: symmetric, asymmetric	11 to 15 mA	40	x 1 daily for 5 days
Lichtbroun 2001	Ear clip electrodes	0.5	Not specified	Biphasic square wave	100 μA	60	x 1 daily for 30 days
Rintala 2010	Ear clip electrodes	Not specified	Not specified	Not specified	100 μA	40	x 1 daily for 6 weeks
Tan 2000	Ear clip electrodes	0.5	Not specified	Not specified	10 to 600 μA	20	12 (timing not specified)
Tan 2006	Ear clip electrodes	Not specified	Not specified	Not specified	100 to 500 μA	60	x 1 daily for 21 days
Tan 2011	Ear clip electrodes	Not specified	Not specified	Not specified	100 μA	60	x 1 daily for 21 days
Taylor 2013	Ear clip electrodes	0.5	Not specified	Modified square‐wave biphasic	100 μA	60	x 1 daily for 8 weeks

Stimulation device, parameters and electrode location

Seven studies of CES used the 'Alpha‐stim' CES device (Electromedical Products International, Inc, Mineral Wells, Texas, USA). This device uses two ear clip electrodes that attach to each of the participant's ears (Cork 2004; Lichtbroun 2001; Rintala 2010; Tan 2000; Tan 2006; Tan 2011; Taylor 2013), and these studies utilised stimulation intensities of 100 μA with a frequency of 0.5 Hz. One study (Capel 2003) used a device manufactured by Carex (Hemel Hempstead, UK) that also used earpiece electrodes and delivered a stimulus intensity of 12 μA.

Two studies used the 'Pulsatilla 1000' device (Pulse Mazor Instruments, Rehavol, Israel) (Gabis 2003; Gabis 2009). The electrode array for this device involved an electrode attached to each of the participant's mastoid processes and one attached to the forehead; current is passed to the mastoid electrodes. One study used the 'Nexalin' device (Kalaco Scientific Inc, Scottsdale, AZ, USA) (Katsnelson 2004). With this device current is applied to a forehead electrode and returned via electrodes placed behind the patient's ears. These three studies utilised significantly higher current intensities than those using ear clip electrodes with intensities of 4 mA (Gabis 2003; Gabis 2009), and 11 to 15 mA (Katsnelson 2004).

All CES studies gave multiple treatment sessions for each treatment group with variation between the number of treatments delivered.

Type of sham

Eight studies utilised inert sham units (Capel 2003; Cork 2004; Lichtbroun 2001; Rintala 2010; Tan 2000; Tan 2006; Tan 2011; Taylor 2013). These units were visually indistinguishable from the active devices. Stimulation at the intensities used is subsensation and as such it should not have been possible for participants to distinguish between the active and sham conditions.

Two studies utilised an "active placebo" treatment unit (Gabis 2003; Gabis 2009). This sham device was visually indistinguishable and delivered a current of much lower intensity (≤ 0.75 mA) than the active stimulator to evoke a similar sensation to ensure patient blinding. Similarly, Katsnelson 2004 utilised a visually indistinguishable sham device that delivered brief pulses of current of < 1 mA. The placebo conditions used in these three studies delivered current at much greater intensities than those used in the active stimulation conditions of the other CES studies.

Studies of tDCS

See Table 3 for a summary of stimulation characteristics utilised in tDCS studies.

See: Summary of findings for the main comparison

Table 3. tDCS studies ‐ characteristics of stimulation

Study	Location of stimulation	Electrode pad size	Intensity (mA)	Anodal or cathodal?	Stimulus duration (min)	Treatment sessions per group
Antal 2010	M1 left hand area	35 cm²	1 mA	Anodal	20	5, x 1 daily
Boggio 2009	M1 contralateral to painful side	35 cm²	2 mA	Anodal	30	1
Fenton 2009	M1 dominant hemisphere	35 cm²	1 mA	Anodal	20	2
Fregni 2006a	M1 contralateral to painful side or dominant hand	35 cm²	2 mA	Anodal	20	5, x 1 daily
Fregni 2006b	M1 and DLPFC contralateral to painful side or dominant hand	35 cm²	2 mA	Anodal	20	5, x 1 daily
Jensen 2013	M1 left	35cm²	2 mA	Anodal	20	1
Mendonca 2011	Group 1: anodal left M1 Group 2: cathodal left M1 Group 3: anodal supraorbital Group 4: cathodal supraorbital Group 5: sham	35 cm²	2 mA	Anodal or cathodal	20	1
Mori 2010	M1 contralateral to painful side	35 cm²	2 mA	Anodal	20	5, x 1 daily
Portilla 2013	M1 contralateral to painful side	35 cm²	2 mA	Anodal	20	x 1 per condition
Riberto 2011	M1 contralateral to painful side or dominant hand	35 cm²	2 mA	Anodal	20	10, x 1 weekly
Soler 2010	M1 contralateral to painful side or dominant hand	35 cm²	2 mA	Anodal	20	10, x 1 daily (week days only)
Valle 2009	M1 and DLPFC contralateral to painful side or dominant hand	35 cm²	2 mA	Anodal	20	5, x 1 daily
Villamar 2013	M1 left	HD‐tDCS 4 x 1‐ring montage	2 mA	Anodal or cathodal	20	x 1 per condition
Wrigley 2014	M1 contralateral to painful side or dominant hand	35 cm²	2 mA	Anodal	20	5, x 1 daily

DLPFC: dorsolateral prefrontal cortex; M1: primary motor cortex

HD‐tDCS: High definition tDCS

Stimulation parameters and electrode location

Two studies of tDCS stimulated the dorsolateral prefrontal cortex in one treatment group (Fregni 2006b; Valle 2009). Thirteen studies stimulated the motor cortex (Antal 2010; Boggio 2009; Fenton 2009; Fregni 2006a; Fregni 2006b; Jensen 2013; Mori 2010; Portilla 2013; Riberto 2011; Soler 2010; Valle 2009; Villamar 2013; Wrigley 2014). Of these, nine stimulated the cortex contralateral to the side of worst pain (Boggio 2009; Fregni 2006a; Fregni 2006b; Mori 2010; Portilla 2013; Riberto 2011; Soler 2010; Villamar 2013; Wrigley 2014), of which six studies stimulated the opposite hemisphere to the dominant hand where pain did not have a unilateral dominance (Fregni 2006a; Fregni 2006b; Jensen 2013; Riberto 2011; Soler 2010; Wrigley 2014). Three studies stimulated the left hemisphere for all participants (Antal 2010; Valle 2009; Villamar 2013). One study of chronic pelvic pain stimulated the opposite hemisphere to the dominant hand in all participants (Fenton 2009). One study specifically investigated the use of tDCS in conjunction with transcutaneous electrical nerve stimulation (TENS) therapy (Boggio 2009). We extracted data comparing active tDCS and sham TENS with sham tDCS and sham TENS for the purposes of this review. One applied anodal or cathodal stimulation to the left motor cortex or to the right supraorbital area (Mendonca 2011).

Six studies delivered a current intensity of 2 mA for 20 minutes once a day for five days (Antal 2010; Fregni 2006a; Fregni 2006b; Mori 2010; Valle 2009; Wrigley 2014). One study applied a current intensity of 1 mA once a day for two days (Fenton 2009), and four studies applied one treatment per stimulation condition at an intensity of 2 mA for 20 minutes (Boggio 2009; Mendonca 2011; Jensen 2013; Villamar 2013). One study delivered 10 stimulation sessions of 20 minutes at 2 mA once weekly for 10 weeks (Riberto 2011), and another delivered 10 sessions once a day, with a visual illusion condition or a sham visual illusion condition for 10 consecutive weekdays (Soler 2010).

All studies of tDCS utilised a sham condition whereby active stimulation was ceased after 30 seconds without the participants' knowledge.

Excluded studies

See Characteristics of excluded studies.

In our original review we excluded 11 studies after consideration of the full study report. Of these, one was not a study of brain stimulation (Frentzel 1989), two did not assess self reported pain as an outcome (Belci 2004; Johnson 2006), four were not restricted to participants with chronic pain (Evtiukhin 1998; Katz 1991; Longobardi 1989; Pujol 1998), one study was unclear on the duration of participants' symptoms (Avery 2007), two were single case studies (Silva 2007; Zaghi 2009), one study presented duplicate data from a study already accepted for inclusion (Roizenblatt 2007, duplicate data from Fregni 2006b), and one did not employ a sham control (Evtiukhin 1998).

For this update we excluded a further 17 reports, after consideration of the full study report. Nine reports referred to studies which had already been included in the previous version of this review, one was not a study of brain stimulation (Carraro 2010), two were not clearly in a chronic pain population (Choi 2012a; Choi 2012b), one was not a randomised controlled trial (O'Connell 2013), one reported uncontrolled long‐term follow‐up data from an included study (Hargrove 2012a), one employed an intervention that was not designed to alter cortical activity directly through electrical stimulation (Nelson 2010), and one included some participants who did not meet our criterion of chronic pain (Bolognini 2013). A final study was screened by a Russian translator and excluded on the basis that it did not employ a sham control for tDCS (Sichinava 2012). Finally one citation referred to a booklet of conference proceedings which contained no relevant citations.

Risk of bias in included studies

Risk of bias varied across studies for all of the assessment criteria. For a summary of 'Risk of bias' assessment across studies see Figure 2.

Figure 2

Methodological quality summary: review authors' judgements about each methodological quality item for each included study.

Sequence generation

For the criterion 'adequate sequence generation' we awarded cross‐over trials a judgement of 'low risk of bias' where the study report mentioned that the order of treatment conditions was randomised. Since this criterion has a greater potential to introduce bias in parallel designs we only awarded a judgement of 'low risk of bias' where the method of randomisation was specified and adequate.

We judged 14 trials as having an unclear risk of bias (Antal 2010; Carretero 2009; Cork 2004; Defrin 2007; Hargrove 2012; Katsnelson 2004; Lee 2012; Mendonca 2011; Picarelli 2010; Riberto 2011; Rintala 2010; Tan 2006; Taylor 2013; Tzabazis 2013), as they did not specify the method of randomisation used or the description was not clear. We judged two studies as having a high risk of bias for this criterion (Ahmed 2011; Khedr 2005), as the reports suggested that patients were allocated depending on the day of the week on which they were recruited, which we did not judge as being genuinely random.

Allocation concealment

We only considered for the criterion 'Adequate concealment of allocation' studies with parallel designs or from which only data from the first phase of the study were included (i.e. we them considered as parallel studies). Seventeen studies did not report concealment of allocation and we judged them as 'unclear' (Antal 2010; Carretero 2009; Cork 2004; Defrin 2007; Fregni 2011; Hargrove 2012; Katsnelson 2004; Lee 2012; Mendonca 2011; Passard 2007; Picarelli 2010; Riberto 2011; Rintala 2010; Soler 2010; Tan 2006; Taylor 2013; Tzabazis 2013), and we judged two studies as having a high risk of bias for this criterion since the method of randomisation employed would not have supported concealment of allocation (Ahmed 2011; Khedr 2005).

Blinding

Blinding of assessors

While many studies used self reported pain outcomes we considered that the complex nature of the intervention, and the level of interaction this entails between participants and assessors, suggested that a lack of blinding of the researchers engaged in the collection of outcomes might potentially introduce bias. As such, where blinding of assessors was not clearly stated we made a judgement of 'unclear' for this criterion.

Sixteen studies did not specify whether they blinded outcome assessors (André‐Obadia 2011; Borckardt 2009; Hirayama 2006; Irlbacher 2006; Lee 2012; Lefaucheur 2001a; Lefaucheur 2001b; Lefaucheur 2004; Lefaucheur 2006; Onesti 2013; Picarelli 2010; Pleger 2004; Rollnik 2002; Saitoh 2007; Tan 2000; Tzabazis 2013), while we judged the majority of studies of tDCS at unclear risk of bias on this criterion (Boggio 2009; Fregni 2006a; Fregni 2006b; Jensen 2013; Mori 2010; Portilla 2013; Riberto 2011; Soler 2010; Valle 2009; Villamar 2013; Wrigley 2014), since there is evidence that assessor blinding may be compromised at the stimulation intensities used (O'Connell 2012).

Blinding of participants

rTMS studies

All studies attempted to blind participants. However, due to the difficulties involved in producing a robust sham control in rTMS studies (see Assessment of risk of bias in included studies) we made an assessment of sham credibility. Where the coil was angulated or angulated and elevated away from the scalp, this is potentially distinguishable both visually and by the sensory effects of stimulation. Two studies simultaneously electrically stimulated the scalp during rTMS stimulation to mask the differences in sensation between conditions (Hirayama 2006; Saitoh 2007). However, by angulating the coil away from the scalp participants may have been able to visually distinguish between the conditions. Where sham coils were utilised they usually did not control for the sensory aspects of stimulation. We assessed most rTMS studies as having sub‐optimal sham control conditions and we therefore assessed them as having an 'unclear' risk of bias. Four rTMS studies included in this update utilised modern sham coils that are visually indistinguishable, emit the same noise during stimulation and elicit similar scalp sensations (Avery 2013; Fregni 2011; Onesti 2013; Short 2011). These studies met the criteria for an optimal sham condition and as such we judged them at low risk of bias for participant blinding.

