A multicentred randomised controlled trial of a primary-care based cognitive behavioural programme for low back pain. The Back Skills Training (BeST) trial

Article history

The research reported in this issue of the journal was commissioned by the HTA programme as project number 01/75/01. The contractual start date was in October 2003. The draft report began editorial review in January 2009 and was accepted for publication in August 2009. As the funder, by devising a commissioning brief, the HTA programme specified the research question and study design. The authors have been wholly responsible for all data collection, analysis and interpretation, and for writing up their work. The HTA editors and publisher have tried to ensure the accuracy of the authors’ report and would like to thank the referees for their constructive comments on the draft document. However, they do not accept liability for damages or losses arising from material published in this report.

Declared competing interests of authors

None

Copyright statement

© 2010 Queen’s Printer and Controller of HMSO. This journal is a member of and subscribes to the principles of the Committee on Publication Ethics (COPE) (http://www.publicationethics.org/). This journal may be freely reproduced for the purposes of private research and study and may be included in professional journals provided that suitable acknowledgement is made and the reproduction is not associated with any form of advertising. Applications for commercial reproduction should be addressed to: NETSCC, Health Technology Assessment, Alpha House, University of Southampton Science Park, Southampton SO16 7NS, UK.

2010 Queen’s Printer and Controller of HMSO

Background

Low back pain (LBP) is a major public health problem in Western industrialised societies. 1 In the UK, the annual period prevalence is approximately 37%2,3 and LBP is so common that it affects almost everyone at some time during his or her lifetime. 4 For the majority, LBP follows a recurrent, fluctuating time course, with about 70% of patients having at least one recurrence within 12 months. 5,6 Annually around one in three people have an acute bout of LBP, but symptoms resolve quickly and pose no ongoing problem. 7 A significant proportion of people self-manage LBP without consulting the NHS. Estimates suggest that between 7% and 20% of the adult population who experience LBP consult a general practitioner (GP)8 and this results in 2.6 million additional consultations annually. 9 Of these people, 75% have symptoms 1 year later10,11 and about 30% develop persistent disabling LBP.

The direct health-care costs associated with LBP in 1998 were £1628 million; the majority of this expenditure was on physiotherapy and general practice. 1 During the years 1994 and 1995, 116 million production days were lost in the UK as the result of LBP, costing an estimated £10,668 million in production and informal care costs. 1 The majority of these costs are generated by those with the most chronic symptoms, and as a consequence the prevention and amelioration of chronic disabling pain is now the focus of research and clinical activity in this field. 12

This introduction provides a background to our sampling strategy, risk factors for developing chronic LBP and reasons why cognitive behavioural approaches (CBAs) may be helpful in LBP, an overview of current management of LBP, and the evidence to support CBA at the time the trial was designed.

Sampling approach and risk factors for the development of chronic low back pain

The population of people who suffer from low back pain is highly diverse. There have been a number of attempts to define different subgroups but none are satisfactory or in widespread use. We focused our trial on non-specific LBP, which is a term used to describe the majority of presentations, i.e. those in which no serious cause (infection, cancer or fracture) for LBP can be identified. The primary focus of our trial was the prevention or amelioration of disabling LBP in those with established (subacute or chronic) symptoms. We defined LBP according to Croft et al. 11 and used the International Society of Pain (ISAP) definitions of acute, subacute and chronic duration (Table 1). The ISAP provides a simple and widely used definition of back pain syndromes, although this does not include activity limitation.

Considerable advances have been made in identifying risk factors and understanding the processes involved in the development of chronic LBP. 14 Psychological, social and behavioural risk factors have been found consistently to be stronger predictors of chronic disability than the physical factors associated with the initial onset of pain. 15 This does not mean that physical symptoms are unimportant; for example, pain radiating down the leg is acknowledged to be a risk factor for developing chronic symptoms. 16,17 However, a recent prospective study found only a weak association between structural changes and pain and no association with disability or future health care. 18

Psychological risk factors play an important role in the progression of LBP to chronic disabling LBP. A review of psychological risk factors found that coping strategies, beliefs, distress, depressive mood and somatisation were associated with the development of persisting pain and disability. 19,20 Of these, ‘fear of movement’ (which is a negative health belief about the relationship between pain and re-injury such that if I increase my back pain, I am causing further damage, so I should avoid movement), has been that most consistently and strongly associated with the progression to chronic disability. 21,22 Participation in physical recreational activity is associated with lower LBP and a reduced risk of progression to disability. 23 There is now a widely held belief that increasing physical activity and/or exercise participation is important in managing LBP. 24 Physical activity is usually characterised as a behavioural factor.

Social factors are also important. A systematic review of social risk factors identified low social support in the workplace, heavy manual work and low job satisfaction as strong risk factors for the occurrence of chronic back pain. 25 More recent studies have also found strong correlations between chronic back pain and low socioeconomic status, educational level, work satisfaction and female gender. 6,26,27

Epidemiological evidence suggests that interventions that address this range of psychological, behavioural and social risk factors offer hope in reducing the burden of LBP. Recent clinical guidelines6,26 stress the importance of assessing psychosocial factors and emotional distress, and suggest that such assessments might assist in targeting intervention more appropriately. Cognitive behavioural approaches encompass a range of interventions that aim to directly change behaviour using models of learning, and to indirectly change behaviour by changing beliefs and behavioural risk factors However, it is broadly acknowledged that there is insufficient evidence to determine the optimal methods of assessment and intervention and that large-scale evaluation is required.

Current management in the UK

The majority of LBP is managed in primary care, with a small number of people referred to secondary care, often only for a single consultation and with little consistency or rationale. 28 About 9% of the population with LBP attend for physiotherapy treatment, usually after primary-care consultation. 1 Chiropractic and osteopathy are used less often. 1

Since 2000, there has been a major change in the approach to managing LBP in primary care known as the active management (AM) strategy, which forms the core of all international guidelines. 29–31 The AM strategy discourages bed rest for LBP and instead advocates physical activity. 10 Appropriate medication is encouraged, although many classes of pain medication have been shown to have limited long-term benefits and are recommended most often for the management of acute LBP. 32,33 An information booklet called The Back Book34 has facilitated implementation of the AM guidelines in the UK. The Back Book was designed to challenge negative beliefs and behaviours, rather than merely to impart factual information. A randomised controlled trial evaluating The Back Book showed a positive effect on patients’ beliefs and clinical outcomes, maintained at 1 year, for those with both acute and recurrent back pain. 25,35

A 2007 systematic review36 evaluating the effectiveness of advice for the management of LBP compared advice offered to acute, subacute and chronic LBP patients. The review suggests that advice to stay active is sufficient for acute LBP patients and could be more widely implemented in practice. There is uncertainty about the management of both subacute and chronic LBP. No conclusions could be drawn from the evidence base about the optimal method of advice, and although there are suggestions in the literature that additional treatments may be beneficial, differences in case definitions of LBP and the treatments tested make it difficult to draw substantial conclusions about what type of advice is optimal for various presentations of LBP.

