European Journal of Pain 15 (2011) 775.e1–775.e11

Contents lists available at ScienceDirect

European Journal of Pain

journal homepage: www.EuropeanJournalPain.com

Review

Can we identify how programmes aimed at promoting self-managementin musculoskeletal pain work and who benefits? A systematic reviewof sub-group analysis within RCTs

Clare L. Miles a, Tamar Pincus a,⇑, Dawn Carnes b, Kate E. Homer b, Stephanie J.C. Taylor b,Stephen A. Bremner b, Anisur Rahman c, Martin Underwood d

a Royal Holloway University of London, Department of Psychology, Egham, Surrey TW20 0EX, UKb Queen Mary University of London, Barts & The London School of Medicine and Dentistry, Centre for Health Sciences, London, UKc University College London, Centre for Rheumatology Research, UKd Warwick Medical School, Clinical Trials Unit, Coventry, UK

a r t i c l e i n f o a b s t r a c t

Article history:Received 18 August 2010Received in revised form 4 January 2011Accepted 31 January 2011Available online 26 February 2011

Keywords:ModeratorSub-groupsPredictorMediatorSelf-managementRCTChronic painMusculoskeletal painSystematic review

1090-3801/$36.00 � 2011 Published by Elsevier Ltd.doi:10.1016/j.ejpain.2011.01.016

⇑ Corresponding author. Tel.: +44 1784 443523; faxE-mail address: t.pincus@rhul.ac.uk (T. Pincus).

Background: There are now several systematic reviews of RCTs testing self-management for those withchronic musculoskeletal pain. Evidence for the effectiveness of self-management interventions in chronicmusculoskeletal pain is equivocal and it is not clear for which sub-groups of patients SM is optimallyeffective.Aims: To systematically review randomized controlled trials of self-management for chronic musculo-skeletal pain that reported predictors, i.e., ‘baseline factors that predict outcome independent of anytreatment effect’; moderators, i.e., ‘baseline factors which predict benefit from a particular treatment’;or mediators i.e., ‘factors measured during treatment that impact on outcome’ of outcome.Method: We searched relevant electronic databases. We assessed the evidence according to the method-ological strengths of the studies. We did meta-regression analyses for age and gender, as potential mod-erators.Results: Although the methodological quality of primary trials was good, there were few relevant studies;most were compromised by lack of power for moderator and mediator analyses. We found strong evi-dence that self-efficacy and depression at baseline predict outcome and strong evidence that pain catas-trophizing and physical activity can mediate outcome from self-management. There was insufficient dataon moderators of treatment.Conclusions: The current evidence suggests four factors that relate to outcome as predictors/mediators,but there is no evidence for effect moderators. Future studies of mediation and moderation should bedesigned with ‘a priori’ hypotheses and adequate statistical power.� 2011 Published by Elsevier Ltd. on behalf of European Federation of International Association for the

Study of Pain Chapters.

1. Introduction

Self-management programmes have been promoted as one ofthe most important, non-pharmacological ways of helping peo-ple with intractable chronic conditions (Newman et al., 2004).However evidence for the effectiveness of self-management pro-grammes in chronic musculoskeletal pain is less clear. One po-tential explanation is that interventions directed at improvingpatient’s self management may be more effective for sub-groupswith particular characteristics. Indeed research in other chronicconditions suggests that some individuals benefit disproportion-

on behalf of European Federation o

: +44 1784 434347.

ately from such interventions (Monninkhof et al., 2004). Turkhas argued that the ‘‘myth’’ of patient homogeneity may explainwhy treatment outcomes are often disappointing amongst pa-tients with different chronic pain syndromes (Turk, 2005). Iden-tifying relevant sub-groups may improve the effectiveness andcost-effectiveness of self-management interventions (Kennedyet al., 2007).

The growing interest in sub-group analysis to explore processeswithin treatment and their impact on outcome has resulted ininformative findings in the field of cognitive-behavioural interven-tions for pain patients. Turner et al. (2007) demonstrated thatchanges in self-efficacy and perceived pain control contributed sig-nificantly to treatment outcome and Turk found that fibromyalgiapatients with different psychosocial profiles responded differently

f International Association for the Study of Pain Chapters.

775.e2 C.L. Miles et al. / European Journal of Pain 15 (2011) 775.e1–775.e11

to a multi-component FM management programme (Turk et al.,1998). However, Vlaeyen and Morley (2005) caution that suchanalysis must be theory-driven.

Outcome for people with chronic musculoskeletal pain maybe affected by a wide range of demographic, clinical, psycholog-ical and social factors. These may be: predictors, i.e., ‘baselinefactors that predict outcome independent of any treatment ef-fect’; moderators, i.e., ‘baseline factors which predict benefitfrom a particular treatment’; or mediators i.e., ‘factors measuredduring treatment that impact on outcome’(Kraemer et al., 2002).There is considerable evidence from prospective cohorts report-ing predictors, but far less from randomized controlled trials(RCTs) reporting moderators and mediators (Pincus et al., 2002,2006; Hayden et al., 2009). RCTs are the best study design to ex-plore moderators and mediators. However, RCTs that includeplanned sub-group analysis need very large samples (Hedgesand Pigott, 2004; Aguinis et al., 2005). The sub-group analysesmust also be based on good theoretical reasoning and previousevidence to support the hypothesis that the correct a priorisub-groups have been identified (Higgins and Green, 2008).Although individually underpowered, pooling of data from multi-ple trials may identify consistent findings across these trials, andwhere pooling cannot be carried out, consistency of findings mayprovide evidence to direct future better designed trials. The pur-pose of this review is to summarise the RCT evidence about pre-dictors, mediators and moderators in patients with chronicmusculoskeletal pain who take part in self-management pro-grammes. By self-management programmes (defined in detail la-ter), we mean multi component interventions aimed principallyat improving patient’s self management of their chronic muscu-loskeletal pain. Other systematic reviews have looked at predic-tors for the effectiveness of single interventions (e.g., Hoffmanet al., 2007), but multi component self-management pro-grammes, with their distinct approach, have not been the subjectof previous reviews of this nature.