Similarly with tDCS studies, due to evidence that blinding of participants to the stimulation condition may be compromised at intensities of 1.5 mA and above, we judged the majority of tDCS studies at unclear risk of bias on this criterion (Boggio 2009; Fregni 2006a; Fregni 2006b; Jensen 2013; Mori 2010; Portilla 2013; Riberto 2011; Soler 2010; Valle 2009; Villamar 2013; Wrigley 2014).

We assessed all studies of CES as having a low risk of bias for this criterion.

Incomplete outcome data

We assessed 11 studies as having an unclear risk of bias for this criterion (Ahmed 2011; André‐Obadia 2006; André‐Obadia 2011; Boggio 2009; Cork 2004; Fregni 2011; Hargrove 2012; Katsnelson 2004; Lefaucheur 2006; Lichtbroun 2001; Tzabazis 2013). Ahmed 2011 and Fregni 2011 did not report the level of drop‐out from their studies. In the study of André‐Obadia 2006, two participants (17% of the study cohort) did not complete the study and this was not clearly accounted for in the data analysis. This was also the case for Boggio 2009, where two participants (25% of the cohort) failed to complete the study. Five studies did not clearly report levels of drop‐out (Cork 2004; Katsnelson 2004; Lefaucheur 2006; Lichtbroun 2001; Tzabazis 2013), of which one reported recruiting 16 participants in the full study report (Tzabazis 2013), but an earlier abstract report of the same study reported the recruitment of 45 participants (Schneider 2012). We assessed three studies as having a high risk of bias for this criterion (Antal 2010; Irlbacher 2006; Tan 2000). In the Antal 2010 study, of 23 participants recruited only 12 completed the full cross‐over. In the study by Irlbacher 2006, only 13 of the initial 27 participants completed all of the treatment conditions. In the studies of Lee 2012 and Rintala 2010, attrition exceeded 30% of the randomised cohort. In the study by Tan 2000, 17 participants did not complete the study (61% of the cohort) and this was not clearly accounted for in the analysis. We considered this level of withdrawal unsustainable.

Selective reporting

We assessed studies as having a high risk of bias for this criterion where the study report did not produce adequate data to assess the effect size for all groups/conditions, and these data were not made available upon request. We assessed 11 studies as having a high risk of bias for this criterion (Capel 2003; Cork 2004; Fregni 2005; Fregni 2011; Katsnelson 2004; Lichtbroun 2001; Mendonca 2011; Onesti 2013; Portilla 2013; Tzabazis 2013; Valle 2009). We judged two studies as being at unclear risk of bias (Fregni 2006a; Fregni 2006b). In the reports of these studies data were not presented in a format that could be easily interpreted. On request data were available from these two studies for the primary outcome at baseline and short‐term follow‐up but not for other follow‐up points. We assessed the remaining studies as having a low risk of bias for this criterion. For this update, we first made requests for data (by email where possible) in February 2013, with repeat emails sent where necessary in March, April and June 2013. For studies identified in the second round of searches we made requests in June 2013 and we made the final round of requests on 1 August 2013. If these data are made available in time for future updates then we can revise judgements on this criterion accordingly.

Study size

We rated three studies at unclear risk of bias (Hosomi 2013; Lefaucheur 2004; Tan 2011), with all remaining studies rated at high risk of bias on this criterion.

Study duration

We rated seven studies at low risk of bias on this criterion (Ahmed 2011; Avery 2013; Gabis 2009; Mhalla 2011; Passard 2007; Picarelli 2010; Valle 2009), 19 studies at unclear risk of bias (André‐Obadia 2008; André‐Obadia 2011; Antal 2010; Borckardt 2009; Carretero 2009; Defrin 2007; Fenton 2009; Fregni 2006a; Fregni 2006b; Fregni 2011; Hosomi 2013; Kang 2009; Khedr 2005; Lee 2012; Mori 2010; Onesti 2013; Soler 2010; Tzabazis 2013; Wrigley 2014), and the remaining studies at high risk of bias (André‐Obadia 2006; Boggio 2009; Capel 2003; Cork 2004; Fregni 2005; Gabis 2003; Hargrove 2012; Hirayama 2006; Irlbacher 2006; Jensen 2013; Katsnelson 2004; Lefaucheur 2001a; Lefaucheur 2001b; Lefaucheur 2004; Lefaucheur 2006; Lefaucheur 2008; Lichtbroun 2001; Mendonca 2011; Pleger 2004; Portilla 2013; Riberto 2011; Rintala 2010; Rollnik 2002; Saitoh 2007; Short 2011; Tan 2000; Tan 2006; Tan 2011; Taylor 2013; Villamar 2013) .

Other potential sources of bias

Carry‐over effects in cross‐over trials

We judged one study as unclear on this criterion as no pre‐stimulation data were provided and no investigation of carry‐over effects was discussed in the study report (Fenton 2009). In one cross‐over study baseline differences between the sham and the 10 Hz stimulation condition were notable (Saitoh 2007). A paired t‐test did not show a significant difference (P > 0.1) and we judged this study as having a low risk of bias for carry‐over effects. We judged another study at unclear risk of bias on this criterion as the necessary data were not available in the study report from which to make a judgement (Portilla 2013).

Other sources of bias

Two studies did not present baseline data for key outcome variables and we judged them as 'unclear' (Fregni 2011; Tzabazis 2013). Three studies demonstrated baseline imbalances: one study on pain intensity levels (Defrin 2007), one study on Brief Pain Inventory pain interference, SF‐36 pain sub‐scale and coping strategies (Tan 2011) and one study on duration of pain, education, age and economic activity (Riberto 2011). We judged these studies at unclear risk of bias for these reasons. One study of CES did not clearly present relevant baseline group characteristics of the included participants and we judged it as being at high risk of bias for this criterion (Katsnelson 2004). One study of CES also applied electrical stimulation to the painful body area as part of the treatment, which may have affected the final outcomes (Tan 2000). Two studies of CES used an "active placebo condition" that delivered a level of cortical stimulation that was greater than that used in the active arm of other CES studies (Gabis 2003; Gabis 2009). It is possible that delivering cortical stimulation in the sham group might mask differences between the sham and active condition. Also such a large difference in current intensity compared with other studies of CES might be a source of heterogeneity. We judged these three studies as 'unclear' on this criterion. We judged one study at high risk of bias on this criterion due to imbalances between the groups at baseline on the duration of pain, education, age and economic activity (Riberto 2011).

Effects of interventions

For a summary of all core findings see summary of findings Table for the main comparison.

Primary outcome: pain

Repetitive transcranial magnetic stimulation (rTMS) for short‐term relief of chronic pain

The primary meta‐analysis (Analysis 1.1) pooled data from all rTMS studies with low or unclear risk of bias (excluding the risk of bias criteria 'study size' and 'study duration') where data were available (n = 528), including cross‐over and parallel designs, using the generic inverse variance method (André‐Obadia 2006; André‐Obadia 2008; André‐Obadia 2011; Borckardt 2009; Carretero 2009; Defrin 2007; Hirayama 2006; Hosomi 2013; Kang 2009; Lefaucheur 2001a; Lefaucheur 2001b; Lefaucheur 2004; Lefaucheur 2006; Lefaucheur 2008; Mhalla 2011; Passard 2007; Pleger 2004; Pleger 2004; Rollnik 2002; Saitoh 2007; Short 2011). We excluded the studies by Ahmed 2011, Khedr 2005, Irlbacher 2006 and Lee 2012, as we classified them as having a high risk of bias on at least one criterion. We were unable to include data from five studies (Fregni 2005; Fregni 2011; Onesti 2013; Picarelli 2010; Tzabazis 2013 combined n = 86), as the necessary data were not available in the study report or upon request by the submission date of this update. We imputed the correlation coefficient used to calculate the standard error (SE) (standardised mean difference (SMD)) for cross‐over studies (0.764) from data extracted from André‐Obadia 2008 (as outlined in Unit of analysis issues) and we entered the SMD (SE) for each study into a generic inverse variance meta‐analysis. We divided the number of participants in each cross‐over study by the number of comparisons made by that study entered into the meta‐analysis. For parallel studies we calculated the standard error of the mean (SEM) from the 95% confidence intervals of the standardised mean difference (SMD) and entered both the SMD and the SEM into the meta‐analysis. We then entered this into the meta‐analysis with the SMD using the generic inverse variance method.

We observed substantial heterogeneity (I² = 67%, P < 0.01) and investigated this using pre‐planned subgroup analysis. Categorising studies by high (≥ 5 Hz) or low (< 5 Hz) frequency rTMS demonstrated a significant difference between subgroups (P < 0.01) and reduced heterogeneity in the low‐frequency group (n = 81, I² = 0%). In this group there was no evidence of an effect of low‐frequency rTMS for short‐term relief of chronic pain (SMD 0.15, 95% confidence interval (CI) ‐0.01 to 0.3, P = 0.07). While high‐frequency stimulation demonstrated a significant effect (SMD ‐0.27, 95% CI ‐0.35 to ‐0.20, P < 0.01), we observed substantial heterogeneity in this (n = 447, I² = 64%). Separating studies that delivered a single treatment per condition from those that delivered multiple treatment sessions did not reduce heterogeneity substantially in multiple‐dose studies (n = 225, I² = 75%) or single‐dose studies (n = 303, I² = 61%) (Analysis 1.2).

There were insufficient data to support the planned subgroup analysis by the type of painful condition as planned. However, when the analysis was restricted to studies including only well‐defined neuropathic pain populations (Analysis 1.3 excluding Carretero 2009; Mhalla 2011; Passard 2007; Pleger 2004; Rollnik 2002; Short 2011), there was little impact on heterogeneity (I² = 71% P < 0.01). In the subgroup of non‐neuropathic pain studies overall heterogeneity remained significant and high (I² = 56%, P = 0.04) (Analysis 1.4).

rTMS motor cortex

Restricting the analysis to single‐dose studies of high‐frequency stimulation of the motor cortex (n = 233) reduced heterogeneity (I² = 31%, P = 0.13) (Analysis 1.5). In this group the pooled SMD was ‐0.39 (95% confidence interval (CI) ‐0.51 to ‐0.27, P < 0.01). We back‐transformed the SMD to a mean difference using the mean standard deviation of the post‐treatment sham group score of the studies included in this analysis (1.87). We then used this to estimate the real percentage change on a 0 to 100 mm visual analogue scale (VAS) of active stimulation compared with the mean post‐stimulation score from the sham groups of the included studies (6.2). This equated to a reduction of 7.3 mm (95% CI 5 mm to 9.5 mm), or a percentage change of 12% (95% CI 8% to 15%) of the control group outcome. This estimate does not reach the pre‐established criteria for a minimal clinically important difference (≥ 15%). Of the included studies in this subgroup, nine did not clearly report blinding of assessors and we awarded them a judgement of 'unclear' risk of bias for this criterion (André‐Obadia 2011; Hirayama 2006; Lefaucheur 2001a; Lefaucheur 2001b; Lefaucheur 2004; Lefaucheur 2006; Pleger 2004; Rollnik 2002; Saitoh 2007). Sensitivity analysis removing these studies reduced heterogeneity to I² = 0% although only three studies were preserved in the analysis (André‐Obadia 2006; André‐Obadia 2008; Lefaucheur 2008). There remained a statistically significant difference between sham and active stimulation although the SMD reduced to ‐0.31 (95% CI ‐0.49 to ‐0.13). This equates to a percentage change of 9% (95% CI 4% to 15%) in comparison with sham stimulation. For multiple‐dose studies of high‐frequency motor cortex stimulation heterogeneity was high (n = 157, I² = 71%, P < 0.01), but the pooled effect was not significant (SMD ‐0.07, 95% CI ‐0.41 to 0.26, P = 0.68).

When the analysis was restricted to studies of single‐dose, high‐frequency motor cortex stimulation in well‐defined neuropathic pain populations (excluding data from Pleger 2004; Rollnik 2002), there was little effect on the pooled estimate (SMD ‐0.43, 95% CI ‐0.57 to ‐0.30) or heterogeneity (I² = 31%, not significant). When we applied the same process to multiple‐dose studies of high‐frequency motor cortex stimulation (excluding data from Passard 2007) heterogeneity remained high (I² = 62%, P = 0.03) with no significant pooled effect.

Sensitivity analysis

To assess whether the imputation of standard errors for cross‐over studies was robust we repeated the analysis with the correlation coefficient reduced to 0.66 and increased to 0.86. This had no marked effect on the overall analysis (Analysis 1.6; Analysis 1.7). The same process was applied to the subgroup analysis of single‐dose studies of high‐frequency motor cortex stimulation (Analysis 1.8; Analysis 1.9). This had a negligible impact on the effect size or the statistical significance of this subgroup.

To assess the impact of excluding the studies of Ahmed 2011, Irlbacher 2006, Khedr 2005 and Lee 2012, we performed the analysis with data from these studies included (Analysis 1.10). While this produced a modest increase in the SMD it increased heterogeneity from 69% to 74%. Inclusion of Ahmed 2011, Khedr 2005 and Lee 2012 to the multiple‐dose studies of high‐frequency motor cortex stimulation subgroup increased heterogeneity (I² = 88%, P < 0.01), though the subgroup demonstrated an effect that approached statistical significance (SMD ‐0.50, 95% CI ‐0.99 to ‐0.01, P = 0.05) (Analysis 1.11). Inclusion of the Irlbacher 2006 study in the single‐dose studies of high‐frequency motor cortex stimulation subgroup caused a slight decrease in the pooled effect size (SMD ‐0.36, 95% CI ‐0.48 to ‐0.24) with no impact on heterogeneity.