Unlike most guidance, and at the time of planning this trial, most UK guidelines tentatively recommended early referral to physical treatments, including acupuncture, spinal manipulation and exercise. 1 The evidence for these treatments has strengthened during the lifetime of the Back Skills Training (BeST) trial. More recent studies have raised some doubt about the effectiveness of traditional physiotherapy approaches to LBP; for example a comparison of routine physiotherapy with a single session of advice for LBP found no difference in treatment outcomes. 37 An Australian study of participants with subacute LBP, found physiotherapist-directed exercise and advice were each slightly more effective than placebo, with greatest effectiveness when combined. 38 A Cochrane review concludes that exercise therapy appears to be slightly effective at decreasing pain and improving function in adults with chronic LBP. 39 A meta-analysis of the effectiveness of spinal manipulation therapy for LBP concluded that there is no evidence to suggest that manipulation is superior to other standard treatments for acute LBP. 40 However, a recent large UK trial comparing manipulation with exercise and a combination of manipulation and exercise found small medium-term benefits in each treatment arm, and greater benefits if treatments were combined. 24

We selected the active management strategy and The Back Book as the intervention for the control arm of the trial, as these were recommended as best practice strategies within UK guidance at the time of starting and during the conduct of the trial.

Cognitive behavioural approaches in low back pain

The CBA includes a range of therapies deriving from cognitive and behavioural psychological models that aim to teach an individual to tackle their problems using specific psychological and practical skills. 41

Cognitive behavioural approaches for managing LBP were first introduced in the UK as part of inpatient pain management programmes for those with very chronic/severe LBP. Intensive (> 100 hours of therapy) multidisciplinary bio-psychosocial rehabilitation programmes were found to improve pain and function for patients in the secondary-care setting. 42 Systematic reviews suggest that the type of CB treatments for chronic LBP (> 12 weeks) that could be delivered in primary care appear to have short-term benefits but, possibly, no sustained long-term benefits. 43

Although there has been no formal systematic review, trials of CBA in populations with acute and subacute LBP report a mixed picture, including some improvement in disability. 44–46

Although some trials pointed to the potential for CBA in treatment of subacute and chronic LBP, there was a mixed picture of results. Few programmes found sustained benefits. Differences in results may be attributable to poor research design, but more likely they are the result of variable adherence to the principles of CBA and differences in how the programmes are delivered. These include the amount of contact time, level of expertise, components included in programmes and method of delivery. Indications are that the important attributes of effective interventions are ensuring that the health-care professionals who deliver the interventions are able to elicit psychosocial risk factors, implementing a CB framework that results in modification of beliefs as well as behaviours (as opposed to delivering skills alone), and delivering a treatment that is credible to patients. 47,48

The aim of this study was therefore to develop and test a group-based CBA intervention that could be delivered within the UK NHS and that could be accessed from primary care. We carried out a definitive randomised controlled trial with a parallel economic and qualitative study that would provide a comprehensive evaluation of CBA as a treatment for subacute and chronic LBP, and provide the NHS with evidence to support decision-making in this area of clinical practice.

Introduction

A detailed description of the theoretical basis of the intervention, as well as dose and mode of delivery is an essential step in the reporting of randomised trials of complex interventions. 49 Our approach was to draw together the essential elements of a CBA, ensuring that the intervention was consistent with the principles of CB therapy, and that we targeted health behaviours and beliefs that are broadly accepted as being on the causal pathway between LBP and disability. The intervention was developed by systematically reviewing experimental and observational literature, and linking the results of these reviews into a CB framework. We also considered the optimal delivery method to balance clinical effectiveness and cost-effectiveness.

The CB model states that the way a person thinks about their problem will produce emotions, including associated physical sensations, which then drive behaviour. 50 Often, the behaviour will inadvertently maintain the thoughts or beliefs, so creating a maintenance or vicious cycle effect. The following sections provide a rationale for the risk factors selected as treatment targets for the CBA.

Identifying the targets for a cognitive behavioural intervention for low back pain

The key modifiable risk factors appear to be psychological and behavioural factors that have a mediating effect on activity levels. Psychological constructs, including catastrophising, passive coping, fear avoidance and depression lead to decreased activity levels or, for some, overactivity. These changes in activity levels are implicated in the development of chronic disabling LBP through a pathway of deconditioning, and worsening pain. We specified the targets of the CB intervention:

• to increase activity levels

• to manage periods of overactivity

• to specifically address catastrophising and fear avoidance

• to improve coping skills.

There is consistent evidence across various reviews that coping skills, catastrophising and fear avoidance are key factors in the progression of acute LBP to chronic disability. Distress has also been identified as an important risk factor;51 however, it is difficult to define distress separately from the other psychological constructs and mood. 6,18,52

Many studies have pointed to the importance of coping strategies and beliefs held by patients. 19,53 A sense of personal control and self-efficacy are associated with active coping strategies (e.g. taking exercise). A lack of personal control and feelings of helplessness are associated with passive (maladaptive) coping strategies such as rest and catastrophising. 54–57

The strong link between beliefs predicting behaviour has been shown in two studies. 3,58 A reduction in patients’ belief that they were disabled and that increased pain signified harm and the need to restrict activity, was strongly associated with a reduction in pain behaviours, physical disability and depression. These beliefs are commonly referred to as ‘catastrophic beliefs’ and lead to avoidance of the feared activity or pain. This behaviour has been labelled ‘fear avoidance’ and has been consistently and strongly associated with the progression of acute LBP to disability. 21,22,59–61 In addition, catastrophising is associated with hypervigilance for symptoms, which increases pain perception. 62,63 Interventions that have used CBA to target catastrophising and fear avoidance behaviours appear to reduce disability in the short term. 64

At the other end of the activity level spectrum, there are people who increase their activity levels in response to pain. 65 This overactivity has been linked to mood and to unhelpful beliefs66 and can lead to poor control over pain and subsequent activity.

