The Effects of Graded Motor Imagery and Its Components on Chronic Pain: A Systematic Review and Meta-Analysis K. Jane Bowering, *Neil E…. [628850]

Critical Review
The Effects of Graded Motor Imagery and Its Components on
Chronic Pain: A Systematic Review and Meta-Analysis
K. Jane Bowering, *Neil E. O’Connell,yAbby Tabor, *,zMark J. Catley, *Hayley B. Leake, *
G. Lorimer Moseley, *,xand Tasha R. Stanton *,x
*Sansom Institute for Health Research, University of South Australia, Adelaide, South Australia, Australia.
yCentre for Research in Rehabilitation, Brunel University, Middlesex, United Kingdom.
zKing’s College, London, United Kingdom.
xNeuroscience Research Australia, Randwick, New South Wales, Australia.
Abstract: Graded motor imagery (GMI) is becoming increasingly used in the treatment of chronic
pain conditions. The objective of this systematic review was to synthesize all evidence concerning
the effects of GMI and its constituent components on chronic pain. Systematic searches were conduct-ed in 10 electronic databases. All randomized controlled trials (RCTs) of GMI, left/right judgmenttraining, motor imagery, and mirror therapy used as a treatment for chronic pain were included.Methodological quality was assessed using the Cochrane risk of bias tool. Six RCTs met our inclusioncriteria, and the methodological quality was generally low. No effect was seen for left/right judgmenttraining, and conflicting results were found for motor imagery used as stand-alone techniques, butpositive effects were observed for both mirror therapy and GMI. A meta-analysis of GMI versus usual
physiotherapy care favored GMI in reducing pain (2 studies, n = 63; effect size, 1.06 [95% confidence
interval, .41, 1.71]; heterogeneity, I
2= 15%). Our results suggest that GMI and mirror therapy alone
may be effective, although this conclusion is based on limited evidence. Further rigorous studies areneeded to investigate the effects of GMI and its components on a wider chronic pain population.
Perspective: This systematic review synthesizes the evidence for GMI and its constituent compo-
nents on chronic pain. This review may assist clinicians in making evidence-based decisions on man-aging patients with chronic pain conditions.
ă2013 by the American Pain Society
Key words: Graded motor imagery, GMI, mirror therapy, motor imagery, left/right judgments, chronic
pain, systematic review.
Rapid advances in our understanding of the role of
the brain in chronic pain have seen the develop-ment of treatments for chronic pain that directly
target cortical reorganization.
30,44The first of these
treatments was developed in response to remarkablefindings in amputees with phantom limb pain (PLP),which showed that pain was associated withreorganization of the primary sensory cortexcontralateral to the amputated limb. The normalrepresentation of the amputated hand had been
invaded by the representation of the lip.
11This cortical
reorganization has also been demonstrated for chronic
low back pain, in which representation of the painfulside of the back was enlarged and shifted medially ascompared with representation in healthy controls.
10
That primary sensory cortex receptive fields can be mod-ified by tactile stimuli with a behavioral relevance (forexample, eating or braille) is now well accepted.
12Flor
et al aimed to exploit this plasticity in amputees withPLP by 2 weeks of sensory discrimination training, inwhich participants discriminated between stimuli of dif-
ferent frequencies and at different locations on their
stump.
9,13Their randomized controlled trial (RCT)G.L.M. is supported by an NHMRC Senior Research Fellowship. T.R.S. is
supported by the Canadian Institutes of Health Research PostdoctoralTraining Fellowship [ID 223354]. This work supported by NHMRC projectgrant ID 1008017.
The authors have no conflicts of interest to report.Supplementary data related to this article can be found at http://dx.doi.
org/10.1016/j.jpain.2012.09.007 .
Address reprint requests to Tasha R. Stanton, University of South Aus-
tralia, Sansom Institute for Health Research, Adelaide, South Australia
5000, AU. E-mail: Tasha.stanton@unisa.edu.au
1526-5900/$36.00
ă2013 by the American Pain Society
http://dx.doi.org/10.1016/j.jpain.2012.09.007
3The Journal of Pain, Vol 14, No 1 (January), 2013: pp 3-13
Available online at www.jpain.org andwww.sciencedirect.com

showed normalization of cortical organization and
a clinically important reduction of pain. This process,from discovery of altered sensory cortex organizationto targeted sensory discrimination training for clinicalbenefit, has been repeated in complex regional painsyndrome (CRPS).
15,17,32,34
As well as physiological evidence of disrupted somato-
topic representation in chronic pain, there is alsobehavioral evidence of disrupted spatial representa-
tion—disrupted processing of stimuli delivered to
healthy body parts held in the affected space,
31the ab-
normality of the perceived size of the painful bodypart,
19,20,27,29and poor voluntary movement and motor
imagery performance.1,5,6,25,28,37-39One treatment that
was developed to directly target these corticaldisruptions is graded motor imagery (GMI), a 3-stagetreatment that aims to gradually engage cortical motornetworks without triggering the protective response ofpain. This treatment gets its theoretical frameworkfrom the principle established in the physical therapies,
of graded increase in activity. This principle is adapted
in GMI to cater to both the overly sensitive nociceptionsystem and the disrupted cortical mechanisms men-tioned above. GMI was developed initially for an applica-tion to chronic limb pain or PLP but has been extendedclinically to chronic back pain, where a component ofGMI has been used for some time.
43
The first stage of the GMI program is left/right judg-
ments of photographs that depict the affected area.For limb pain, this involves viewing an image of a limband judging whether that image depicts a left or a rightlimb. Functional brain imaging studies in healthy sub-
jects have shown that this task selectively activates the
premotor cortex without activating primary motorareas.
35,41,45The second stage, motor imagery, requires
imagined movement of the area. These imaginedmovements have been demonstrated to activate motorcortical areas similar to those activated in the actualexecution of that movement.
8For the final stage, mirror
therapy, patients place their affected limb inside a mirrorbox and watch movements of their nonaffected limb inthe mirror, giving the illusion of a moving, but pain-free, affected limb. This task activates the motor cortex
and also provides a strong visual input to the cortex
that the movements are occurring normally and withoutimpediment.