Small study effects/publication bias

We investigated small study effects using Egger's test. The results are not suggestive of a significant influence of small study effects.

rTMS prefrontal cortex

Restricting the analysis to studies that stimulated the dorsolateral pre‐frontal cortex (DLPFC) included four studies (n = 68) (Avery 2013; Borckardt 2009; Carretero 2009; Short 2011) (Analysis 1.12). We excluded the study by Lee 2012 due to its high risk of bias. The pooled effect was non‐significant (P = 0.36) with substantial heterogeneity (I² = 82%, P < 0.01). Restricting the analysis to high‐frequency studies (Avery 2013; Borckardt 2009; Short 2011), the effect remained non‐significant (P = 0.33) with high heterogeneity (I² = 85%, P < 0.01). The only remaining low‐frequency study (Carretero 2009, n = 26) was not suggestive of a significant effect (SMD 0.16, 95% CI ‐0.29 to 0.61). It is worthy of note that the only study in the analysis which individually demonstrated a significant effect was very small (n = 4) and its removal from the analysis makes heterogeneity non‐significant (Borckardt 2009).

Sensitivity analysis

To assess the impact of excluding the study of Lee 2012, we performed the analysis with data from this study included (Analysis 1.13). The overall effect remained non‐significant (P = 0.27) with high heterogeneity (I² = 76%, P < 0.01). Restricting this to low‐frequency studies (Carretero 2009; Lee 2012) brought heterogeneity down to a non‐significant level (I² = 16%, P = 0.28), though the effect remained non‐significant. Restricting the analysis to high‐frequency studies (Borckardt 2009; Lee 2012; Short 2011), the effect remained non‐significant (P = 0.25) though heterogeneity remained high (I² = 74%, P < 0.01). Restricting the analysis to low‐frequency studies (Carretero 2009; Lee 2012), the effect remained non‐significant (P = 0.92) with no heterogeneity (I² = 16%, P = 0.28).

rTMS for medium‐term relief of chronic pain (< 6 weeks post‐treatment)

Seven studies provided data on medium‐term pain outcomes (Avery 2013; Carretero 2009; Hosomi 2013; Lefaucheur 2001a; Kang 2009; Passard 2007; Short 2011). We excluded the studies by Ahmed 2011, Khedr 2005 and Lee 2012 as we classified them as having a high risk of bias. The analysis included 184 participants (Analysis 1.14). Overall heterogeneity was high (I² = 57%, P = 0.02) and no significant effect was observed (SMD ‐0.18, 95% CI ‐0.43 to 0.06, P = 0.15). Restricting the analysis to studies of prefrontal cortex stimulation (Avery 2013; Carretero 2009; Short 2011) demonstrated no significant effect (SMD ‐0.03, 95% CI ‐0.52 to 0.35). Studies of motor cortex stimulation also demonstrated no significant effect (SMD ‐0.22, 95% CI ‐0.52 to 0.07, P = 0.14) although heterogeneity was high (I² = 72%, P < 0.01). We performed sensitivity analysis to assess the impact of excluding the studies by Ahmed 2011, Khedr 2005 and Lee 2012 on the basis of risk of bias (Analysis 1.15). Including these studies increased heterogeneity (I² = 76%, P < 0.01) though the effect reached significance overall (SMD ‐0.43, 95% CI ‐0.76 to ‐0.10) and specifically for high‐frequency studies (SMD ‐0.48, 95% CI ‐0.83 to ‐0.13) (I² = 79%, P < 0.01).

rTMS for long‐term relief of chronic pain (≥ 6 weeks post‐treatment)

Three studies provided data for long‐term pain relief (Avery 2013; Kang 2009; Passard 2007) (Analysis 1.16). The analysis included 59 participants. There was no heterogeneity (I² = 0%, P = 0.95). The analysis demonstrated no significant effect (SMD ‐0.12, 95% CI ‐0.46 to 0.21, P = 0.47). Sensitivity analysis to assess the impact of excluding the study of Ahmed 2011 due to its high risk of bias continued to demonstrate no significant effect, though heterogeneity was introduced (Analysis 1.17, I² = 68%, P = 0.03).

Cranial electrotherapy stimulation (CES) for short‐term pain relief

Six studies provided data for this analysis (Gabis 2003; Gabis 2009; Rintala 2010; Tan 2006; Tan 2011; Taylor 2013) (Analysis 2.1, n = 270). We excluded the study by Rintala 2010 due to high risk of attrition bias. All studies utilised a parallel‐group design and so we used a standard inverse variance meta‐analysis using SMD. Four studies did not provide the necessary data to enter into the analysis (Capel 2003; Cork 2004; Katsnelson 2004; Lichtbroun 2001, combined n = 228) and we classified two studies as being at high risk of bias on criteria other than 'free of selective outcome reporting' (Katsnelson 2004; Tan 2000). The studies by Gabis 2003 and Gabis 2009 differ substantially from the other included studies on the location of electrodes and the intensity of the current provided. Despite this, there was no heterogeneity (I² = 0%). No individual study in this analysis demonstrates superiority of active stimulation over sham and the results of the meta‐analysis do not demonstrate statistical significance (SMD ‐0.24, 95% CI ‐0.48 to 0.01, P = 0.06). Sensitivity analysis, including the study by Rintala 2010, did not meaningfully affect the results (SMD ‐0.21, 95% CI ‐0.45 to 0.02, P = 0.07).

There were insufficient data to perform a meta‐analysis for medium or long‐term pain outcomes for CES.

Transcranial direct current stimulation (tDCS) for short‐term pain relief

Adequate data were available from 11 studies (Antal 2010; Boggio 2009; Fenton 2009; Fregni 2006a; Fregni 2006b; Jensen 2013; Mori 2010; Riberto 2011; Soler 2010; Villamar 2013; Wrigley 2014) for this analysis (n = 193). We were unable to include data from Mendonca 2011 and Valle 2009 (combined n = 71) as the necessary data were not reported in the study report or available upon request to the authors. We only included first‐stage data from the study of Antal 2010 (n = 12) due to the unsustainable level of attrition following this stage. We analysed data using the generic inverse variance method. We imputed the correlation coefficient (0.635) used to calculate the SE (SMD) for cross‐over studies from data extracted from Boggio 2009 (see Unit of analysis issues). One study compared two distinct active stimulation conditions to one sham condition (Fregni 2006b). Combining the treatment conditions was considered inappropriate as each involved stimulation of different locations and combination would hinder subgroup analysis. Instead we included both comparisons separately with the number of participants in the sham control group divided by the number of comparisons. The overall meta‐analysis did not demonstrate a significant effect of active stimulation (SMD ‐0.18, 95% CI ‐0.46 to 0.09, P = 0.19) (Analysis 3.1), but heterogeneity was significant (I² = 49%, P = 0.02). Subgrouping studies by multiple or single dose did not demonstrate a significant subgroup difference (test for subgroup differences P = 0.89) and decreased heterogeneity in the single‐dose subgroup (I² = 0%, P = 0.53) but increased heterogeneity in the multiple‐dose subgroup (I² = 62%, P < 0.01). Analysis restricted to comparisons of active motor cortex stimulation (single and multiple‐dose studies (n = 183, Analysis 3.2) reduced heterogeneity substantially (I² = 33%, P = 0.13) but did not demonstrate a statistically significant effect (SMD ‐0.23, 95% CI ‐0.48 to 0.01, P = 0.06). This lack of effect was consistent for the subgroups of single‐dose studies (SMD ‐0.18, 95% CI ‐0.41 to 0.05, P = 0.13) and multiple‐dose studies (SMD ‐0.35, 95% CI ‐0.79 to 0.09, P = 0.12).

To assess whether the imputation of standard errors for cross‐over studies was robust we repeated the analyses with the imputed correlation coefficient reduced and increased by a value of 0.1 (Analysis 3.3; Analysis 3.4; Analysis 3.5; Analysis 3.6). When the correlation was decreased the analysis including both single and multiple‐dose studies of motor cortex tDCS stimulation only approached, but did not reach, statistical significance (SMD ‐0.24, 95% CI ‐0.48 to 0.00, P = 0.05).

Small study effects/publication bias

We investigated small study effects using Egger's test. The results are not suggestive of a significant influence of small study effects.

tDCS for medium‐term pain relief (1 to < 6 weeks post‐treatment)

Five studies provided adequate data for this analysis (Antal 2010; Fenton 2009; Mori 2010; Soler 2010, Wrigley 2014, pooled n = 87) (Analysis 3.7). There was no significant heterogeneity (I² = 31%, P = 0.21) and the pooled effect was not statistically significant (SMD ‐0.20, 95% CI ‐0.63 to 0.24, P = 0.37).

Reduced impedance non‐invasive cortical electrostimulation (RINCE) for short‐term pain relief

The one study that investigated RINCE stimulation demonstrated a positive effect on pain (mean difference (0 to 10 pain scale) ‐1.41, 95% CI ‐2.48 to ‐0.34, P < 0.01) (Analysis 4.1; Hargrove 2012).

Secondary outcomes: disability and quality of life

rTMS for disability/pain interference: short‐term follow‐up

Five studies provided data on disability/pain interference at short‐term follow‐up (Avery 2013; Kang 2009; Mhalla 2011; Passard 2007; Short 2011). Pooling of these studies (Analysis 1.18; n = 119) demonstrated no significant effect on pain interference (SMD ‐0.29, 95% CI ‐0.87 to 0.29, P = 0.33) with substantial heterogeneity (I² = 71%, P < 0.01). All of these studies delivered multiple doses of high‐frequency stimulation. Two studies stimulated the DLPFC (Avery 2013; Short 2011) and three stimulated the motor cortex (Kang 2009; Mhalla 2011; Passard 2007). Subgrouping studies by stimulation site had no impact on heterogeneity.

rTMS for disability/pain interference: medium‐term follow‐up (1 to < 6 weeks post‐treatment)

Four studies provided data on disability/pain interference at medium‐term follow‐up (Avery 2013; Kang 2009; Mhalla 2011; Passard 2007). Pooling of these studies (Analysis 1.19; n = 99) demonstrated no significant effect (SMD ‐0.37, 95% CI ‐1.07 to 0.33, P = 0.3) with significant heterogeneity (I² = 78%, P < 0.01). All studies delivered multiple sessions of high‐frequency stimulation. Of these, one study stimulated the DLPFC (Avery 2013) and the remaining studies stimulated the motor cortex (Kang 2009; Mhalla 2011; Passard 2007). Removing the study of Avery 2013 did not decrease heterogeneity (I² = 85%, P < 0.01).

rTMS for disability/pain interference: long‐term follow‐up (≥ 6 weeks post‐treatment)

Three studies provided data on disability/pain interference at long‐term follow‐up (Avery 2013; Kang 2009; Passard 2007). Pooling of these studies demonstrated no significant effect (SMD ‐0.23, 95% CI ‐0.62 to 0.16, P = 0.24) without significant heterogeneity (I² = 15%, P = 0.31) (Analysis 1.20).

rTMS for quality of life: short‐term follow‐up

Three studies provided data on quality of life at short‐term follow‐up (Mhalla 2011; Passard 2007; Short 2011). We were unable to include data from Tzabazis 2013, as the size of the treatment groups was not clear from the study report. All studies used the Fibromyalgia Impact Questionnaire so we were able to use the mean difference as the measure of effect. Pooling data from these studies (Analysis 1.21; n = 80) demonstrated a significant effect (mean difference (MD) ‐10.38, 95% CI ‐14.89 to ‐5.87, P < 0.01) with no heterogeneity (I² = 0%, P = 0.99). Expressed as a percentage of the mean post‐stimulation score in the sham groups from the included studies (58.3) this equates to a 18% (95% CI 10% to 26%) reduction in fibromyalgia impact.

rTMS for quality of life: medium‐term follow‐up (1 to < 6 weeks post‐treatment)

The same three studies provided data on quality of life at medium‐term follow‐up (Mhalla 2011; Passard 2007; Short 2011). All studies used the Fibromyalgia Impact Questionnaire so we were able to use the mean difference as the measure of effect. Pooling data from these studies (Analysis 1.22; n = 80) demonstrated a significant effect (MD ‐11.49, 95% CI ‐17.04 to ‐5.95, P < 0.01) with no heterogeneity (I² = 0%, P = 0.63). Expressed as a percentage of the mean post‐stimulation score in the sham groups from the included studies (57.8) this equates to a 20% (95% CI 10% to 29%) reduction in fibromyalgia impact.

rTMS for quality of life: long‐term follow‐up (≥ 6 weeks post‐treatment)

Data were only available from one study (Passard 2007, n = 30) for quality of life at long‐term follow‐up. This study demonstrated no significant effect (MD ‐0.61, 95% CI ‐1.34 to 0.12) (Analysis 1.23).