Depression is associated with the risk of developing chronic LBP,67 with a hypothesised pathway of apathy, demotivation and low mood resulting in decreased activity and poor outcome. Exercise is effective in the management of depression68 and has the potential to reverse this cycle. 69

Pain intensity has been positively and consistently associated with a poor outcome. 70 However, pain intensity is not independent of other psychological constructs such as catastrophising. 71 Education on pain mechanisms appears effective in changing beliefs and improving physical functioning. 72

Low levels of physical activity have been shown to correlate with future episodes of persistent LBP. 23,73,74 In addition, a significant proportion of the LBP population will reduce their activity levels in response to developing LBP. 75 The resultant ‘deconditioning or disuse syndrome’ describes the physical decline in strength, mobility, endurance and coordination that is postulated to contribute to ongoing pain. 76

Several systematic reviews have investigated the efficacy of exercises in the treatment of LBP and found modest improvements in pain and function. 77–78 Hayden et al. 79 went on to perform a meta-regression to identify features of the exercise programmes associated with successful outcomes. They concluded that the most effective exercises were conducted as part of an individualised supervised programme and included stretching and strengthening exercises. Although counterintuitive, particularly for those suffering LBP, exercise does not increase recurrence of pain. 80–82 There have been relatively few interventions that have focused on improving general or functional activities. 78

The intervention did not intend to target social factors that are not modifiable within the context of a primary care-based intervention (e.g. educational level, work-related risk factors, job satisfaction). Models of pain based on disc degeneration, posture, injury, weight and leg-length discrepancy were excluded because there is little evidence to support a relationship between these factors and chronic LBP. 83,84

Framework and structure of the cognitive behavioural intervention

The British Association of Behavioural and Cognitive Psychotherapies (BABCP) has developed definitions of CB therapy41 and suggests that the following components are essential:

• The intervention should be delivered collaboratively, i.e. intervention draws on the expertise of therapist and client and direction of treatment is determined jointly.

• The intervention should use a goal-oriented approach, i.e. treatment is centred around achieving the client’s goals.

• The intervention should use the CB model, i.e. intervention links thoughts, feelings and behaviours.

• The intervention should explore unhelpful beliefs through CB questioning techniques, e.g. the use of open questions to help the client consider alternative ways of thinking.

• The intervention should focus on the development of specific psychological and practical skills to enable the client to tackle problems independently, e.g. problem-solving and relaxation skills.

• The intervention should use homework to achieve skill development, e.g. practising skills discussed during treatment sessions.

We designed an intervention comprising an assessment followed by six group sessions based on these principles. The CB model we developed to underpin the interventions is shown in Figure 1. The figure depicts the concept that back pain leads to altered activity and associated physical changes, which in turn leads to maintenance or worsening of pain. These relationships are mediated by unhelpful thoughts and feelings. The shaded boxes indicate the key skills developed by the BeST intervention, and their hypothesised mode of action.

FIGURE 1.

Integrated model of LBP. Text in boxes indicates how sessions target components.

Group structure

Each session was structured in line with standard CB therapy approaches to include agenda setting, topics and homework review. The work of the previous session was reviewed briefly at the beginning of each session, and all sessions included a 10-minute break midway through to allow participants to move around and exercise (if they wished to). Each session started with agenda setting, which included asking participants what they wanted to cover in the sessions. Homework was reviewed at some point during each session to allow for group problem-solving.

Each session focused on one or more of the components of the CB model shown in Figure 1. We used simple CB maintenance models for fear avoidance and hypervigilance to provide a structure and rationale for the skills being taught in the sessions (Figures 2 and 3). These provide a conceptual framework for the relationship between thoughts, feelings and behaviours which could be communicated to participants, and form the foundation of discussions to enable problem-solving to break vicious cycles.

FIGURE 2.

Fear avoidance cycle used in BeST session 4.

FIGURE 3.

Hypervigilance cycle used in BeST session 5.

We prespecified a benchmark in attendance to define those who we believed had been compliant with the intervention, and those who had not. This threshold was attendance at the assessment session and at least three of the six group sessions. Although not evidence based, we hypothesised that this was the minimum number of sessions required to cover the key components of CBA as defined by the BABCP. 41

Participant information folder

A high-quality participant information folder was designed for the intervention and provided to each participant at the first assessment appointment (Figure 4). The folder contained an overview of the sessions, how and where the groups would be run, therapist contact details, what participants should expect and exercise sheets. During the assessment these exercise sheets were personalised and the goal sheets were filled in. To support participants in working towards their goals while waiting for the group sessions to start, there were information sheets on goals and how to set baselines (starting level for activity). In addition, there were information sheets on sleep, medication and communicating with health professionals.

FIGURE 4.

Participant information folder.

At each group session participants were given sheets to add to their folders that summarised the content of the session and provided details of their ‘homework’ task. If a participant missed any sessions, either the inserts were provided at subsequent sessions or they were sent through the post if the participant was unable to attend any other sessions. As the folder was only provided at the therapist assessment, any participant who did not attend for this session received only the active management intervention.

Therapist recruitment and training

A 2-day programme was developed to train registered health professionals (physiotherapists, psychologists, nurses or occupational therapists). The length of training reflects the typical length of informal training programmes for qualified staff currently available in the UK. We trained a spectrum of health professionals. This was a pragmatic decision – LBP is a common condition and there are unlikely to be sufficient numbers of psychologists to meet demand. It is recognised that generic pain management skills cross professional boundaries. 92 For example, when CBAs have been applied in diabetes self-management, the treatment effect sizes are found to be the same for psychological specialists and non-specialist clinicians. 93 Although there are high rates of distress and low mood among LBP sufferers in primary care, these symptoms are generally not severe. Therefore, a prolonged training programme in psychological management for the therapists in the trial was not felt necessary.

With this in mind the training was designed to ensure that individuals with different professional backgrounds would be equipped with the same basic knowledge and skills and that the intervention was effective regardless of which health-care professional delivered it. This included equipping psychologists with knowledge of LBP, physical activity and exercise prescription.

The training covered an understanding of LBP and the risk factors associated with chronicity, understanding the CB model, developing basic CB skills such as questioning techniques, developing group facilitation skills and learning the topics to be covered in each session, including pain management techniques. Training was delivered by a cotrained CBT therapist/physiotherapist and a clinical psychologist and was documented in an extensive training manual. 49

Therapist support

In addition to the training already described, each therapist received a comprehensive treatment manual detailing the rationale and content of each session, including suggested dialogue. In response to feedback from the first training session a website and DVD were developed. The website contained all support materials, including trial updates and a ‘frequently asked questions’ section; a screen print is shown in Figure 5. The website also had a forum for posting questions to be answered by the trial intervention team and by other therapists. This section was not used; therapists preferred to e-mail or phone the trial team directly.

FIGURE 5.

Screen print of BeST Therapist webpage.

The DVD was a recording of sessions 1 and 3 run for volunteers from the research department as group participants. Feedback from therapists indicated that they found the DVD very useful before running the programme for the first time.

