Improving the self-management of chronic pain: COping with persistent Pain, Effectiveness Research in Self-management (COPERS)

A note on terminology

Although it is obvious that only the affected individual can ‘self-manage’ his or her condition, courses to promote better self-management are commonly referred to as ‘self-management courses’, whereas a more accurate term might be ‘self-management support courses’. For brevity we have used the term ‘self-management course’ throughout this report but it should be understood that by this we really mean ‘self-management support course’.

Chronic musculoskeletal pain

The focus of this research programme is on chronic, non-malignant musculoskeletal pain. The International Association for the Study of Pain (IASP) defines chronic pain as that which has persisted beyond normal tissue healing time, usually interpreted as 3 months. 17 Estimates of the prevalence of chronic pain vary, but one estimate is that 7.8 million people in the UK suffer moderate to severe pain that has lasted for > 6 months. 2

Musculoskeletal health care is costly. In 2011 in the UK, NHS trusts (secondary care services) spent around 10% of their patient care expenditure on musculoskeletal disorders (around £5.16B), and in 2009/10 there were around 21 million primary care consultations for musculoskeletal conditions in the UK. 6,18 Musculoskeletal pain is more commonly reported by women and those from socially or financially disadvantaged groups. 19 Chronic pain can cause considerable disruption to people’s lives. Around one-quarter of those who have chronic pain report severe disruption (> 2 weeks in the last 3 months) to their usual activities and chronic pain was associated with poorer mental health and well-being and lower levels of happiness and higher levels of anxiety and depression.

Psychology and chronic pain

The presence of chronic pain is strongly associated with adverse psychological factors. 20 European guidelines on the management and prevention of low back pain21 include psychological criteria to identify those at risk of poor outcomes, known as ‘psychosocial yellow flags’. These were developed to determine whether or not patients required more detailed assessment and to identify those for whom physical intervention could be less appropriate because of the dominance of psychological problems that would affect a successful outcome of the treatment. Psychological factors that consistently predict poor outcome include:

• beliefs that back pain is harmful and potentially disabling

• fear-avoidant behaviour and reduced activity levels

• low mood and reduced social interaction

• expectation that passive rather than active participation in treatment will help.

These factors can be identified during a consultation or with screening questionnaires. 22,23

Treatment approaches for people living with chronic low back pain that address some of the psychological issues include cognitive–behavioural therapy (CBT) and reactivation and reassurance strategies. 24 These encourage new behaviours and activity to overcome and change the psychological constructs limiting recovery and activity. Self-management courses incorporate some of the same strategies but more work is needed to maximise and quantify the clinical effectiveness and cost-effectiveness of these programmes. Redirecting resources to develop appropriate psychological interventions, with potentially more sustained benefits and fewer side effects than drug treatment, may have long-term clinical and economic advantages.

The biomedical model as opposed to the biopsychosocial model might be of limited benefit, especially for patients with high levels of health anxiety. 25 Linton et al. 24 suggest that a key component in establishing reassurance in patients is empathy and emotional support from the clinician. Emotional support outweighs the need for information and explanations in patients with unexplained pain. 26 A systematic review, however, found that cognitive reassurance was more effective in sustaining long-term improvements in patient outcomes than affective reassurance. 27 Cognitive reassurance includes objective information giving with clear explanations, whereas affective reassurance includes concepts such as empathy and warmth. The components of an effective self-management intervention should be designed in the knowledge that individual factors will determine different responses by patients to different components. Conceptually, the identification of groups of people for whom the same components will prove effective must be considered before implementing interventions. Indeed, the failure to adapt specific components that target the needs of different subgroups may explain the negative findings in some recent trials of behavioural interventions. 28

Self-management and evidence

We use the following broad definition of self-management throughout:

Self-management education programmes are distinct from simple patient education or skills training, in that they are designed to allow people with chronic conditions to take an active part in the management of their own conditions.

Foster et al. 14

The UK national evaluation of the original Expert Patients Programme for the self-management of chronic conditions reported a statistically significant increase in patients’ self-efficacy (which can be interpreted as self-confidence in relation to a specific context) and self-reported energy levels but no reduction in health-care utilisation. 15 Others found that the beneficial effects of lay-led self-management programmes for chronic conditions were modest in the short term and demonstrated a paucity of evidence on long-term benefit. 14 Two subsequent systematic reviews13,29 have reported improvements in patients’ confidence to manage their symptoms and small effects on pain and disability; both studies concluded that the benefits were too small to have any meaningful effect. Possible explanations for these findings included:

1. Suboptimal content of interventions.

2. Suboptimal delivery of interventions.

3. Programmes are effective only for some patients.

4. Measuring outcomes that may not be relevant to the intervention – the majority of studies measured change in pain symptoms, which are unlikely to change in chronic conditions. Self-management interventions may have a greater effect on confidence, positive outlook and coping methods than pain.

5. Poor targeting of the interventions (i.e. to those least likely to benefit).

6. Supporting self-management is an inherently ineffective approach.

Predictors, moderators and mediators

In many areas of health-related research, attention has started to focus on better matching of subgroups of patients to interventions. This aims to improve the effectiveness of treatment by targeting those most likely to benefit and avoiding offering treatment to groups for whom treatment is neither acceptable nor beneficial.

For example, in research in the musculoskeletal pain and mental health population, identification of subgroups is considered a priority. 22 The terminology around subgroup effects can be confusing. We have adopted the distinctions between predictors, moderators and mediators to describe how participant factors affect outcomes in randomised controlled trials (RCTs) suggested by Kraemer et al.:32

• Predictors are baseline variables that affect outcome (significant main effect only) but do not interact with treatment. Such factors significantly predict outcome equally for target interventions and control conditions.

• Moderators are baseline variables (such as patient baseline characteristics) that interact with treatment to change outcome for each subgroup. These specify for whom and under what conditions treatment works.

• Mediators are variables measured during treatment (factors that change during the intervention) that impact on outcome, with or without interaction with treatment, for example mood might be a mediator for a different outcome such as employment status. Mediators help inform the process and potential mechanisms (including causal mechanisms) through which treatment might work and therefore can be used to improve subsequent interventions through strengthening the components that best influence the identified mediators.

There is evidence from prospective cohort studies reporting predictors of outcomes for people with chronic musculoskeletal pain, but far less from RCTs reporting moderators and mediators. 20,33–35 Most of what is published, at least for low back pain, is of poor quality. 36 RCTs are the best study design to explore moderators and mediators but those that include planned subgroup analysis require very large samples. 37,38 Such subgroup analyses must also be based on good theoretical reasoning and previous evidence to support the hypothesis that the correct subgroups have been identified a priori. 39 In practice, these analyses are usually undertaken as secondary analyses and, consequently, are statistically underpowered, leading to a lack of robust data for mediators and moderators in self-management interventions.

Part I: development, design and feasibility testing of the intervention

The development phase consisted of two systematic reviews, a qualitative study, modelling, the design of the intervention and a feasibility study. Chapter 8 summarises the findings from part I and is not shown as part of the study design illustrated in Figure 1.

Part II: clinical effectiveness and cost-effectiveness of the intervention

The testing phase consisted of a RCT and a cost-effectiveness analysis and study of the fidelity of the delivery of the intervention in the RCT.

At the end of part II we bring together the findings and discuss these in relation to current thinking in the field of chronic pain and self-management.

Patient and public involvement

We included patients and the public in phases 1 and 2 of the research. In phase 1 we recruited four people (one male and three females) with a chronic condition to a patient advisory group. These people gave advice and made comments on all of our patient-related documentation, resulting in substantial improvements to the documentation. In addition, they played a role in the outcomes study (see Chapter 5) by reviewing outcome measurement tools and commenting on their acceptability, brevity, comprehension and ease of completion (in retrospect we feel that we could have included more patient advisors in this phase of the study). People with experience of chronic pain were heavily involved in the development and refinement of the intervention, particularly in terms of their collaboration during the feasibility study when we collected data from all participants and from the lay facilitators on their experiences of every session (using a questionnaire) and at focus groups and interviews following the completion of the courses. The free and frank discussions at the focus groups enabled us to refine the intervention and the training to deliver the intervention. We also consulted extensively with the Bangladeshi community through interviews during the feasibility study of the intervention delivered in Sylheti. Two professional bilingual Bangladeshi advocates also provided extensive advice on patient-related material, running the courses in Sylheti and outcome measures in the Sylheti-speaking community.

In phase 2 we included two patient representatives with a chronic condition on the Trial Steering Committee (TSC), one female and one male. Both were experienced representatives who had previously sat on National Institute for Health Research (NIHR) research priorities panels. These people gave valuable advice to the TSC and the excellent recruitment rates and low attrition seen in the study are, in part, a reflection of their contribution.

In addition, Social Action for Health [see www.safh.org.uk (accessed 11 April 2016)] was a coapplicant in the study. Social Action for Health is a community interest company providing socially orientated services to the local community. Members of Social Action for Health were part of the trial study team and represented the patient perspective for decisions made throughout the progress of the study.

Finally, and perhaps most importantly, patients were integral to the design and delivery of the intervention as we recruited patients with experience of chronic pain to deliver our intervention for the feasibility study and the main trial.

Outcome measures

The outcomes that we were interested in were extracted and grouped into the following categories:

• global health measures

• pain intensity

• physical/functional capability

• quality of life

• self-efficacy

• anxiety

• depression

• social role/function

• Short Form questionnaire-36 items (SF-36)62 general mental health (excluded in update review)

• number of visits to HCPs (excluded in update review)

• fatigue (excluded in update review).

The mean and standard deviation (SD) of the final value and/or change scores for each group at each follow-up interval were extracted.

Search method

Assessment of study bias

Two review authors independently assessed trial quality according to the following criteria modelled on The Cochrane Collaboration methods. 39 We asked:

1. Did the study have an adequate randomisation sequence?

2. Was allocation concealment carried out?

3. Was there a description (i.e. numbers provided) of withdrawals and dropouts?

4. Was outcome assessment blinded?

5. Was there an intention-to-treat (ITT) analysis?

Each question was rated as yes/no or unclear (see Table 3).

Data extraction

The two reviewers working independently each extracted data about country of origin, number randomised to each arm, who delivered the course (HCP or lay tutor or a combination), whether the course was delivered to groups or individuals or was self-delivered, setting [community, medical, occupational, remote (telephone/internet) or mixed], total number of sessions and contact hours, course duration, course components (see Table 2), description of control group and the description and results of any moderator analyses. Data were extracted, when possible, at four time points: baseline, short-term follow-up (< 4 months), medium-term follow-up (4–8 months), long-term follow-up (> 8 months) or a mixture of follow-up points.

Description of self-management components

To handle the vast number of data arising from the studies, we categorised self-management interventions into psychological, mind–body therapy, physical, lifestyle or medical education components, as described in Table 2. Each study was coded so that the intervention arm was described using two or more components from the list. We accept that these categorisations represent our interpretation of the published reports of studies and that some components may well have overlapped between our broad categories.

Final study selection

Following data extraction, the RCTs were further selected to include only comparisons of the self-management intervention(s) with waiting list controls or usual care.

Meta-analysis methods

The meta-analyses were carried out using Review Manager v5 (The Cochrane Collaboration, The Nordic Cochrane Centre, Copenhagen, Denmark). Calculations were based on final values as these were the most commonly reported data. Change score data were also collected when possible. When studies reported p-values for change from baseline for each group, this enabled a SD for the change score to be calculated. 39 Change scores were analysed in the same way as the final value data for the outcomes when there were sufficient data and compared with the final value results for the same outcomes as a sensitivity analysis. We used a random-effects model because of the heterogeneity in study populations and interventions.

When there was more than one measurement tool for an outcome we combined data across studies using a SMD meta-analytical approach (see section 9.2.3.2 of the Cochrane Handbook for Systematic Reviews of Interventions39), where

For those instruments for which an increase in score indicates improvement we reversed the sign on the mean score to enable us to combine these as a pooled SMD with measures from instruments for which a decrease in score is beneficial.

The resulting SMDs were interpreted using Cohen’s d63,64 (where d is derived from the difference between two independent means divided by the within-population SD as above). The effect sizes were conventionally defined as follows: ‘minor’ < 0.2, ‘small’ ≥ 0.2, ‘moderate’ ≥ 0.5 and ‘large’ ≥ 0.8.

Meta-analysis comparisons

We tested:

1. Overall effectiveness. Total effect size or SMD for self-management interventions with regard to our prespecified outcome categories compared with the control.