18While functional brain imaging studies
have supported the proposed cortical activation foreach stage of GMI in healthy subjects, no studies have in-vestigated cortical activation of GMI stages in pain pa-tients. These imaging studies nonetheless providesupport for the possibility that similar sequential activa-tion of cortical areas within each stage of the GMI pro-gram could occur in pain patients.
Both GMI and its components have been used in the
clinical setting to treat chronic pain conditions such asCRPS, PLP, and back pain. However, an issue that re-mains to be addressed is whe ther the evidence sup-
ports or negates the use of GMI or its components inthe treatment of a wider chronic pain population. Arecent systematic review evaluating interventions fortreating CRPS supported the use of GMI.
7However,a recent clinical audit of CRPS multimodal manage-ment including but not limited to GMI clearly showedno benefit of treatment.
14These conflicting findings,
and that GMI has not, to our knowledge, been empir-ically evaluated in a wider chronic pain population,highlight the importance of systematic evaluation ofthe entire literature concerning GMI and its compo-nents. The aim of this review and meta-analysis wasto synthesize all available literature regarding the
efficacy of GMI programs, or any of the 3 constituent
components, on chronic pain. The results of thissystematic review will enable clinicians to makeevidence-based decisions on the use of GMI withchronic pain patients.
Methods
Data Sources
For this review, several health-based databases were
searched from their relative inception through January2012. The electronic search was performed using thefollowing databases: Medline (via OvidSP), Embase
(via Ovid SP), Cumulative Index to Nursing and Allied
Health Literature (CINAHL), Scopus, Academic SearchPremier, Web of Science, Allied and ComplementaryMedicine, PubMed, the Cochrane Collaboration, andthe Physiotherapy Evidence Database (PEDro). A sensi-tive search was completed using a combination of keywords and relevant subject headings for GMI, its com-ponents, and chronic pain. The relevant subject head-ings were determined specific to each database. Thecomplete Medline search strategy is provided inAppendix A . Searches were limited to English lan-
guage and humans only. To attempt to identify greyliterature (specifically nonindexed published trials,conference abstracts, and book chapters), expertswere contacted and asked to contribute any materialsnot identified by database search. The references of allrelevant articles were also hand-searched for furtherarticles. We did not search clinical trials registers forunpublished studies.
Study Selection
Four reviewers (K.J.B., A.T., M.J.C., and H.B.L.) were
paired and each pair independently screened the titlesand abstracts of half of the potential studies—thus, allpapers were screened by 2 people. Results of the screen-ing process were compared within pairs. In this process,studies were retained if they evaluated GMI or at least 1
component of GMI. Following initial screening, the full
texts of potentially relevant studies were retrieved andreviewed independently by 2 reviewers (K.J.B. and A.T.).Studies were retained if they met the following criteria:human adult subjects (>18 years of age); clinically vali-dated pain measure used; RCT; and subjects all hada chronic pain condition lasting longer than 3 months.No restrictions were placed on the comparison groupused (ie, placebo, wait list control, or other active treat-ment). Any discrepancies were resolved through4
The Journal of Pain Effect of Graded Motor Imagery on Chronic Pain

discussion, or if necessary, through consultation with
a third independent reviewer.
Outcome Measures
Pain intensity ratings were the primary outcome of
interest for this review. This included self-reportedmeasures such as the McGill Pain Questionnaire, a visualanalog scale (VAS), a numerical rating scale (NRS), a neu-ropathic pain scale, or a categorical rating of pain (suchas mild, moderate, severe). A rating of pain using 1 ofthese measures was required immediately preinterven-tion and immediately postintervention. Follow-up painratings were a secondary outcome of interest for this re-
view.
Risk of Bias Assessment and Data
Extraction
Two reviewers (K.J.B. and A.T.) independently assessed
the risk of bias of included studies using the Cochrane
Collaboration’s risk of bias tool. For the category of‘‘other’’ sources of bias, the reviewers were particularlyconcerned with similarity of pain scores at baseline, asthis is recommended by other quality assessment toolssuch as PEDro.
36In the ‘‘other’’ source of bias category
we also included evaluation of sample sizes (ie, lessthan 50 participants per treatment arm considereda high risk of bias).
22These items were added as we antic-
ipated that studies identified were likely to be small and,as such, these factors were more likely to represent a sig-nificant source of bias.
For all eligible studies, data extraction was completed
independently by 2 reviewers using a customized dataextraction form. This data extraction form was pilotedbefore use. Data extracted included participant charac-teristics such as age, gender, pain condition, and length
of pain; the outcome measure used; the control and
treatment intervention choices and their length (min-utes per each session), frequency (sessions per day/week), and total duration (weeks of intervention); base-line and immediate postintervention pain scores; andfollow-up pain scores if provided. Any disagreements re-garding risk of bias or data extraction were resolvedthrough discussion or, if necessary, through consultationwith a third independent reviewer. If necessary, authorswere contacted to provide further information.
Data Synthesis
We sought to pool data for pain relief from studies
where adequate data were available. We planned a priorito pool data from studies comparing GMI programs with
usual care or no treatment, and to perform separate
meta-analyses for studies that investigated similar indi-vidual components of GMI.
Data were pooled using Review Manager 5 software
4
using a random effects inverse-variance approach. A ran-dom effects model was chosen as it was anticipated andsubsequently confirmed that there would be differencesin the populations and interventions studied that wouldsuggest that the effects might differ somewhat acrossstudies. Using the postintervention means of each groupand the pooled postintervention standard deviations ofpain scores, the standardized mean difference (Hedge’sg) was calculated for each study to allow comparison be-tween studies. Effect sizes were interpreted according toCohen
40(#.2 small, .5 moderate, $.8 large). We used the
chi-square test to detect statistically significant hetero-geneity and the I
2statistic to estimate the amount of het-
erogeneity. When heterogeneity was high, we did notpool the outcomes. Further, we considered it inappropri-
ate to pool data from studies that used full GMI pro-
grams with those that used individual components ofGMI because it does not follow that the different typesof interventions should be estimating the same effectsize. We therefore planned separate meta-analyses forthese types of studies considering both short-term (im-mediately postintervention or the closest measurepresented to that point) and follow-up (>4 weeks postin-tervention) time points. We undertook a sensitivity anal-ysis to investigate the influence of using a random effectsmodel by reanalyzing the data using a fixed effects
model.