Within psychological therapies it is usual practice to have regular supervision sessions with a senior therapist to discuss cases. This is not normal practice within most of the other professions allied to medicine. For the purposes of the trial we allowed for flexible supervision, which consisted of discussions either face-to-face on-site visits or via phone or e-mail, whichever was convenient and appropriate for the issues to be discussed. Supervision was provided during the course of the trial by the clinical research fellow, which was on average 1.5 hours per group run. This supervision usually centred on difficulties encountered in the groups, for example difficulties in setting goals with some individuals.

Assessment of treatment fidelity

Quality of the intervention delivery was checked via three processes:

• a site visit was made to a group session for each therapist during their first programme of intervention

• subsequent groups had one session randomly selected for audio recording

• treatment records were screened for each participant.

The session content and skills demonstrated by the therapist were assessed via these processes using a checklist that included items shown in Table 3. The audio recordings were used as an observational tool as part of a process of evaluation to determine treatment fidelity, and were not used for further training and feedback.

Summary

A CB intervention was designed that targeted known modifiable risk factors for the development of chronic disabling LBP. The intervention adhered strongly to the principles of a CBA and was structured with sufficient flexibility to allow delivery in a variety of settings and by a range of different health professionals. The training and support package was practical to allow for roll out within the NHS if the intervention proved effective.

Aims

There were two aims:

• To estimate the clinical effectiveness of AM in general practice versus AM in general practice plus a group-based, professionally-led CB package (AM+CBA) for subacute and chronic LBP in terms of:

  1. – reduction in disability associated with LBP

  2. – reduction of pain or improved tolerance of pain symptoms

  3. – reduction of further medical, rehabilitation or surgical treatment for LBP

  4. – improvements in quality of life.

• To measure the cost of each strategy, including treatment and subsequent health-care costs, over a period of 12 months and to estimate cost-effectiveness. The methods employed in the economic analysis are detailed in Chapter 6.

In addition, we planned to interview a selection of participants to gain insight into the experience of LBP and of the treatments delivered as part of the trial (for the methods used, see Chapter 5).

Research methods

Treatments

Study procedures

Outcome assessments

Other treatments

There was a possibility that participants would seek other forms of treatment during the follow-up period. The resource use questionnaire allowed us to monitor changes in the amount or types of analgesia used, use of physical treatments (osteopathy, chiropractic or physiotherapy), alternative therapies, or referral to secondary-care services. Participants and GPs were encouraged to refrain from referral of participants to other treatments where possible during the first 3 months after randomisation and while participants were attending the CBA course.

Randomisation

Quality assurance

We implemented a quality assurance protocol to monitor and ensure that research nurses were adherent to the trial protocol. Each research nurse was visited on at least two occasions by a senior research nurse from the MRC GPRF, and observed taking consent and delivering the active intervention component of the intervention. Problems were minimal.

Compliance

We measured compliance with the intervention by the number of sessions attended. This information was ascertained from records collected by the therapist providing the treatment.

Formal approvals

The original ethics approval for the project was dated 26 March 2003 with a substantial amendment approving protocol amendments on 22 July 2004. Further amendments for the qualitative interview study and consent, change to patient information sheets and promotional posters were all approved on 5 March 2005. An amendment to the randomisation procedure was submitted on 24 May 2005 and approved via chairman’s action on 14 July 2005. Further chairman’s action approval was received on 7 September 2005 for a change in trial personnel. A substantial amendment was approved on 25 January 2006 for version control. Chairman’s action was sought on 5 July regarding follow-up procedures and approval was received on 1 September 2006.

Adverse events

Sample size

The primary outcome measures were the RMQ and MVK assessed over 12 months.

Increasingly it has been recognised that advances in modern health care are most likely to yield moderate improvements, but in the context of highly prevalent conditions like LBP, these are considered worthwhile. 113

Pilot study

We undertook a series of pilot studies to refine the study procedures and ensure that the intervention was acceptable and deliverable in the format intended.

The intervention was piloted at the Nuffield Orthopaedic Centre NHS Trust in Oxford in the first instance. Two CBA cycles were completed and participants provided feedback on the contents and revised contents of the intervention. We then piloted the intervention in primary care along with the study procedures.

We planned to carry out the pilot study in three general practices in the Coventry area and to recruit 30–40 participants. In fact it was carried out in only one of the practices for the following reasons:

• one practice dropped out when they learnt more about the trial

• we were forced to run the pilot over the July/August period because of a 7-week delay in the Central Office for Research Ethics Committees confirming ethical approval because of an internal Central Office for Research Ethics Committees communication problem (as a consequence the research nurse who was secured to run the second practice was unavailable – she had to cover leave, went on holiday and was subsequently stranded abroad by hurricanes; however, this practice participated in the main trial).

We tested one of the two recruitment strategies – identification of participants through participant record searches but not prospective identification of participants via the GPs. In general the pilot study had been successful and useful in refining the study methods and in particular, methods related to the approach to participants.

Data management

All the databases were developed in Microsoft access 2002.

The prerandomisation and randomisation information was captured on laptops, whereas the baseline and follow-up data were collected using postal clinical research forms or by telephone and then entered into the database manually. Computerised validation checks were incorporated into the data sets to minimise data errors.

Database specifications were set up by the statistician and programmer for each variable collected at baseline and follow-up assessments.

Statistical analysis

Monitoring and time line

Participant approach and characteristics of the randomised sample

A total of 9771 people were identified from primary contacts or from primary-care record searches and were sent an initial approach screening questionnaire. Of these people, 3604 returned the screening questionnaire and 3496 provided data that could be used to assess potential eligibility.

Of these, 1465 appeared eligible. The most common reasons for ineligibility were that people no longer had back pain (n = 481), their back pain was not of sufficient frequency (n = 611) or troublesomeness (n = 479), or they were pregnant (n = 20). Some participants had several reasons for ineligibility.

All 1465 apparently eligible people were invited to attend the first nurse assessment check: 971 attended the first nurse assessment, and of these 833 were eligible for the second nurse assessment; 754 people attended the second nurse assessment and 705 were randomised. Baseline data were not received for four of these and they have been excluded from our analyses.

There were no important differences between those participants who were determined eligible and randomised, and those who were determined eligible and were not randomised. All randomised participants met the eligibility criteria.

The characteristics of the randomised sample are shown in Table 7. In summary, the majority of the randomised sample (70%) experienced back pain every day. The most common presentation was pain in the back or buttocks, and stiffness and restricted range of motion. Just over half of the randomised sample reported their back pain to be moderately troublesome (55%), the remainder reporting very or extremely troublesome pain. The average age of the sample was 54 years, and nearly 60% of participants were women. Most had left full-time education before the age of 15, and approximately half of the sample reported that they were currently working. If participants were working then they were generally in full-time work (working between 25 and 40 hours a week), were in professional or skilled manual work and were being paid. The majority of those not working were retired.