2. The effect of course delivery mode. We grouped the studies at each follow-up interval into different course delivery modes [group, individual, mixed (group and individual) or remote (internet, mail, telephone)] and compared the pooled SMDs for each subgroup.

3. The effect of course leader. We grouped the studies and outcomes at each follow-up interval into those that were delivered by a HCP, those that were delivered by a lay tutor and those using a mix of delivery methods and compared the pooled SMDs for each subgroup.

4. The effect of course setting. We grouped the studies at each follow-up interval into those delivered in the community, those delivered in a medical setting (primary care or hospital) and those delivered in an occupational setting and compared the pooled SMDs for each subgroup.

5. The effect of course duration. We grouped the studies at each follow-up interval into those with courses of ≤ 8 weeks and those with courses of > 8 weeks and compared the pooled SMDs for each subgroup.

6. The effect of contributing self-management components. We tested whether or not the presence of a particular self-management component improved the effectiveness of the interventions. We compared the pooled SMDs of studies with psychological, mind–body therapy, physical, lifestyle and medical education components with the pooled SMDs of studies without these components.

7. The effect of number of components. We tested whether or not the number of components (two, three, four or five) influenced the effect size, comparing the pooled SMDs at short-, medium- and long-term follow-up intervals.

We produced forest plots of final value data for each comparison showing the pooled SMD for each subgroup.

Assessment of publication bias

Funnel plots were generated using Review Manager v5 for the outcomes with the most studies. The funnel plots were visually examined for symmetry about the y-axis and resemblance to an inverted funnel to denote absence of bias.

We present data from the original review meta-analysis and the updated review data.

Literature search

The initial search produced 4676 results and of these we included 46 RCTs. When we updated the search to September 2013 we included a further 18 trials in the meta-analyses for the overall effectiveness of self-management interventions.

Figure 2 shows the flow chart for the initial search and Figure 3 shows the flow chart for the updated search.

FIGURE 2.

Searches for the systematic review 1994–2009. CMP, chronic musculoskeletal pain; SM, self-management.

FIGURE 3.

Systematic review update June 2009–September 2013. CMP, chronic musculoskeletal pain; FU, follow-up; SM, self-management; UC, usual care; WLC, waiting list control.

Effectiveness analyses

For our original meta-analyses we included 46 RCTs54,65–111 with final-value data comparing self-management programmes with usual care or waiting list controls (n = 8539) (Table 3). Thirteen RCTs were conducted in the USA,72,74,77,83,85,86,91–93,95,104,106,108 seven in the Netherlands,69,81,94,96,98,102,110 five in the UK,54,67,70,88,97,103 five in Canada,68,76,80,87,107 three each in Sweden,82,89,99 Norway75,78,79,84 and China,100,105,111 two in Germany71,73 and one each in Turkey,65 Iran,66 Switzerland,90 Spain101 and Brazil. 109 Of these studies, 13 (28%) were on OA (hip or knee),74,81,88,91,92,94–97,101,104,105,111 12 (26%) were on low back pain,54,66,71,73,77–79,83,89,93,102,109,110 nine (20%) were on mixed pain,68,70,75,76,80,84,99,100,106 five (11%) were on fibromyalgia,69,82,85,87,90 three (7%) were on mixed arthritis (OA and rheumatoid arthritis)72,107,108 and one (2%) each was on temporomandibular joint disorder,86 osteoporosis,65 upper limb pain98 and knee pain. 67,103 The mean age of participants for the 44 studies reporting age was 55 years (range 38–82 years). In the 41 studies reporting gender breakdown, 72% were female participants, with two studies having exclusively female patients. Thirty-six studies were health care professionally led, six were mixed health care and lay led and four were lay led. Twenty-seven studies were conducted in a medical setting, sixteen in the community and three in occupational settings. Twenty-seven were delivered in groups, five were delivered remotely via the internet and five were delivered individually; nine used mixed group and individual delivery.

For the update review we included an additional 19 sets of data; one trial67 was included in the original meta-analyses but published longer-term follow-up data after our 2009 cut-off103 and we included this study in the update analyses. Of the 18 new studies, five were from the USA,113,114,118,119,123 three from the UK,51–53 two each from Australia112,124 and Spain120,126 and one each from Italy,121 Germany,115 Sweden,117 Norway,116 Taiwan125 and Canada122 (n = 3745). Five focused on low back pain,53,114,115,121,123 five on OA,112,117,122,124,125 five on fibromyalgia,116,118–120,126 two on mixed pain52,113 and one on knee pain only. 51 Fourteen were health care professionally led,51–53,113,115–118,120–122,124–126 one was mixed health care and lay led112 and three were self-led. 114,119,123 Nine were conducted in a medical setting,112,115–118,120,121,125,126 eight in the community51,52,113,114,119,122–124 and one in a mixed setting. 53 Ten were delivered in groups,53,116–118,120–122,124–126 four were delivered remotely via the internet52,114,119,123 and three were delivered individually;51,113,115 one used a mixed method of delivery. 112 The mean age of participants for 17 of the 18 studies reporting age was 54 years (range when specified 25–84 years) and 2509 (67%) of the 3745 participants were female (see Table 3).

Quality assessment

Around half of the original 46 studies (54%) reported an adequate randomisation sequence, and in the remainder of the studies this was unclear. Allocation concealment was present in nine studies (20%), absent in two (4%) and unclear in 35 (78%). Masked outcome assessment was reported in 19 studies (41%), with the remainder unclear. Nearly all studies (91%) reported reasons for dropping out of the study and, in the 44 studies reporting this, the mean attrition rate across all study arms was 18% (range 6–47%). One study reported a 100% completion rate and 11 studies had an attrition rate of > 20%. Only 22 studies (48%) reported that they had analysed their data on an ITT basis, using last observation carried forward or imputation methods to fill in missing data (see Table 3). Quality assessment in the updated articles showed that 17 out of 18 studies reported adequate randomisation procedures, 9 out of 18 used concealed allocation, 10 out of 18 reported withdrawals, in 14 out of 18 researchers were masked to outcomes and in 15 out of 18 ITT analyses were conducted (see Table 3).

Overall effectiveness of self-management programmes

We used the data from the original review (up to 2009) to inform our intervention design (Table 4). The final column of this table includes the updated meta-analyses. The addition of studies from the updated search made little difference to these findings with the exception that there are more studies reporting depression and those reporting medium- and longer-term results show small effects on most outcomes. The differences in results between 2009 and 2013 showed changes in effect sizes for anxiety (small significant medium effect size in the long term) and social function (in the medium term) (see Appendix 1 for forest plots). In summary, these data suggest that the interventions studied have small beneficial effects on global health, pain intensity, physical function, quality of life, anxiety and social function in the short term and sometimes the medium term but that these effects are much reduced in studies reporting longer-term follow-up (beyond 8 months). For quality of life and anxiety, the effects in studies reporting longer-term follow-up remain small rather than ‘minor’, but closer examination reveals that in each case there was only one small study supplying longer-term follow-up data, raising the possibility of publication bias. For depression the beneficial effects are ‘minor’ in the short term but small in the medium term; however, there are far fewer studies reporting medium-term (or long-term) effects. Unlike the other outcomes, there appears to be a small beneficial effect on self-efficacy in studies reporting short-, medium- and long-term follow-up.

Effectiveness of the different characteristics of self-management programmes

The data for these analyses come from the studies identified in the original search (1994–2009).

For ease of reading we present the statistically significant SMD effect sizes [with 95% confidence intervals (CIs)] only for data favouring self-management over waiting list control or usual care for each outcome subgroup comparison except for fatigue, SF-36 general mental health and visits to HCPs. Tables showing all of the results are available from the corresponding author.

In Tables 5–7 we present small (≥ 0.2), moderate (≥ 0.5) and large (≥ 0.8) effect sizes for different outcomes at different follow-up intervals. Results that favoured the control arm, favoured neither subgroup, were non-estimable or resulted in minor effect sizes are presented in the accompanying text for each outcome.

Effectiveness of the different components of self-management programmes

In Tables 8–13 we present small (≥ 0.2), moderate (≥ 0.5) and large (≥ 0.8) effect sizes for different outcomes at different follow-up intervals. Results that favoured the control arm, favoured neither subgroup, were non-estimable or resulted in minor effect sizes are presented in the accompanying text.

Effectiveness of self-management programmes

This review identified 64 studies (65 papers) providing usable final-value outcome data for meta-analysis. Overall, it showed ‘small’ statistically significant effect sizes in the short term (< 4 months) for pain intensity, physical/functional capability, self-efficacy, SF-36 general mental health, global health status and SF-36 social role. These effect sizes became ‘minor’ (SMD < 0.2) and/or statistically insignificant in the medium term (4–8 months) and long term (≥ 8 months) for all outcomes except for self-efficacy (small statistically significant effect size for all time intervals) and global health status (small statistically significant effect size for medium-term follow-up).

Study limitations

Definitions

The definitions of predictors, moderators and mediators have been refined in recent years. We adopted the approach of Kraemer et al. 32 in which three types of subgroups are described and clearly defined:

Predictors of treatment outcome are defined as baseline variables that affect outcome (significant main effect only) but do not interact with treatment. Such factors significantly predict outcome equally for target interventions and control conditions.

Effect moderators (or moderators) are variables measured at baseline (such as patient baseline characteristics) that interact with treatment to change outcomes. The interaction should be related to outcome in the linear model with or without a main effect. These specify for whom and under what conditions treatment works and can improve power in subsequent trials by better selection of target groups for stratification.

Mediators are variables measured during treatment (such as change-in-process factors) that impact on outcome, with or without interaction with treatment. Mediators help inform the process and potential mechanisms (including causal mechanisms) through which treatment might work. They can be used to improve subsequent interventions through strengthening the components that best influence the identified mediators. Mediators should not be a component of the treatment or outcome. There should be a clear distinction between the constructs measured by the proposed mediators and treatment outcome.

Predictors and moderators may include:31,32,42

• demographic status [age, gender, education, marital status, lifestyle (alcohol consumption, exercise, smoking)]

• clinical status (e.g. disability, duration of pain, pain intensity)

• psychological status (e.g. catastrophising, depression, fear avoidance and beliefs)

• work-related factors [e.g. employment status, type of work, reasons for not working, number of sick days taken over previous year, financial factors (pending compensation, sickness benefit, insurance and duration on current benefits), job satisfaction, social support at work, a sense of control at work]

• medication history.

Quality assessment of included studies

For the overall quality of the papers we used the assessment carried out for the review of main effects. Early in the project we sought established methodological criteria for the assessment of subgroup analyses within RCTs. We were unable to identify any such criteria suitable for our purpose. Consequentially, as part of this programme of work we carried out a literature review and Delphi study to determine a consensus on methodological criteria for the evaluation of studies reporting moderator analyses within RCTs. We used these assessment criteria to assess the quality of our moderator studies and grade the evidence for moderators. 135

We applied the following criteria to included papers:

1. Was there an a priori specification of the subgroup?

2. Was there a theoretical or an evidence-based rationale for the selection of subgroup factors?

3. For moderator and predictor analyses only: was the measurement of subgroup factors carried out prior to randomisation?

4. Was the measurement of subgroup factors adequate (reliable and valid) and appropriate for the target population?

5. For moderator and mediator analyses only: does the analysis contain an explicit test of the interaction between subgroups and treatment?

Papers were classified as providing either confirmatory findings or exploratory findings. Confirmatory findings refer to studies that include a priori hypotheses in relation to subgroups, for which support was obtained through adequate statistical testing. Exploratory findings inform future research (hypothesis generating) and are the product of post hoc testing. Only studies that satisfied all of the above five criteria relevant to the analysis were regarded as providing confirmatory evidence. Papers satisfying criteria 3–5 were categorised as exploratory findings. Studies that did not satisfy these criteria were regarded as having insufficient findings. For this review we applied these same criteria to predictor and mediator analyses as appropriate.

We included any RCTs identified in the searches in Chapter 2 that reported predictor, moderator or mediator analyses. For selected RCTs that did not report any subgroup analyses but which had ≥ 80% completion rates for their primary analysis and > 200 participants in each arm, the authors were contacted and asked if they had carried out, or were now able to carry out, such analyses. We also contacted authors when any aspects of their reported subgroup analyses were unclear.