CES for quality of life: short‐term follow‐up

Two studies provided quality of life data for this analysis (Tan 2011; Taylor 2013). One study used the physical component score of the SF‐12 and the other used the Fibromyalgia Impact Questionnaire. However, one study demonstrated a baseline imbalance of the SF‐12 that exceeded in size any pre‐post stimulation change (Tan 2011). Therefore we considered it inappropriate to enter this into a meta‐analysis. The study by Taylor 2013 (n = 36) demonstrated a positive effect on this outcome (SMD ‐1.25, 95% CI ‐1.98 to ‐0.53) (Analysis 2.3).

tDCS for quality of life

Two studies provided adequate data for this analysis (Mori 2010; Riberto 2011, pooled n = 32). Of these, Mori 2010 used the Multiple Sclerosis Quality of Life 54 scale (MS‐QoL‐54) and Riberto 2011 used the SF‐36 (total score). The pooled effect was significant (SMD 0.88, 95% CI 0.24 to 1.53, P < 0.01) with no heterogeneity (I² = 0%, P = 0.41) (Analysis 3.9). At medium‐term follow‐up only Mori 2010 (n = 19) provided data and the effect of tDCS on quality of life was not significant.

RINCE for quality of life

The one study of RINCE therapy demonstrated no significant effect on quality of life (Fibromyalgia Impact Questionnaire) (Analysis 4.2) .

Adverse events

rTMS

Minor

Of the rTMS studies that reported adverse events, nine studies reported none (André‐Obadia 2006; André‐Obadia 2008; Fregni 2005; Hirayama 2006; Lefaucheur 2001a; Lefaucheur 2001b; Lefaucheur 2004; Onesti 2013; Saitoh 2007). Carretero 2009 reported neck pain or headache symptoms in six out of 14 participants in the active stimulation group compared with two out of 12 in the sham group. One participant in the active stimulation group reported worsening depression and four participants in the sham group reported symptoms of nausea and tiredness. Passard 2007 reported incidence of headaches (four out of 15 participants in the active group versus five out of 15 in the sham group), feelings of nausea (one participant in the active group), tinnitus (two participants in the sham group) and dizziness (one participant in the sham group). Rollnik 2002 reported that one participant experienced headache, but it is unclear in the report whether this was following active or sham stimulation. Avery 2013 reported a range of reported sensations including headache, pain at the stimulation site, muscle aches/fatigue, dizziness and insomnia, though there were no clear differences in the frequency of these events between the two groups. Mhalla 2011 reported that nine patients (five following active stimulation and four following sham stimulation) reported transient headache, and one participant reported transient dizziness after active stimulation. Picarelli 2010 found six reports of headache following active stimulation and four following sham stimulation, and two reports of neck pain following active stimulation with four reports following sham stimulation. Short 2011 reported that there were few side effects and Hosomi 2013 reported no difference between real and sham rTMS for minor adverse events. In the study by Fregni 2011, the incidence of headache and neck pain was higher in the active stimulation group than in the sham group. Forty‐one participants reported headache after active stimulation compared to 19 after sham and 18 participants reported neck pain after active stimulation compared with three after sham. Following four‐coil rTMS, Tzabazis 2013 reported no serious adverse events. The incidence of scalp pain, headache, lightheadedness, back pain, otalgia, hot flashes and pruritis was more commonly reported following sham stimulation than active stimulation. Neck pain (14% of participants following active stimulation versus no participants following sham) and nausea (19% of participants following active stimulation verus 11% following sham) were more common with active stimulation.

Major

Both Lee 2012 and Picarelli 2010 reported one incidence of seizure following high‐frequency active stimulation.

CES

Four studies of CES reported the incidence of adverse events (Capel 2003; Gabis 2003; Rintala 2010; Tan 2011). In these studies no adverse events were reported. Rintala 2010 reported no major adverse events. In the active stimulation group they reported incidences of pulsing, tingling, tickling in ears (three participants), tender ears (one participant) and pins and needles feeling near bladder (one participant). In the sham group they reported drowsiness (one participant), warm ears (one participant) and headache after one session (one participant). Tan 2011 reported only mild adverse events with a total of 41 reports in the active stimulation group and 56 in the sham group. Of note, sensations of ear pulse/sting/itch/electric sensations or ear clip tightness seemed more common in active group than the sham group (12 versus six incidents). Through correspondence with the authors of Taylor 2013, we confirmed that there were no adverse events reported.

tDCS

Most studies of tDCS reported the incidence of adverse events. Of these, four studies reported none (Fregni 2006a; Mendonca 2011; Mori 2010; Portilla 2013). Boggio 2009 reported that one participant experienced headache with active stimulation. The study by Fenton 2009 reported three cases of headache, two of neck ache, one of scalp pain and five of a burning sensation over the scalp in the active stimulation group versus one case of headache in the sham stimulation group. Fregni 2006b reported one case of sleepiness and one of headache in response to active stimulation of the DLPFC, three cases of sleepiness and three of headache with active stimulation of M1 and one case of sleepiness and two of headache in response to sham stimulation. Soler 2010 recorded three reports of headache, all following active stimulation. Villamar 2013 reported that the vast majority of participants reported a mild to moderate tingling or itching sensation during both active and sham stimulation that faded over a few minutes but no other adverse effects. Valle 2009 reported "minor and uncommon" side effects, such as skin redness and tingling, which where equally distributed between active and sham stimulation. Antal 2010 recorded reports of tingling, moderate fatigue, tiredness, headache and sleep disturbances, though there were no large differences in the frequency of these between the active and sham stimulation groups. Wrigley 2014 reported only "mild to moderate" side effects with no significant difference between active and sham over the 24‐hour post‐stimulation period. These included sleepiness (70% of participants following active, 60% following sham), fatigue, inertia (60% of participants following active, 30% following sham), lightheadedness (20% of participants during active and sham treatment) and headache (10% of participants during active and sham treatment).

Four studies monitored for possible effects on cognitive function using the Mini Mental State Examination questionnaire (Boggio 2009; Fregni 2006a; Fregni 2006b; Valle 2009) and three of these also used a battery of cognitive tests including the digit‐span memory test and the Stroop word‐colour test (Boggio 2009; Fregni 2006a; Fregni 2006b) and simple reaction time tasks (Fregni 2006a). No studies demonstrated any negative influence of stimulation on these outcomes. No studies of tDCS reported severe or lasting side effects. Jensen 2013 and Riberto 2011 did not consider adverse events in their study reports.

RINCE

Hargrove 2012 reported a low incidence of side effects from RINCE stimulation including short‐lived headache (two participants in the active group, one in the sham group), eye movement/flutter during stimulation (one active, one sham), restlessness (one active and none sham) and nausea (one active and none sham).

GRADE judgements

GRADE judgements for all core comparisons of the primary outcome can be found in Table 4. For all comparisons the highest rating of the quality of evidence was 'low'.

Table 4. GRADE judgements for core comparisons

Comparison	Result	Limitations of studies	Inconsistency	Indirectness	Imprecision	Publication bias	GRADE judgement
rTMS
Pain: short‐term
Low‐frequency rTMS all	Ineffective SMD 0.15 (‐0.01 to 0.31)	Down one < 75% at low risk of bias	None (I² = 0%, P = 0.78)	None	Down one, n = 81	No direct evidence	Low
High‐frequency TMS all	Effective SMD ‐0.27 (‐0.35 to ‐0.20)	Down one < 75% studies at low risk of bias	Down one (I² = 64%, P < 0.01)	None	None, n = 447	No direct evidence	Low
Single‐dose, high‐frequency rTMS applied to the motor cortex on chronic pain	Effective SMD ‐0.39 (‐0.51 to ‐0.27)	Down one < 75% studies at low risk of bias	None (I² = 31%, P = 0.13)	None	Down one, n = 233	No direct evidence	Low
Multiple‐dose, high‐frequency rTMS applied to the motor cortex on chronic pain	Ineffective SMD ‐0.07 (‐0.41 to 0.26)	Down one < 75% studies at low risk of bias	Down one (I² = 71%, P < 0.01)	None	Down one, n = 157	No direct evidence	Very low
rTMS prefrontal cortex	Ineffective SMD ‐0.47 (‐1.48 to 0.54)	Down one < 75% studies at low risk of bias	Down one (I² = 82%, P < 0.01)	None	Down one, n = 68	No direct evidence	Very low
Pain: medium‐term
rTMS all	Ineffective SMD ‐0.15 (‐0.41 to 0.11)	Down one < 75% studies at low risk of bias	Down one (I² = 60%, P = 0.01)	None	Down one, n = 184	No direct evidence	Very low
Pain: long‐term
rTMS all	Ineffective SMD ‐0.12, (‐0.46 to 0.21)	Down one < 75% studies at low risk of bias	None (I² = 0%, P = 0.95)	None	Down one, n = 59	No direct evidence	Low
CES
Pain: short‐term
CES all	Ineffective SMD ‐0.24 (‐0.48 to 0.01)	Down one < 75% studies at low risk of bias	None (I² = 0%, P = 0.43)	None	Down one, n = 270	No direct evidence	Low
tDCS
Pain: short‐term
tDCS all	Ineffective SMD ‐0.18 (‐0.46 to 0.09)	Down one < 75% studies at low risk of bias	Down one (I² = 49%, P = 0.02)	None	Down one, n = 183	No direct evidence	Very low
tDCS motor cortex	Ineffective SMD ‐0.23 (‐0.48 to 0.01)	Down one < 75% studies at low risk of bias	None (I² = 33%, P = 0.13)	None	Down one, n = 182	No direct evidence	Low
tDCS motor cortex multiple‐dose studies	Ineffective SMD ‐0.35 (‐0.79 to 0.09)	Down one < 75% studies at low risk of bias	Down one (I² = 51%, P = 0.05)	None	Down one, n = 129	No direct evidence	Very low
Pain: medium‐term
tDCS all	Ineffective SMD ‐0.32 (‐0.76 to 0.11)	Down one < 75% studies at low risk of bias	None (I² = 40%, P = 0.14)	None	Down one n = 87	No direct evidence	Low
RINCE
Pain: short‐term	Effective SMD ‐1.41 (‐2.48 to ‐0.34) P = 0.01	Down one ‐ study at unclear risk of bias	n/a ‐ single study only	None	Down two, as only a single study available	No direct evidence ‐ only a single study	Very low

CES: cranial electrotherapy stimulation
RINCE: reduced impedance non‐invasive cortical electrostimulation
rTMS: repetitive transcranial magnetic stimulation
SMD: standardised mean difference
tDCS: transcranial direct current stimulation
TMS: transcranial magnetic stimulation

Discussion

Summary of main results

This update has included a substantial number of new studies. Despite this, for rTMS and CES our findings have not altered substantially from the previous version of this review. However, for tDCS the inclusion of these new data have altered the outcome of our analyses, which no longer suggest a statistically significant effect of tDCS over sham. We recommend that previous readers should re‐read this update.

Repetitive transcranial magnetic stimulation (rTMS) for chronic pain

Meta‐analysis of all rTMS studies in chronic pain demonstrated significant heterogeneity. Predetermined subgroup analysis suggests a short‐term effect of single‐dose, high‐frequency rTMS applied to the motor cortex on chronic pain. This effect is small and does not conclusively exceed the threshold of minimal clinical significance. The evidence from multiple‐dose studies of rTMS demonstrates conflicting results with substantial heterogeneity both overall and when the analysis is confined to high‐frequency motor cortex studies. Low‐frequency rTMS does not appear to be effective. rTMS applied to the pre‐frontal cortex does not appear to be effective. That the majority of studies in this analysis are at unclear risk of bias, particularly for participant blinding, suggests that the observed effect sizes might be exaggerated. While there is substantial unexplained heterogeneity the available evidence does not suggest a significant effect of rTMS in the medium term. The limited evidence at long‐term follow‐up consistently suggests no effect of rTMS.

Cranial electrotherapy stimulation (CES) for chronic pain

The evidence from trials where it is possible to extract data is not suggestive of a significant beneficial effect of CES on chronic pain. While there are substantial differences within the trials in terms of the populations studied and the stimulation parameters used, there is no measurable heterogeneity and no trial shows a clear benefit of active CES over sham stimulation.

Transcranial direct current stimulation (tDCS) for chronic pain

Meta‐analysis of all tDCS studies in chronic pain demonstrated significant heterogeneity. Predetermined subgroup analyses did not demonstrate a statistically significant effect of tDCS on chronic pain despite many of the studies included in this review being at unclear risk of bias for participant and assessor blinding. The evidence available at medium‐term follow‐up does not suggest a significant effect of tDCS.

Reduced impedance non‐invasive cortical electrostimulation (RINCE) stimulation for chronic pain

There is one small trial suggesting a positive effect of RINCE stimulation over sham for chronic pain. This trial is at unclear risk of bias due to possible attrition bias. As such, further research is needed to confirm this exploratory finding.

Adverse effects

rTMS, CES, tDCS and sham stimulation are associated with transient adverse effects such as headache, scalp irritation and dizziness, but reporting of adverse effects was inconsistent and did not allow for a detailed analysis. There were two incidences of seizure following active rTMS, which occurred in separate studies. For all forms of stimulation adverse events reporting is inconsistent across studies.

Secondary outcome measures

The available evidence does not suggest an effect of rTMS on disability/pain interference levels at any follow‐up point. There is insufficient evidence from which to draw conclusions regarding CES or tDCS for pain interference or disability.

Limited evidence suggests that rTMS and tDCS have positive effects on quality of life. This finding in rTMS is difficult to interpret as it arises from multiple‐dose studies which together do not demonstrate an effect on pain intensity levels. Any hypothesised effects of non‐invasive brain stimulation techniques on quality of life would presumably be through the reduction of pain. Given this inconsistency between outcomes for rTMS and the limited amount of data available to these analyses, we would recommend that this finding should be interpreted with caution.