In studies that evaluated a comprehensive GMI pro-
gram, the effect sizes for the first component (ie, left/
right judgments stage) were also calculated using post-intervention scores when individual participant datawere present. It was decided, a priori, that effect sizeswould not be calculated for the second or third GMItreatment components (motor imagery and mirrortherapy, respectively) because in these latter compo-nents, the methodological tenets of the RCT studydesign do not hold. Specifically, participants are notre-randomized following each component stage, so
there are preintervention pain differences between
groups in the latter stages. That the responses of thelatter components were due to carryover effects orcontinuing improvement from the previous treatmentcould therefore not be ruled out. We did not establishany a priori sensitivity or subgroup analyses because weanticipated identifying inadequate data to support thisprocess.
Results
Study Description
The initial literature search yielded 6,160 records fol-
lowing the removal of duplicates. Six thousand fifteenstudies were excluded in the initial screening of titleand abstracts. One hundred thirty-nine studies werethen excluded following review of the full text. Themost prevalent reason for exclusion was that articles
did not include primary research data; primarily, these
were reviews, conference abstracts, and book chapters,all presented in a narrative form. Other reasons for exclu-sion were studies that recruited sample populationswithout chronic pain or did not evaluate pain outcomemeasures, were not of RCT design, were non-Englishstudies, and that recruited children. The screening andreview process is shown in a PRISMA flow-diagram inFig 1 . Key data of the remaining 6 RCTs included are sum-
marized in Table 1 .Bowering et al
The Journal of Pain 5

Characteristics of Included Studies
Three studies evaluated the effects of GMI on
chronic pain.23,24,26Two of these studies compared
a 6-week program of GMI to usual physiotherapy
care.23,24The third study compared an ordered
program of GMI to an unordered program of GMI.26
Participants were instructed to spend 10 minutes ofeach waking hour on the intervention. All studies col-lected follow-up data: 1 study at 6 weeks postinterven-
tion,
241 study at 12 weeks postintervention,26and 1
study at 6 months postintervention.23These studies
used varying methods of collecting participant pain
scores. The author of each study was contacted, andNRS data for each participant’s pain level was pro-vided. These NRS data were used in the analyses.Only 1 study
23provided data on adherence to the
treatment program. This study found that both GMIand usual care groups had adherence rates of 75%.
Three other studies evaluated individual components
of GMI.
2,3,21No studies primarily evaluated left/right
judgments; however, 2 studies23,24evaluating GMI
provided sufficient data to enable calculation of effectsizes for the 2 weeks of left/right judgment training.Two studies
2,3evaluated the effects of motor imagery.
Three studies2,3,21evaluated the effects of mirror
therapy on chronic pain. The time spent on theintervention differed between studies. In 1 study,participants completed 5 1-hour sessions of mirror ther-apy a week.
21In the second study, participants spent 30
minutes per day doing either mirror therapy or motorimagery, depending on their group allocation.2In the
third study, participants spent 15 minutes per day doingeither mirror therapy or motor imagery, depending ontheir group allocation.
3Follow-up data from these stud-
ies were collected at either 4 weeks2,3or 6 months.21All 3
studies used 100-mm VAS data to report participants’pain levels.
Characteristics of Included Populations
The participants in each study had experienced
pain for greater than 3 months. The chronic painconditions included CRPS,
2,23,24,26PLP,3,23and pain
following stroke.21Studies including children were
excluded from this review. The mean age in each studyranged from 32 to 57 years. Overall, there were more fe-males (n = 90) than males (n = 81) in the included studies.
Risk of Bias of Included Studies
The results of the risk of bias assessment are shown in
Table 2 (see also the Supplementary graph for a repre-
sentation of risk of bias results). The study appraisedto be at lowest risk of bias was that by Moseley,
23which
met every criterion except the blinding of therapistsand participants and the ‘‘other’’ category, for its smallsample size. None of the 6 included RCTs met the blind-ing of therapists and participants criterion. In therapytrials such as these, direct participant-therapist involve-ment means that blinding is not feasible; hence, all 6RCTs had nonblinded therapists and participants. While
Figure 1. The PRISMA flow-diagram describing the screening and review process.6 The Journal of Pain Effect of Graded Motor Imagery on Chronic Pain

blinding in these trials is not feasible, it is still an inher-
ent source of bias that must be highlighted for everystudy. No study was free of additional bias, as all studieshad sample sizes less than 50. Michielsen et al
21pre-
sented additional bias in that they failed to reportany baseline similarities or differences between groupson pain scores. Two other studies also failed to report
whether groups had similar baseline pain levels.
2,3The
lack of this information has implications for the
validity of the observed effect sizes as it is uncertainwhether differences found between groups may havebeen influenced by baseline group differences. Thesesame studies also failed to provide informationregarding whether the person who determinedparticipant eligibility was blinded to treatmentallocation. Given the lack of participant/therapistblinding due to nature of the interventions within thestudies, all studies were considered to have someinherent bias.
Outcomes
Four authors were contacted to gain additional infor-
mation required to calculate the effect size of their inter-
vention.2,3,21,23,24,26One author could not be contacted,
so the effect size for this study could not be calculated.2
The effect sizes for the remaining studies are presented
inTable 3 .