Data extraction

We extracted the following data from each RCT:

• country and setting

• population: brief description of participants, including size of eligible population identified and actual uptake

• intervention/control

• baseline factors measured

• outcomes measured

• description of predictor, moderator and mediator analyses

• results of predictor, moderator and mediator analyses.

Finally, an independent statistician scrutinised the included studies to determine whether or not there were appropriate data to include in the meta-regression and whether any moderator or mediator analyses were carried out appropriately. The statistician checked for the presence of a reported power calculation; whether there were any statistician authors or whether the authors described seeking statistical advice; and the statistical methods used and the quality of reporting, paying particular attention to moderator and mediator analyses. We also conducted a meta-regression to identify potential moderators. We included any RCTs identified in our original search (see Chapter 2) whose final-value data were suitable for inclusion in a meta-regression. For all potential moderators reported in ≥ 10 studies we performed a random-effects meta-regression. 39 We collapsed outcomes into the following domains: pain intensity, physical/functional capability, self-efficacy, depression and global health status. When a variety of measurement tools had been reported for a domain, we calculated SMDs (difference in mean outcome between groups/SD of outcome among participants). 39 To make the best use of the available data, and reduce the possibility of making a type 1 error, we collapsed the different follow-up time points (early and late) to obtain one average effect size per outcome. We considered results from the meta-regression to be statistically significant if p < 0.10. We used this criterion because of a potential type II error as a result of the limited number of studies in these effect size calculations. 42,136 We used an I² statistic to estimate the percentage of residual variation attributable to between-study heterogeneity and an adjusted R² statistic to estimate the proportion of between-study variance explained by the covariate. We produced scatter diagrams using circles as plotting symbols in which the areas of the circles indicated the value of a third variable. 137 We fitted values and predicted random effects against age and gender separately, with 95% confidence.

Methodological quality

In the first instance we applied quality assessment criteria to the trial methodology, using criteria adapted from The Cochrane Collaboration methods39 and described in Chapter 2. We independently analysed and coded each publication separately, even though some studies were further analyses of previously published RCTs; hence, in Table 14 we present the quality appraisal for 20 reports covering 16 studies. 67,75,93,104,107,110,138–151 Nine RCTs were of high quality, six were of medium quality104,110,138,139,145,148 and one was of low quality. 146 Eight of the subgroup studies provided confirmatory evidence, one study provided exploratory evidence93 and seven studies provided insufficient evidence. For subgroup analyses we applied the criteria described in Quality assessment of included studies. Only three studies67,107,142 used high-quality trial methodology and carried out preplanned theoretically driven subgroup analysis, using correct statistical analysis. However, there were no two trials in this category (for any predictors, moderators and mediators) that examined the same subgroup.

We did not find any high-quality studies appropriately reporting moderator analyses.

Predictors

Hurley et al. 67 found that higher levels of depression at baseline predicted poorer physical functioning at 6 months in people with chronic knee pain (effect size 0.48; p = 0.011), whereas higher levels of self-efficacy at baseline, measured by positive exercise beliefs (effect size –0.24; p = 0.001) and confidence in the ability to exercise (effect size –0.62; p = 0.001), predicted better physical functioning at 6 months, regardless of intervention arm.

Mediators

Smeets et al. 142 found that reduced levels of pain catastrophising during treatment led to a post-treatment decrease in patient-specific complaints, disability and pain in people with chronic low back pain. Patients in the intervention group scored, on average, 1.3 points lower for disability (out of 24) than patients in the control arm, after adjusting for pain catastrophising. For current pain, the difference was 4.7 units on a visual analogue scale (out of 100). For patient complaints, the difference was 6.7 (out of 100).

Laforest et al. 107 reported that increases in physical activity mediated greater decreases in helplessness in people with arthritis; however, the data were not available to quantify this effect. This effect was defined by Laforest et al. 107 as a moderator, although from the authors’ description it is a mediator. Because of these limitations, we recommend that the findings for the mediating effects of physical activity be reviewed with caution.

Meta-regression to identify potential moderators

All RCTs identified from the original search that supplied full data on age and gender at baseline against at least one of our selected outcomes were included (n = 46/126).

We used bivariable meta-regression to determine whether or not the baseline characteristics (age and gender) explained the variation in treatment outcomes. Age and gender were selected because they are the most frequently reported demographic characteristics. 39 They were the only variables reported in at least 10 studies.

For one outcome, general mental health, a single measurement tool had been used, the SF-36. 62 We intended to combine the data as a weighted mean difference but because of anomalies in score values between studies we judged that SMD analyses would be more robust. A variety of measurement tools was reported for each domain; for these we also calculated SMDs.

Forty-six studies were included in the meta-regression and eight in the 16 subgroup meta-regressions. Gender was significantly associated with effect size for SF-36 general mental health and global health status, indicating that self-management interventions were more effective for groups with a higher proportion of females (all p < 0.10 and p ≥ 0.05) (Table 15). Inspection of bubble graphs (see Appendix 2, Meta-regression) suggested a positive association between studies with an increasing proportion of males (note that a fall in measures indicated improvement in our analyses). Gender was not statistically significantly associated with effect size for pain intensity, physical/functional capability, self-efficacy or depression. Age was associated with effect size for physical/functional capability and self-efficacy (all p < 0.10 and p > 0.05). The bubble graphs (see Appendix 2, Meta-regression) suggested a positive association between effect size and these outcomes, indicating that self-management interventions might be more effective in younger samples. Age was not associated with effect sizes for the other outcomes.

Summary of findings for predictors, moderators and mediators

There were only three studies with sufficient methodological rigour, including a priori planned subgroup analysis, to inform on predictors and mediators. The findings from these studies suggest that high levels of depression at baseline are associated with lower function and that self-efficacy, especially about the ability to exercise, is associated with improved function 6 months later, regardless of any intervention. 67 There was also evidence that reduction in catastrophic thinking directly after the intervention was associated with improved function 6 months later. 142 Finally, engaging in exercise during the intervention was associated with better function at outcome. 107

Findings from the meta-regression looking at age and gender in relation to outcomes showed no significant associations with pain intensity, physical/functional capability, self-efficacy, depression, SF-36 general mental health or global health status. As mentioned in the methods section, because of the small number of studies for each outcome, marginally significant associations (p < 0.10), were considered as hypothesis generating for future studies to explore possible relationships. 42

Recommendations

Based on the above findings we can make the following recommendations:

• Findings for the predictive effect of self-efficacy on outcomes and the mediating effects of pain catastrophising on function indicate that interventions should incorporate strategies to improve self-efficacy and coping techniques.

• Based on the currently available evidence, we do not recommend targeting the intervention to a particular age group or gender.

• There is currently no evidence to support an association between duration of pain/complaint, pain intensity, work status, disability level or diagnosis of depression at baseline and outcome. Therefore, access to current interventions should not be limited based on these factors.

• We recommend that study authors use the criteria that we identified for assessing the quality of moderator studies to guide their methodology.

Limitations/considerations of this review

Because of the small numbers of studies carrying out appropriate analyses, we were unable to quantitatively assess predictors, moderators and mediators of self-management programme outcome success, other than age and gender. Overall, the evidence is insufficient to inform on moderators, mediators and predictors of intervention success, other than for self-efficacy as a predictor and catastrophic thinking as a mediator. We have only considered RCT data here rather than including observational data. This is because we were specifically interested in factors to consider in the context of a RCT rather than an observational cohort study.

Because of the lack of consistent reporting of potential predictors, moderators and mediators in research, the meta-regression featured only two potential moderators, age and gender. These were selected as they are the most commonly reported characteristics in studies and data were readily available. Only when researchers report potential moderating variables as a standard can meta-analytical techniques be used to calculate a more accurate estimate of the variance between studies.

Studies reporting significant predictor, moderator or mediator effects of self-management interventions are more likely to be published than those studies that do not report significant effects. We did not include unpublished studies or dissertations in this review and therefore the studies identified for this review are likely to reflect those that escaped the ‘file drawer’ problem,152 which might be a result of publication bias.

An additional limitation, which most meta-analytical studies face, involves combining scores from different tests. For example, several different tests measuring self-efficacy were combined and it was unclear whether or not they were suitable for merging until an item analysis had been carried out on the measures. It could be argued that only constructs that are measured using the same instrument should be merged, to avoid this problem.

The analysis showed heterogeneity of findings for some outcomes. For example, a mild to moderate level of heterogeneity of findings across studies was reported in the meta-regression. Heterogeneity may result from the small number of studies available for each test. Other factors that have not yet been identified, or not yet included in moderator research, might also explain some of this heterogeneity. It would be premature to conclude that intervention effects do not vary by patient or study characteristics. We were unable to examine several moderator effects because of insufficient data (i.e. education, self-efficacy, pain duration, ethnicity, disability and depression).

We aimed to identify moderators, predictors and mediators of efficacy of self-management programmes for those with musculoskeletal pain. We believe that this systematic summary of the literature identifying moderators, predictors and mediators, and meta-analysis to examine influences of moderating factors provides useful recommendations based on evidence and we have also provided a checklist of methodological criteria for assessing moderator studies.

Recruitment and sampling: individual interviews

We aimed for a diverse sample, with a total anticipated sample size of 20 (or fewer if data saturation was reached earlier). We recruited people living with persistent pain who had attended self-management courses. Two Expert Patients Programme community interest companies, one in London and one in Warwick, and Social Action for Health in London approached participants who had attended their self-management programmes and invited them to be interviewed by the study team. The invitation letter contained a consent to contact form and a reply-paid envelope so that the researchers could approach those willing to participate. We also invited people to participate via an internet chat forum for people who had attended self-management courses.

Participant inclusion and exclusion criteria are shown in Table 16.

We used a purposive, diversity sampling method to obtain a good representation of different genders, age groups, socioeconomic areas (by postcode) and self-management course attendance record. Further data about the participants were collected after they had consented to participate in the study.

Conducting the interviews

Face-to-face interviews were arranged by the researchers at a time and place convenient to the participants. With permission, interviews were audio-recorded, transcribed and anonymised, with each interview lasting approximately 1 hour. The content of the topic guide was informed by knowledge of the literature, the specific needs of this study and the study team’s past experience of issues in other trials and evaluations of self-management programmes. The topic guide covered referral processes, motivation to attend courses, memorable aspects of courses, most effective and least effective components, positive and negative aspects of support/networking, follow-up, strategies for coping and long-term effects/outcomes of courses. Most interviews were conducted by a medical anthropologist; as a substantial proportion of first-generation Bangladeshi residents who live in Tower Hamlets are not fluent in English some interviews were carried out by a health researcher fluent in Bengali/Sylheti and English.

The focus groups

We convened two focus groups (maximum six participants in each), one with UK-based self-management academic researchers and opinion leaders in this field (the study team identified key personnel to approach based on their own knowledge of researchers and policy-makers in this field) and one with self-management course lay tutors and providers (a mix of programme leaders and tutors was approached). The focus groups were facilitated by a health researcher familiar with self-management research and a musculoskeletal clinician. We fed the findings from the individual interviews into the focus groups to test the ‘acceptability’ of our findings and to generate discussion about participant receipt of self-management courses and the issues raised by participants who attended courses.

Analysis

All focus group conversations and individual interviews were audio-recorded and transcribed. NVivo software version 9 (QSR International, Warrington, UK) was used to manage the analysis of the data. We carried out thematic analysis using the framework method; this involves familiarisation with the literature and the development of a framework based on emergent themes and subthemes. 154 A third experienced qualitative researcher was involved as an independent reviewer; to promote study validity, he commented on the framework of themes and subthemes based on a sample of interviews and independently coded a transcript. Three researchers coded the same three transcripts to test the framework and to compare the inter-rater reliability of our coding. We used this as a training exercise to improve the consistency and reliability of coding. The results from the focus groups with experts and lay tutors were also compared and contrasted with the findings from the interview study.

Ethical approval was granted by the East London and the City Research Ethics Committee Alpha (reference number 09/H0704/24).

Sample

Twenty-six people participated in this study. We conducted 16 one-to-one interviews, at which point we felt that we had reached data saturation. Twelve interviews were conducted with English-speaking participants and four were conducted in Bengali/Sylheti. Four interviews were conducted in Warwickshire and the remainder were conducted in East London. Table 17 shows the characteristics of the interview sample.