Description of sample by cluster

The study recruited from 56 general practices from seven regions across England: Norwich (27.8% of the sample), Coventry and Solihull (26.8%), South Warwickshire (18.6%), Birmingham (12.6%), Langbaurgh (7.5%) and North Warwickshire (6.7%). The characteristics of each regional sample were generally consistent (Table 8). The only exception was ethnicity. Although overall the sample was representative of the UK population ethnic mix, the regional samples were different in their ethnic make-up. The Birmingham cluster provided the majority of Asian participants. Birmingham also had the greatest proportion of people who were not working.

Success of randomisation

The severity of back pain was used as a stratification variable along with the regional centre and, as shown in Table 9, the randomisation resulted in a satisfactory balance in the stratification variables over the treatment groups.

Flow and characteristics of randomised participants

Figure 6 presents the flow diagram for the study. The number of participants randomised and in whom a baseline assessment was received by the trial office was 701 (233 on the AM and 468 on AM+CBA). An additional four people provided consent and were randomised but no baseline questionnaire was received for them. Hence they have been excluded from the analysis (Figure 6).

FIGURE 6.

Flow chart of the trial.

In summary, 78% of the participants provided data at 3 months, 83% at 6 months and 85% at 12 months. Approximately 1% of the clinical research forms were returned blank over the course of the trial. A greater amount of the data was collected by telephone follow-up at 12 months (14%) compared with 3 months (4%) and 6 months (8%). There was no difference in the amount of telephone follow-up between the two arms of the trial at any of the follow-up time points (p = 0.13). There was no significant difference in the amount of non-response between treatments over time (p = 0.24; Table 10).

Just over 5% (38/701) of participants withdrew over the course of the trial. There was no association between withdrawal and treatment allocation at any of the time points (p = 0.66). The usual reason for withdrawal was that participants did not want to complete questionnaires – 15 participants (2%).

Missing data and characteristics of the randomised sample

There was no statistically significant difference in the return rates between the two arms of the trial at any of the follow-up time points (p = 0.67). There was no evidence of a systematic difference in the baseline characteristics of participants who provided follow-up data and those who did not (see Appendix 12). The most common pattern of missing data was for data to be missing at the first follow-up point, and then complete for the remainder (Table 11). We obtained some follow-up data on over 90% of participants. Reasons for withdrawal are given in Table 12.

The demographic characteristics of all randomised participants (by treatment group) are shown in Table 13. Demographic and baseline characteristics were well balanced across the treatment arms.

Treatments received

Intervention arm (AM+CBA)

One participant was randomised to AM but received CBA in addition to AM.

Details of therapists for cognitive behavioural approach

We used a variety of methods to recruit the therapists and although we targeted all allied health and nursing professionals, it was mainly physiotherapists who responded. We recruited and trained 19 therapists: 14 physiotherapists, two occupational therapists, one nurse, one clinical psychologist and one health psychologist. In total 77.3% (n = 378) of participants attended groups facilitated by physiotherapists, 6% (n = 30) saw occupational therapists, 9% (n = 45) saw psychologists and 3% (n = 15) saw nurses.

Details of the therapists are shown in Table 14. The therapists had a range of experience with years since professional qualification ranging from 2 to 32 years with a mean of 13.6 years (SD 9.6). Some therapists had experience of running similar groups in the past.

TABLE 14 - Characteristics of therapists who delivered the BeST intervention (based on data for 2006)

Grade provides the level at which the therapist is working within the NHS, with increasing levels of responsibility and expertise the titles of the grades are Junior, Senior II/Band 6, Senior I, Clinical specialist/extended scope practitioner (ESP)/Band 7.

Therapist number	Profession	Years qualified	Grade	Gender	Experience of running similar groups	Number of groups run in trial	Total participants seen in trial (% of total number in CBA arm)
1	Physiotherapist	9	Senior I	Female	Yes	12	91 (19)
2	Physiotherapist	10	ESP	Male	Yes	5	30 (6)
3	Physiotherapist	10	Senior I	Male	No	1	7 (2)
4	Physiotherapist	8	Clinical specialist	Female	Yes	1	5 (1)
5	Physiotherapist	14	Senior I	Male	No	4	32 (7)
6	Physiotherapist	9	Senior I	Female	No	5	43 (9)
7	Physiotherapist	31	Senior I	Female	Yes	6	52 (11)
8	Occupational therapist	4	Senior II	Female	No	3	22 (5)
9	Physiotherapist	23	ESP	Female	No	1	8 (2)
10	Clinical psychologist	2	Band 7	Female	No	2	16 (3)
11	Physiotherapist	12	Clinical specialist	Female	Yes	3	18 (4)
12	Physiotherapist	21	Clinical specialist	Female	Yes	3	24 (5)
13	Health psychologist	3	Band 6	Female	No	4	29 (6)
14	Occupational therapist	32	Senior I	Female	No	1	8 (2)
15	Physiotherapist	2	Senior II	Male	No	1	5 (1)
16	Physiotherapist	7	Senior I	Female	No	2	13 (3)
17	Nurse	24	Band 7	Female	No	2	15 (3)
18	Physiotherapist	25	Senior I	Female	No	3	26 (6)
19	Physiotherapist	12	Senior II	Female	No	3	24 (5)

Attendance for assessment and sessions in the trial

Attendance for the assessment and group sessions is summarised in Table 15. Nearly two-thirds (63%) of the participants randomised to receive CBA attended the assessment session and at least three of the six group sessions, and were therefore considered to have received the basic elements of the treatment. Just over 10% of people randomised to CBA did not attend any of the programme. Attendance at all six sessions was achieved by 25% of the sample. There was no evidence to suggest that therapist characteristics were associated with attendance – neither the professional background (Table 16), previous experience with CBA nor years since qualification were associated with attendance rates (Table 15).

TABLE 15 - Number (and percentage) of participants attending the assessment and group sessions

The number of sessions attended does not imply consecutive sessions.

	No. (%) of participants
Did not attend for assessment	50 (11)
Attended assessment	418 (89)
Attended assessment plus ≥ 1 session	358 (77)
Attended assessment plus ≥ 2 sessions	325 (69)
Attended assessment plus ≥ 3 sessions	294 (63)
Attended assessment plus ≥ 4 sessions	275 (59)
Attended assessment plus ≥ 5 sessions	225 (48)
Attended assessment plus ≥ 6 sessions	121 (26)

TABLE 16 - Number (and percentage) of patients seen by each therapist profession with attendance rates

Profession	Attended < 3 group sessions (%)	Attended ≥ 3 group sessions (%)	Total (%)
Physiotherapists	136 (36)	242 (64)	378 (100)
Occupational therapists	11 (37)	19 (63)	30 (100)
Nurses	7 (47)	8 (53)	15 (100)
Psychologists	20 (44)	25 (56)	45 (100)

There were few differences between the groups of participants who were adherent and those who were not. On average, participants who adhered were 4.5 years older than those who did not (95% CI 1.8 to 7.5), and had slightly lower pain scores at baseline [MVK (pain) mean difference 4.5, 95% CI 0.94 to 8.19]. There were no differences in RMQ, MVK (disability), sex, troublesomeness of pain or fear avoidance beliefs at baseline.