2. Methods

2.1. Searches

Two reviewers searched independently using the same agreedfree text terms supplemented by electronic database indexingterms where possible, using both British and American spellingsand names (English language only, details of the search strategyare available upon request). We searched relevant electronic dat-abases from 1984 to April 2009 (MEDLINE, EMBASE, PsycINFO,CINAHL, AMED, Web of Science and the Cochrane Library) toidentify RCTs of self-management programmes for chronic mus-culoskeletal pain. We based our search strategies on free textterms supplemented by electronic database indexing termswhere possible. Search terms included: chronic musculoskeletalpain, back pain, neck pain, shoulder pain, knee pain, hip pain,fibromyalgia and osteoarthritis. These were combined withsearches on terms such as self-management, self-care, self-effi-cacy, self-help, self-improvement, patient education, patientteaching, patient training, expert patient, lay-led, peer-led andprofessionally-led. We also tracked citations in identified sys-tematic reviews for RCTs.

Two researchers working independently identified those papersthat included sub-group analyses, including analysis for predictors,moderators or mediators. Finally a statistician scrutinised the in-cluded studies to determine if the authors had conducted appropri-ate predictor, moderator or mediator analyses. We approached theauthors of studies with more than 200 participants in each arm andwith 80% or more completion rates that did not report sub-group

analyses or had done sub-group analyses that did not provide suf-ficient information and asked if they could carry out a sub-groupanalysis and inform us of their findings.

We have adapted the approach from Kraemer (Kraemer et al.,2002):

Effect predictors are defined as baseline variables that affect out-come (significant main effect only) but do not interact with treat-ment. Such factors significantly predict outcome equally for targetinterventions and control conditions.

Effect moderators represent variables measured at baseline(such as patient baseline characteristics) that interact with treat-ment to change outcome for each sub-group. These specify forwhom and under what conditions treatment works.

Effect mediators are variables measured during treatment (fac-tors that change during the intervention) that impact on outcome,with or without interaction with treatment. Mediators help informthe process and potential mechanisms (including causal mecha-nisms) through which treatment might work, and therefore canbe used to improve subsequent interventions through strengthen-ing the components that best influence the identified mediators.

2.2. Selection of studies

We included RCTs that compared a self-management pro-gramme intervention with waiting list control (WLC) or usualcare (UC). We extracted data on country, population, interven-tion/control, and the components of interventions baseline mea-sures, outcome measures, description of moderator analyses,and results of moderator analyses. The programme had to con-tain at least two components from the following five groupsagreed by our steering group: psychological (including behav-ioural or cognitive therapy), mind–body therapies (MBT) (includ-ing such as relaxation, meditation or guided imagery), physicalactivity (any form of exercise), lifestyle (such as dietary adviceand sleep management) and medical education (such as under-standing their condition and how to take medication effectively).We identified components by author descriptions of interven-tions from their published reports and classified content byconsensus.

2.2.1. Type of participantsWe included studies of adults (age P 18 years) with non-

specific musculoskeletal pain with or without comorbidities,degenerative joint disease, chronic widespread pain, arthritis,osteoarthritis, fibromyalgia and unexplained, and non-pathologi-cal neuropathic painful conditions. We excluded studies of pa-tients with migraine, headaches, facial pain, eye pain, irritablebowel syndrome, angina, chronic obstructive pulmonary disease,non-cardiac chest pain, inflammatory joint conditions such asrheumatoid arthritis or ankylosing spondylitis, and chronic fati-gue syndrome/myalgic encephalopathy (CFS/ME). We also ex-cluded studies of patients with chronic pain arising frommalignant disease.

2.2.2. Self-management programmeThere is no universally accepted definition of a self-manage-

ment programme so we developed our own working definition.Single component interventions (such as CBT) were not consid-ered programmes and were excluded (see above). We consideredstudies to be directed at self management if they had the broadgoal of improving participants’ health status or quality of lifeand where there was scope for improvement in patients manag-ing their own health. Carers or tutors may have been involvedbut we only included programmes principally directed atpatients. Included programmes had to be structured with ataught or self-taught component that aimed to increase

C.L. Miles et al. / European Journal of Pain 15 (2011) 775.e1–775.e11 775.e3

participants’ skills and knowledge and to enable participants todeploy these enhanced skills in aspects of their lives beyond theintervention. Psychological elements had to be behavioural and/orcognitive and, or structured (not psychodynamic therapy). Studieswere excluded if they did not provide a clear description of theintervention.

2.2.3. Types of outcomeWe extracted data on health behaviours (such as health care

utilization, days off work, medication consumption, social anddaily activities), health status (such as pain intensity and disabil-ity), self-efficacy and adverse effects. We only included outcomemeasures that had previously published evidence for good validityand reliability in this population.

Table 1Included RCTs (moderators, predictors and mediators).

Author (reference) Country Population

Gallagher et al. (1997) North America OA

Haas et al. (2005) North America CLBP

Haldorsen et al. (1998) Norway Patients with muscle pain (mixed)(sick-listed for 8 weeks)

Haugli et al. (2000) Norway Employees with a diagnosis ofmusculoskeletal pain that occurreseveral days a week for more than4 weeks during the past 12 month

Haugli et al. (2003) Norway Employees with a diagnosis ofmusculoskeletal pain that occurreseveral days a week for more than4 weeks during the past 12 month

Heapy et al. (2005) North America Chronic pain patients. Aetiology othe complaint was broadly eithermusculoskeletal or neurologic

Hurley et al. (2007)a UK Patients with mild, moderate, orsevere knee pain of >6 months

Jensen et al. (2001) Sweden Participants suffered from long-ternon-specific spinal pain

Jensen et al. (2005) Sweden Participants suffered from long-ternon-specific spinal pain

Kole-Snijders et al. (1999) Netherlands CLBP

Laforest et al. (2008) Canada Participants were suffering from R(38%) or OA (62%)

Lemstra and Olszynski (2005) Canada Patients with fibromyalgia, havingchronic widespread pain

Lindh et al. (1997) Sweden A general population with 90 dayssick-leave due to non-specific MSpain (Swedish and foreign citizens

Lorig et al. (2002) North America CLBP

Martire et al. (2007) North America Individuals who were married anddiagnosed with hip or knee OA

Nour et al. (2006) Canada Participants were suffering from R(38%) or OA (62%)

Smeets et al. (2006) Netherlands Non-specific CLBP

Spinhoven et al. (2004) Netherlands CLBP

van der Hulst et al. (2008) Netherlands Non-Specific CLBP

Veenhof et al. (2007)a Netherlands Patients with OA of the hip or kne

WLC: waiting list control; UC: usual care control; OA: osteoarthritis; RA: rheumatoid arenabling self-management and coping with arthritic knee pain through Exercise.

a Cluster randomized controlled trial.