We ran two focus groups. The tutor focus group included five self-management course tutors or facilitators active in east London, south London and Essex (FG1.1–1.5). The expert focus group also consisted of five participants (one primary care trust service commissioner, three academic researchers and one self-management course service provider) (FG2.1–2.5). When the focus group discussion reflected the thoughts and beliefs of the interviewees we have added this into the text. We also report dissonant data from the focus group discussion when applicable.

Themes and subthemes from the interviews

We initially derived eight main themes (Table 18, first column). When we initially tested this framework we found a lot of duplication of data in themes 3 and 8. There were many references to social aspects of life (theme 3) but these were nearly always in the context of barriers to and/or promoters of change (theme 8); these themes were therefore merged. Additionally, the original themes 1 and 4 were collapsed to behaviour and thoughts, including empowerment and ‘negative’ behaviours, including self-absorption, anger and frustration. We also identified a new theme about outcome expectations. The last two columns in Table 18 show the final framework used for the analysis of the interviews.

Findings

Theme IV: course characteristics

In theme IV, we separated the data into four categories: marketing, referrals and recruitment; course delivery and impact of tutor; attendance; and mix and character of attendees.

Theme IVa: marketing, referrals and recruitment

Two schools of thought emerged about recruitment and marketing. The participants did not want to be told or coerced to go on a course, but they did want legitimisation of the course through recommendation by their general practitioner (GP). They also felt that there was a need for a self-referral route, coupled with increased marketing and information for the public, as many GPs were not aware of the different courses. The tutors, however, were keen for recruitment to take place through GP registers and GPs themselves, as this allowed for better targeting of those with greater need and gave access to a ready source of potential participants.

Encouragement from those involved with the course from the initial point of contact was important. Building rapport with potential participants helped manage expectations and encouraged engagement with learning and the process. The presence of the recruiter at the first session was welcomed. The tutors thought that the recruitment process was very important for managing expectations. Both unrealistically high and very low expectations were seen by tutors:

And I think that’s a gap where some health care professionals will not understand what it is.

FG1.4

we are getting access to the GPs’ register, i.e. we have got a standard letter saying: a self management course is coming to the area, and so on, and we have asked the GP practice would they use that letter, you know, put their letterhead on it and also sign it and so on . . . And we are getting a really, really good response that way.

FG1.6

Mile End itself phoned up: a new unit’s opened up dealing with persistent pain, we wondered if you’d like to attend. So I thought, I’ll go anywhere where . . . there’s always ways of learning things.

P2

I must be honest, if that physiotherapist had not have put me forward, I would not have known about it.

P4

you cannot lose anything but you might gain. I can tell her, but that’s me telling her; if somebody else tells her, it’s totally different. It comes over different, you know what I mean?

P7

But I do think doctors need to know, because I do not think many doctors know that EPPs [Expert Patients Programmes] exist!

P8

I think you have to be ready to be open-minded about helping yourself, and if you have gone there because you have been referred, not because you want to go, I think that would be a bad reason.

P3

When they are sort of at the stage where they are asking: I really want to do something. I want to . . . I’m fed up of waiting for everything and I’d like to help. I think that would be when doctors should refer them to EPP [Expert Patients Programme].

P4

They are professional people, they would think and say whether you’re doing it wrong or right and advise you how else it would be good. We would comply with them.

P5

Because of the fact that some participants tend to be isolated in their everyday life, they do kind of look for somebody to kind of latch on to, you get me?

FG1.5

Theme IVb: course delivery and impact of tutor

Both professional and lay leadership was valued. The tutor was generally perceived as a strong role model, often with ‘more right’ to give advice and guidance than their GP:

Yeah, and the fact that the tutors themselves has an action plan as well. Yeah, the role modelling is important.

FG1.3

Well, by relating, they were more or less relating to . . . they definitely knew about pain themselves, that’s the main thing.

P1

The reason why because they know, they could understand the highs and lows.

P7

Participants generally had a need to feel ‘listened to’ and to be taken seriously. Some would have preferred more time to tell ‘their story’, whereas others felt that this would incur too much competition. The tutors felt that it was an important part of their job to contain comparisons between people. There were lengthy discussions about mixed-disease courses and disease-specific courses. There were pros and cons of both, with no clear conclusions possible.

Effective course tutors were described as friendly, open, honest and non-judgemental; they respected others’ opinions, made people laugh, enjoyed the sessions and were relaxed; they listened and were flexible, informal and engaging; they controlled the input of those who were very negative or those who were overdominant and curtailed those who talked too much about themselves; they related to the group culturally; and they had experience of chronic pain.

Conversely, the descriptions of poor tutoring involved reading from a crib sheet; rigidly sticking to timescales at the expense of learning; not having experience of chronic pain; and not managing disruptive, negative or self-absorbed people.

Theme IVc: attendance

A number of factors were mentioned with regard to attendance and non-attendance on courses. Attendance was influenced by personal motivation to attend the course, the quality of the tutors and receptiveness to information and participation in group activity and social bonding. When all or some of these were mentioned the participants demonstrated good attendance. Those with poor attendance were those with a low or depressed mood, a negative attitude to the course, lack of positive support and demanding friends and family. Other factors that affected attendance were medical appointments, sickness and holidays – these were common to those with both low and high rates of attendance (Figure 4). The interviewees and the tutors in the focus group indicated that people with low self-esteem and those who felt intimidated about talking in front of others were unlikely to fare well on courses:

They’d be negative against everything and would not give things a try, but it could be they are actually chronically, clinically depressed, but they actually left.

P9

but they have not got that something to help them, to drive, to push probably.

P9

It was nice to share ideas with other people and get ideas, get new ideas, fresh ideas, some of which I have taken on board and some I have not! I wanted it to stop my illnesses, to be beneficial for my body, that’s what I thought, it would be enough.

P10

There are seven of us in this house including my husband and I, it’s a two-bedroom house, and we are living in such difficulties [she goes on to describe the problems encountered getting to the course] . . . It was quiet a distance for me. I just went the one day.

P3

They do not get me to do exercise nor do they do it themselves. You have to do things yourself.

P4

FIGURE 4.

Factors influencing attendance.

Theme IVd: mix and character of attendees

Attending a course that was delivered to a group appeared to have an impact on several factors. It gave participants motivation to achieve goals as they had to feed back progress to the group and it reduced levels of isolation – some were inspired by others and some made friends and all appeared to learn from each other or the course itself. Conversely, some participants felt that the action planning and feedback could be false, that is, an exercise in creative storytelling, and others insinuated that the goal-setting could potentially set them up for failure if they consistently did not achieve in front of the group and indeed this was suggested as a reason for people not attending. ‘Eureka’ moments were described, such as the realisation that others felt the same and had the same issues to deal with, and some found this inspiring and motivating:

It is that proper group work and I think that might be where the gap is where . . . maybe more emphasis on forming as a group, like group activities, before just throwing at them the information, might help in that process.

P4

You get ideas when they ask you questions but now I can’t say, you get it when talking, you can learn things too. If you say your idea then that’s one way and I’ll think another way.

P8

Theme VI: change

From the data we identified three main explanations/models of behaviour about change: behaviours of those who remain fixated and absorbed in pain (Figure 5), factors that appear to promote and sustain beneficial change (Figure 6) and the process of change as lifestyles expand, broaden and evolve to include more things.

FIGURE 5.

Observed behaviours and attitudes towards pain.

FIGURE 6.

Factors promoting and sustaining change behaviour.

Figure 5 shows the behaviours associated with those who appeared to be fixated on their condition; their lives and their identity revolved around being in pain, with few non-pain-related activities. As a result, ‘sickness’-related activity and behaviours were maintained. These participants were generally isolated, bored and frustrated and they had a low mood and were dependent on family and friends. Conversely, those who were not fixated on their pain used distraction techniques to broaden their activity and life experience, often creating new activities and hobbies, leading to new social networks and a self-identity associated with the activity rather than with their pain.

Figure 6 shows the factors that we found to be associated with participants who had managed to change and sustain the changes that enhanced their quality of life. Those participants who had adapted and coped well were motivated, were engaged in the process, had support and reported having good, inspiring tutors.

Each figure illustrates the potential impact that group-related activities may have.

Barriers to change included depression, lack of motivation, physical capability/disability, change in benefit payments, change in family dynamics, stress/pressure/confusion and change in behaviour conflicting with a ‘sick role’.

Additional findings from the expert focus group

A key consideration for the expert focus group was the legitimacy and credibility of findings from some studies and the transferability of some concepts in the Stanford self-management model9 to a UK audience:

It makes me wonder why people go on doing it [Expert Patients Programme] when there are better things available.

FG2.8

if we had a drug which on balance produced no effect, would we be working so hard to get GPs to put people on it?

FG2.5

Identifying those who do well and those who do not was discussed extensively but no firm conclusions were made about subgroups and screening. Selecting a ‘choose all’ strategy to include those with varying degrees of severity was, however, seen as a good way of recruiting:

Some people do benefit enormously . . ., it doesn’t make it unique to EPP [Expert Patients Programme] . . . a good regime of analgesics leaves some people self managing.

FG2.5

Rather than working hard to try and understand who does benefit and how we can target them better . . . what can we do about those who didn’t [benefit].

FG2.1

Cognitive reframing and linking pain and mood were flagged as lacking in current approaches to self-management. The group postulated that without personal identity reframing and acknowledgement of negative behaviours and cognitions no change would be made.

The involvement of GPs and other HCPs in referrals and recruitment lent credibility to the courses and was seen as integral to the success of any programme; additionally, medical staff involvement encouraged and motivated participants. However, the counter-argument that many chronic pain patients were disillusioned with their GP was not seen as detrimental to the self-management courses as GPs themselves were not running them.

the GP has assigned it, so it does make a difference.

FG2.5

All agreed that there was a need for aftercare or support. The Alcoholics Anonymous model was discussed, as were approaches to smoking cessation. Buddying and mentoring concepts were discussed and were thought to be helpful, but it was also thought that such schemes would be difficult to organise and maintain and that it would be difficult to provide support for the ‘buddies/mentors’. Longer-term thinking and support groups were discussed:

Education is an incredibly weak way to change behaviour . . . otherwise, you know, who would smoke?

FG2.4

Distraction is such a short-term thing . . . but developing interests and involvement which really compete for space in someone’s life.

FG2.1

it took about a year to get them to own it [the support group], it’s a very different dynamic with a key core of people.

FG2.5

Additional findings from the tutor/facilitator focus group

The tutor/facilitator focus group was characterised by the strength of the participants’ beliefs in the concept and process of the Expert Patients Programme. The terminology used and the phrases used from the Expert Patients Programme literature were common to all participants in this group:

We do not want cloned tutors . . . where you expand too big and you have got tutors for tutors sake, and the empathy is not there.

FG1.3

Tutors relayed stories relating to very practical considerations, such as not being able to access rooms because they were locked or centres were closed or had no disabled facilities. Participants had few criticisms of the Expert Patients Programme courses; however, they did suggest that providing more time for general discussion would be beneficial.

Statement of principal findings

Our data indicate that those who got the most out of the courses were those who were motivated to change at the outset and who became engaged in the process. The quality of the tutoring influenced participants’ perceptions of the courses. Factors that helped people were social support (family, new and old friends) and undertaking new activities that extended beyond the course; conversely, factors that did not help were a low, depressed or negative mood at the outset and a reluctance to alter behaviour, activities or lifestyle.

Strengths and weaknesses of the study

Our study focused on participants with chronic pain who had untaken self-management courses as well as tutors and experts. Each group’s perspectives were ‘triangulated’ with the perspectives of the other groups to assess responses and reactions. Discordant data were mainly obtained from the ‘experts’, who questioned the overall evidence for effectiveness for lay-led self-management in general, and, conversely, the tutors, who provided very positive narratives about the effectiveness of the courses that they had run.

One-third of our sample was Bangladeshi, reflecting our London population. We were able to identify some cultural issues, for example running gender-specific courses and not organising sessions that coincided with Friday afternoon prayers for men, but we were not able to draw any conclusions about cognitive ethnic differences from this study as it was not designed for this purpose. A further limitation was that we were not able to interview those who were invited to attend a self-management course but who chose not to go. Our sampling strategy generated a range of views that enabled us to consider issues relevant to the design and delivery of self-management programmes and some speculative data about the traits of people who seem to respond or not to self-management courses. The model adds to the existing literature by consolidating common findings about social interaction and activity and depression and behaviour change.