There were 174 (37%) participants who did not comply (i.e. had not received three or more group-based sessions of the therapy) with the AM+CBA therapy.

Therapists were asked to document the reasons for non-attendance if possible. A reason was documented in 80 cases (Table 17).

TABLE 17 - Reasons provided by participants for not attending

	No. of participants (%)
Unknown	95 (54)
Unwell	18 (10)
Changes to work	16 (9)
Family issues	15 (9)
Decided format of group ‘not for them’	10 (6)
Difficulty attending – other	10 (6)
Decided no benefit to intervention	4 (2)
Moved	3 (2)

In addition four participants were excluded from the groups by the therapists (Table 18).

TABLE 18 - Reasons provided by the therapists for excluding the four participants from the group

Therapist	Reason
8	Patient has other musculoskeletal disorders and also very resistant to change. Participant and therapist agreed that the group was not for him
6	Wife aggressive to therapist and wanted to attend sessions. Participant had learning difficulties
7	Excluded after randomisation by GP
5	Excluded as participant was receiving private physiotherapy

Time from randomisation to starting treatment

Table 19 provides summary statistics for the time scale of treatment delivery. The target time for attendance at the therapist assessment was set at 45 days after randomisation. Of those participants who attended the assessment, 271 out of 418 (65%) received their assessment within the target time and 313 (75%) had completed the group treatment by the 3-month follow-up time point (Table 19).

TABLE 19 - Time from randomisation to assessment, the first group session and attendance at last CB group

	No. of days from randomisation to assessment	No. of days from assessment to session one	No. of days from randomisation to session one	No. of days from randomisation to last CB session attended
Mean	25.7	17.7	43.1	82.1
Median	21.0	14.0	38.0	77.0
SD	22.3	13.7	26.0	26.4

Meeting the target time proved more difficult to achieve for some of the groups. This was mainly because of delays at one site: 16% of participants at this site were assessed before 45 days compared with an average of 80% across the other sites (Figure 7). There did not appear to be pressures on venue or staff time at this site but there were consistent delays in assessing participants and starting groups. A minority of other groups were delayed by last minute changes in therapist availability.

FIGURE 7.

Number of days from randomisation to receiving assessment.

Thirteen participants attended a group at a later date than the first one they were randomised to due to changes in their personal circumstances after randomisation. No participants switched to a different group once sessions had commenced. There was no change-over in therapist during any of the 6-week group sessions. The attendance rate for those who started the group in less than 45 days was the same as for those who waited longer than 45 days (71% and 70% respectively).

Number and size of groups

Group size was difficult to standardise because of variations in recruitment rate. In total, 62 groups were run with a mean size of eight (SD 1.62, range 4–12). A breakdown of the group sizes and attendance rates is given in Table 20. Group size was not associated with attendance rate (p = 0.17).

TABLE 20 - Group size with number of patients and attendance rates at group sessions

Group size	No. of groups run at this size	Total participants	Attended ≥ 3 group sessions (%)
4	2	8	6 (75)
5	3	15	10 (67)
6	11	66	44 (67)
7	14	98	60 (61)
8	14	112	70 (64)
9	10	90	52 (58)
10	6	60	40 (68)
11	1	11	5 (46)
12	1	12	7 (64)

Group intervention venue and time

During the trial, a variety of premises were used and a range of times were offered to cater for the needs of the participants. The venue setting did not impact on attendance compliance, with 63% attendance for both PCT premises and private community premises. The day and time of the group did impact on compliance. Groups run on a Friday had an attendance of 70% compared with around 61% on other days. Groups running around lunchtime and after work saw the best compliance, with 68% and 65% respectively. Evening groups were surprisingly poorly attended (51%) despite being requested by participants.

Thirty-five group sessions (56% of all groups run) were assessed for quality assurance either with an assessor in the session or via audio recordings. The results are shown in Table 21.

TABLE 21 - Quality assurance tool with percentage achieved during visits or via recordings

Item		Not achieved (% of total)	Partially achieved (% of total)	Satisfactorily achieved (% of total)
Content	Set agenda	5 (14)	8 (23)	22 (63)
	Homework reviewed	0 (0)	1 (3)	34 (97)
	Topics covered	0 (0)	3 (9)	32 (91)
	Break	0 (0)	0 (0)	35 (100)
	Homework set	0 (0)	5 (14)	30 (86)
	Feedback elicited	6 (17)	7 (20)	22 (63)
	Exercises checked in break	0 (0)	1 (3)	34 (97)
Style	Encouraged group participation	0 (0)	4 (11)	31 (89)
	Listened appropriately	0 (0)	0 (0)	35 (100)
	Empathy demonstrated	2 (6)	1 (3)	32 (91)
	Elicited beliefs/thoughts	2 (6)	11 (31)	22 (63)
	Questioning style demonstrated	2 (6)	8 (23)	25 (71)
	Referred to CB model	4 (11)	4 (11)	27 (77)
	Appropriate pacing of session	1 (3)	7 (20)	27 (77)
	Appeared professional	0 (0)	0 (0)	35 (100)
Environment	Comfortable	0 (0)	1 (3)	34 (97)
	Spacious	0 (0)	0 (0)	35 (100)

As many of the therapists were new to the CB skills of questioning, agenda setting, eliciting beliefs and feedback we expected these items to be more difficult to achieve, and this was reflected in the quality assurance scores. The remaining skills were evidenced in most groups.

Description of treatment effectiveness

Data summaries by treatment group are given in Table 22 for the primary outcomes and in Tables 23 and 24 for the secondary outcomes.

TABLE 22 - Summary statistics for the primary clinical outcome measures at each follow-up time point by treatment (unadjusted)

MVK, Modified Von Korff Scale; RMQ, Roland Morris Disability Questionnaire.