2.3. Quality assessment

Overall quality of included studies was assessed by two reviewauthors independently. The methodological quality assessment forthe RCTs was modelled on the Cochrane methods (details in Carneset al., in preparation) using some of their criteria (adequaterandomization sequence, adequate allocation concealment,description of withdrawals and dropouts, blinding of outcomeassessment and >20 participants in each arm). Inter-raterreliability for assessing the studies was checked on a 10% sampleof studies and there were few discrepancies. Trials received a scoreof a rating out of 5:1 (for each positively scored criterion) or 0(information not reported or unclear). The quality of trials was cat-egorised as high (4–5); medium (2–3) or low (1).

Intervention Follow-up Total N

Social support and education vs.WLC

Annually for 3 years 363

Chronic Disease Self-Management Programme vs.WLC

6 months 109

Multimodal CBT vs. UC 6 and 12 months 469

d

s

Educational programme vs. UC Short-term 174

d

s

Educational programme vs. UC Follow up of Haugli et al.(2000) at 1 year

174

f PRIME CBT vs. UC None 89

ESCAPE (individual and group)vs. UC

6 months 418

m, CBT and physical therapy vs. UC 18- months and 3-years 214

m, CBT and physical therapy vs. UC Follow up of Jensen et al.(2001) at 3 years

214

Operant conditioning pluscognitive coping skills vs. WLC

6 and 12 months 159

A I’m taking charge of my arthritis’’program vs. WLC

Follow up to Nour et al.(2006) at 8 weeks

113

Multidisciplinary rehabilitationvs. UC

15 months 79

of

)

Multidisciplinary rehabilitationvs. UC

Every 3 months for 5 years 464

Back pain email discussion groupvs. UC

12 months 580

Couple and patient-orientededucation and supportintervention vs. UC

6 months 242

A I’m taking charge of my arthritis’’program vs. WLC

8 weeks 113

Active Physical Therapy + CBT vs.WLC

None 211

Operant conditioning pluscognitive coping skills vs. WLC

6 and 12 months (Samestudy as Kole Snijders(1999))

148

The Roessingh BackRehabilitation Program (basedon the Swedish Back School andmultidimensional painprograms) vs. UC

8 weeks and 6 months 163

e Behavioural Graded Activity vs.UC

13, 39 and 65 weeks 200

thritis; CLBP: chronic low back pain; CBT: cognitive behavioural therapy; ESCAPE:

775.e4 C.L. Miles et al. / European Journal of Pain 15 (2011) 775.e1–775.e11

2.4. Quality of sub-group analyses

Due to the lack of an established standard for assessing thequality of studies with sub-group analyses we used the followingcriteria, based on guidance from the Cochrane handbook and a con-sensus study of international experts (Pincus et al., 2011).

(1) Was the sub-group analysis specified a priori?(2) Was the selection of sub-group factors for analysis theory/

evidence driven?(3) Were sub-group factors measured prior to randomization?(4) Was measurement of sub-group factors measured by ade-

quate (reliable and valid) measurements, appropriate forthe target population?

(5) Does the analysis contain an explicit test of the interactionbetween moderator and treatment?

We classified studies complying with all five criteria as provid-ing confirmatory evidence, those complying with criteria three,four and five as providing exploratory evidence. All other studieswere classified as providing insufficient evidence.

2.5. Levels of evidence

In line with the Cochrane Handbook that states that differences insub-groups observed within studies are more reliable than analysesof subsets of studies, we deemed the strongest evidence for sub-group/moderator analysis to exist where there was as at least onehigh quality trial (using Cochrane criteria for the general qualityassessment of trials (Higgins and Green, 2008) that met all of the cri-teria listed above for assessing sub-group analysis. Due to the inher-ent problem of lack of power in sub-group analyses, we did notconsider or report negative findings as providing evidence. We re-garded the remaining studies that did not meet the above criteria,to be methodologically weak and providing only exploratory data.

Table 2Quality assessment of the studies (predictors, moderators and mediators).

Quality appraisal forsub-group studies

1. Was theanalysisa priori?

2. Was selectionof factors for analysistheory/evidencedriven

3. Were sub-gromeasured prior trandomization?

Gallagher et al. (1997) Yes Yes N/A gender as mHaas et al. (2005) No No YesHaldorsen et al. (1998) No No YesHaugli et al. (2000) Yes Yes UnclearHaugli et al. (2003) Yes. Yes UnclearHeapy et al. (2005) Yes Yes YesHurley et al. (2007) Yes Yes N/A cluster randJensen et al. (2001) Yes Yes N/A gender as mJensen et al. (2005) Yes Yes N/A gender as mKole-Snijders et al. (1999) Yes No NoLaforest et al. (2008) Yesa Yes YesLemstra and Olszynski (2005) No No UnclearLindh et al. (1997) Yes Yes N/A demographiLorig et al. (2002) Yes Yes YesMartire et al. (2007) Yes Yes N/A demographiNour et al. (2006) Yes Yes YesSmeets et al. (2006) Yes Yes YesSpinhoven et al. (2004) Yes Yes Novan der Hulst et al. (2008) Yes Yes YesVeenhof et al. (2007) Yesa Yes N/A cluster rand

Confirmatory evidence: The study fulfils all of the quality assessment criteria for moderarandomization, adequate measurement of baseline factors and explicit test of the interaExploratory evidence: Fulfilling the last three quality assessment criteria. Insufficient evInsufficient evidence: The study did not carry out an explicit test of interaction or mea

a Analysis is weak due to multiple testing.b nb: the significant findings were from the exploratory analysis only.