Strengths and weaknesses in relation to other studies

Conclusion

From our study the most important factor that appeared to be associated with better coping and improved quality of life was shifting individual focus away from pain to other activities. However, this was not easy as chronic pain patients were often very occupied with their pain, which superseded other activities. Acceptance and self-efficacy appeared to be important factors in the process of positive change and enhanced quality of life.

Selection of domains

The following measurement domains were included after brainstorming and consideration of the IMMPACT and MMICS recommendations:

• pain and pain-related disability

• pain intensity, interference

• pain catastrophising

• recovery

• depression and anxiety

• health-related quality of life

• perception of social support/social integration

• self-efficacy

• fear avoidance

• coping and acceptance

• patient satisfaction at follow-up

• health-care resource use.

Selection of candidate instruments

We were able to immediately select candidate measurement tools based on the IMMPACT and MMIC recommendations for the following domains (the tool chosen is given in brackets):

• pain intensity (numerical rating scale)

• pain interference (Brief Pain Inventory168)

• health-related quality of life (either SF-3642 or EQ-5D169)

• recovery (numerical rating scale)

• pain catastrophising (Pain Catastrophising Questionnaire170)

• patient satisfaction at follow-up (the Patient Global Impression of Change171).

For the domain of fear avoidance, the only systematic review in this area looked at fear avoidance and prognosis in back pain. 34 The two main contenders for measuring fear avoidance from this review were the Tampa Scale of Kinesiophobia172 and the Fear Avoidance Beliefs Questionnaire (FABQ). 173 The MMICS study31 reported a slight advantage for the FABQ if work-related fear is considered important. These two measures were presented at the expert consensus meeting.

Another immediate selection was based on the fact that there was only one suitable measure for the domain of pain acceptance: the CPAQ. 174

For disability, the MMICS recommendations included only instruments specific to low back pain. The IMMPACT recommendations included the Brief Pain Inventory. The Brief Pain Inventory, however, measures both pain and pain-related disability. Furthermore, it is a measure principally designed for measuring acute pain. For this study we were interested in making a difference to long-term outcomes and so a measure that measures pain and its impact on just 1 day was not appropriate for our current purpose. For this reason we carried out a further search to identify measures of chronic pain and chronic pain-related disability. We identified the CPG as a well-established measure with good clinimetric properties when used to measure chronic pain and chronic pain-related disability. 175 We added this measure to our potential pool of measures to assess in the pilot study.

For depression, the Delphi study produced conflicting opinions from experts (see Appendix 3). The main reasons given for endorsing measures were absence of confounding somatic items, brevity and clarity, and widespread use in research. The consensus was around the use of four depression measures. The HADS176 was thought to be a good candidate, although there was some concern about responsiveness. Reasons for using the Beck Depression Inventory (BDI)177 were around compatibility with other studies in our population. The Center for Epidemiologic Studies Depression (CES-D) scale178 allows international comparisons and would be a useful measure. Finally, the Patient Health Questionnaire for Depression and Anxiety (PHQ)-4179 was thought to be a useful screening tool. The overall recommendation based on the Delphi study was to use the HADS as both a baseline descriptor and an outcome measure as the simplest option. The HADS, CES-D scale and PHQ-4 were presented to the expert consensus meeting for selection of a single measure.

For self-efficacy it was not possible to make a selection of outcome measures based on the MMICS and IMMPACT guidelines, and we could not identify a recent relevant systematic review. We conducted an original systematic review for self-efficacy and identified 37 different self-efficacy questionnaires (see Appendix 3 for methodological details and a description of the results). We selected the two most commonly used measures (where > 10 articles had used these measures): (1) the Arthritis Self-Efficacy Scale (ASES)180 with its four variants (for chronic disease and shorter versions) and (2) the PSEQ. 181 We searched for further information on the clinimetric and psychometric properties of these instruments. We reviewed the clinimetric properties of the questionnaires (full details are provided in Appendix 3). Construct validity was evident for most measures. There was evidence for content validity for the ASES-20,180 ASES-11183 and PSEQ181 questionnaires. There was some evidence of test–retest reliability for the ASES-20,180 Chronic Disease Self-Efficacy Scale-33 (CDSES-33)182 and PSEQ. There was (limited) evidence for responsiveness for the ASES-20, CDSES-33 and PSEQ.

For perceived social support there were no recommendations from the MMICS or IMMPACT guidelines, nor were we able to identify a systematic review comparing measures. We carried out a literature search to identify candidate instruments (see Appendix 3 for the methods and results). There were 30 social support questionnaires identified in the first broad search and a further nine additional social support measures identified from specialist text books (see Appendix 3). At this stage we created a reduced list of six measures that focused on our target population and which included items representing the domains that we had previously identified, with special emphasis on social (re)integration, while considering the psychometric properties of the questionnaires (see Appendix 3). Two of these six questionnaires were most closely aligned with our aims of tapping into use of both health resources and social relationships for social support. However, we decided to delay a full clinimetric assessment of the questionnaires because we felt that they did not quite capture our remit in reference to social (re)engagement. It was decided to present these two measures to the focus group meeting prior to clinimetric assessment.

Consensus meeting

The consensus meeting included five members from the project team (TP, ST, DC, CM and KH) and four other participants: two clinical psychologists (one from Mile End Persistent Pain Service and one from Whitechapel Health Centre), one researcher from Patient Reported Outcomes Measurement Group at Oxford University and one chronic pain patient and lay representative.

The measurement domains discussed and the decisions made with regard to recommending tools are described in the following sections.

Social learning theory

Bandura’s10 model of social learning suggests that behaviour is learned through the process of observation of the environment and the social world to which we are exposed. Bandura identified the importance of learning from role models and peers. He suggested that imitation and social reactions to those imitations (positive and negative reinforcement) can influence future behaviour. Our systematic reviews and our qualitative research also reflected this, with group courses having better outcomes than individual and remote (web-based) courses and participants recalling other participants who coped well with their pain (and, conversely, those who coped particularly poorly). We decided that a group approach, in which participants could learn from each other and try techniques in the company of and with the support of others, was appropriate.

Theory of planned behaviour/reasoned action

The theory of planned behaviour and reasoned action suggests that a person’s behaviour is determined by an intention to perform the behaviour, which is based on an individual’s attitude towards the behaviour, his or her subjective norms and his or her readiness to perform the behaviour (i.e. whether or not the individual feels that he or she has control and the ability to perform the behaviour). 197

Typically, the more favourable the attitude and the subjective norm, and the greater the perceived control, the stronger the person’s intention to perform the behaviour in question should be. There are many coping behaviours that those with chronic pain adopt, and these behaviours may be beneficial or detrimental. Other behaviours that could be advantageous to pain management are not adopted at all. Raising an individual’s awareness of his or her own behaviour may change attitudes towards existing behaviours if he or she is exposed to, or becomes aware of, alternative behaviours that may be beneficial.

We wanted our self-management course to provide an environment that could promote positive attitudes, challenge and or change inappropriate subjective norms and empower and motivate people to realise that they have the ability to change.

Cognitive–behavioural concepts and acceptance

Cognitive–behavioural therapy evolved out of behaviour therapy and cognitive psychology research. Cognitive therapy has the premise that some behaviours are not simply influenced by rational thoughts but are also controlled by automatic thoughts. 191 In rational emotive therapy, developed by Ellis,198 it was proposed that once people are made aware of their thoughts and behaviours they can rationalise their emotions towards them and modify their behaviour accordingly. These two therapies informed the basis of CBT.

Cognitive–behavioural treatment focuses on individuals’ thoughts, images, beliefs and attitudes (cognitions) and how these impact on behaviour and emotions. The therapeutic process facilitates individual reflection on negative patterns of thinking or behaviour that may cause difficulties in living. Once these behaviours are addressed, this, in turn, is expected to change the way that individuals feel about their issues. In our study the focus was on pain-related behaviours, thoughts and emotions.

We used the fundamentals of modern CBT, which incorporates problem-solving (identifying problem behaviours), goal-setting and action planning, to underpin our approach to help individuals raise their consciousness about how they feel, think and behave towards their pain. Within the action planning stage we also considered ideas surrounding graded exposure used to overcome fear-avoidant behaviour. 173 Modern CBT also encompasses relaxation training, acceptance and commitment therapy, and mindfulness. 192 These are known as third-generation behavioural therapies, with the first- and second-generation therapies being traditional behaviour therapy and CBT. These third-generation therapies focus more on thoughts and feelings rather than behaviour. 199

Acceptance and commitment therapy has six core principles designed to help develop psychological flexibility:

1. cognitive defusion – learning methods to reduce the tendency to reify thoughts, images, emotions and memories (in other words when people ‘overvalue/prioritise’ thoughts and images and/or fit and fix them into misleading mental models)

2. acceptance – allowing thoughts to come and go without struggling with them

3. contact with the present moment – awareness of the here and now, experienced with openness, interest and receptiveness

4. observing the self – accessing a transcendent sense of self, a continuity of consciousness that is unchanging

5. values – discovering what is most important to one’s true self

6. committed action – setting goals according to values and carrying them out responsibly. 200

Pain catastrophising theory

Pain catastrophising describes a maladaptive thinking/cognitive style often seen in patients with chronic pain who have associated anxiety and depressive disorders. 201,202 It is characterised by the tendency to magnify the future threat of a predicted pain stimulus, regardless of whether or not it will occur. Chronic pain patients who catastrophise lose their ability to inhibit pain-related thoughts in anticipation of, during or following a painful encounter.

Figure 8 illustrates the relationships between the theories that we considered and our desired outcomes.

FIGURE 8.

Conceptual model.

Day 1: Living with and dealing with pain

• Aims: introduce aims of the course and concept of group work, and increase understanding of pain and reasons for it; introduce concept of acceptance and no cure; introduce the idea of recognising different moods and their effects on pain.

• Learning outcomes: know where pain comes from and why we have it and, in chronic pain, why it persists; the participants should be able to be able to describe why they have persistent pain to somebody else and be able to identify some of their own beliefs about their pain and identify some negative thoughts and behaviours, for example when is pain OK/manageable and, when it is at its worst, how this can relate to mood.

• Teaching methods: group introductions/presentation, facilitated discussion, watching a DVD of frequently asked questions, pain/mood exercise.

Day 2: Doing something about life with pain

• Aim: identify opportunities to change and understand when change is possible and when it is not.

• Learning outcomes: be able to break down issues into manageable chunks and set simple, measurable, achievable, realistic goals within a suitable time frame; be able to identify negative behaviours and thoughts and spot errors in thinking.

• Teaching methods: group discussion, self-reflection and practice.

Day 3: Communication and relationships

• Aim: promote effective utilisation of health-care services and improve communication skills.

• Learning outcomes: moderate expectations; communicate effectively.

• Teaching methods: group discussion and role play.

Follow-up at 2 weeks

• Aim: managing setbacks: knowing what to do when experiencing a setback or a flare-up.

• Learning outcomes: improve bonding, group cohesion and understanding of learning; embedding learning.

Recruitment

We aimed to recruit participants to the English-speaking courses by targeting patients with chronic pain who were registered with the local persistent pain service, physiotherapy department or two local general practices. We selected the community-based pain and physiotherapy services as they reported in excess of 1000 new chronic pain patients per year. We also involved general practices as clinicians estimated that around 10% of GP consultations in Tower Hamlets were for chronic pain. 204 One of the general practices was chosen because around 90% of patients were Sylheti speaking (in other local practices the equivalent figure was around 50%).

Randomisation

On receipt of the signed study consent form the English-speaking participants were randomised to either the intervention or the control group. An independent statistician based at the Pragmatic Clinical Trials Unit (PCTU) at Queen Mary University of London performed the randomisation using minimisation, with gender as a minimisation factor, in a ratio of 3 : 1 favouring the intervention. Participants were then telephoned by the study team and informed of their allocation. Those randomised to the intervention group were enrolled on the earliest available convenient course and asked to complete a baseline questionnaire prior to the course. The Bangladeshi participants were not randomised but booked on to courses directly.

The intervention

The structure of the course and the rationale for its design are described in Chapter 6.

We aimed to book courses on alternating days when possible, for example Monday, Wednesday and Friday, in easily accessible, familiar and local places, with a large room, kitchen and toilet facilities. The course content was modified to be culturally appropriate for the local Bangladeshi population but remained as similar in design and structure as possible to test whether or not our approach was directly transferable to this particular ethnic group. The modifications included single-sex courses and avoiding a clash with Friday prayers.