Overall completeness and applicability of evidence

For rTMS we were unable to include data from five full published studies (Fregni 2005; Fregni 2011, Onesti 2013; Picarelli 2010; Tzabazis 2013, combined n = 86). In addition, we identified six studies of rTMS published in abstract format for which we have not been able to acquire full study reports. A conservative estimate of the combined number of participants that those studies might add, assuming that some reports refer to the same study, is 243.

We were unable to extract the relevant data from four studies of CES (Capel 2003; Cork 2004; Katsnelson 2004; Lichtbroun 2001). This may have impacted upon the results of our meta‐analysis although one of those studies would have been excluded from the meta‐analysis as we judged it as being at risk of bias on criteria other than selective outcome reporting (Katsnelson 2004).

We were also unable to extract the relevant data from two studies of tDCS (Mendonca 2011; Valle 2009), and these data were not made available upon request to the study authors. These data would have significantly contributed to the power of the meta‐analysis by the introduction of a further 71 participants and may have altered our conclusions. In addition, we identified three studies of tDCS (Acler 2012; Albu 2011; Knotkova 2011, combined n = 87) published in abstract format, one of which is currently being re‐analysed by the study authors and as such the data were not available (Knotkova 2011), and for two of which we were unsuccessful in our efforts to contact the authors (Acler 2012; Albu 2011).

For both rTMS and tDCS there are a number of ongoing studies identified through the trials registers searches. Of these, two registered trials that were identified in the original version of this review have not yet been published and our attempts to contact the authors were unsuccessful (NCT00947622; NCT00815932). We hope that future updates of this review will include the aforementioned data.

Quality of the evidence

Using the GRADE criteria we judged the quality of evidence for all comparisons as low or very low‐quality. In large part this is due to issues of blinding and of precision and to a degree it reflects the early stage of research development that these technique are at. The majority of studies of rTMS were at unclear risk of bias. The predominant reason for this was the use of sub‐optimal sham controls that were unable to control for all possible sensory cues associated with active stimulation. A number of studies did not clearly report blinding of assessors and sensitivity analysis excluding those studies reduced both heterogeneity and the pooled effect size. It could be reasonably argued that the presence of a subgroup of single‐dose studies of high‐frequency stimulation specific to the motor cortex that does demonstrate superiority over sham with acceptable levels of heterogeneity is evidence for a specific clinical effect of rTMS. It should be considered, however, that high‐frequency rTMS is associated with more intense sensory and auditory cues that might plausibly elicit a larger placebo response, and many of the included studies were unable to control conclusively for these factors. The pooled effect size for the high‐frequency studies of motor cortex rTMS does not meet our predetermined threshold for clinical significance. This estimate is based solely on studies that delivered a single dose of rTMS. It is feasible that a single dose may be insufficient to induce clinically meaningful improvement. These single‐dose studies included in the analysis are best characterised as proof of principle studies which sought to test whether rTMS could modulate pain, rather than full‐scale clinical studies with the aim of demonstrating clinical utility. However the combined evidence from studies of rTMS that delivered multiple doses (excluding studies judged as being at high risk of bias), while demonstrating substantial heterogeneity, does not indicate a significant effect on pain.

Similarly, we judged no study of tDCS as having a low risk of bias on all criteria. While there is evidence that the sham control used in tDCS does achieve effective blinding of participants at stimulation intensities of 1 mA (Gandiga 2006), evidence has emerged since the last version of this review which indicates that at 1.5 mA the sensory profile of stimulation differs between active and sham stimulation (Kessler 2013), and at 2 mA participant and assessor blinding may be compromised (O'Connell 2012). Meta‐epidemiological evidence demonstrates that incomplete blinding in controlled trials that measure subjective outcomes may exaggerate the observed effect size by around 25% (Wood 2008). It is therefore reasonable to expect that incomplete blinding may have exaggerated the effect sizes seen in the current analyses of rTMS and tDCS. The non‐significant trend towards a positive effect of CES and tDCS over sham should be considered in this light.

No study of CES could be judged as having a low risk of bias across all criteria. Despite this, no study from which data were available demonstrated a clear advantage of active over sham stimulation. There was substantial variation in the stimulation parameters used between studies. Notably three studies utilised an "active placebo" control in which stimulating current was delivered but at much lower intensities (Gabis 2003; Gabis 2009; Katsnelson 2004). These intensities well exceed those employed in the active stimulation condition of other studies of CES devices and as such it could be hypothesised that they might induce a therapeutic effect themselves. This could possibly disadvantage the active stimulation group in these studies. However, the data available in the meta‐analysis do not suggest such a trend and statistical heterogeneity between studies entered into the analysis was low.

All of the included studies may be considered to be small in terms of sample size and we reflected this in our 'Risk of bias' assessment. The prevalence of small studies increases the risk of publication or small study bias, wherein there is a propensity for negative studies to not reach full publication. There is evidence that this might lead to an overly positive picture for some interventions (Dechartres 2013; Moore 2012; Nüesch 2010). In a review of meta‐analyses, Dechartres 2013 demonstrated that trials with fewer than 50 participants, which reflects the majority of studies included in this review, returned effect estimates that were on average 48% larger than the largest trials and 23% larger than estimates from studies with sample sizes of more than 50. Similarly, in a recent Cochrane review of amitriptyline neuropathic pain and fibromyalgia (Moore 2012), smaller studies were associated with substantially lower numbers needed to treat (NNTs) for treatment response than larger studies. In their recommendations for establishing best practice in chronic pain systematic reviews, the authors of Moore 2010 suggest that study size should be considered an important source of bias. It is therefore reasonable to consider that the evidence base for all non‐invasive brain stimulation techniques is at risk of bias on the basis of sample size. We did not downgrade any of the GRADE judgements on the basis of publication bias as there was no direct evidence. However, it is accepted that existing approaches to detecting publication bias are unsatisfactory. To an extent our GRADE judgements reflect this risk through the assessment of imprecision and the limitations of included studies. It should be noted that even where a pooled estimate includes a large number of participants, if it is dominated by small studies, as are all comparisons in this review, then it is prone to small study effects.

Potential biases in the review process

There is substantial variation between the included studies of rTMS and tDCS. Studies varied in terms of the clinical populations included, the stimulation parameters and location, the number of treatment sessions delivered and in the length of follow‐up employed. This heterogeneity is reflected in the I² statistic for the overall rTMS and tDCS meta‐analyses. However, pre‐planned subgroup investigation significantly reduced this heterogeneity.

The majority of rTMS and tDCS studies specifically recruited participants whose symptoms were resistant to current clinical management and most rTMS studies specifically recruited participants with neuropathic pain. As such it is important to recognise that this analysis in large part reflects the efficacy of rTMS and tDCS for refractory chronic pain conditions and may not accurately reflect their efficacy across all chronic pain conditions.

One study included in the in the analysis of rTMS studies demonstrated a difference in pain levels between the two groups at baseline that exceeded the size of the difference observed at follow‐up (Defrin 2007). Specifically, the group that received sham stimulation reported less pain at baseline than those in the active stimulation group. The use in the current analysis of a between‐groups rather than a change from baseline comparison is likely to have affected the results although the study contributes only 1.5% weight to the overall meta‐analysis and the study itself reported no difference in the degree of pain reduction between the active and sham stimulation groups.

The method used to back‐transform the pooled standardised mean difference (SMD) to a visual analogue scale and subsequent calculation of the effect as a percentage improvement rests upon the assumption that the standard deviation and the pain levels used are representative of the wider body of evidence and should be considered an estimate at best. Representing average change scores on continuous scales is problematic in chronic pain studies since response to treatments has been found to display a bimodal distribution (Moore 2013). More plainly, some participants demonstrate a substantial response to pain therapies while many demonstrate little or no response with few individual participants demonstrating a response similar to the average. As a consequence the meaning of the average effect sizes seen in this review is difficult to interpret. This had led to the recommendation that chronic pain trials employ responder analysis based on predetermined cut‐offs for a clinically important response (≥ 30% reduction in pain for a moderate benefit, ≥ 50% reduction for a substantial benefit) (Dworkin 2008; Moore 2010). Very few studies identified in this review presented the results of responder analyses and so this type of meta‐analysis was not possible. However, where statistically significant effects were observed in this review they were small, which would indicate that if there is a subgroup of 'responders' to active stimulation who demonstrate moderate or substantial benefits it is likely to include a small number of participants.

Agreements and disagreements with other studies or reviews

The European Federation of Neurological Societies (EFNS) published guidelines on the use of neurostimulation therapy for chronic neuropathic pain in 2007 (Cruccu 2007), following a review of the existing literature. Using a narrative synthesis of the evidence they similarly concluded that there was moderate evidence (two randomised controlled trials) that high‐frequency rTMS (≥ 5 Hz) of the motor cortex induces significant pain relief in central post‐stroke pain and several other neuropathic conditions, but that the effect is modest and short‐lived. They did not recommend its use as a sole clinical treatment but suggested that it might be considered in the treatment of short‐lasting pain.

Leung 2009 performed a meta‐analysis of individual patient data from studies of motor cortex rTMS for neuropathic pain conditions. Whilst the analysis was restricted to studies that clearly reported the neuroanatomical origin of participants' pain (and therefore excluded some of the studies included in the current analysis) the overall analysis suggests a similar effect size of 13.7% improvement in pain (excluding the study of Khedr 2005). The authors also performed an analysis of the influence of the neuro‐anatomical origins of pain on the effect size. They noted a trend suggestive of a larger treatment effect in central compared with peripheral neuropathic pain states although this did not reach statistical significance. While the data in the current review were not considered sufficient to support a detailed subgroup analysis by neuro‐anatomical origin of pain, the exclusion of studies that did not specifically investigate neuropathic pain did not significantly affect the overall analysis and the two multiple‐dose studies of motor cortex rTMS for central neuropathic pain that were included failed to demonstrate superiority of active over sham stimulation (Defrin 2007; Kang 2009).

All but one of the included studies in the review by Leung 2009 delivered high‐frequency (≥ 5 Hz) rTMS and no clear influence of frequency variations was observed within this group. The authors suggest that the number of doses delivered may be more crucial to the therapeutic response than the frequency (within the high‐frequency group), based on the larger therapeutic response seen in the study of Khedr 2005 that was excluded from the current analysis. This review preceded the studies by Defrin 2007 and Kang 2009 that did not demonstrate superiority of active over sham stimulation. While there are limited data to test this proposition robustly the result of our subgroup analysis of studies of high‐frequency motor cortex rTMS does not suggest a benefit of active stimulation over sham.

Lima and Fregni undertook a systematic review and meta‐analysis of motor cortex stimulation for chronic pain (Lima 2008). They pooled data from rTMS and tDCS studies. While the report states that data were collected on mean between‐group pain scores they are not presented. The authors present the pooled data for the number of responders to treatment across studies. They conclude that the number of responders is significantly higher following active stimulation compared with sham (risk ratio 2.64, 95% confidence interval (CI) 1.63 to 4.30). In their analysis the threshold for treatment response is defined as a global response according to each study's own definition and as such it is difficult to interpret and may not be well standardised. They note a greater response to multiple doses of stimulation, an observation that is not reliably reflected in the current review. Additionally they included the study of Khedr 2005 (excluded from this review due to high risk of bias) and Canavero 2002 (excluded on title and abstract as it is not a randomised or quasi‐randomised study). The current review also includes a number of motor cortex rTMS studies published since that review (André‐Obadia 2008; Defrin 2007; Kang 2009; Lefaucheur 2006; Lefaucheur 2008; Passard 2007; Saitoh 2007). Neither the review of Leung 2009 nor Lima 2008 applied a formal quality or 'Risk of bias' assessment. While the current review also suggests a small, significant short‐term benefit of high‐frequency motor cortex rTMS in the treatment of chronic pain the effect is small, appears short‐term and although the pooled estimate approaches the threshold of minimal clinical significance it is possible that it might be inflated by methodological biases in the included studies.

A recent systematic review of tDCS and rTMS for the treatment of fibromyalgia concluded that the evidence demonstrated reductions in pain similar to US Food and Drug Administration (FDA) approved pharmaceuticals for this condition and recommended that rTMS or tDCS should be considered, particularly where other therapies have failed (Marlow 2013). This review included randomised and non‐randomised studies, did not undertake meta‐analysis and took a "vote‐counting" approach to identifying significant effects based primarily on each included study's report of statistical testing. While our analysis did not specifically investigate a subgroup of studies in fibromyalgia participants, we would suggest that the methodology chosen by Marlow 2013 does not offer the most rigorous approach to establishing effect size, particularly in light of the inconsistency seen among the included studies of that review. Indeed given the degree of uncertainty that remains regarding the efficacy these interventions it could be suggested that the application of tDCS or rTMS for this or other conditions would ideally be limited to the clinical research situation.