2.6. Meta-regression

We included studies that did not necessarily report sub-groupanalyses, but could contribute to such analysis by a meta-regres-sion of their final value data. Studies from the original search thatsupplied full data on age and gender at baseline against at least oneof our selected outcomes were included (n = 46). For potentialmoderators reported in 10 or more studies we did a random effectsmeta-regression (Higgins and Green, 2008). We collapsed out-comes into the following categories: pain intensity, physical/func-tional capability, self-efficacy, depression and global health status.For one outcome, general mental health, a single measurement toolhad been used SF36 (36-item short form health survey, Ware andSherbourne, 1992). As a variety of measurement tools had been re-ported for each other outcome, we calculated standardized meandifferences (difference in mean outcome between groups/standarddeviation of outcome among participants) (Higgins and Green,2008). For SF36 general mental health data, where we intendedto combine the data using a weighted mean difference, we notedanomalies in score values between studies and so we also analysedthese using standardized mean differences. To make the best use ofavailable data, and reduce possibility of making a Type 1 error wecollapsed the different follow-up time points (early and late) toobtain one average effect size per outcome. We considered resultsfrom the meta-regression to be statistically significant if P < 0.10.This criterion was adopted because of a potential Type II error asa result of the limited number of studies in these effect size calcu-lations (Armitage et al., 2002; Hoffman et al., 2007). We used an I2

statistic to estimate the percentage of residual variation attribut-able to between-study heterogeneity and an adjusted R-squaredstatistic to estimate the proportion of between-study varianceexplained by the covariate. We produced bubble plots (scatter dia-grams using circles as plotting symbols in which the areas of thecircles indicate the value of a third variable) (Upton and Cook,2002). We fitted values and predicted random effects against age

upso

4. Adequatequality ofmeasurementof baselinefactors?

5. Contains an explicittest of the interactionbetween sub-groupand treatment (e.g.,regression)?

Strength ofevidence

Quality ofunderlyingtrial

oderator Yes Yes Confirmatory MediumYes Yes Exploratory HighYes Unclear Insufficient HighYes Yes Confirmatory MediumYes Yes Confirmatory MediumYes No Insufficient High

omization Yes Yes Confirmatory Highoderator Yes No Insufficient Highoderator Yes No Insufficient High

Yes Yes Insufficient HighYes Yes Confirmatory HighYes No Insufficient High

c moderators Yes No Insufficient LowYes No Insufficient Medium

c moderators Yes Yes Confirmatory MediumYes Unclear Insufficient HighYes Yes Confirmatory HighYes Yes Insufficient HighYes Yes Confirmatory Medium

omization Yes Yesb Confirmatory High

tor studies (a priori analysis, factors evidence driven, moderators measured prior toction between moderator and treatment).idence: The studies failed to provide adequate statistical analysis of the moderators.surement of the sub-groups was reported to take place post randomization.

C.L. Miles et al. / European Journal of Pain 15 (2011) 775.e1–775.e11 775.e5

and gender separately, with 95% confidence and prediction inter-vals. Each bubble is the estimate of association taken from eachstudy and the size of the bubble is the precision of the estimate:the larger the variance the smaller the bubble and bigger bubbleshave greater influence on the model (Fig. S1. See the online versionat 10.1016/j.ejpain.2011.01.016). We used Stata 10.1 for the meta-regression analyses and bubble plot graphs (StataCorp, 2007;Harbord and Higgins, 2008).

3. Results

3.1. Literature search

We found 126 relevant RCTs of self-management. Of these, 16studies, reported in 20 papers, with 4047 participants met ourinclusion criteria and included appropriate analyses of moderatorsand/or mediators (Table 1). There were a further 46/126 RCTs withfinal value data that used WLC or UC control groups that contrib-uted to the meta-regression (Table 2). These 46 studies will onlybe discussed in reference to the meta-regression. All other results

Potential RCTs and systematic reviews frsearch (n=4676).

RCTs – primary pape(n=126).

All studies underwenquality assessment

RCTs with waiting list control or usual care comparison (n=53)

Quantitative synthesis of the RCTs with final value data (n=46)

Exclusions:RCTs that did not have final value data (n=7)

Fig. 1. Study searc

refer to the 16 studies that carried out appropriate sub-groupanalyses. No additional data was obtained from contacted authorsdata (Fig. 1).

3.2. Characteristics of predictor, moderator and mediator studies

These came from six countries: USA five (Gallagher et al., 1997;Lorig et al., 2002; Haas et al., 2005; Heapy et al., 2005; Martireet al., 2007), Netherlands four (Kole-Snijders et al., 1999; Smeetset al., 2006; Veenhof et al., 2007; van der Hulst et al., 2008), Canadatwo (Lemstra and Olszynski, 2005; Laforest et al., 2008), Norwaytwo (Haldorsen et al., 1998; Haugli et al., 2000), Sweden two(Lindh et al., 1997; Jensen et al., 2001), and UK one (Hurley et al.,2007). Three trials included participants aged 30–39 years, six in-cluded participants aged 40–49, one included participants aged50–59 and seven included participants aged 60 or over.

3.3. Methodological quality

Nine RCTs were high quality; six were medium quality and onelow quality. Eight of the sub-group studies provided confirmatory

om

Exclusions: Not chronic musculoskeletal pain, not self-management, conference abstracts, commentaries and literature reviews, studies validating outcome measures, not RCT, cost-effectiveness studies, pilot studies or sample size <20, not relevant outcome data, systematic reviews superseded or updated.

rs

t

Statistician verified a selected 20/126 articles (16 trials) that included predictor, moderator or mediator analyses

Quality assessment of predictor, moderator and mediator studies

Descriptive analysis of predictor, moderator and mediator studies. Eight of these papers had final value data and were included in the meta-regression

h and process.

775.e6 C.L. Miles et al. / European Journal of Pain 15 (2011) 775.e1–775.e11

evidence, one study provided exploratory evidence, and sevenstudies provided insufficient evidence (Table 3). Only three studieshad high quality methodology, carried out pre-planned theoreti-cally-driven sub-group analysis, using correct statistical analysis.However, there were no two trials in this category that examinedthe same sub-group, thus, strong evidence was restricted to reportsfrom single studies (Fig. 1).

3.4. Strong evidence: findings from single studies

Three studies (four papers) were of high quality, and fulfilled allcriteria for assessing the sub-group analysis (Nour et al., 2006;Smeets et al., 2006; Hurley et al., 2007; Laforest et al., 2008). Hur-ley found that higher levels of depression at baseline predictedpoorer physical functioning at 6 months (effect size = 0.48)whereas higher levels of self-efficacy at baseline, measured by po-sitive exercise beliefs (effect size �0.24), and confidence in theability to exercise (effect size �0.62), predicted better physicalfunctioning at 6 months, regardless of intervention arm (Hurleyet al., 2007). Smeets found that reduced levels of pain catastrophiz-ing during treatment led to a post-treatment decrease in patient-specific complaints, disability and pain. Patients in the interventiongroup scored, on average, 1.3 points lower on disability (out of 24)than patients in the control arm, after adjusting for pain catastoph-ising. For current pain, the difference was 4.7 units on the visualanalogue scale (out of 100). For patient complaints, the differencewas 6.7 (out of 100). Laforest found that increases in physical activ-ity mediated greater decreases in helplessness, however the datawas not available to quantify this effect. This effect was definedby Laforest as a moderator, although from the description it is amediator (Laforest et al., 2008). Because of these limitations, werecommend that the findings for the mediating effects of physical

Table 3Quality assessment of the studies (predictors, moderators and mediators).