Control arm

Participants randomised to the control arm of the study received usual GP care plus best practice advice in the form of a 20-page booklet called The Pain Toolkit, developed by Frances Cole and Pete Moore. 207 We asked the control participants to refrain from attending any self-management courses for the duration of the study but encouraged them to contact their local Expert Patients Programme after the COPERS follow-up period had elapsed.

Data collection

We collected both quantitative and qualitative data as part of the feasibility study. Quantitative data included questionnaire-based self-reported outcomes and participant feedback. Qualitative data included observational field notes from the courses, free response questions from participant feedback forms, a focus group with team members and COPERS facilitators, and in-depth interviews with some of the participants.

Data analysis

Recruitment sources and participant identification

Recruitment and delivery of the intervention took place over 9 months (February to October 2010). We identified 526 potential participants, of whom 32% (n = 167) expressed an initial interest in participating. Of these, 42% (n = 70) were allocated a study ID number and sent a questionnaire and consent form to participate in the English-speaking study, and 24% (n = 40) were enrolled into the Sylheti-speaking feasibility study, representing an overall 21% recruitment rate (Table 24).

Reasons for declining to take part were:

• a preference for ‘physical’ treatment such as physiotherapy or injections

• inability to get time off work, especially for teachers

• inability to attend because of childcare issues

• language barriers (English fluency not sufficient to take part in a group process).

Consenting participants were randomised using minimisation with a 60 : 40 female/male split expected and a 3 : 1 ratio in favour of the intervention. However, we abandoned allocation to the control arm as the initial randomisation sequence was top heavy with control participants and did not yield sufficient participants in the intervention arm to fill the first prebooked course. We continued to run the randomisation sequence and eventually the randomisation sequence was as expected overall. We modified our randomisation strategy for the main study to include randomly permuted blocks to ensure that a steady flow of participants was randomised to the active intervention.

We ran nine courses in total starting in the last week of March 2010 and continuing to mid-July 2010, with around seven participants on each course. There were six mixed-gender English-speaking courses, one Sylheti-speaking male course and three Sylheti-speaking female courses. Fifty-six participants [n = 30/56 (54%) female participants] were booked onto the English-speaking courses. Fourteen participants (25%) did not attend any sessions. Four participants attended day 1 only and 38 attended ≥ 2 days (Table 25). Four male participants attended the Sylheti-speaking course, with three attending ≥ 3 days. The first Sylheti-speaking female course had 20 participants booked but four (20%) did not attend any sessions. Of the remaining 16 participants all attended at least 2 days. The second Sylheti-speaking female course had 14 participants booked but eight (57%) did not attend any sessions.

There were 10 participants on the English-speaking courses who were not from a white British/Irish ethnic group and who spoke English with varied fluency. They coped with the course but those who were not very fluent did struggle to maintain full involvement with the discussions, often preferring to listen rather than talk.

There were more female than male participants interested in being part of the study (75 females vs. 34 males), particularly for the Sylheti-speaking group (33 females vs. 7 males) (Table 26). The mean number of participants attending each English-speaking course was seven and the mean number of participants attending each Sylheti-speaking courses was nine. The median age range of participants was 41–50 years (see Table 26).

Questionnaire for self-reported outcomes

Of 56 potential participants allocated a study ID number and booked on a course, 48 (86%) returned usable baseline questionnaire and study consent forms. The response rate at 3 months was 52% (n = 25). Of 41 Bangladeshi participants booked onto courses, 18 (44%) returned their baseline questionnaire and three (7%) retuned their follow-up questionnaire.

The participants found the questionnaire extremely difficult to complete because it was long and repetitive; this was particularly so for the Bangladeshi group. We provided assistance and extra time prior to the courses being run to help participants complete the questionnaire but participants and the assistants found the task arduous and slow. Table 27 shows that, despite providing the questionnaire in both English and standard Bengali to the Bangladeshi group, more English questionnaires were returned. Our qualitative work explained that help to complete the questionnaire was often given by participants’ children who were fluent in English and but who did not read Bengali.

There was a high proportion of missing data in the participant questionnaires, meaning that the data should be interpreted with caution (Table 28). At baseline, differences in employment were seen between the groups, with few of the Bangladeshi group being employed and the majority of responders being female home/family carers. The other between-group difference related to education: 7/14 (50%) of the Bangladeshi participants who supplied data on age completing education had either no formal education or were educated only to the age of 11 years, compared with 2% of the English-speaking group.

The mean scores for pain extent, intensity, depression and social integration were similar between the groups at baseline (Table 29). EQ-5D, anxiety, self-efficacy and coping showed differences between the groups, indicating worse states in the Bangladeshi group. This is consistent with other evidence from Tower Hamlets. 153

At follow-up, overall, almost all mean outcomes showed a tendency towards improvement but, when confined to those for whom we had longitudinal data, as anticipated, no statistically significant changes in scores were seen in these pilot data (see Table 29).

Recruitment and related issues are summarised in Table 30. These issues were identified from feedback from the facilitators and from observations by the study team. Detailed feedback data are provided in Appendix 5.

Table 31 shows the number of respondents recording each satisfaction score for each question. Overall, the average response rate was 72%.The mean overall satisfaction score was high (4.3). There was little variation in overall scores; however, the ranges indicate that there was less satisfaction with time allocated for discussion, socialising and each topic. At least four people found the ‘taster’ unsatisfactory.

Sample

We approached 18 course participants who were potentially exposed to the intervention for interview; five declined [reasons included work commitments (n = 2); having an operation; bereavement; and ‘no time’] and we therefore interviewed 13 participants, including four males. Three interviews were conducted in Sylheti and the remainder were conducted in English. Four researchers conducted the interviews; all were members of the study team.

Themes

We organised the data into 12 themes: clarity of aims, motivation, positive aspects, negative aspects, learning, social interaction, effect of others, repercussions/outcomes post course, suggested changes, facilitation, ‘buddying’, course material; these themes are detailed in Appendix 6. There was no need to conduct second-order analysis to interpret the data in greater depth as the aim of the evaluation was to find out what modifications needed to be carried out to the intervention programme.

In summary, participants wanted not only the aims of course made clearer but also the rationale for each session. Participants greatly valued the social interaction and wanted more time for informal interaction. They valued good facilitation, which was seen as good group control, that is, managing difficult people and situations; this had implications for facilitator training. Different participants preferred different sessions but they really liked the relaxation sessions and requested audio-recordings of the scripts used. ‘Buddying’ was wanted by some and not others and those interviewed strongly recommended that it be left up to individuals in the groups to identify other participants to ‘buddy up’ with if they so wished.

Study team meeting for modifications and changes

Two study team meeting sessions were conducted because of the number of data, one for facilitator/observer feedback and one for participant feedback. The recommendations from the study team meetings are provided in Table 32.

Overall summary of the results of the feasibility study

The process evaluation of the feasibility study gave clear indications of areas for improvement for the main RCT. The key areas are summarised as follows:

• The feasibility study demonstrated that it is feasible and acceptable to translate and deliver the intervention in another language; however, it also demonstrated that it was not feasible to collect sufficient outcome data from the non-English-speaking Bangladeshi population. The main trial should be conducted only among people fluent in English (but the results could probably be extrapolated to non-English-speaking people).

• A credible control is required to encourage participants to enter into a definitive trial. Usual care and a pain education booklet were not sufficient to attract participants.

• More time was required to secure recruitment sites. A phased approach would assist in the planning of recruitment to courses, along with recruitment monitoring against an expected rate per source to flag up problems early in the study.

• A more specific search strategy needed to be developed for general practice medical record systems to find suitable participants. GPs do not have time to look through lists of hundreds of patients. General practices may need assistance from the study team in performing the searches and preparing invitation packs and reminder letters.

• At least 8 weeks should be allowed between posting out participant invitation letters and the start of a local course to allow time for potential participants to complete and return the baseline questionnaire and complete the enrolment process.

• A block system for randomisation would ensure that groups of participants are randomised in sufficient numbers to provide enough participants for a course.

• The questionnaire was too long.

• Good facilitators, and hence good facilitator training, were absolutely key to the successful delivery of the intervention and the retention and satisfaction of course participants.

Consideration of our outcome measures and choice of a primary outcome measure for the main trial

It is usual in trials of treatment for chronic painful disorders to see a substantial improvement between baseline and follow-up in the control arm. 210,211 This is because of a combination of two factors. First, the natural history of the disease is to improve as people will be more likely to be motivated to join a trial if they are in a period of relatively more severe pain; chronic musculoskeletal pain tends to wax and wane over time. Second, one would expect to see some regression to the mean in the results for any outcome measure, irrespective of any change in the underlying disorder. That the EQ-5D169 did not identify this expected change in our pilot study gave the study team serious concerns that it would not be sufficiently responsive to any change in participants’ underlying disorder. The team therefore considered that an alternative primary outcome was needed. The closest measure in the portfolio that we had evaluated (see Chapter 5) was the pain-related disability component of the CPG. This is a disorder-specific outcome that is directly related to overall quality of life. It has much greater disorder-specific clinical relevance than other measures in our portfolio. For this reason the study team chose to use the pain-related disability subscale of the CPG as the primary outcome in the main trial.

Results in the context of other research

Past research on self-management courses has shown that self-management can have small effects on self-efficacy14 and that group approaches can benefit participants. The group process encourages active involvement, cheering successes and offering and receiving advice and support. 49 Our findings support the ‘power’ of the group process as an active part of learning and behaviour change. Self-efficacy seems to enable and contribute to and/or evolve out of the group process. Participants could be empowered either by those who were seemingly worse off than themselves or by those who appeared to be proactive and/or worthy role models. We also found in our systematic review (see Chapter 2) that group-based self-management programmes provided more beneficial effects for participants than remote or individually delivered programmes.

Providing a meaning for learning about the complex interactions between the body, pain, mind and emotions, be it internal (via self) or external (via others), was observed in participants who went on to illustrate examples of better coping strategies. Heightened self-awareness potentially could lead to hypervigilance or, conversely, self-awareness could promote coping strategies that related to personal life situations and were therefore more meaningful to the participant.

Cognitive–behavioural approaches for pain management have been systematically reviewed48 and found to be effective for short-term pain relief. However, it is still unclear which types of cognitive approaches or which elements of the cognitive approaches are more effective than others. Based on our interview data we postulate that self-reflection on unhelpful behaviours and goal-setting have most impact on change. These elements of CBT are often linked with plans to carry out activities in the future; if these plans are carried out they can have an impact on physical and mental health. Other systematic reviews conclude that exercise (any type) is beneficial for persistent non-specific pain. 33 In the COPERS programme we encouraged activity as opposed to exercise to avoid raising existing negative cognitions about exercise, but ironically the posture and stretch sessions were very well received despite us trying to avoid the concept of exercise as much as we could.

Strengths and limitations

We specifically sought to deliver the intervention in a language other than English to a minority ethnic group. This research had its origins in a collaboration with Social Action for Health, a Tower Hamlets-based community organisation that seeks to promote health and well-being in the Bengali community of Tower Hamlets. Social Action for Health had identified the management of chronic musculoskeletal pain as a community priority. We had also demonstrated the higher prevalence of chronic musculoskeletal pain in first-generation Bengali migrants living in Tower Hamlets153 and were aware of the issues through our clinical work. Developing a programme to help this group was therefore a priority for both us and our patient partners.

The large Bengali community in Tower Hamlets meant that it was plausible to develop and a deliver an intervention tailored to this group. This information is locally important. However, we cannot necessarily extrapolate from these findings to other minority ethnic groups or establish how practical it might be to deliver such a programme in areas that do not have the same dominant minority ethnic group. Nevertheless, we have for the first time demonstrated that such an intervention is deliverable to such groups. We had no a priori expectation that this experience would not be transferable to other ethnic groups living in the UK.

Our evaluation was designed to gather information about the COPERS course and its content to determine whether or not any improvements needed to be made and how to improve delivery. We systematically collected data from participants, observers and facilitators. The data were gathered and analysed by different team members to try and avoid bias in interpretation; however, some bias may have occurred. We demonstrated that it was feasible to deliver the COPERS course in Sylheti, that the courses were acceptable to participants and that there was a relatively low rate of attrition. However, this pilot showed that it would be very difficult to determine the effectiveness of the Sylheti language course in a definitive effectiveness trial because most participants were unwilling or unable to complete questionnaires administered by post. An alternative would be to formally evaluate English-language courses for those fluent in English of all ethnicities and extrapolate the data to non-English-language courses. Delivering courses in multiple languages poses an issue in ethnically diverse areas because there is no one dominant language group; however, it is feasible in areas with predominantly one ethnic group, as we have shown. There is also the additional problem that some of our outcome measures have not been validated in other languages.