GMI Program
Three studies evaluated the effects of a 6-week GMI pro-
gram on chronic pain, with all finding that GMI reducedpain when compared to usual physiotherapy care
23,24
and unordered GMI.26The 2 studies comparing GMI toTable 1. Study Characteristics Data for Randomized Controlled Trials of Graded Motor Imagery or
Its Components for Chronic Pain
STUDY PARTICIPANTS CONDITION INTERVENTIONOUTCOME
MEASURES
Studies evaluating the components of GMI
Michielsen et al21n=4 0
Mean age = 57 *
Gender = 50% maleChronic pain following
stroke (mean time
since stroke 3.9 years)Exp: 6-week bilateral hand movement
with mirror therapy program. Practiced5x/week, 1 hour per session.
Con: 6-week bilateral hand movements.
Practiced 5x/week, 1 hour per session.100-mm VAS y
Follow-up:
6 months
Cacchio et al
2n=2 4Median age = 62
(53 to 71) z
Gender = 46% maleCRPS Exp: 4-week mirror therapy program,
30 min daily.
Con: 4-week covered mirror program,
30 min daily.
Exp2: 4 weeks of motor imagery,
30 min daily.100-mm VASFollow-up:
4 weeks
Chan et al
3n=2 2Mean age = 29 68.8x
Gender = 100% malePLP Exp: 4-week mirror therapy program,
15 min daily.
Con: 4-week covered mirror program,
15 min daily.
Exp2: 4 weeks of motor imagery,
15 min daily.100-mm VAS y
Follow-up:
4 weeks
Studies evaluating GMI
Moseley
23{ n=5 0
Mean age = 41 616x
Gender = 36% maleCRPS, PLP following
amputation or brachial
plexus avulsionExp: laterality retraining, motor imagery,
mirror therapy. 2 weeks each component,10 min for each waking hour.
Con: usual physiotherapy/other treatment.MPQ, NRS y
Follow-up:
6 months
Moseley
26n=2 0
Mean age = 32 611x
Gender = 30% maleCRPS type 1 Exp: sequential GMI. 2 weeks each
component, 10 min for each waking hour.
Con: nonsequential GMI: MI, left/right, MI.
2 weeks each component, 10 min for each
waking hour.
Con2: nonsequential GMI: left/right, mirror,
left/right. 2 weeks each component,10 min for each waking hour.NPS, NRS y
Follow-up:
12 weeks
Moseley
26{ n=1 3Mean age = 57 619x
Gender = 30% maleCRPS type 1 Exp: sequential GMI. 2 weeks each component,
10 to 15 min for each waking hour.
Con: usual physiotherapy/other treatment.NPS, NRS y
Follow-up:
6 weeks
Abbreviations: Exp, experimental group; Exp2, secondary control group; Con, control group; Con2, secondary control group; n, number recruited (prior to drop-out or
loss to follow-up); MPQ, McGill Pain Questionnaire; NPS, neuropathic pain scale; MI, motor imagery; left/right, left/right judgments; mirror, mirror therapy.
*Range or standard deviation not provided.
yData used to calculate effect sizes.
zRange.
xStandard deviation.
{Due to the presence of individual participant postintervention data, the left/right judgments component of treatment was also examined.Bowering et al The Journal of Pain 7

usual physiotherapy care both found large effect sizes
(1.70 [95% confidence interval (CI), .36, 3.04]24and .89
[95% CI, .31, 1.47]23). In the study that compared a course
of GMI to an unordered course of GMI,26moderate-to-
large effects in favor of the ordered GMI were found
(.73 [95% CI, /C0.41, 1.87] and .99 [95% CI, /C0.19, 2.17]).
The immediate postintervention results of the 2 studies
comparing GMI with usual care were pooled.23,24The
results of the study evaluating GMI versus unorderedGMI
26were not included in the meta-analysis because
the control group intervention had pronounced differ-ences; this heterogeneity meant that pooling of these
data was not appropriate. The heterogeneity of the
pooled studies was low (I
2= 15%) and produced a large
pooled effect size (1.06 [95% CI, .41, 1.71]; Fig 2 ). While
the statistical heterogeneity of the studies was low, itmust be noted that the chronic pain population in eachstudy differed slightly; 1 included only CRPS participants
24and the other a mix of CRPS, PLP , and pain after brachial
plexus avulsion.23Sensitivity analysis using fixed effects,
rather than random effects, meta-analysis had no substan-
tive impact on our findings (I2= 0%; effect size, .97 [95%
CI, .52, 1.42]; test for overall effect, P<. 0 0 0 1 ) .
Follow-up data also suggest an effect of GMI further re-
ducing pain, with large effect sizes reported at 6 monthsfor GMI when compared to usual physiotherapy care(1.59 [95% CI, .28, 2.90]
24and 1.68 [95% CI, 1.02, 2.33]),23
and also at 12 weeks for GMI when compared to an unor-dered GMI program (1.35 [95% CI, .09, 2.60] and 1.31 [95%CI, .06, 2.55]).
26Pooling of these effect estimates was not
considered appropriate as the follow-up in each studywas conducted at a markedly different time point.
Left/Right Judgments
No studies were found that evaluated left/right judg-
ments as the primary intervention, although 2 studiesTable 2. Risk of Bias Assessment of Included Randomized Controlled Trials
RANDOM
ALLOCATIONCONCEALED
ALLOCATIONBLINDING OF
PARTICIPANTS /
THERAPISTSOUTCOME
ASSESSORSINCOMPLETE
DATANOSELECTIVE
OUTCOME REPORTINGFREE OF
ADDITIONAL BIAS
Michielsen et al21YY NY Y Y N
Cacchio et al2UU NN U Y N
Chan et al3UU NN U Y N
Moseley23YY NY Y Y N
Moseley26YU NY Y Y N
Moseley24YU NY Y N N
Abbreviations: Y, yes, low risk of bias; N, no, high risk of bias; U, unclear, uncertain risk of bias.