Quality appraisal forsub-group studies

1. Was theanalysisa priori?

2. Was selectionof factors foranalysis theory/evidence driven

3. Were sub-groupsmeasured prior torandomization?

Gallagher et al. (1997) Yes Yes N/A gender as modeHaas et al. (2005) No No YesHaldorsen et al. (1998) No No YesHaugli et al. (2000) Yes Yes UnclearHaugli et al. (2003) Yes. Yes UnclearHeapy et al. (2005) Yes Yes YesHurley et al. (2007) Yes Yes N/A cluster randomiJensen et al. (2001) Yes Yes N/A gender as modeJensen et al. (2005) Yes Yes N/A gender as modeKole-Snijders et al. (1999) Yes No NoLaforest et al. (2008) Yesa Yes YesLemstra and Olszynski (2005) No No UnclearLindh et al. (1997) Yes Yes N/A demographic mLorig et al. (2002) Yes Yes YesMartire et al. (2007) Yes Yes N/A demographic mNour et al. (2006) Yes Yes YesSmeets et al. (2006) Yes Yes YesSpinhoven et al. (2004) Yes Yes Novan der Hulst et al. (2008) Yes Yes YesVeenhof et al. (2007) Yesa Yes N/A cluster randomi

Confirmatory evidence: The study fulfils all of the quality assessment criteria for moderatrandomization, adequate measurement of baseline factors and explicit test of the interaExploratory evidence: Fulfilling the last three quality assessment criteria. Insufficient evidInsufficient evidence: The study did not carry out an explicit test of interaction or measu

a Analysis is weak due to multiple testing.b nb: The significant findings were from the exploratory analysis only.

activity be reviewed with caution. None of these studies found evi-dence for effect moderators.

3.5. Moderate evidence: findings from meta-regression

Primary papers (n = 46) with final value data (the outcome mea-surement obtained at the follow up interval) that included age andgender as potential moderators were included in the meta-regres-sion (see Table 2 for characteristics of these studies). Eight of thesestudies were included in the 16 sub-group analyses studies. We usedbivariate meta-regression to determine if the baseline characteris-tics (age and gender) explained the variation in treatment outcomes.Age and gender were selected because they are the most frequentlyreported demographics: they were the only variables reported in atleast 10 studies (Higgins and Green, 2008). We found that genderwas significantly associated with effect sizes for SF36 general mentalhealth and global health status (all P < 0.10 and P P 0.05) (Tables S1and S2. See the online version at 10.1016/j.ejpain.2011.01.016).Inspection of bubble graphs suggested a positive association be-tween effect sizes for these outcomes, suggesting that self-manage-ment interventions might be more effective in studies with a greaterproportion of female subjects (Fig. S1). Gender was not associatedwith effect size for pain intensity, physical/functional capability,self-efficacy and depression. Age was significantly associated witheffect size for physical/functional capability and self-efficacy (allP < 0.10 and P > 0.05). Inspection of bubble graphs suggested a posi-tive association between effect size and these outcomes indicatingthat self-management interventions might be more effective inyounger samples (Fig. S1). Age was not associated with different ef-fect size for other outcomes.

4. Adequatequality ofmeasurementof baselinefactors?

5. Contains an explicittest of the interactionbetween sub-groupand treatment(e.g., regression)?

Strength ofevidence

Quality ofunderlyingtrial

rator Yes Yes Confirmatory MediumYes Yes Exploratory HighYes Unclear Insufficient HighYes Yes Confirmatory MediumYes Yes Confirmatory MediumYes No Insufficient High

zation Yes Yes Confirmatory Highrator Yes No Insufficient Highrator Yes No Insufficient High

Yes Yes Insufficient HighYes Yes Confirmatory HighYes No Insufficient High

oderators Yes No Insufficient LowYes No Insufficient Medium

oderators Yes Yes Confirmatory MediumYes Unclear Insufficient HighYes Yes Confirmatory HighYes Yes Insufficient HighYes Yes Confirmatory Medium

zation Yes Yesb Confirmatory High

or studies (a priori analysis, factors evidence driven, moderators measured prior toction between moderator and treatment).ence: The studies failed to provide adequate statistical analysis of the moderators.rement of the sub-groups was reported to take place post randomization.

C.L. Miles et al. / European Journal of Pain 15 (2011) 775.e1–775.e11 775.e7

3.6. Weak evidence: findings from an exploratory, descriptive analysis

There were some otherwise high quality studies that did notfulfill all five sub-group assessment criteria. We therefore haveproduced a list of factors that warrant further investigation basedon this review:

� Predictors: Depression (Haugli et al., 2003; van der Hulst et al.,2008); disability (van der Hulst et al., 2008); self-efficacy (Loriget al., 2002).� Moderators: Immigrant status (Lindh et al., 1997); pain intensity

(Veenhof et al., 2007; van der Hulst et al., 2008); depression(Nour et al., 2006; van der Hulst et al., 2008).� Mediators: Pain catastrophizing (Smeets et al., 2006); self-effi-

cacy (Lorig et al., 2002); Outcome expectations were measuredin one otherwise high quality study (Laforest et al., 2008) butthe Cronbach’s alpha, a measure of internal reliability, in thisstudy was low (0.58) and no information was given on its validity.