Luedtke 2012 systematically reviewed studies of tDCS for chronic pain and experimental pain. Unlike our review they excluded the study by Fenton 2009, as it was judged to be at high risk of bias on the grounds of unclear randomisation procedure and due to a lack of clarity of participant withdrawal, and Boggio 2009 due to the level of drop‐out. The results of their meta‐analysis are broadly consistent with those presented in the last iteration of this review and similarly conclude that the evidence is insufficient to allow definite conclusions but that there is low‐level evidence that tDCS may be effective for chronic pain. However, the inclusion of new studies in this update has rendered these analyses non‐significant. Moreno‐Duarte 2013 recently reviewed the evidence for a variety of electrical and magnetic neural stimulation techniques for the treatment for chronic pain following spinal cord injury, including rTMS, tDCS and CES, including both randomised and non‐randomised studies. They found that the results varied across studies, though trials of tDCS were consistently positive, and concluded that further research is needed and that there is a need to develop methods to decrease the variability of treatment response to these interventions. However, it is worth noting that this review did not include the recent negative study of tDCS for post‐spinal cord injury pain by Wrigley 2014, and also that variability in observed treatment "responses" may simply represent the play of chance rather than evidence of a specific group of responders.

Kirsch 2000 reviewed studies of CES in the management of chronic pain and concluded in favour of its use. The review did not report any formalised search strategy, inclusion criteria or quality assessment and discussed a number of unpublished studies that remain unpublished at the time of the current review. Using a more systematic methodology and including papers published since that review, we found that the data that were available for meta‐analysis do not suggest a statistically or clinically important benefit of active CES over sham. Our analysis included 270 participants. While this is not particularly large it does suggest that if there is an effect of CES on chronic pain it is either small, or that the number of responders is likely to be small.

Figure 1

Study flow diagram for updated search.

Figure 2

Methodological quality summary: review authors' judgements about each methodological quality item for each included study.

Analysis 1.1

Comparison 1 Repetitive transcranial magnetic stimulation (rTMS), Outcome 1 Pain: short‐term follow‐up.

Analysis 1.2

Comparison 1 Repetitive transcranial magnetic stimulation (rTMS), Outcome 2 Pain: short‐term follow‐up, subgroup analysis: multiple‐dose vs single‐dose studies.

Analysis 1.3

Comparison 1 Repetitive transcranial magnetic stimulation (rTMS), Outcome 3 Pain: short‐term follow‐up, subgroup analysis, neuropathic pain participants only.

Analysis 1.4

Comparison 1 Repetitive transcranial magnetic stimulation (rTMS), Outcome 4 Pain: short‐term follow‐up, subgroup analysis, non‐neuropathic pain participants only.

Analysis 1.5

Comparison 1 Repetitive transcranial magnetic stimulation (rTMS), Outcome 5 Pain: short‐term follow‐up, subgroup analysis: motor cortex studies only, low‐frequency studies excluded.

Analysis 1.6

Comparison 1 Repetitive transcranial magnetic stimulation (rTMS), Outcome 6 Sensitivity analysis ‐ imputed correlation coefficient increased. Pain: short‐term follow‐up.

Analysis 1.7

Comparison 1 Repetitive transcranial magnetic stimulation (rTMS), Outcome 7 Sensitivity analysis ‐ imputed correlation coefficient decreased. Pain: short‐term follow‐up.

Analysis 1.8

Comparison 1 Repetitive transcranial magnetic stimulation (rTMS), Outcome 8 Sensitivity analysis ‐ imputed correlation increased. Pain: short‐term follow‐up, subgroup analysis: motor cortex studies only, low‐frequency studies excluded.

Analysis 1.9

Comparison 1 Repetitive transcranial magnetic stimulation (rTMS), Outcome 9 Sensitivity analysis ‐ imputed correlation decreased. Pain: short‐term follow‐up, subgroup analysis: motor cortex studies only, low‐frequency studies excluded.

Analysis 1.10

Comparison 1 Repetitive transcranial magnetic stimulation (rTMS), Outcome 10 Sensitivity analysis ‐ inclusion of high risk of bias studies. Pain: short‐term follow‐up.

Analysis 1.11

Comparison 1 Repetitive transcranial magnetic stimulation (rTMS), Outcome 11 Sensitivity analysis ‐ inclusion of high risk of bias studies. Pain: short‐term follow‐up, subgroup analysis: motor cortex studies only, low‐frequency studies excluded.

Analysis 1.12

Comparison 1 Repetitive transcranial magnetic stimulation (rTMS), Outcome 12 Pain: short‐term follow‐up, subgroup analysis: prefrontal cortex studies only.

Analysis 1.13

Comparison 1 Repetitive transcranial magnetic stimulation (rTMS), Outcome 13 Sensitivity analysis ‐ inclusion of high risk of bias studies. Pain: short‐term follow‐up, subgroup analysis: prefrontal cortex studies only.

Analysis 1.14

Comparison 1 Repetitive transcranial magnetic stimulation (rTMS), Outcome 14 Pain: medium‐term follow‐up.

Analysis 1.15

Comparison 1 Repetitive transcranial magnetic stimulation (rTMS), Outcome 15 Sensitivity analysis ‐ inclusion of high risk of bias studies. Pain: medium‐term follow‐up.

Analysis 1.16

Comparison 1 Repetitive transcranial magnetic stimulation (rTMS), Outcome 16 Pain: long‐term follow‐up.

Analysis 1.17

Comparison 1 Repetitive transcranial magnetic stimulation (rTMS), Outcome 17 Sensitivity analysis ‐ inclusion of high risk of bias studies. Pain: long‐term follow‐up.

Analysis 1.18

Comparison 1 Repetitive transcranial magnetic stimulation (rTMS), Outcome 18 Disability/pain interference: short‐term follow‐up.

Analysis 1.19

Comparison 1 Repetitive transcranial magnetic stimulation (rTMS), Outcome 19 Disability/pain interference: medium‐term follow‐up.

Analysis 1.20

Comparison 1 Repetitive transcranial magnetic stimulation (rTMS), Outcome 20 Disability/pain interference: long‐term follow‐up.

Analysis 1.21

Comparison 1 Repetitive transcranial magnetic stimulation (rTMS), Outcome 21 Quality of life: short‐term follow‐up (Fibromyalgia Impact Questionnaire).

Analysis 1.22

Comparison 1 Repetitive transcranial magnetic stimulation (rTMS), Outcome 22 Quality of life: medium‐term follow‐up (Fibromyalgia Impact Questionnaire).

Analysis 1.23

Comparison 1 Repetitive transcranial magnetic stimulation (rTMS), Outcome 23 Quality of life: long‐term follow‐up.

Analysis 2.1

Comparison 2 Cranial electrotherapy stimulation (CES), Outcome 1 Pain: short‐term follow‐up.

Analysis 2.2

Comparison 2 Cranial electrotherapy stimulation (CES), Outcome 2 Disability/function/pain interference.

Analysis 2.3

Comparison 2 Cranial electrotherapy stimulation (CES), Outcome 3 Quality of life.

Analysis 3.1

Comparison 3 Transcranial direct current stimulation (tDCS), Outcome 1 Pain: short‐term follow‐up.

Analysis 3.2

Comparison 3 Transcranial direct current stimulation (tDCS), Outcome 2 Pain: short‐term follow‐up, subgroup analysis: motor cortex studies only.

Analysis 3.3

Comparison 3 Transcranial direct current stimulation (tDCS), Outcome 3 Pain: short‐term sensitivity analysis: correlation increased.

Analysis 3.4

Comparison 3 Transcranial direct current stimulation (tDCS), Outcome 4 Pain: short‐term sensitivity analysis: correlation decreased.

Analysis 3.5

Comparison 3 Transcranial direct current stimulation (tDCS), Outcome 5 Pain: short‐term follow‐up, subgroup analysis: motor cortex studies only, sensitivity analysis: correlation increased.

Analysis 3.6

Comparison 3 Transcranial direct current stimulation (tDCS), Outcome 6 Pain: short‐term follow‐up, subgroup analysis: motor cortex studies only, sensitivity analysis: correlation decreased.

Analysis 3.7

Comparison 3 Transcranial direct current stimulation (tDCS), Outcome 7 Pain: medium‐term follow‐up.

Analysis 3.8

Comparison 3 Transcranial direct current stimulation (tDCS), Outcome 8 Disability (pain interference): short‐term follow‐up.

Analysis 3.9

Comparison 3 Transcranial direct current stimulation (tDCS), Outcome 9 Quality of life: short‐term follow‐up.

Analysis 3.10

Comparison 3 Transcranial direct current stimulation (tDCS), Outcome 10 Quality of life: medium‐term follow‐up.

Analysis 3.11

Comparison 3 Transcranial direct current stimulation (tDCS), Outcome 11 Pain: short‐term follow‐up, subgroup analysis: motor cortex studies only.

Analysis 4.1

Comparison 4 Reduced impedance non‐invasive cortical electrostimulation (RINCE), Outcome 1 Pain: short‐term follow‐up.

Analysis 4.2

Comparison 4 Reduced impedance non‐invasive cortical electrostimulation (RINCE), Outcome 2 Fibromyalgia Impact Questionnaire total score.

Repetitive transcranial magnetic stimulation (rTMS) compared with sham for chronic pain
Intervention: active rTMS Comparison: sham rTMS
Outcomes: pain (VAS or NRS)
Comparison	No of participants (studies)	Effect size (SMD, 95% CIs)	Relative effect (average % improvement (reduction) in pain (95% CIs) in relation to post‐treatment score from sham group)* *statistically significant outcomes with low heterogeneity only	Quality of the evidence (GRADE)
Pain: short‐term follow‐up Subgroup analysis: low‐frequency rTMS	81 (6)	Ineffective 0.15 (‐0.01 to 0.31) P = 0.07		⊕⊕⊝⊝ low
Pain: short‐term follow‐up subgroup analysis: high‐frequency rTMS	447 (20)	Effective ‐0.27 (‐0.35 to ‐0.20) P < 0.01		⊕⊕⊝⊝ low
Pain: short‐term follow‐up Subgroup analysis: motor cortex studies only, low‐frequency studies excluded, single‐dose studies	233 (12)	Effective ‐0.39 (‐0.51 to ‐0.27) P < 0.01	12% (8% to 15%)	⊕⊕⊝⊝ low
Pain: short‐term follow‐up Subgroup analysis: motor cortex studies only, low‐frequency studies excluded, multiple‐dose studies	157 (5)	Ineffective ‐0.07 (‐0.41 to 0.26) P = 0.68		⊕⊝⊝⊝ very low
Pain: short‐term follow‐up Subgroup analysis: prefrontal cortex studies only	68 (5)	Ineffective ‐0.47 (‐1.48 to 0.11) P = 0.36		⊕⊝⊝⊝ very low
Pain: medium‐term follow‐up rTMS all studies	184 (8)	Ineffective ‐0.18 (‐0.43 to 0.06) P = 0.15		⊕⊝⊝⊝ very low
Pain: long‐term follow‐up rTMS all studies	59 (3)	Ineffective ‐0.12 (‐0.46 to 0.21) P = 0.47		⊕⊕⊝⊝ low
CES compared with sham for chronic pain
Intervention: active CES Comparison: sham CES
Outcomes: pain (VAS or NRS)
Pain: short‐term follow‐up CES all studies	270 (5)	Ineffective ‐0.24 (‐0.48 to 0.01) P = 0.06		⊕⊕⊝⊝ low
tDCS compared with sham for chronic pain
Intervention: active tDCS Comparison: sham tDCS
Outcomes: pain (VAS or NRS)
Pain: short‐term follow‐up tDCS all studies	183 (10)	Ineffective ‐0.18 (‐0.56 to 0.09) P = 0.19		⊕⊝⊝⊝ very low
Pain: short‐term follow‐up Subgroup analysis: motor cortex studies only (single and multiple‐dose studies)	172 (10)	Ineffective ‐0.23 (‐0.48 to 0.01) P = 0.06		⊕⊕⊝⊝ low
Pain: short‐term follow‐up Subgroup analysis: motor cortex studies only (multiple‐dose studies only)	119 (7)	Ineffective ‐0.35 (‐0.79 to 0.09) P = 0.12		⊕⊝⊝⊝ very low
Pain: medium‐term follow‐up tDCS	77 (4)	Ineffective ‐0.20 (‐0.63 to 0.24) P = 0.37		⊕⊕⊝⊝ low
RINCE compared with sham for chronic pain
Intervention: active RINCE Comparison: sham RINCE
Outcomes: pain (VAS or NRS)
Pain: short‐term follow‐up tDCS all studies	91 (1)	Effective ‐1.41 (‐2.48 to ‐0.34) P = 0.01		⊕⊝⊝⊝ very low
GRADE Working Group grades of evidence High quality: Further research is very unlikely to change our confidence in the estimate of effect. Moderate quality: Further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate. Low quality: Further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate. Very low quality: We are very uncertain about the estimate. CES: cranial electrotherapy stimulation; CI: confidence interval; NRS: numerical rating scale; RINCE: reduced impedance non‐invasive cortical electrostimulation; rTMS: repetitive transcranial magnetic stimulation; tDCS: transcranial direct current stimulation; VAS: visual analogue scale
For full details of the GRADE judgements for each comparison see Appendix 6.