Table 3. Effect Sizes (95% CI) for GMI and Its Components on Chronic Pain When Compared to
Control Groups
STUDY CONTROLNUMBER OF PARTICIPANTS POSTINTERVENTION PAIN(MEAN6SD)
EFFECT SIZE(95% CI) CONTROL INTERVENTION CONTROL INTERVENTION
Laterality judgment task
Moseley23Usual care 25 25 54613 48 614 .44 (/C0.12, 1.00)
Moseley24Usual care 6 7 61610 57 615 .29 (/C0.81, 1.39)
MI
Cacchio et al2Covered mirror therapy 8 8 — — —*
Chan et al3Covered mirror therapy 6 6 34622 58 620 /C01.05 ( /C02.30, .19)
Mirror therapy
Michielsen et al21Bilateral hand movements 19 17 9.2 614 8.8 610.8 .03 ( /C0.62, .69)
Cacchio et al2Covered mirror therapy
MI8
88
8—————*
—*
Chan et al
3Covered mirror therapyMI6
66
634622
5862017621
17621.73 (/C0.46, 1.92)
1.85 (.40, 3.29) y
GMI
Moseley
23Usual care 25 25 47616 33 615 .89 (.31 to 1.47) y
Moseley26MI, left/right, MILeft/right, mirror, left/right6
67
740610
42693368
3368.73 (/C0.41, 1.87)
.99 (/C0.19 to 2.17)
Moseley
24Usual care 6 7 58612 38 610 1.70 (.36, 3.04) y
Abbreviations: MI, motor imagery; left/right, left/right judgments; mirror, mirror therapy.
NOTE. The effect sizes are standardized mean differences, calculated using Hedge’s g (ie, the difference in postintervention pain scores between control and interven-
tion groups divided by the pooled standard deviation of the 2 groups, each weighted for sample size). Effect sizes are grouped according to intervention type. Positiveeffect sizes indicate a lower pain score in the intervention group, favoring the intervention group. Negative effect sizes indicate a lower pain score in the control group,
favoring the control group.
*Did not provide postintervention pain data for control or intervention groups.yP< .05; For all Moseley studies, pain scores and effect estimates are for NRS results.8 The Journal of Pain Effect of Graded Motor Imagery on Chronic Pain

investigated the effects of left/right judgments as part of
a GMI program on chronic pain.23,24Neither study found
statistically significant effect estimates for left/rightjudgments reducing pain when compared to usual care.However, the effect estimates produced were positive,
albeit small (.29 [95% CI, /C0.81, 1.39]
24and .44 [95% CI,
/C0.12, 1.00]23). The heterogeneity of the pooled studies
was low (I2= 0%) and produced a similarly small effect
estimate (.41 [95% CI, /C0.09, .91]; Fig 3 ). Sensitivity analysis
using fixed effects, rather than random effects, meta-
analysis again had no substantive impact on our findings(I
2= 0%; effect size, .41 [95% CI, /C0.09, .91]; test for overall
effect, P= .11).
Motor Imagery
None of the included studies had a primary aim of eval-
uating the effects of motor imagery on chronic pain.However, in 2 studies, motor imagery was used as a sec-ondary control group
2,3and was compared to covered
mirror therapy (in which the participant is instructed tolook at a mirror that is covered with a cloth so as tooffer no reflection; controlling for attention). Thesestudies found contrasting results. Chan et al
3found cov-
ered mirror therapy to be much more effective at reduc-
ing pain when compared to motor imagery, with a large
effect size found ( /C01.05 [95% CI, /C02.30, .19]). Interest-
ingly, participants receiving motor imagery treatmenthad increased pain levels (compared to baseline pain).Similar findings were reported by Cacchio et al,
2in which
6 out of 8 participants experienced increased pain levelsfollowing 4 weeks of motor imagery. However, Cacchioet al
2found no difference between motor imagery and
covered mirror therapy (5 of 8 participants had increasedpain in covered mirror therapy group). All pain assess-ments were immediately postintervention; no short- or
long-term follow-up data were available. Both studies
had small sample sizes and had a high risk of bias.
Mirror Therapy
A total of 3 studies evaluated mirror therapy as a stand-
alone treatment in chronic pain; in each study, mirror ther-apy was the primary treatment evaluated.2,3,21All 3
studies found positive effects of mirror therapy inreducing pain, despite using different control groups.The effect sizes ranged from trivial (.03 [95% CI, /C0.62,
.69],
21bilateral hand movement control group) to moder-
ate (.73 [95% CI, /C0.46, 1.92],3covered mirror control
group) to large (1.85 [95% CI, .40, 3.29],3motor imagery
control group). Notably, this final effect size was theonly statistically significant finding in the mirror therapyanalyses. This finding was further supported by Cacchioet al,
2who reported 7 of 8 participants in the mirror ther-
apy group experiencing decreased pain levels (comparedto only 1 of 8 participants in the covered mirror groupand only 2 of 8 participants in the motor imagery grouphaving decreased pain levels).
The pooling of studies of mirror therapy demonstrated
high levels of heterogeneity (I
2= 63%) but no effect ( P=
.07). Visual inspection of the forest plot showed that the1 study that utilized a different comparison condition
3
(motor imagery as opposed to covered mirror therapy)was the most likely source of this variance. Post hoc sen-sitivity analysis removing this study from the analysis re-duced this heterogeneity substantially (I
2= 2%) and
continued to demonstrate no effect ( P= .51). Sensitivity
analysis using fixed effects, rather than random effects,meta-analysis again had no substantive impact on ourfindings (I
2= 63%; effect size, .42 [95% CI, /C0.011, .95];
test for overall effect, P= .12).
Only 1 study presented follow-up data,21reporting
a small, nonsignificant effect size (.34 [95% CI, /C0.29, .96])
of mirror therapy compared to bilateral hand movementsin patients with pain following stroke at 6 months follow-up. All 3 studies were considered to have a high risk of bias.
Discussion
This is the first review to systematically evaluate the ef-
fect of GMI or its components on pain outcomes in peo-ple with chronic pain. The limited number of small RCTsavailable have found mixed results for the effects ofGMI or its components on chronic pain. Of the six RCTs
Figure 2. The pooled effect estimate for GMI versus usual care. Abbreviations: SD, standard deviation; CI, confidence interval.