Recommended changes

The changes required to the COPERS course were not major changes. The suggestions for change could be categorised into three main areas: changes to the course content, changes to processes and changes to facilitator training. We believed that many of the issues that the participants flagged would be avoided by better facilitator training. The facilitator training course was therefore modified and run closer to the start of the courses; when this was not possible we decided to provide ‘top-up’ training to refamiliarise facilitators with the techniques and course material. Facilitators who wanted top-up training would be able to observe other courses and receive individual coaching on a one-to-one basis from the study team (in practice, two facilitators out of 39 (5%) requested additional training and observed a course). Additionally, we would endeavour to always pair a new facilitator with an experienced COPERS facilitator. We also needed to provide a more credible control. Relaxation does not have medium- to long-term effects on pain severity48 but our participants really enjoyed this element of the course. Relaxation can be carried out in groups or alone and so we decided to use this as part of our control as it has limited clinical effects and does not involve any interaction with others.

The use of observer notes for each course was useful to check for ‘treatment drift’ and fidelity. The observers also helped to debrief the facilitators after each day. An audio-recording of each course was used to check quality and ‘treatment drift’ in the main trial.

The evidence

Group-delivered courses that had HCP input had better outcomes than other types of courses and longer courses did not appear to provide additional benefits over shorter courses. We found mixed evidence for the effectiveness of different proposed components of the intervention package. Courses with a psychological component, however, had more evidence of beneficial effects than those without psychological components and increasing the number of components did not appear to improve course effectiveness. We proposed that this indicated that other factors related to the group process, such as socialisation, may be as important as the content in the success of self-management courses.

Our qualitative work with chronic pain patients who had participated in courses also indicated that they gained benefit from the social aspects of courses and learning from others, as well as the need to focus their attention on things other than pain. People who had accepted their pain and had personal coping strategies that enabled them to lead fulfilling lives were most positive about their experience of pain management courses.

The psychological theories that we reviewed supported our qualitative study findings and underpinned the structure and content of the course. The theory underpinning the decisions to include a variety of sessions and behavioural change techniques worked well within the group learning environment; this has also shown to be effective in other studies of chronic pain. 48,53

Informed by the available theoretical and empirical evidence we developed a brief group intervention to be delivered jointly by a HCP and a lay expert.

The course

The course ran over 3 days in 1 week with a top-up session 2 weeks later. The target attendance was 10–12 participants.

The learning sequence that we adopted enabled each session to build on the previous session and in many cases the participants were able to predict the next phase of learning in advance. The learning and flow of information was pitched at a level at which participants could follow the structure and understand the content. This was shown in the daily feedback sheets that asked participants what they had learned: their learning mapped well onto the learning objectives.

In our feasibility study we demonstrated that the course was feasible, acceptable to participants and deliverable in a NHS context. Participants were positive about the course and the content appeared to be meaningful to them. Attrition was very low over the 3 main days: participants attended on average 85% of the course. Attrition has been reported as an issue in other trials. In one such trial of an Expert Patients Programme run over a 6-week period (intervention arm n = 313), one-quarter of participants failed to attend any session, only 60% were considered to have completed the course (attending at least four of the six sessions) and only one-third attended all six sessions. 15

We tested the feasibility of the intervention in a non-English-speaking Bangladeshi population. In Tower Hamlets in east London 32% of the community are of Bangladeshi origin,212 with some not speaking English or not speaking English well. The prevalence of chronic pain in the Bangladeshi population (arrived in UK after the age of 14 years) is higher than that in the white British and British Bangladeshi population living in Tower Hamlets, at around 75% compared with around 50%. 153 We demonstrated that the intervention can be delivered in Sylheti to a group who may find services hard to access. Although we could not collect validated participant-reported outcomes, our evidence from interviews and feedback from the courses indicated that the participants valued the courses and appreciated the effort made to accommodate their needs, for example running single-sex courses. From a national perspective the ethnic distribution varies, supporting our decision to run the trial in two areas. The non-white population in London makes up about 40% of the total population, whereas in the West Midlands it is around 17%. 212 In the census 8% report that English is not their main language and of these 21% do not speak English well (this equates to 1.7% of the total English and Welsh population). 213 Our experience running the non-English speaking courses was that they were feasible and well received but it was hard to recruit sufficient people onto the courses to make them viable and hard to collect outcome data, which may affect the analysis and sample size required for the main trial (attrition and incomplete data).

The facilitation and group process may have optimised the learning process as discussion embedded participant thinking. All of the course evaluation material suggested that good facilitation skills were crucial for positive participant perception. Comprehensive facilitator training was essential for courses to run effectively. We found that it was possible to train both laypeople and non-psychologists to facilitate the courses and deliver the cognitive–behavioural approach that we had developed. The facilitator training included four components: facilitation skills training, course content familiarisation and practice, trial management protocols and evaluation.

Delivery styles varied, indicating a need to embed fidelity assessment from the outset to measure both adherence and the competence of those delivering the intervention. Measuring whether or not an intervention has been delivered as efficiently and effectively as possible has been advocated by others;214,215 however, published reports of fidelity assessment, or the methods used to perform it, are sparse.

We found that the course withstood the inexperience of our facilitators in delivering an entirely new course; the content in terms of the discussions, information and handouts was robust enough to make an impression regardless of delivery style. We recommended that inexperienced personnel be partnered with experienced personnel in the main trial.

Recruitment and participants

Recruitment to the feasibility study was challenging; the conversion rate from invitation to course attendance was lower than we had hoped (approximately 13% of those invited) but was in line with other studies of this nature recruiting patients from primary care with chronic conditions. 53,83,98 Nevertheless, we had sufficient interest from patients to run nine courses and this demonstrated a demand for learning about non-pharmacological approaches to managing pain. Procedures for future recruitment need to be enhanced by increasing the number of invitations sent and devising and testing a comprehensive and inclusive electronic search strategy for patients with chronic pain. To address this need we developed and tested an electronic patient record search strategy using repeat prescription data for chronic pain-related medication, musculoskeletal Read codes and contact within the last 3 months. 206

In our feasibility study participants reported poor quality of life, low self-efficacy to manage their chronic pain, relatively high levels of social isolation, poor coping skills and a tendency towards anxiety and depression. This is consistent with the findings of a European survey carried out in 2005216 that asked respondents about the impact that pain had on their daily lives. In total, 32% reported that they were no longer able to work outside the home or attend social activities and 21% reported having a diagnosis of depression as a result of having chronic pain (24% in the UK); thus, our secondary outcome measures need to reflect these health and social states.

We noted the levels of depression and considered whether depression should be addressed with patients prior to, or in conjunction with, attending these types of courses.

The control

The pilot study indicated that the control arm needed to have more credibility; there was little motivation for people to join the trial if they perceived that they would receive no added benefit, that is, usual care. Our systematic review illustrated that mind–body approaches (mostly relaxation) may have some short-term beneficial outcomes but did not produce medium- or long-term benefits and this has also been shown by other reviews of relaxation. 217,218 Our qualitative work indicated that participants really enjoyed this part of the course that they attended and so we decided that if there was little evidence for effectiveness but positive appeal for relaxation this may be an incentive for recruiting people into the trial who might otherwise decline.

Strengths and weaknesses

We designed a novel approach to analyse the effectiveness of courses by content and characteristics by looking at interventions with and without prespecified components and characteristics. Although many studies were available, some subgroups had few data available and the conclusions drawn were tentative. However, looking at all of the studies, and the data indicating beneficial outcomes, we were able to identify characteristics and components that seemed to elicit better effects than others. This allowed us to translate the findings directly into our intervention design, enabling us to put together an evidence-based intervention at a level of detail that included content and delivery style.

A byproduct of our predictor, mediator and moderator study was that we found that there was no consensus about methodological standards for mediator studies. We carried out an additional expert consensus study that informed and enhanced our systematic review. 135 Additionally, we also tested and piloted an electronic search strategy for identifying chronic pain patients. 206 The strategy that we devised is applicable to other conditions but would need further testing. We involved patients in the selection of outcome measures but, in retrospect, as was pointed out by one of our reviewers at the end of the entire programme, we would have liked more patient involvement in this process. Following the recommendations in MRC guidance,28 as part of the feasibility phase of the project we started to design and develop a methodology for rigorously and systematically assessing and measuring the fidelity of our intervention delivery from the outset,207 so that the assessment was embedded into the training, delivery and implementation of the intervention. Importantly, we were able to demonstrate that we could deliver the course in Sylheti for the Bangladeshi community of Tower Hamlets. We were not able to include this group in the main trial; nevertheless, showing that the course can be simply translated and delivered to another ethnic group indicates that our findings are likely to be generalisable to other non-English-speaking groups in the UK. We suggest that running a parallel non-randomised arm of the study for groups who may find services hard to access may be a sustainable and credible option for researchers to demonstrate the generalisability of an intervention for groups who cannot be included in a randomised comparison.

Ethical approval

The study was approved by the Cambridgeshire Ethics Committee on 18 March 2011 (reference no. 11/EE/046). Additional ethical approval was obtained to give participants unconditional £5 high-street shop vouchers with the 6- and 12-month questionnaires and, on the advice of the TSC, to add questions about non-NHS, pain-related personal costs at 12 months (7 October 2011). We sought further approval for additional questions at 12 months enquiring about attendance at other similar or pain-related courses, new hobbies or activities undertaken and any psychological help participants may have sought during the trial period (August 2011–July 2012).

Recruitment

Inclusion criteria

• People aged ≥ 18 years with musculoskeletal pain of > 3 months’ duration.

• Available to participate in an intervention course if randomised to the intervention arm.

• The IASP17 defines chronic pain as that which has persisted beyond normal tissue healing time, usually interpreted as 3 months. Causes/types of pain included, but were not restricted to, OA, back pain, chronic widespread pain and fibromyalgia.

Exclusion criteria

• Unable to give informed consent.

• Not fluent in English.

• Life expectancy of < 6 months.

• Presence of chronic pain arising from active malignant disease.

• Presence of inflammatory arthritis such as rheumatoid arthritis.

• Presence of a serious comorbidity that was more disabling than chronic pain.

• Poorly controlled serious mental health illness that would make it difficult to participate in the intervention.

• Misusing substances to an extent that would make it difficult to participate in the intervention.

We excluded people living with chronic pain arising from malignant disease because this requires specific management. However, chronic pain in people living with or beyond cancer may arise from non-malignant causes and such patients were eligible for our study.

We restricted the study to those who were fluent in spoken English for practical reasons:

• The interactive, participatory nature of the intervention meant that it was unsuitable for delivery through an interpreter or advocate.

• Our systematic review (see Chapter 3) identified that lack of fluency in the language of the programme was associated with lack of clinical effect.

• The only language other than English that was sufficiently common to consider running courses in the study recruitment areas was Sylheti. We piloted delivering the intervention in Sylheti and found that it was not feasible to include a Sylheti language stream in the evaluation of the intervention (see Chapter 7).

• The validity and reliability of the outcome measures translated into languages other than English have not always been established.

People with serious mental health problems or substance abuse problems were able to join the study. They were excluded only if their GP, or other screening clinician, judged that their current problems were so severe that they were likely to cause difficulties within the group sessions or if patients themselves felt that this might be the case.

Randomisation

The randomisation ratio between the intervention and control arms was 1.33 : 1 (see Sample size for an explanation). Randomisation was overseen and implemented by the PCTU at Queen Mary University of London. Participants were randomised after the study team had received signed informed consent and the completed baseline questionnaire. Strict allocation concealment was maintained through the use of Sheffield University’s Clinical Trials Research Unit’s web-based, real-time, randomisation programme,222 which was accessed by the study team while on the telephone to participants to avoid having to call them again. Randomisation was performed using random permuted blocks stratified by site of recruitment, with randomly varying block lengths of seven and 14. All randomisations were logged using an online audit trail of use according to each user ID.

Interventions

Quality control and fidelity of intervention delivery

To promote fidelity of intervention delivery the first day of each course was observed by a member of the study team experienced in delivering the intervention. In the case of first-time facilitators or facilitators who expressed a lack of confidence the course was observed in its entirety. When necessary, immediate feedback was given to facilitators during breaks in the course and at the end of each day, enabling them to modify and improve their performance.