4. Discussion and conclusions

To our knowledge this is the first attempt to systematicallyidentify moderators and mediators for self-management pro-grammes directed at chronic pain. Our findings suggest that thefollowing factors impact on outcome: Self-efficacy, depression,pain catastrophizing and physical activity. Depression and self-effi-cacy predict outcome irrespective of intervention, suggesting thatthese should be targeted at early stages to prevent the transitionto chronic disability. This is supported by previous reviews (e.g.,Pincus et al., 2002). The evidence for pain catastrophizing, and,with caution (see Section 3), increased physical activity as mediat-ing factors, suggest that self-management interventions that focuson modifying these factors appear most likely to improve out-comes in patients experiencing chronic musculoskeletal pain. Wefound no evidence for effect moderators, thus, current data donot allow us to recommend targeting interventions at particulargroups. That self-efficacy is both a predictor and a mediator of out-come does not necessarily mean that targeting those with the low-est levels of self-efficacy is the way forward. We hypothesise thatthe crucial measure is one of the likelihood of low self-efficacyimproving with intervention; this improvement then mediatingthe clinical outcome. It may be that people with very low self-effi-cacy are the ones least likely to change, conversely those with highself-efficacy cannot make further positive change. It might be that,in fact, those in the middle range of self-efficacy scores are mostlikely to benefit (i.e., they have the capacity to change and arepotentially changeable). Further research is needed also to exam-ine the influence of immigrant status, baseline disability, baselinepain intensity, age and gender on self-management programmesuccess as the state of evidence on these factors is currently weak.A final recommendation would be that study authors should usethe criteria for assessing the quality of moderator studies to guidetheir methodology. There needs to be significant guidance ondesigning, collecting data and analysing sub-groups in the contextof RCTs.

4.1. Limitations of the current study

We have attempted to conduct our review of the evidence withsystematic rigor, but we note that this has implications to thevalidity of the findings in real life situations, in which treatmentis often subtly adapted to meet patients’ needs, state and respon-siveness. We recognise that this limitation is present in most trials,in which adherence to protocol may compromise the creativity andresponsiveness of therapists, thus, failing to capture more fluidmediation mechanisms.

This review was confined to a consideration of self-manage-ment programmes. As we elaborated, although these are quitecommon in practice, they are ill-defined and we developed ourown working definition to operationalise the review. In order tobe both systematic and inclusive we grouped components into fivebroad categories and relied on published accounts of interventionsto do this. We recognise that other groups might have defined self-management programmes differently or categorised componentsdifferently and therefore included a slightly different group ofstudies in this review (Barlow et al., 2002; Chodosh et al., 2005).Furthermore our judgement may have been hampered by the lim-ited descriptions of the individual interventions in the includedRCTs – a problem which is well recognised (Petticrew, 2003).

In addition, we recognise that content delivered in componentsmay not match the label, either because of lack of integrity indelivery (which is seldom included in the reporting), or becauseof the non-specific characteristics bound in all treatment. Thus,an intervention labelled as delivering psychology may deliver littlepsychology or do so poorly, while aspects of prescribing and advis-ing on exercise may include counselling, empathy and implicit tar-geting of self-efficacy. Some of these problems may be resolved bybetter measurement of the explicit goals of delivering each compo-nent, and by measuring integrity of delivery within trials.

The evidence was insufficient to clearly determine the influenceof predictors, moderators and mediators of intervention successdue to the lack of consistent reporting across studies and themeta-regression being limited to the examination of age and gen-der (as the most commonly reported characteristics). Only whenresearchers report potential moderating variables as a standardwill meta-analytic techniques be able to calculate a more accurateestimate of the variance between studies. With respect to the gen-eralisability of our findings, patients under the age of the 30 werenot featured in any of the studies reviewed.

We also note that some of the usual predictors reported in pro-spective cohorts, which are the preferred means for identifyingpredictors (e.g., Pincus et al., 2002) were only partially supportedby this review, but this may be a result of the general paucity ofincluding theory-driven baseline factors.

5. Conclusion

Our results suggested that self-efficacy, depression, pain catas-trophizing and physical activity are important factors in influenc-ing patients’ outcome. The first two, as predictors of outcomeirrespective of intervention, may be most important to target atearly stages of pain. Pain catastrophizing and increased physicalactivity, as mediators, should be targeted by interventions inchronic musculoskeletal pain. The current evidence on age andgender is tentative but suggests directions for future research.

Conflict of interest

The authors declare no conflicts of interest

Funding statement

MU’s contribution to this project benefited from facilitiesfunded through Birmingham Science City Translational MedicineClinical Research and infrastructure Trials platform, with supportfrom Advantage West Midlands.

Acknowledgements

This work was supported by the National Institute for HealthResearch (NIHR) under its Programme Grants for Applied Research

775.e8 C.L. Miles et al. / European Journal of Pain 15 (2011) 775.e1–775.e11

scheme (Grant #: RP-PG-0707-10189). The views expressed in thispublication are those of the author(s) and not necessarily those ofthe NHS, the NIHR or the Department of Health.

References

Aguinis H, Beaty JC, Boik RJ, Pierce CA. Effect size and power in assessingmoderating effects of categorical variables using multiple regression: a 30-year review. J Appl Psychol 2005;90:94–107.

Armitage P, Berry G, Matthews J. Statistical Methods in Medical Research. 4th ed.;2002.

Barlow J, Wright C, Sheasby J, Turner A, Hainsworth J. Self-management approachesfor people with chronic conditions: a review. Patient Educ Couns2002;48:177–87.

Carnes D, Homer KE, Miles CL, Pincus T, Underwood M, Taylor SJC. Which deliverycharacteristics in self management interventions are most effective for chronicmusculoskeletal pain: a systematic review, in preparation.

Chodosh J, Morton SC, Mojica W, Maglione M, Suttorp MJ, Hilton L, et al. Meta-analysis: chronic disease self-management programs for older adults. AnnIntern Med 2005;143:427–38.

Gallagher RA, Miller C, Cronan TA, Groessl E. Gender differences in participation andresponsiveness to a health intervention for older Americans. Women Health1997;25:63–82.

Haas M, Groupp E, Muench J, Kraemer D, Brummel-Smith K, Sharma R, et al. Chronicdisease self-management program for low back pain in the elderly. JManipulative Physiol Ther 2005;28:228–37.

Haldorsen EM, Kronholm K, Skouen JS, Ursin H. Multimodal cognitive behavioraltreatment of patients sicklisted for musculoskeletal pain: a randomizedcontrolled study. Scand J Rheumatol 1998;27:16–25.

Harbord RM, Higgins JPT. Meta-regression in Stata. Stata J 2008;8:493–519.Haugli L, Steen E, Laerum E, Finset A, Nygaard R. Agency orientation and chronic

musculoskeletal pain: effects of a group learning program based on the personalconstruct theory. Clin J Pain 2000;16:281–9.

Haugli L, Steen E, Laerum E, Nygard R, Finset A. Psychological distress andemployment status: effects of a group learning programme for patients withchronic musculoskeletal pain. Psychol Health Med 2003;8:135–48.