Table 1. rTMS studies ‐ characteristics of stimulation

Study	Location of stimulation	Coil orientation	Frequency (Hz)	Intensity (% RMT)	Number of trains	Duration of trains	Inter‐train intervals (sec)	Number of pulses per session	Treatment sessions per group
Ahmed 2011	M1 stump region	45° angle from sagittal line	20	80	10	10 sec	50	2000	5, x 1 daily
André‐Obadia 2006	M1 contralateral to painful side	Posteroanterior	20, 1	90	20 Hz: 20 1Hz: 1	20 Hz: 4 sec 1 Hz: 26 min	20 Hz: 84	1600	1
André‐Obadia 2008	M1 contralateral to painful side	Posteroanterior Medial‐lateral	20	90	20	4 sec	84	1600	1
André‐Obadia 2011	M1 hand area, not clearly reported but likely contralateral to painful side	Not specified	20	90	20	4 sec	84	1600	1
Avery 2013	Left DLPFC	Not specified	10	120	75	4	26	3000	15
Short 2011	Left DLPFC	Para‐sagittal	10	120	80	5 sec	10 sec	4000	10, x 1 daily (working days) for 2 weeks
Borckardt 2009	Left PFC	Not specified	10	100	40	10 sec	20	4000	3 over a 5‐day period
Carretero 2009	Right DLPFC	Not specified	1	110	20	60 sec	45	1200	Up to 20 on consecutive working days
Defrin 2007	M1 midline	Not specified	5	115	500	10 sec	30	? 500*	10, x 1 daily
Fregni 2005	Left and right SII	Not specified	1	90	Not specified	Not specified	Not specified	1600	1
Fregni 2011	Right SII	Not specified	1	70% maximum stimulator output intensity (not RMT)	1	Not specified	Not specified	1600	10, x 1 daily (week days only)
Hirayama 2006	M1, S1, PMA, SMA	Not specified	5	90	10	10 sec	50	500	1
Hosomi 2013	M1 corresponding to painful region	Not specified	5	90	10	10 sec	50	500	10, x 1 daily (week days only)
Irlbacher 2006	M1 contralateral to painful side	Not specified	5, 1	95	Not specified	Not specified	Not specified	500	1
Kang 2009	Right M1	45º postero‐lateral	10	80	20	5 sec	55	1000	5, x 1 daily
Khedr 2005	M1 contralateral to painful side	Not specified	20	80	10	10 sec	50	2000	5, x 1 daily
Lee 2012	Right DLPFC (low‐frequency) Left M1 (high‐frequency)	Not specified	10, 1	10 Hz: 80 1 Hz: 110	10 Hz:25 1 Hz: 2	10 Hz: 8 sec 1 Hz: 800 sec	10 Hz: 10 1 Hz: 60	10 Hz: 2000 1 Hz: 1600	10, x 1 daily (week days only)
Lefaucheur 2001a	M1 contralateral to painful side	Not specified	10	80	20	5 sec	55	1000	1
Lefaucheur 2001b	M1 contralateral to painful side	Posteroanterior	10, 0.5	80	10 Hz: 20 0.5 Hz: 1	10 Hz: 5 sec 0.5 Hz: 20 min	10 Hz: 55	10 Hz: 1000 0.5 Hz: 600	1
Lefaucheur 2004	M1 contralateral to painful side	Posteroanterior	10	80	20	5 sec	55	1000	1
Lefaucheur 2006	M1 contralateral to painful side	Posteroanterior	10, 1	90	10 Hz: 20 1 Hz: 1	10 Hz: 6 sec 1 Hz: 20 min	10 Hz: 54	10 Hz: 1200 1 Hz: 1200	1
Lefaucheur 2008	M1 contralateral to painful side	Posteroanterior	10, 1	90	10 Hz: 20 1 Hz: 1	10 Hz: 6 sec 1 Hz: 20 min	10 Hz: 54	10 Hz: 1200 1 Hz: 1200	1
Mhalla 2011	Left M1	Posteroanterior	10	80	15	10 sec	50	1500	14, 5 x 1 daily (working days), then 3 x 1 weekly, then 3 x 1 fortnightly, then 3 x 1 monthly
Onesti 2013	M1 deep central sulcus	H‐coil	20	100	30	2.5 sec	30	1500	5, x 1 daily on consecutive days
Passard 2007	M1 contralateral to painful side	Posteroanterior	10	80	25	8 sec	52	2000	10, x 1 daily (working days)
Picarelli 2010	M1 contralateral to painful side	Posteroanterior	10	100	25	10 sec	60	2500	10, x 1 daily (working days)
Pleger 2004	M1 hand area	Not specified	10	110	10	1.2 sec	10	120	1
Rollnik 2002	M1 midline	Not specified	20	80	20	2 sec	Not specified	800	1
Saitoh 2007	M1 over motor representation of painful area	Not specified	10, 5, 1	90	10 Hz; 5 5 Hz: 10 1 Hz: 1	10 Hz: 10 sec 5 Hz: 10 sec 1 Hz: 500 sec	10 Hz: 50 5 Hz: 50	500	1
Tzabazis 2013	Targeted to ACC	4‐coil configuration	1 Hz (10 Hz data excluded as not randomised)	110	Not reported	Not reported	Not reported	1800	20, x 1 daily (working days)
ACC: anterior cingulate cortex; DLPFC: dorsolateral prefrontal cortex; M1: primary motor cortex; PFC: prefrontal cortex; PMA: pre‐motor area; RMT: resting motor threshold; dS1: primary somatosensory cortex; SII: secondary somatosensory cortex; SMA: supplementary motor area

Table 1. rTMS studies ‐ characteristics of stimulation

Table 2. CES studies ‐ characteristics of stimulation

Study	Electrode placement	Frequency (Hz)	Pulse width (msec)	Waveform shape	Intensity	Duration (min)	Treatment sessions per group
Capel 2003	Ear clip electrodes	10	2	Not specified	12 μA	53	x 2 daily for 4 days
Cork 2004	Ear clip electrodes	0.5	Not specified	Modified square wave biphasic	100 μA	60	? daily for 3 weeks
Gabis 2003	Mastoid processes and forehead	77	3.3	Biphasic asymmetric	≤ 4 mA	30	x 1 daily for 8 days
Gabis 2009	Mastoid processes and forehead	77	3.3	Biphasic asymmetric	≤ 4 mA	30	x 1 daily for 8 days
Katsnelson 2004	Mastoid processes and forehead	Not specified	Not specified	2 conditions: symmetric, asymmetric	11 to 15 mA	40	x 1 daily for 5 days
Lichtbroun 2001	Ear clip electrodes	0.5	Not specified	Biphasic square wave	100 μA	60	x 1 daily for 30 days
Rintala 2010	Ear clip electrodes	Not specified	Not specified	Not specified	100 μA	40	x 1 daily for 6 weeks
Tan 2000	Ear clip electrodes	0.5	Not specified	Not specified	10 to 600 μA	20	12 (timing not specified)
Tan 2006	Ear clip electrodes	Not specified	Not specified	Not specified	100 to 500 μA	60	x 1 daily for 21 days
Tan 2011	Ear clip electrodes	Not specified	Not specified	Not specified	100 μA	60	x 1 daily for 21 days
Taylor 2013	Ear clip electrodes	0.5	Not specified	Modified square‐wave biphasic	100 μA	60	x 1 daily for 8 weeks

Table 2. CES studies ‐ characteristics of stimulation

Table 3. tDCS studies ‐ characteristics of stimulation

Study	Location of stimulation	Electrode pad size	Intensity (mA)	Anodal or cathodal?	Stimulus duration (min)	Treatment sessions per group
Antal 2010	M1 left hand area	35 cm²	1 mA	Anodal	20	5, x 1 daily
Boggio 2009	M1 contralateral to painful side	35 cm²	2 mA	Anodal	30	1
Fenton 2009	M1 dominant hemisphere	35 cm²	1 mA	Anodal	20	2
Fregni 2006a	M1 contralateral to painful side or dominant hand	35 cm²	2 mA	Anodal	20	5, x 1 daily
Fregni 2006b	M1 and DLPFC contralateral to painful side or dominant hand	35 cm²	2 mA	Anodal	20	5, x 1 daily
Jensen 2013	M1 left	35cm²	2 mA	Anodal	20	1
Mendonca 2011	Group 1: anodal left M1 Group 2: cathodal left M1 Group 3: anodal supraorbital Group 4: cathodal supraorbital Group 5: sham	35 cm²	2 mA	Anodal or cathodal	20	1
Mori 2010	M1 contralateral to painful side	35 cm²	2 mA	Anodal	20	5, x 1 daily
Portilla 2013	M1 contralateral to painful side	35 cm²	2 mA	Anodal	20	x 1 per condition
Riberto 2011	M1 contralateral to painful side or dominant hand	35 cm²	2 mA	Anodal	20	10, x 1 weekly
Soler 2010	M1 contralateral to painful side or dominant hand	35 cm²	2 mA	Anodal	20	10, x 1 daily (week days only)
Valle 2009	M1 and DLPFC contralateral to painful side or dominant hand	35 cm²	2 mA	Anodal	20	5, x 1 daily
Villamar 2013	M1 left	HD‐tDCS 4 x 1‐ring montage	2 mA	Anodal or cathodal	20	x 1 per condition
Wrigley 2014	M1 contralateral to painful side or dominant hand	35 cm²	2 mA	Anodal	20	5, x 1 daily
DLPFC: dorsolateral prefrontal cortex; M1: primary motor cortex HD‐tDCS: High definition tDCS

Table 3. tDCS studies ‐ characteristics of stimulation

Table 4. GRADE judgements for core comparisons

Comparison	Result	Limitations of studies	Inconsistency	Indirectness	Imprecision	Publication bias	GRADE judgement
rTMS
Pain: short‐term
Low‐frequency rTMS all	Ineffective SMD 0.15 (‐0.01 to 0.31)	Down one < 75% at low risk of bias	None (I² = 0%, P = 0.78)	None	Down one, n = 81	No direct evidence	Low
High‐frequency TMS all	Effective SMD ‐0.27 (‐0.35 to ‐0.20)	Down one < 75% studies at low risk of bias	Down one (I² = 64%, P < 0.01)	None	None, n = 447	No direct evidence	Low
Single‐dose, high‐frequency rTMS applied to the motor cortex on chronic pain	Effective SMD ‐0.39 (‐0.51 to ‐0.27)	Down one < 75% studies at low risk of bias	None (I² = 31%, P = 0.13)	None	Down one, n = 233	No direct evidence	Low
Multiple‐dose, high‐frequency rTMS applied to the motor cortex on chronic pain	Ineffective SMD ‐0.07 (‐0.41 to 0.26)	Down one < 75% studies at low risk of bias	Down one (I² = 71%, P < 0.01)	None	Down one, n = 157	No direct evidence	Very low
rTMS prefrontal cortex	Ineffective SMD ‐0.47 (‐1.48 to 0.54)	Down one < 75% studies at low risk of bias	Down one (I² = 82%, P < 0.01)	None	Down one, n = 68	No direct evidence	Very low
Pain: medium‐term
rTMS all	Ineffective SMD ‐0.15 (‐0.41 to 0.11)	Down one < 75% studies at low risk of bias	Down one (I² = 60%, P = 0.01)	None	Down one, n = 184	No direct evidence	Very low
Pain: long‐term
rTMS all	Ineffective SMD ‐0.12, (‐0.46 to 0.21)	Down one < 75% studies at low risk of bias	None (I² = 0%, P = 0.95)	None	Down one, n = 59	No direct evidence	Low
CES
Pain: short‐term
CES all	Ineffective SMD ‐0.24 (‐0.48 to 0.01)	Down one < 75% studies at low risk of bias	None (I² = 0%, P = 0.43)	None	Down one, n = 270	No direct evidence	Low
tDCS
Pain: short‐term
tDCS all	Ineffective SMD ‐0.18 (‐0.46 to 0.09)	Down one < 75% studies at low risk of bias	Down one (I² = 49%, P = 0.02)	None	Down one, n = 183	No direct evidence	Very low
tDCS motor cortex	Ineffective SMD ‐0.23 (‐0.48 to 0.01)	Down one < 75% studies at low risk of bias	None (I² = 33%, P = 0.13)	None	Down one, n = 182	No direct evidence	Low
tDCS motor cortex multiple‐dose studies	Ineffective SMD ‐0.35 (‐0.79 to 0.09)	Down one < 75% studies at low risk of bias	Down one (I² = 51%, P = 0.05)	None	Down one, n = 129	No direct evidence	Very low
Pain: medium‐term
tDCS all	Ineffective SMD ‐0.32 (‐0.76 to 0.11)	Down one < 75% studies at low risk of bias	None (I² = 40%, P = 0.14)	None	Down one n = 87	No direct evidence	Low
RINCE
Pain: short‐term	Effective SMD ‐1.41 (‐2.48 to ‐0.34) P = 0.01	Down one ‐ study at unclear risk of bias	n/a ‐ single study only	None	Down two, as only a single study available	No direct evidence ‐ only a single study	Very low
CES: cranial electrotherapy stimulation RINCE: reduced impedance non‐invasive cortical electrostimulation rTMS: repetitive transcranial magnetic stimulation SMD: standardised mean difference tDCS: transcranial direct current stimulation TMS: transcranial magnetic stimulation

Table 4. GRADE judgements for core comparisons

Comparison 1. Repetitive transcranial magnetic stimulation (rTMS)

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Pain: short‐term follow‐up Show forest plot	21		Std. Mean Difference (Fixed, 95% CI)	‐0.20 [‐0.26, ‐0.13]