		RMQ		MVK (disability)		MVK (pain)
		AM	AM+CBA	AM	AM+CBA	AM	AM+CBA
Baseline	N	232	468	228	455	230	463
	Mean	8.5	8.8	46.2	48.5	59.4	59.3
	SD	4.72	4.99	23.79	23.85	19.51	19.24
	Median	7.5	8.0	46.7	50.0	60.0	56.7
	Range	0–22	0–23	0–100	0–100	10–100	6.67–100
Month 3	N	179	335	187	345	189	354
	Mean	7.3	6.5	38.8	35.3	52.9	46.76
	SD	5.28	5.27	24.68	24.33	22.74	23.13
	Median	6.0	5.0	36.7	30.0	53.3	46.7
	Range	0–24	0–24	0–100	0–96.67	0–100	0–100
Month 6	N	177	352	183	381	187	387
	Mean	7.4	6.1	41.8	33.9	53.0	44.5
	SD	5.34	5.44	24.72	25.94	23.29	25.16
	Median	6.0	4.0	43.3	30.0	53.3	43.3
	Range	0–23	0–24	0–100	0–100	0–100	0–100
Month 12	N	159	339	189	374	195	392
	Mean	6.9	6.1	40.0	33.3	51.0	44.5
	SD	5.12	5.62	25.22	26.22	23.93	26.04
	Median	6.0	4.0	40.0	26.7	50.0	46.7
	Range	0–22	0–24	0–100	0–100	3.33–100	0–100

TABLE 23 - Summary statistics for the health-related quality of life outcome measures at each follow-up time point and treatment

SF-12, Short Form 12-item health survey.

		SF-12 (physical)		SF-12 (mental)
		AM	AM+CBA	AM	AM+CBA
Baseline	N	217	421	217	421
	Mean	46.1	44.5	37.5	37.1
	SD	11.03	11.51	10.08	9.31
	Median	46.4	43.9	37.4	37.3
	Range	17–73	11–69	12–61	7–62
Month 3	N	176	332	176	332
	Mean	46.4	47.0	39.1	40.7
	SD	11.25	11.44	10.38	11.05
	Median	47.6	48.9	39.8	42.4
	Range	12–68	11–71	11–62	7–61
Month 6	N	177	362	177	362
	Mean	46.1	47.6	39.2	41.0
	SD	10.61	11.45	10.33	11.26
	Median	47.0	49.6	39.7	41.7
	Range	17–66	11–71	10–69	7–68
Month 12	N	187	375	187	375
	Mean	47.0	46.4	38.6	41.9
	SD	11.35	11.51	11.31	11.77
	Median	47.9	48.3	39.3	43.5
	Range	12–68	13–71	13–63	7–70

TABLE 24 - Summary statistics for the secondary clinical outcome measures at each follow-up time point and treatment

FABQ, Fear Avoidance Beliefs Questionnaire.

		FABQ		Pain self-efficacy
		AM	AM+CBA	AM	AM+CBA
Baseline	N	219	443	223	453
	Mean	14.2	13.6	41.2	39.6
	SD	6.22	6.34	12.54	13.39
	Median	15.0	14.0	43.0	41.0
	Range	0–24	0–24	5–60	2–60
Month 3	N	173	319	173	325
	Mean	13.3	10.3	40.6	43.0
	SD	6.07	6.37	13.51	13.31
	Median	14.0	11.0	43.0	46.0
	Range	0–24	0–24	3–60	2–60
Month 6	N	168	336	170	333
	Mean	13.8	10.4	40.4	43.3
	SD	6.09	6.54	13.38	13.54
	Median	14.0	10.5	42.0	46.0
	Range	0–24	0–24	2–60	3–60
Month 12	N	152	320	155	317
	Mean	13.1	9.9	41.3	43.2
	SD	6.18	6.26	13.28	14.08
	Median	14.0	10.0	43.0	47.0
	Range	0–24	0–24	2–60	4–60

Therapist and group effects

The ICCs were very small and there was no evidence of significant clustering effects related either to group sessions or to therapist (Table 25), the implication being that group session and therapist effects were negligible. Estimates of the treatment effects were not altered by the inclusion of therapist or group effects and hence results are based on the linear models adjusted for age, sex and the baseline value of the variable being tested. Within the adjusted models, only age and baseline measures were statistically significant. This was consistent across all the outcomes and time points.

TABLE 25 - Intracluster correlation coefficient for RMQ score and MVK questionnaire (using change from baseline)

ICC, intracluster correlation coefficient; MVK, Modified Von Korff scale; RMQ, Roland Morris Disability Questionnaire.

Cluster	Month	RMQ		MVK (disability)		MVK (pain)
		ICC	95% CI	ICC	95% CI	ICC	95% CI
Group session	3	– 0.01	– 0.02 to 0.001	– 0.02	– 0.04 to – 0.02	0.01	– 0.02 to 0.03
	6	0.09	– 0.02 to 0.19	– 0.04	– 0.06 to – 0.01	– 0.02	– 0.03 to – 0.02
	12	– 0.002	– 0.02 to 0.02	– 0.02	– 0.02 to – 0.02	0.02	– 0.02 to 0.06
Therapist	3	0.01	– 0.04 to 0.05	– 0.02	– 0.04 to 0.01	0.01	– 0.03 to 0.05
	6	0.03	– 0.02 to 0.09	– 0.004	– 0.04 to 0.03	– 0.001	– 0.03 to 0.03
	12	– 0.0001	– 0.04 to 0.04	– 0.002	– 0.04 to 0.03	– 0.003	– 0.03 to 0.03

Adjusted effects are given in Table 26. Both tables provide the estimate of difference in change between the intervention groups with a 95% CI.

TABLE 26 - Clinical outcome measures (using change from baseline) – adjusted (for age, sex and baseline) linear regression models analysis

FABQ, Fear Avoidance Beliefs Questionnaire; MVK, Modified Von Korff Scale; RMQ, Roland Morris Disability Questionnaire; SF-12, Short Form 12-item health survey.