Hayden JA, Chou R, Hogg-Johnson S, Bombardier C. Systematic reviews of low backpain prognosis had variable methods and results: guidance for future prognosisreviews. J Clin Epidemiol 2009;62:781–796 e1.

Heapy A, Otis J, Marcus KS, Frantsve LM, Janke EA, Shulman M, et al. Intersessioncoping skill practice mediates the relationship between readiness for self-management treatment and goal accomplishment. Pain 2005;118:360–8.

Hedges LV, Pigott TD. The power of statistical tests for moderators in meta-analysis.Psychol Methods 2004;9:426–45.

Higgins J, Green S. Cochrane handbook for systematic reviews of interventionsversion 5.0.1. The Cochrane collaboration, 2008 <http://www.cochrane-handbook.org> [accessed 26.09.09].

Hoffman BM, Papas RK, Chatkoff DK, Kerns RD. Meta-analysis of psychologicalinterventions for chronic low back pain. Health Psychol 2007;26:1–10.

Hurley MV, Walsh NE, Mitchell HL, Pimm TJ, Patel A, Williamson E, et al. Clinicaleffectiveness of a rehabilitation program integrating exercise, self-management, and active coping strategies for chronic knee pain: a clusterrandomized trial. Arthritis Rheum 2007;57:1211–9.

Jensen IB, Bergström G, Ljungquist T, Bodin L, Nygren AL. A randomized controlledcomponent analysis of a behavioral medicine rehabilitation program for chronicspinal pain: are the effects dependent on gender? Pain 2001;91:65–78.

Kennedy A, Reeves D, Bower P, Lee V, Middleton E, Richardson G, et al. Theeffectiveness and cost effectiveness of a national lay-led self care supportprogramme for patients with long-term conditions: a pragmatic randomisedcontrolled trial. J Epidemiol Community Health 2007;61:254–61.

Kole-Snijders AMJ, Vlaeyen JWS, Goossens MEJB, Rutten-van Molken MPMH, HeutsPHTG, van Breukelen G, et al. Chronic low-back pain: what does cognitivecoping skills training add to operant behavioral treatment? Results of arandomized clinical trial. J Consult Clin Psychol 1999;67:931–44.

Kraemer HC, Wilson GT, Fairburn CG, Agras WS. Mediators and moderators oftreatment effects in randomized clinical trials. Arch Gen Psychiatry2002;59:877–83.

Laforest S, Nour K, Gignac M, Gauvin L, Parisien M, Poirier MC. Short-term effects ofa self-management intervention on health status of housebound older adultswith arthritis. J Appl Gerontol 2008;27:539–67.

Lemstra M, Olszynski WP. The effectiveness of multidisciplinary rehabilitation inthe treatment of fibromyalgia: a randomized controlled trial. Clin J Pain2005;21:166–74.

Lindh M, Lurie M, Sanne H. A randomized prospective study of vocational outcomein rehabilitation of patients with non-specific musculoskeletal pain: amultidisciplinary approach to patients identified after 90 days of sick-leave.Scand J Rehabil Med 1997;29:103–12.

Lorig KR, Laurent DD, Deyo RA, Marnell ME, Minor MA, Ritter PL. Can a back pain E-mail Discussion Group improve health status and lower health care costs? Arandomized study. Arch Intern Med 2002;162:792–6.

Martire LM, Schulz R, Keefe FJ, Rudy TE, Starz TW. Couple-oriented education andsupport intervention: effects on individuals with osteoarthritis and theirspouses. Rehabil Psychol 2007;52:121–33.

Monninkhof E, van der Aa M, van der Valk P, van der Palen J, Zielhuis G, Koning K,et al. A qualitative evaluation of a comprehensive self-management programmefor COPD patients: effectiveness from the patients’ perspective. Patient EducCouns 2004;55:177–84.

Newman S, Steed L, Mulligan K. Self-management interventions for chronic illness.Lancet 2004;364:1523–37.

Nour K, Laforest S, Gauvin L, Gignac M. Behavior change following a self-management intervention for housebound older adults with arthritis: anexperimental study. Int J Behav Nutr Phys Act 2006;30(3):12.

Petticrew M. Why certain systematic reviews reach uncertain conclusions. BMJ2003;326:756–8.

Pincus T, Burton AK, Vogel S, Field AP. A systematic review of psychological factorsas predictors of chronicity/disability in prospective cohorts of low back pain.Spine 2002;27:E109–20.

Pincus T, Vogel S, Burton AK, Santos R, Field AP. Fear avoidance and prognosis inback pain: a systematic review and synthesis of current evidence. ArthritisRheum 2006;54:3999–4010.

Pincus T, Miles C, Froud R, Underwood M, Carnes D, Taylor SJ. Methodologicalcriteria for the assessment of moderators in systematic reviews of RandomisedControlled Trials: a consensus study. BMC Med Res Methodol 2011;11:14.

Smeets RJE, Vlaeyen JWS, Kester ADM, Knottnerus JA. Reduction of paincatastrophizing mediates the outcome of both physical and cognitive-behavioral treatment in chronic low back pain. J Pain 2006;7:261–72.

StataCorp. Stata statistical software. Release 10.1 2007, College Station, TX.Turk DC. The potential of treatment matching for sub-groups of patients with

chronic pain: lumping versus splitting. Clin J pain 2005;21(1):44–55.Turk DC, Okifuji A, Sinclair JD, Starz TW. Differential responses by psychosocial

subgroups of fibromyalgia syndrome patients to an interdisciplinary treatment.Arthritis Care Res 1998;11(5):397–404.

Turner JA, Holtzman S, Mancl L. Mediators, moderators, and predictors oftherapeutic change in cognitive-behavioral therapy for chronic pain. Pain2007;127(3):276–86.

Upton G, Cook I. Oxford dictionary of statistics. Oxford: Oxford University Press; 2002.van der Hulst M, Vollenbroek-Hutten MM, Groothuis-Oudshoorn KGM,

Hermens HJ. Multidisciplinary rehabilitation treatment of patients withchronic low back pain: a prognostic model for its outcome. Clin J Pain2008;24:421–31.

Veenhof C, van den Ende CHM, Dekker J, Kiike AJA, Oostendorp RA, Bijlsma JWJ.Which patients with osteoarthritis of hip and/or knee benefit most frombehavioral graded activity? Int J Behav Med 2007;14:86–92.

Vlaeyen J, Morley S. Cognitive-behavioral treatments for chronic pain: what worksfor whom? Clin J Pain 2005;21(1):1–8.