Our assessment of the actual fidelity of delivery of the intervention is described in detail elsewhere223 and in Chapter 10. In summary, every course was audio-recorded. After all the courses had been delivered a random selection of audio-taped sessions was chosen for formal evaluation of fidelity. The evaluators used a checklist to record adherence to structure and content and facilitator competence.

Outcome measures

The rationale for our final choice of outcome measures is described in detail in Chapters 5 and 7. All selected outcome measures were used with approval and licence when required.

Schedule of data collection

Participants received follow-up questionnaires by post at 12 weeks and 6 and 12 months post randomisation. Based on our systematic review225 (see Chapter 3), we hypothesised that improving self-efficacy was the principal mediator of the expected benefits from a self-management intervention. Improving self-efficacy was a key aim for the COPERS intervention. To ensure that we had evidence for any effect that our intervention had on this hypothesised mediator we sent participants the PSEQ at 12 weeks.

At baseline and 6 and 12 months we collected all of our primary and secondary outcomes from participants. We sent a postal reminder after 2 weeks if no response was received. If there was still no response we collected the primary clinical outcome and the EQ-5D data needed for the health economic analysis by telephone. We sent participants a £5 high-street shop voucher that was redeemable in multiple stores with their 6- and 12-month questionnaires to encourage a response. The vouchers were given on a non-conditional basis. This expression of appreciation has been shown to improve questionnaire return rates, with a systematic review reporting that ‘the odds of response were more than doubled when a monetary incentive was used (odds ratio 2.02; 95% confidence interval 1.79 to 2.27) and almost doubled when incentives were not conditional on response (1.71; 1.29 to 2.26)’. 226

At 12 months we collected health-care activity data from participants’ medical records for the 12 months post randomisation and the 3 months prior to randomisation for our health economic analyses.

Figure 13 illustrates the data collected at each time point and how the data were collected.

FIGURE 13.

Follow-up data collection.

Analysis of drug data

Medications used over a 15-month period were collected from participants’ medical records. We extracted drug names and strength used plus quantity and dates prescribed. We used the Prescription Cost Analysis (PCA) database227 to attach a cost to each individual preparation used. Using the World Health Organization (WHO) DDD228 for each drug we generated the number of days of each medication used organised by British National Formulary (BNF) chapter and subchapter. 229 The WHO does not provide DDDs for topical NSAIDs or rubefaciants and so we used a previously published report to define these. 56

The total DDD consumed for each drug was defined as:

The total DDD for a group of medications (e.g. the total DDD for opioids) was the sum of the total DDD for each drug within that medication group (e.g. each drug that was considered an opioid). For example, if there were three opioid drugs (drugs A, B and C), the total DDD_opioid was defined as:

The DDD (used in the denominator of the calculation for the total DDD) was determined in the first instance using the WHO register, then by precedent in other trials56,230 and then by clinician consensus. For compound drugs, for example co-codamol, we separated out the components (paracetamol and codeine) and worked out the DDD for each component drug.

Adverse event reporting

An adverse event was defined as any untoward physiological or psychological occurrence in a subject to whom the intervention or control intervention was administered.

We defined serious adverse events as:

• death

• life-threatening events

• events necessitating hospitalisation

• events resulting in persistent or significant disability or incapacity

• events otherwise considered medically significant by the chief investigators.

Minor adverse incidents, for example a participant being tearful and distressed during a session, were logged and fed back to the study team by the end of the course.

If a course participant experienced an adverse event the facilitators were required to notify the study manager and, if necessary, the chief investigators immediately. The study team would then inform the Research Ethics Committee and the sponsor if, in the opinion of the chief investigators, the event was serious and (1) related to the COPERS intervention or (2) unexpected and possibly related to the COPERS intervention.

Details of any adverse events were recorded and stored in the trial master file.

Sample size

The sample size calculation was based on detecting a SMD of 0.3 in pain-related disability between the intervention group and the control group, with 80% power and at a 5% significance level. This effect size was commensurate with the largest change seen in a systematic review of Expert Patients Programmes14 and also with the sort of change effected by interventions for other chronic pain syndromes, such as low back pain, on any continuous outcome measure. An initial sample size calculation indicated that we would require data on 350 subjects (175 in each group). We inflated the sample size because of the possibility of a ‘clustering’ effect in the group intervention arm and chose a ratio between intervention participants and control participants to maximise statistical efficiency. 231 Using an intracluster correlation coefficient (ICC) of 0.1 and assuming an average of nine individuals in each group (from an average of 13 individuals recruited to each group, with a 30% loss to follow-up) meant that 480 individuals were needed, with 275 in the intervention group and 205 in control the control group (ratio of 1.33 : 1 between the intervention group and the control group). Allowing for this very conservative estimated 30% loss to follow-up we originally sought to randomise 685 participants (391 intervention participants and 294 control participants).

Previous research221 and electronic record searches in the pilot study indicated that around 5% of adults on GP registers consult with chronic musculoskeletal pain. Based on our pilot study we estimated that 10% of these may be interested in participating in the trial and around half of these would be recruited into the trial.

Around 80% of the population consists of adults aged > 18 years [www.statistics.gov.uk (accessed 19 April 2016)]. This meant that to recruit 685 participants we needed a population base of around 342,000 registered patients, assuming that 80% (approximately 274,000) would be adults. Of these, 5% would be identified from searches as having chronic pain (n = 13,700), of whom around 10% might express an interest in participating in the study (n = 1370). Half of these might be recruited and enrolled (n = 685). Using an average total practice list size of 7000, this equates to around 49 practices. We estimated that we would recruit between 12 and 14 patients per practice.

Blinding and protection from bias

All baseline data were collected by self-completed questionnaire prior to randomisation. After allocation, however, it was impossible to blind participants to treatment allocation because of the nature of the intervention. General practices and clinicians were informed of their patients’ enrolment into the trial but not their treatment allocation, although participants were free to divulge this information to their clinicians. We felt that this information in itself would have little impact on their care. Outcome data were collected using a participant-completed postal questionnaire. Complete blinding of the research team was not possible because of the need to manage course attendance and the unbalanced randomisation. Whenever possible we blinded the research team to treatment allocation during outcome data collection, for example, for those who did not return their questionnaire after the second reminder, primary outcome data and the EQ-5D instrument were collected over the telephone by research personnel blind to treatment allocation using a set script, asking participants not to divulge their allocation prior to collecting their data. To achieve this blinding London staff contacted Midlands patients and vice versa. Patient record data, such as GP consultations, hospital admissions and drug usage, were extracted by personnel blind to treatment allocation. All data were coded according to participants’ study ID and notes were anonymised as much as possible by clinical staff at the practices.

Data management

All data were managed in line with Queen Mary University of London’s PCTU standard operating procedures and were subject to review by PCTU staff audit and the Data Monitoring and Ethics Committee. All electronic participant data were stored in encrypted and/or password-protected files in a secure environment. A database was designed to manage the data input to ensure consistency of practice and coding, with a built-in audit trail that enabled us to track all entries and changes. Regular audit and double-checking of all primary outcome data was conducted to ensure the accuracy of data entry and a further random 10% of all data were double entered.

Statistical analysis

Our statistical analysis plan has been published elsewhere. 232 We briefly summarise the analytical approach used here.

The main analysis for each outcome followed ITT principles, meaning that all participants with a recorded outcome were included in the analysis233 and were analysed according to the treatment group to which they were randomised.

All analyses accounted for clustering by course in the intervention arm,234,235 Participants in the control arm (who did not attend courses) acted as their own cluster (i.e. we analysed the data as if each participant in the control arm belonged to a ‘course’ in which he or she was the only member).

Site of recruitment (London or Midlands), age, gender and HADS depression score at baseline were included as covariates in each analysis. 236–239 Additionally, for continuous outcomes (CPG pain-related disability, CPG pain intensity, PSEQ, HADS anxiety, HADS depression, CPAQ, heiQ and EQ-5D), the outcome measured at baseline was included in the analysis. Continuous covariates (age, HADS depression score at baseline) were assumed to have a linear relationship with the outcome.

Ethical considerations

There were few ethical concerns with this study. To maintain patient confidentiality only the clinical staff and primary care research network were able to search clinical records and invite suitable patients to participate. Research staff provided guidance and advice when needed and when confidentiality could be maintained. Only potential participants who had provided their contact details to the research team were approached. Patients received one reminder letter if they did not respond.

The risks to the participants in this study were low; however, the study team was aware that the course could trigger emotional reactions. We therefore ensured that the facilitator training course trained facilitators in distress management. Each course had two facilitators so if any participant became unduly distressed he or she was helped by a facilitator who, if necessary, and with the participant’s agreement, withdrew the participant from the group and helped him or her until he or she was ready to return to the group, go home or seek further help from a more suitable HCP. Under no circumstances was a participant left alone while distressed. If the facilitators felt that a participant was a danger to him- or herself or others, they sought permission to contact the participant’s GP or take him or her to an emergency department.

We ensured that another member of the study team was always available by mobile phone for the duration of any course should any emergency advice be needed.

Microcosting of the intervention

A microcosting of the self-management course for primary care patients with chronic musculoskeletal pain included a bottom-up construction of the costs associated with setting up and delivering the programme. The course running costs included salaries, room rental, course materials, facilitators’ travel expenses and administration costs. The cost of training facilitators included salaries and travel expenses (for both trainers and facilitators), room rental, course materials and administration costs. Courses were run in multiuse settings and the same daily rate was used for all venues. Trainers and facilitators were paid a fixed fee per session. Assumptions used in the microcosting are summarised in Appendix 6. The costs of the intervention were estimated as a cost per course and a cost per participant. Costs associated with usual care included the costs of the pain education booklet and the relaxation CD.

Data analysis

Data analyses were conducted using Microsoft Excel^® 2012 (Microsoft Corporation, Redmond, WA, USA) and Stata 12.1. 254 The base-case cost-effectiveness analysis was conducted for the ITT population on the imputed data set using a multilevel model (MLM). Sensitivity analyses were conducted using a generalised linear model (GLM) and a seemingly unrelated regression (SUR) model. The secondary analyses were performed on the non-imputed data set using a GLM, SUR and MLM. Per-protocol analysis was conducted on the imputed data set using a MLM. Subgroup analyses included participants with different levels of compliance and exposure to treatment. Additional subgroup analysis excluded participants with high service use costs (top 5%). All subgroup analyses were conducted on the imputed data set using a MLM. The intervention and follow-up period lasted for 12 months only so a discount factor of 1 was applied to the costs and benefits following standard discounting practice. 255

Setting and data collection

The research team identified seven of the 24 course components (or sessions) that they considered to be the most important in terms of effecting participant behaviour change (Table 34). These components focused on participant education and theoretically driven behaviour change techniques and strategies in contrast to other components, which encouraged social interaction, relaxation and postural awareness. All of the courses were audio-recorded and intervention integrity was assessed by examination of the recordings of the components listed in Table 35.

Developing the intervention integrity measures

We used the monitoring and assessment tools from three previously published trials to inform the development of the COPERS measures. 53,275,276 The learning outcomes outlined in the COPERS facilitator training course manual helped to design a provisional set of criteria to measure adherence and competence. To develop the measures we used a two-stage pilot testing process.

Our adherence measure was designed to assess the delivery of key elements of each component as described in the facilitators’ manual. The generic competence measure was designed to determine the extent to which the facilitators created an environment in which participants could share their experiences and learn new skills. An overall impression score was designed to reflect the extent to which the aims and objectives of the component were achieved and how the material was received by the group.

We tested a variety of scoring systems for adherence, competence and the overall impression score. We found that each method of assessment had its own strengths and weaknesses. Numerical scales and Likert scales seemed, intuitively, to be more suitable for measuring degrees of competence but they had low levels of intra- and inter-rater reliability. Frequency methods of assessment were resource intensive and time-consuming, had low levels of intra- and inter-rater reliability and were challenging to verify from audio recordings only. Dichotomous response categories (such as yes/no, present/absent or occurred/did not occur), when used to evaluate adherence items, were time efficient and had high intra- and inter-rater reliability.

The research team revised and amended the evaluation forms. The final agreed measures consisted of component-specific adherence forms, a generic competence form and an ‘overall impression’ scoring sheet (see Appendix 7).

The option to transcribe the audio recordings was unrealistic because of the volume involved and potential cost; evaluators were therefore asked to provide supportive quotations and or comments to justify their ratings.

Data analysis and presentation