Figure 3. The pooled effect estimate for left/right judgments versus usual care. Abbreviation: L/R, left/right.Bowering et al The Journal of Pain 9

identified, all contained some inherent bias. A key find-
ing of this review was that the majority of studies evalu-ated the effect of GMI or its components in CRPS or PLP,so it is unclear how GMI might relate to other chronicpain conditions. We will first consider our findings withrespect to individual components of GMI and then con-sider our findings with respect to full GMI programs.
Effect of Individual GMI Components on
Pain
Left/Right Judgments
Left/right judgments as a sole treatment appear to
have no effect on chronic pain.23,24That all effect sizes
were positive raises the possibility that even the pooled
data were underpowered to detect an effect, but one
might conclude that such a small effect is of little
clinical consequence.
Because left/right judgments have never been used as
a stand-alone treatment for chronic pain, there havebeen no studies that evaluate only left/right judgmentsas a treatment for chronic pain. Because only data fromthe first stage of a GMI program can currently be usedto evaluate the effect of left/right judgments, there areno data available on the long-term effect of this treat-ment. While left/right judgments alone may not producestatistically significant effects, they are an integral part
of the sequential GMI program that our results suggest
may be effective. Nonetheless, the clinical importancefor left/right judgments per se remains to be shown.
Motor Imagery
Motor imagery appears less effective at treating chronic
pain than covered mirror therapy.2,3Covered mirror
therapy was utilized in these studies as an inactivecontrol condition. That 2 studies found an increase inpain relative to baseline following motor imagery and 1observed greater improvements in an inactive control
group suggests that motor imagery might have the
potential to increase pain intensity. These findings areconsistent with those of a separate pre-/posttreatmenttrial not included in this review, in which motor imageryincreased pain and swelling in those with chronic armpain
33and speaks against the use of motor imagery alone
as a treatment for chronic pain.
Mirror Therapy
Mirror therapy is arguably the most studied compo-
nent of GMI in terms of its effects on pain; however,much of the available literature concerns case studies,
which were excluded from this review. The results of
the included studies were consistently positive in favorof mirror therapy reducing pain
2,3,21although there is
wide variance in the reported effect sizes.
This variance may reflect differences between studies
in the patient group and the choice of control treatment.For example, Michielsen et al
21recruited chronic pain pa-
tients with very low baseline pain scores, which are atyp-ical of chronic pain populations and provide minimalroom for improvement, creating the possibility of a flooreffect. In contrast, the baseline pain scores for partici-pants in the Chan et al
3study were high, providing the
opportunity for greater pain reductions and thereforea larger effect size. Both the Chan et al
3and Cacchio
et al2studies suggest that mirror therapy is substantially
more effective than motor imagery. However, motor im-agery appeared to increase participants’ pain levels, sothe difference might reflect both the worsening in thecontrol motor imagery group and the improvement in
the mirror therapy group.
One important consideration when interpreting the
effect of mirror therapy relative to a covered mirror con-
trol condition is the possible impact of variable placeboeffects. That is, covering the mirror might imply to thepatient that the mirror is the powerful component oftreatment and, as such, the covered mirror conditionmight not be perceived as credible by the patient. Asstated, blinding of therapists and participants in therapyinterventions such as mirror therapy is nearly impossible.Through matching the frequency and duration of ther-
apy sessions for both the covered and active mirror
groups, all studies achieved structural equivalence,which is particularly important in situations where indis-tinguishable placebo controls are not possible.
16While
covered mirror therapy as a control may not be ideal, itis a pragmatic control.
Effect of Full GMI Programs on Pain
Our results suggest that a GMI program likely has mod-
erate effects when compared to unordered GMI26and
large effects when compared to usual physiotherapycare.
23,24Both of the 2 identified studies evaluating
GMI versus usual physiotherapy found a large effectsize
23,24and clearly support the efficacy of GMI, at least
as delivered within 1 clinical center.
Recently published clinical audit data appear to con-
tradict the GMI findings of this review. Prospective auditdata from 32 patients treated at 2 interdisciplinary cen-ters showed no reduction in pain after a multimodal ap-proach that included GMI
14; indeed, some patients (30%
in 1 center and 50% in the other) actually reported an in-crease in their pain intensity following treatment. Theauthors proposed that variations in GMI protocol fromother studies and logistic constraints may have led tothe poor result. Nonetheless, this study, while less robustthan an RCT, highlights that independent replication ofthe results of Moseley
24and Moseley23in controlled tri-
als remains a research priority.
That GMI produced moderate effects when compared
to an unordered program of GMI26is interesting. The or-
der of GMI components seems to be important, which is
consistent with its proposed mechanism.42Moreover,
that there is such an effect relative to an unordered
treatment control group23,24suggests against the
possibility that the effects of GMI are largely due toa placebo response. That is, unordered GMI might bea more appropriate placebo control treatment infuture studies because it would capture much of thenovelty of GMI, but it appears to have little effect. Thatthis finding arises from a single small trial indicatesthat it also requires independent replication.10
The Journal of Pain Effect of Graded Motor Imagery on Chronic Pain

Given the limited data available, it is difficult to draw
firm conclusions, but these data and those relating to
the ordering of GMI components suggest that thegradual and progressive nature of GMI may be clinicallyimportant. Motor imagery particularly demands atten-tion. Not only was no significant benefit observed withmotor imagery, but unlike with left/right judgments,there was no suggestion in the data of a trend towardpain relief with this intervention and some evidence to
suggest a worsening of pain. This leads to the inevitable
question of whether GMI might be more effective with-out a motor imagery stage. To our knowledge, no studyhas currently investigated this.
The majority of the evidence pertains to patients with
CRPS, and we identified little evidence pertaining to theefficacy of GMI for other chronic pain conditions. Cau-tion is advised when extrapolating these findings tothe broader chronic pain population.