Ware Jr JE, Sherbourne CD. The MOS 36-item short-form health survey (SF-36). I.Conceptual framework and item selection. Med Care 1992;30:473–83.

Table S1Included RCTs in the meta-regression.

Study Country Population Total N FU QA 1 QA 2 QA 3 QA 4 QA5

Alp et al. (2007) Turkey Osteopor 50 ST MT Y U U Y YBasler et al. (1997) Germany LBP 94 MT U U Y U YBernaards et al. (2006) Netherlands Upper limb pain 314 MT LT Y N U Y YBrattberg (2006) Sweden Mix pain 60 ST LT U U Y U YBuhrman et al. (2004) Sweden LBP 56 ST Y U Y U YCedraschi et al. (2004) Switzerland Fibro 164 MT Y Y Y U YCorey et al. (1996) Canada Mix pain 200 LT U U Y Y YCurrie et al. (2000) Canada Mix pain 60 ST Y U Y U YDworkin et al. (2002) USA TMD 124 ST, MT LT U U Y U YErsek et al. (2008) USA Mix pain 256 ST MT LT Y U Y U YFries et al. (1997) USA OA + RA 809 MT LT Y U N U YHaas et al. (2005) USA LBP 109 MT Y Y Y U YHaldorsen et al. (1998) Norway Mix pain 469 LT U Y Y Y YHaugli et al. (2001) Norway Mix pain 174 ST LT U U Y Y YHeuts et al. (2005) Netherlands OA 273 ST LT Y U Y Y YHopman-Rock and Westhoff (2000) Netherlands OA 120 ST MT U U Y Y YHughes et al. (2004) USA OA 150 ST MT LT Y U Y U YHurley et al. (2007) UK Knee 418 MT Y U Y Y YJohnson et al. (2007) UK LBP 234 ST LT Y U Y N YKeller et al. (1997) Germany LBP 65 ST U U Y U YKing et al. (2002) Canada Fibro 124 ST Y U Y Y YLaforest et al. (2008) Canada OA + RA 113 ST Y U Y Y YLeFort et al. (1998) Canada Mix pain 110 ST U Y Y U YLi et al. (2006) China Mix pain 64 ST Y U Y U YLonn et al. (1999, 2001) Norway LBP 81 LT Y U Y U YLorig et al. (2008) USA OA + RA 866 MTLT U U Y U YMannerkorpi et al. (2000) Sweden Fibro 69 MT U U Y Y YMartire et al. (2007) USA OA 143 ST MT U U Y U YMazzuca et al. (1997) USA OA 211 MT LT U Y Y U YMazzuca et al. (2004) USA OA 186 STMT LT U U Y U YMoore et al. (2000) USA LBP 266 ST MT LT U U Y U YNúñez et al. (2006) Spain OA 100 LT Y U Y U YOliver et al. (2001) USA Fibro 400 LT U U Y U YPariser and O’Hanlon (2005) USA OA 92 ST U U N U YQuilty et al. (2003) UK OA 87 MT LT Y Y Y U YRibeiro et al. (2008) Brazil LBP 60 ST MT Y Y Y Y YSmeets et al. (2006) Netherlands LBP 111 ST Y Y Y Y YTak et al. (2005) Netherlands OA 109 ST Y U Y Y YTavafian et al. (2007) Iran LBP 102 ST U N Y U Yvan der Hulst et al. (2008) Netherlands LBP 163 ST MT Y U Y U YVictor et al. (2005) UK OA 193 ST LT U U Y Y YVlaeyen et al. (1996) Netherlands Fibro 131 ST U U Y U YVon Korff et al. (1998) USA LBP 255 STMT LT U U Y Y YWilliams et al. (1996) UK Mix pain 78 ST Y U Y Y YYip et al. (2007) China OA 182 ST MT Y U Y Y YYip et al. (2008) China OA 95 ST MT LT Y U Y Y Y

OA = osteoarthritis, Fibro = fibromyalgia, Osteopor = osteoporosis, TMD = temporomandibular disorder, RA = rheumatoid arthritis, ST = short term, MT = medium term,LT = long-term. QA1 = adequate randomization sequence, QA2 = allocation concealment, QA3 = clear description of withdrawals, QA4 = masked outcome assessment,QA5 = >20 people in each arm, Y = yes, N = no, U = unclear.

C.L. Miles et al. / European Journal of Pain 15 (2011) 775.e1–775.e11 775.e9

Table S2Table of meta-regression results for age and %male as moderators.

Measure N studies (sample size) Moderator I2 (%) Adjusted-R2 (%) Regression coefficient 95% CI for Regression coefficient P-value

Pain intensity 39 (6067) Gender 47.7 �12.2 �0.0006 �0.0062 to 0.0051 0.840Functional capability 27 (4790) Gender 51.8 �3.6 �0.0019 �0.0084 to 0.0047 0.560Self-efficacy 17 (2576) Gender 32.4 �20.2 �0.0017 �0.0115 to 0.0082 0.732Depression 16 (1902) Gender 24.2 �18.7 �0.0025 �0.0108 to 0.0058 0.533SF36 General mental health 11 (1117) Gender 51.6 35.5 0.0097 �0.0021 to 0.0214 0.095a

Global health status 14 (1801) Gender 59.1 13.5 0.0114 �0.0003 to 0.0230 0.055a

Pain intensity 39 (6012) Age 43.1 20.7 0.0004 �0.0114 to 0.0121 0.116Functional capability 28 (4873) Age 45.8 28.3 0.0078 �0.0008 to 0.0164 0.074a

Self-efficacy 17 (2576) Age 17.3 46.6 0.0081 �0.0004 to 0.0165 0.060a

Depression 16 (1902) Age 13.3 42.3 0.0060 �0.0025 to 0.0144 0.156SF36 General mental health 11 (1117) Age 53.2 35.4 0.0118 �0.0082 to 0.0317 0.176Global health status 14 (1801) Age 61.2 12.7 0.0159 �0.0085 to 0.0402 0.223

a Significant effect (P < 0.10).

775.e10 C.L. Miles et al. / European Journal of Pain 15 (2011) 775.e1–775.e11

Gender by ES (GMH) Gender by ES (GHS)

Age by ES (functional capability) Age by ES (self-efficacy)

Fig. S1. Bubble graphs for measures with gender and age as moderators.

C.L. Miles et al. / European Journal of Pain 15 (2011) 775.e1–775.e11 775.e11