1.1 Low‐frequency ≤ 1 Hz	6		Std. Mean Difference (Fixed, 95% CI)	0.15 [‐0.01, 0.31]
1.2 High‐frequency ≥ 5 Hz	20		Std. Mean Difference (Fixed, 95% CI)	‐0.27 [‐0.35, ‐0.20]
2 Pain: short‐term follow‐up, subgroup analysis: multiple‐dose vs single‐dose studies Show forest plot	21		Std. Mean Difference (Random, 95% CI)	‐0.19 [‐0.33, ‐0.06]

2.1 Single‐dose studies	12		Std. Mean Difference (Random, 95% CI)	‐0.23 [‐0.37, ‐0.09]
2.2 Multiple‐dose studies	9		Std. Mean Difference (Random, 95% CI)	‐0.12 [‐0.47, 0.23]
3 Pain: short‐term follow‐up, subgroup analysis, neuropathic pain participants only Show forest plot	14		Std. Mean Difference (Fixed, 95% CI)	‐0.20 [‐0.27, ‐0.12]

3.1 Low‐frequency ≤ 1 Hz	5		Std. Mean Difference (Fixed, 95% CI)	0.15 [‐0.02, 0.32]
3.2 High‐frequency ≥ 5 Hz	14		Std. Mean Difference (Fixed, 95% CI)	‐0.27 [‐0.35, ‐0.19]
4 Pain: short‐term follow‐up, subgroup analysis, non‐neuropathic pain participants only Show forest plot	6		Std. Mean Difference (Fixed, 95% CI)	‐0.19 [‐0.44, 0.05]

4.1 Low‐frequency ≤ 1 Hz	1		Std. Mean Difference (Fixed, 95% CI)	0.16 [‐0.29, 0.61]
4.2 High‐frequency ≥ 5 Hz	5		Std. Mean Difference (Fixed, 95% CI)	‐0.34 [‐0.63, ‐0.05]
5 Pain: short‐term follow‐up, subgroup analysis: motor cortex studies only, low‐frequency studies excluded Show forest plot	17		Std. Mean Difference (Random, 95% CI)	‐0.32 [‐0.46, ‐0.17]

5.1 Single‐dose studies	12		Std. Mean Difference (Random, 95% CI)	‐0.39 [‐0.51, ‐0.27]
5.2 Multiple‐dose studies	5		Std. Mean Difference (Random, 95% CI)	‐0.07 [‐0.41, 0.26]
6 Sensitivity analysis ‐ imputed correlation coefficient increased. Pain: short‐term follow‐up Show forest plot	23		Std. Mean Difference (Random, 95% CI)	‐0.21 [‐0.34, ‐0.08]

6.1 Low‐frequency ≤ 1 Hz	7		Std. Mean Difference (Random, 95% CI)	0.15 [0.01, 0.29]
6.2 High‐frequency ≥ 5 Hz	22		Std. Mean Difference (Random, 95% CI)	‐0.30 [‐0.44, ‐0.16]
7 Sensitivity analysis ‐ imputed correlation coefficient decreased. Pain: short‐term follow‐up Show forest plot	22		Std. Mean Difference (Random, 95% CI)	‐0.20 [‐0.34, ‐0.06]

7.1 Low‐frequency ≤ 1 Hz	6		Std. Mean Difference (Random, 95% CI)	0.17 [‐0.03, 0.37]
7.2 High‐frequency ≥ 5 Hz	21		Std. Mean Difference (Random, 95% CI)	‐0.28 [‐0.42, ‐0.13]
8 Sensitivity analysis ‐ imputed correlation increased. Pain: short‐term follow‐up, subgroup analysis: motor cortex studies only, low‐frequency studies excluded Show forest plot	17		Std. Mean Difference (Random, 95% CI)	‐0.33 [‐0.47, ‐0.20]

8.1 Single‐dose studies	12		Std. Mean Difference (Random, 95% CI)	‐0.41 [‐0.53, ‐0.29]
8.2 Multiple‐dose studies	5		Std. Mean Difference (Random, 95% CI)	‐0.08 [‐0.39, 0.23]
9 Sensitivity analysis ‐ imputed correlation decreased. Pain: short‐term follow‐up, subgroup analysis: motor cortex studies only, low‐frequency studies excluded Show forest plot	17		Std. Mean Difference (Random, 95% CI)	‐0.31 [‐0.46, ‐0.17]

9.1 Single‐dose studies	12		Std. Mean Difference (Random, 95% CI)	‐0.38 [‐0.49, ‐0.27]
9.2 Multiple‐dose studies	5		Std. Mean Difference (Random, 95% CI)	‐0.11 [‐0.48, 0.25]
10 Sensitivity analysis ‐ inclusion of high risk of bias studies. Pain: short‐term follow‐up Show forest plot	25		Std. Mean Difference (Fixed, 95% CI)	‐0.23 [‐0.31, ‐0.16]

10.1 Low‐frequency ≤ 1 Hz	9		Std. Mean Difference (Fixed, 95% CI)	0.09 [‐0.05, 0.24]
10.2 High‐frequency ≥ 5 Hz	23		Std. Mean Difference (Fixed, 95% CI)	‐0.34 [‐0.42, ‐0.26]
11 Sensitivity analysis ‐ inclusion of high risk of bias studies. Pain: short‐term follow‐up, subgroup analysis: motor cortex studies only, low‐frequency studies excluded Show forest plot	21		Std. Mean Difference (Random, 95% CI)	‐0.39 [‐0.56, ‐0.23]

11.1 Single‐dose studies	14		Std. Mean Difference (Random, 95% CI)	‐0.36 [‐0.48, ‐0.24]
11.2 Multiple‐dose studies	8		Std. Mean Difference (Random, 95% CI)	‐0.50 [‐0.99, ‐0.01]
12 Pain: short‐term follow‐up, subgroup analysis: prefrontal cortex studies only Show forest plot	4		Std. Mean Difference (Random, 95% CI)	‐0.47 [‐1.48, 0.54]

12.1 Multiple‐dose studies	4		Std. Mean Difference (Random, 95% CI)	‐0.47 [‐1.48, 0.54]
13 Sensitivity analysis ‐ inclusion of high risk of bias studies. Pain: short‐term follow‐up, subgroup analysis: prefrontal cortex studies only Show forest plot	5		Std. Mean Difference (Random, 95% CI)	‐0.48 [‐1.32, 0.37]

13.1 Multiple‐dose studies	5		Std. Mean Difference (Random, 95% CI)	‐0.48 [‐1.32, 0.37]
14 Pain: medium‐term follow‐up Show forest plot	7		Std. Mean Difference (Random, 95% CI)	‐0.18 [‐0.43, 0.06]

14.1 Low‐frequency ≤ 1 Hz	1		Std. Mean Difference (Random, 95% CI)	0.36 [‐0.41, 1.13]
14.2 High‐frequency ≥ 5 Hz	6		Std. Mean Difference (Random, 95% CI)	‐0.23 [‐0.48, 0.03]
15 Sensitivity analysis ‐ inclusion of high risk of bias studies. Pain: medium‐term follow‐up Show forest plot	10		Std. Mean Difference (Random, 95% CI)	‐0.43 [‐0.76, ‐0.10]

15.1 Low‐frequency ≤ 1 Hz	2		Std. Mean Difference (Random, 95% CI)	‐0.08 [‐1.26, 1.10]
15.2 High‐frequency ≥ 5 Hz	9		Std. Mean Difference (Random, 95% CI)	‐0.48 [‐0.83, ‐0.13]
16 Pain: long‐term follow‐up Show forest plot	3		Std. Mean Difference (Random, 95% CI)	‐0.12 [‐0.46, 0.21]

17 Sensitivity analysis ‐ inclusion of high risk of bias studies. Pain: long‐term follow‐up Show forest plot	4		Std. Mean Difference (Random, 95% CI)	‐0.46 [‐1.10, 0.17]

18 Disability/pain interference: short‐term follow‐up Show forest plot	5		Std. Mean Difference (Random, 95% CI)	‐0.29 [‐0.87, 0.29]

19 Disability/pain interference: medium‐term follow‐up Show forest plot	4		Std. Mean Difference (Random, 95% CI)	‐0.37 [‐1.07, 0.33]

20 Disability/pain interference: long‐term follow‐up Show forest plot	3		Std. Mean Difference (Random, 95% CI)	‐0.23 [‐0.62, 0.16]

21 Quality of life: short‐term follow‐up (Fibromyalgia Impact Questionnaire) Show forest plot	3	80	Mean Difference (IV, Random, 95% CI)	‐10.38 [‐14.89, ‐5.87]

22 Quality of life: medium‐term follow‐up (Fibromyalgia Impact Questionnaire) Show forest plot	3	80	Mean Difference (IV, Fixed, 95% CI)	‐11.49 [‐17.04, ‐5.95]

23 Quality of life: long‐term follow‐up Show forest plot	1		Std. Mean Difference (Random, 95% CI)	Totals not selected

Comparison 1. Repetitive transcranial magnetic stimulation (rTMS)

Comparison 2. Cranial electrotherapy stimulation (CES)

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Pain: short‐term follow‐up Show forest plot	5	270	Std. Mean Difference (IV, Random, 95% CI)	‐0.24 [‐0.48, 0.01]

2 Disability/function/pain interference Show forest plot	1		Mean Difference (IV, Fixed, 95% CI)	Subtotals only

3 Quality of life Show forest plot	1		Std. Mean Difference (IV, Random, 95% CI)	Subtotals only

Comparison 2. Cranial electrotherapy stimulation (CES)

Comparison 3. Transcranial direct current stimulation (tDCS)

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Pain: short‐term follow‐up Show forest plot	11		Std. Mean Difference (Random, 95% CI)	‐0.18 [‐0.46, 0.09]

1.1 Single‐dose studies	3		Std. Mean Difference (Random, 95% CI)	‐0.18 [‐0.41, 0.05]
1.2 Multiple‐dose studies	8		Std. Mean Difference (Random, 95% CI)	‐0.22 [‐0.69, 0.25]
2 Pain: short‐term follow‐up, subgroup analysis: motor cortex studies only Show forest plot	11		Std. Mean Difference (Random, 95% CI)	‐0.23 [‐0.48, 0.01]

2.1 Single‐dose studies	3		Std. Mean Difference (Random, 95% CI)	‐0.18 [‐0.41, 0.05]
2.2 Multiple‐dose studies	8		Std. Mean Difference (Random, 95% CI)	‐0.35 [‐0.79, 0.09]
3 Pain: short‐term sensitivity analysis: correlation increased Show forest plot	11		Std. Mean Difference (Random, 95% CI)	‐0.20 [‐0.47, 0.06]

4 Pain: short‐term sensitivity analysis: correlation decreased Show forest plot	11		Std. Mean Difference (Random, 95% CI)	‐0.23 [‐0.51, 0.06]

5 Pain: short‐term follow‐up, subgroup analysis: motor cortex studies only, sensitivity analysis: correlation increased Show forest plot	11		Std. Mean Difference (Random, 95% CI)	‐0.23 [‐0.48, 0.02]

5.1 Single‐dose studies	3		Std. Mean Difference (Random, 95% CI)	‐0.18 [‐0.41, 0.05]
5.2 Multiple‐dose studies	8		Std. Mean Difference (Random, 95% CI)	‐0.35 [‐0.79, 0.10]
6 Pain: short‐term follow‐up, subgroup analysis: motor cortex studies only, sensitivity analysis: correlation decreased Show forest plot	11		Std. Mean Difference (Random, 95% CI)	‐0.24 [‐0.48, ‐0.00]

6.1 Single‐dose studies	3		Std. Mean Difference (Random, 95% CI)	‐0.18 [‐0.41, 0.05]
6.2 Multiple‐dose studies	8		Std. Mean Difference (Random, 95% CI)	‐0.36 [‐0.79, 0.07]
7 Pain: medium‐term follow‐up Show forest plot	5		Std. Mean Difference (Random, 95% CI)	‐0.20 [‐0.63, 0.24]

8 Disability (pain interference): short‐term follow‐up Show forest plot	1		Mean Difference (IV, Random, 95% CI)	Subtotals only

9 Quality of life: short‐term follow‐up Show forest plot	2	42	Std. Mean Difference (IV, Random, 95% CI)	0.88 [0.24, 1.53]

10 Quality of life: medium‐term follow‐up Show forest plot	1		Std. Mean Difference (IV, Random, 95% CI)	Totals not selected

11 Pain: short‐term follow‐up, subgroup analysis: motor cortex studies only Show forest plot	11		Std. Mean Difference (Random, 95% CI)	‐0.26 [‐0.49, ‐0.03]

11.1 Single‐dose studies	3		Std. Mean Difference (Random, 95% CI)	‐0.18 [‐0.41, 0.05]
11.2 Multiple‐dose studies	8		Std. Mean Difference (Random, 95% CI)	‐0.38 [‐0.80, 0.03]

Comparison 3. Transcranial direct current stimulation (tDCS)

Comparison 4. Reduced impedance non‐invasive cortical electrostimulation (RINCE)

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Pain: short‐term follow‐up Show forest plot	1		Mean Difference (IV, Fixed, 95% CI)	Subtotals only

2 Fibromyalgia Impact Questionnaire total score Show forest plot	1		Mean Difference (IV, Fixed, 95% CI)	Subtotals only

Comparison 4. Reduced impedance non‐invasive cortical electrostimulation (RINCE)