Limitations
Non-English studies were not included due to lack of
translation resources, and we did not search clinical trialsregisters for unpublished studies. However, experts inthe area of GMI/chronic pain were consulted regardingany missing relevant publications or active researchgroups and did not identify any relevant contributions,so we would suggest that the chance of missing a studywould seem low. The number of RCTs included was small,
and the majority had a high risk of bias. The limited num-
ber of studies published in this area also raises the possi-bility of publication bias.
In terms of the evidence of the effectiveness of full GMI
programs for reducing chronic pain, perhaps the strongestlimitation is that all of the included trials were completedby 1 research group with which we ourselves are affili-ated.
23,24,26To increase confidence in our findings, the
need for further trials of GMI by independent researchgroups cannot be overstated. There was significantheterogeneity between the included study populations;the type and duration of chronic pain varied, and studiesused a range of methods for sourcing and recruitingparticipants. Lastly, there were very few long-term fol-
low-ups (ie, all follow-ups were 6 months or earlier), which
suggests that the effectiveness of these treatments in thelonger term remains unknown.
In conclusion, while the results of this systematic re-
view suggest that the effectiveness of GMI and its com-ponents is encouraging in CRPS and PLP, no evidenceexists for these treatments in a wider chronic pain popu-lation. It is critical to acknowledge that more work is re-quired—the theoretical framework underlying thesetreatments suggests the value of additional trials ina wider chronic pain population. It is difficult to be cer-
tain of the findings because there are very few studies
of mixed risk of bias available. Differing methodologiesand samples within each study significantly limits thegeneralizability of these findings to people with CRPSor PLP, although there seems to be good reason to ex-tend this line of investigation into different chronicpain populations.
Supplementary Material
Supplementary data related to this article can be
found at http://dx.doi.org/10.1016/j.jpain.2012.09.007 .
References
1. Bray H, Moseley GL: Disrupted working body schema of
the trunk in people with back pain. Br J Sports Med 45:168-173, 2009
2. Cacchio A, De Blasis E, Necozione S, di Orio F, Santilli V:
Mirror Therapy for Chronic Complex Regional Pain Syn-
drome Type 1 and Stroke. N Engl J Med 361:634-636, 2009
3. Chan BL, Witt R, Charrow AP, Magee A, Howard R,
Pasquina PF, Heilman KM, Tsao JW: Mirror therapy for phan-tom limb pain. N Engl J Med 357:2206-2207, 2007
4. The Cochrane Collaboration. 2011. Review Manager
(RevMan), Version 5.1
5. Coslett H, Medina J, Kliot D, Burkey A: Mental motor imag-
ery and chronic pain: The foot laterality task. JINS 16:603, 2010
6. Coslett HB, Medina J, Kliot D, Burkey AR: Mental motor
imagery indexes pain: The hand laterality task. Eur J Pain14:1007-1013, 2010
7. Daly AE, Bialocerkowski AE: Does evidence support phys-
iotherapy management of adult Complex Regional PainSyndrome Type One? A systematic review. Eur J Pain 13:339-353, 2009
8. Decety J: The neurophysiological basis of motor imagery.
Behavl Brain Res 77:45-52, 19969. Flor H, Denke C, Schaefer M, Grusser S: Effect of sensorydiscrimination training on cortical reorganisation and phan-tom limb pain. Lancet 357:1763-1764, 2001
10. Flor H, Elbert T, Braun C, Birbaumer N: Extensive cortical
reorganization in chronic back pain patients. NeuroImage 5:S216, 1997
11. Flor H, Elbert T, Knecht S, Wienbruch C, Pantev C,
Birbaumer N, Larbig W, Taub E: Phantom-limb pain as a per-ceptual correlate of cortical reorganization following armamputation. Nature 375:482-484, 1995
12. Florence SL, Jain N, Kaas JH: Plasticity of Somatosen-
sory Cortex in Primates. Seminars in Neuroscience 9:3-12,1997
13. Huse E, Preissl H, Larbig W, Birbaumer N: Phantom limb
pain. The Lancet 358:1015, 2001
14. Johnson S, Hall J, Barnett S, Draper M, Derbyshire G,
Haynes L, Rooney C, Cameron H, Moseley GL, de C.Williams AC, McCabe C, Goebel A: Using graded motor im-
agery for complex regional pain syndrome in clinical prac-
tice: Failure to improve pain. Eur J Pain 16:550-561, 2012
15. Juottonen K, Gockel M, Sil /C19en T, Hurri H, Hari R, Forss N:
Altered central sensorimotor processing in patients with
complex regional pain syndrome. Pain 98:315-323, 2002
16. Machado L, Kamper S, Herbert R, Maher C, McAuley J:
Imperfect placebos are common in low back pain trials:Bowering et al The Journal of Pain 11

a systematic review of the literature. Eur Spine J 17:889-904,
2008
17. Maihofner C, Handwerker H, Neundorfer B, Birklein F:
Patterns of cortical reorganization in complex regional
pain syndrome. Neurology 61:1707-1715, 2003
18. Matthys K, Smits M, Van der Geest JN, Van der Lugt A,
Seurinck R, Stam HJ, Selles RW: Mirror-Induced Visual Illu-sion of Hand Movements: A Functional Magnetic ResonanceImaging Study. Arch Phys Med Rehabil 90:675-681, 2009
19. Melzack R: Phantom limbs and the concept of a neuro-
matrix. Trends Neurosci 13:88-92, 1990
20. Melzack R, Israel R, Lacroix R, Schultz G: Phantom limbs
in people with congenital limb deficiency or amputation inearly childhood. Brain 120:1603-1620, 1997
21. Michielsen ME, Selles RW, van der Geest JN, Eckhardt M,
Yavuzer G, Stam HJ, Smits M, Ribbers GM, Bussmann JBJ: Mo-tor Recovery and Cortical Reorganization After Mirror Ther-apy in Chronic Stroke Patients. Neurorehabil Neural Repair25:223-233, 2011
22. Moore RA, Eccleston C, Derry S, Wiffen P, Bell RF,
Straube S, McQuay H: ‘‘Evidence’’ in chronic pain – establish-
ing best practice in the reporting of systematic reviews. Pain
150:386-389, 2010
23. Moseley GL: Graded motor imagery for pathologic pain:
a randomized controlled trial. Neurology 67(12):2129-2134,2006
24. Moseley GL: Graded motor imagery is effective for long-
standing complex regional pain syndrome: A randomisedcontrolled trial. Pain 108:192-198, 2004
25. Moseley GL: I can’t find it! Distorted body image and tac-
tile dysfunction in patients with chronic back pain. Pain 140:239-243, 2008
26. Moseley GL: Is successful rehabilitation of complex re-
gional pain syndrome due to sustained attention to theaffected limb? A randomised clinical trial. Pain 114:54-61, 2005
27. Moseley GL, Gallagher L, Gallace A: Neglect-like tactile
dysfunction in chronic back pain. Neurology 79:327-332, 2012
28. Moseley GL: Why do people with complex regional pain
syndrome take longer to recognize their affected hand?Neurology 62:2182-2186, 2004
29. Moseley GL, Acerra N: Complex regional pain syndrome
is associated with distorted body image of the affected part.J Neurol Sci 238:S501, 2005
30. Moseley GL, Flor H: Targeting Cortical Representations
in the Treatment of Chronic Pain. Neurorehabil Neural Re-pair 26:646-652, 2012
31. Moseley GL, Gallace A, Spence C: Space-based, but not
arm-based, shift in tactile processing in complex regionalpain syndrome and its relationship to cooling of the affectedlimb. Brain 132:3142-3151, 2009
32. Moseley GL, Wiech K: The effect of tactile discrimination
training is enhanced when patients watch the reflected im-
age of their unaffected limb during training. Pain 144:314-319, 2009
33. Moseley GL, Zalucki N, Birklein F, Marinus J, van Hilten JJ,
Luomajoki H: Thinking about movement hurts: The effect ofmotor imagery on pain and swelling in people with chronicarm pain. Arthritis Rheum 59:623-631, 2008
34. Moseley GL, Zalucki NM, Wiech K: Tactile discrimination,
but not tactile stimulation alone, reduces chronic limb pain.Pain 137:600-608, 2008
35. Parsons LM, Fox PT, Downs JH, Glass T, Hirsch TB,
Martin CC, Jerabek PA, Lancaster JL: Use of implicit motorimagery for visual shape discrimination as revealed by PET.Nature 375:54-58, 1995
36. Physiotherapy Evidence Database. 1999. PEDro scale.
http://www.pedro.org.au/english/downloads/pedro-scale/
37. Richter HO, R €oijezon U, Bj €orklund M, Djupsj €obacka M:
Long-term adaptation to neck/shoulder pain and perceptual
performance in a hand laterality motor imagery test. Percep-
tion 39:119-130, 2010
38. Schwoebel J, Coslett H, Bradt J, Friedman R, Dileo C:
Pain and the body schema: effects of pain severity on men-tal representations of movement. Neurology 59:775-778,2002
39. Schwoebel J, Friedman R, Duda N, Coslett HB: Pain and
the body schema: Evidence for peripheral effects on mentalrepresentations of movement. Brain 124:2098-2104, 2001
40. Cohen J (ed): Statistical Power Analysis for the Behaviou-
ral Sciences. Mahwah, NJ, Lawrence Erlbaum Associates, Inc,1988
41. Vingerhoets G, de Lange FP, Vandemaele P, Deblaere K,
Achten E: Motor Imagery in Mental Rotation: An fMRI Study.Neuroimage 17:1623-1633, 2002
42. Walz AD, Usichenko T, Moseley GL, Lotze M: Graded Mo-
tor Imagery and the Impact on Pain Processing in a Case ofCRPS. Clin J Pain, 2012 Aug 21 [Epub ahead of print]
43. Wand BM, O’Connell NE, Di Pietro F, Bulsara M: Manag-
ing Chronic Nonspecific Low Back Pain With a SensorimotorRetraining Approach: Exploratory Multiple-Baseline Studyof 3 Participants. Phys Ther 91:535-546, 2011
44. Wand BM, Parkitny L, O’Connell NE, Luomajoki H,
McAuley JH, Thacker M, Moseley GL: Cortical changes inchronic low back pain: Current state of the art and implica-tions for clinical practice. Man Ther 16:15-20, 2011
45. Zacks J, Rypma B, Gabrieli JDE, Tversky B, Glover GH:
Imagined transformations of bodies: an fMRI investigation.Neuropsychologia 37:1029-1040, 199912 The Journal of Pain Effect of Graded Motor Imagery on Chronic Pain

Appendix A
Full Medline search strategy (23/3/11)
1. exp ‘‘Imagery (Psychotherapy)’’/
2. graded motor imagery.mp.3. exp Physical Therapy Modalities/
4. physiotherapy.mp.
5. physical therapy.mp.6. device therapy.mp.7. Occupational Therapy/8. Rehabilitation/9. Functional Laterality/
10. laterality.mp.11. left right judg$.mp.12. exp Pattern Recognition, Visual/13. visual pattern recognition.mp.14. Discrimination (Psychology)’’/
15. discrimination.mp.
16. Imagination/17. imagined movement.mp.18. mental imagery.mp.19. mental movement.mp.20. visual imagery.mp.21. exp Kinesthesis/22. kinaesthetic imagery.mp.23. kinesthetic imagery.mp.24. mirror therapy.mp.25. Feedback, Sensory/
26. mirror visual feedback.mp.
27. user-computer interface/28. Therapy, Computer-Assisted/29. virtual reality therapy.mp.30. user computer interface.mp.31. mirror box therapy.mp.32. 1 or 2 or 3 or 4 or 5 or 6 or 7 or 8 or 9 or 10 or 11 or
12 or 13 or 14 or 15 or 16 or 17 or 18 or 19 or 20 or21 or 22 or 23 or 24 or 25 or 26 or 27 or 28 or 29 or30 or 31
33. Pain/
34. 32 and 33
35. limit 34 to humanBowering et al
The Journal of Pain 13