	Period	N (%)	Mean difference between treatments		p-value
			Mean	95% CI
RMQ	3	513 (73.2)	– 1.1	– 1.74 to – 0.37	0.003
	6	528 (75.3)	– 1.4	– 2.14 to – 0.72	< 0.0001
	12	498 (71.0)	– 1.3	– 2.05 to – 0.55	0.0008
	Longitudinal	1539 (73.2)	– 1.2	– 1.93 to 0.39	0.0001
MVK (disability)	3	519 (74.0)	– 4.3	– 8.20 to – 0.42	0.03
	6	551 (78.6)	– 8.1	– 12.03 to – 4.09	< 0.0001
	12	552 (78.7)	– 8.4	– 12.39 to – 4.35	< 0.0001
	Longitudinal	1622 (77.1)	– 6.5	– 10.54 to – 2.48	< 0.0001
MVK (pain)	3	538 (76.7)	– 6.8	– 10.18 to – 3.46	< 0.0001
	6	569 (81.2)	– 8.0	– 11.73 to – 4.30	< 0.0001
	12	583 (83.2)	– 7.0	– 10.66 to – 3.24	0.0003
	Longitudinal	1690 (80.4)	– 6.7	– 10.44 to – 2.98	< 0.0001
SF-12 (physical)	3	470 (67.0)	2.2	0.72 to 3.68	0.004
	6	497 (70.9)	1.8	0.34 to 3.25	0.016
	12	521 (74.3)	4.1	2.56 to 5.57	< 0.0001
	Longitudinal	1488 (70.8)	4.1	2.08 to 6.04	< 0.0001
SF-12 (mental)	3	470 (67.0)	1.3	– 0.37 to 2.96	0.129
	6	497 (70.9)	2.5	0.78 to 4.26	0.005
	12	521 (74.3)	0.1	– 1.62 to 1.80	0.91
	Longitudinal	1488 (70.8)	0.1	– 2.16 to 2.34	0.91
FABQ (physical activity)	3	471 (67.2)	– 2.6	– 3.64 to – 1.63	< 0.0001
	6	480 (68.5)	– 3.1	– 4.15 to – 2.05	< 0.0001
	12	447 (63.7)	– 3.0	– 4.08 to – 1.88	< 0.0001
	Longitudinal	1398 (66.5)	– 2.8	– 3.98 to – 1.73	< 0.0001
Pain self-efficacy	3	483 (68.9)	3.2	1.34 to 4.93	0.0007
	6	490 (69.9)	4.1	2.30 to 5.98	< 0.0001
	12	456 (65.0)	3.8	1.93 to 5.67	< 0.0001
	Longitudinal	1429 (68.0)	3.9	1.94 to 5.85	< 0.0001

Primary outcomes

Back pain related disability – Roland Morris Disability Questionnaire

Figure 8 shows the change from baseline in the RMQ for each of the follow-up time points by treatment arm. Improvements in LBP-related disability occurred in the CBA and AM arms of the trial, but were of different magnitude and time course. Improvements in the AM arm were on average 1.1 RMQ points, with change occurring between baseline and 3 months and no further change thereafter. The change in the CBA arm was almost double that of the AM arm by 3 months, and the treatment difference continued to widen at 6 and 12 months.

FIGURE 8.

Roland Morris Questionnaire (RMQ) scores (change from baseline). Least square estimates of the mean (95% CI) from the linear regression models (observed data).

The difference between the treatment arms was estimated to be, on average, 1.1 RMQ points at 3 months, rising to 1.4 and 1.3 RMQ points at 6 and 12 months respectively. These differences between treatments were all statistically significant.

Back disability – Modified Von Korff Scale (disability)

Figure 9 shows the change in MVK (disability) by treatment arm. Both treatment arms showed improvements over baseline, but these were of different magnitude and time course. The AM group improved between 0 and 3 months, and declined thereafter. The CBA group showed a greater improvement. At 6 and 12 months the difference between CBA and AM was more than twofold.

FIGURE 9.

Modified Von Korff (MVK) (disability) scores (change from baseline). Least square estimates of the mean (95% CI) from the linear regression models (observed data).

The difference between CBA and AM was estimated to be on average 4.3% at 3 months, 8.1% at 6 months and 8.4% at 12 months. The differences between treatments were statistically significant at all time points.

Back disability – Modified Von Korff Scale (pain)

Figure 10 shows the change in the MVK (pain) by treatment arm. Both treatment arms showed improvements over baseline, but these were of different magnitude and time course. Pain levels in the AM group improved gradually over the 12-month period. Improvements in pain were more than double in the CBA arm at 3 months, and pain continued to improve (although at a slower rate) between 3, 6 and 12 months. The effect on pain peaked at 12 months. These differences between treatments were statistically significant at all time points.

FIGURE 10.

Modified Von Korff (MVK) (pain) scores (change from baseline). Least square estimates of the mean (95% CI) from the linear regression models (observed data).

The difference between CBA and AM was estimated to be on average 6.8% at 3 months, 8% at 6 months and 7% at 12 months.

On average, a participant improved twice as much on the AM+CBA arm as on AM alone.

Secondary and intermediary outcomes

Short Form-12 physical subscale

Figure 11 shows the change in the SF-12 physical subscale by treatment arm. Both treatment arms showed improvements over baseline in the first 6 months of follow-up, but these were of different magnitude and time course. Improvements in the AM arm were very small, and by 12 months there was no difference in comparison with baseline. Improvements in the CBA arm were larger, with the greatest improvement at 12 months.

FIGURE 11.

Short Form 12-item health survey (physical) scores (change from baseline). Least square estimates of the mean (95% CI) from the linear regression models (observed data).

The difference between CBA and AM was estimated to be on average 2.2 at 3 months, 1.8 at 6 months and 4.1 at 12 months. The differences between treatments were statistically significant at all time points.

Short Form-12 mental subscale

Figure 12 shows the change in SF-12 mental subscale by treatment arm. In the AM arm, there was no change in mental-health score across the 12 months. In the CBA arm, there were statistically significant improvements in mental health at 6 months, with the maximum improvement at 6 months. By 12 months, mental-health scores fell, and there was no statistically significant difference between the CBA and AM arms.

FIGURE 12.

Short Form 12-item health survey (mental) scores (change from baseline). Least square estimates of the mean (95% CI) from the linear regression models (observed data).

The difference between CBA and AM was estimated to be on average 1.3 at 3 months, 2.5 at 6 months, and 0.1 at 12 months.

Fear Avoidance Beliefs Questionnaire

Figure 13 shows changes in fear avoidance beliefs by treatment arm. In the AM arm, there was no evidence of change in fear avoidance beliefs across the 12 months. In the CBA arm, there were substantial improvements in fear avoidance at 3, 6 and 12 months. Improvement occurred between 0 and 3 months and was sustained to 12 months.

FIGURE 13.

Fear Avoidance Beliefs Questionnaire (FABQ) (physical activity) scores (change from baseline). Least square estimates of the mean (95% CI) from the linear regression models (observed data).

The difference between CBA and AM was estimated to be on average 2.6 at 3 months, 3.1 at 6 months and 3.0 at 12 months. The differences between treatments were statistically significant at all time points.

Pain self-efficacy

Figure 14 shows pain self-efficacy by treatment arm. There was no discernible change in pain self-efficacy in the AM arm. Self-efficacy improved in the CBA arm, with maximal improvements at 6 months, and sustained improvement at 12 months.

FIGURE 14.

Pain Self-efficacy scores (change from baseline). Least square estimates of the mean (95% CI) from the linear regression models (observed data).

The difference between CBA and AM was estimated to be on average 3.2 at 3 months, 4.1 at 6 months and 3.8 at 12 months. The differences between treatment arms were statistically significant at all time points.

Changes in employment and hours worked

The number and proportion of participants who were receiving any entitlements/benefits were very small for both therapy groups. There was a small difference in the numbers of people changing their hours of work between the groups (in favour of the CBA) but these differences were not statistically significant. A slightly higher proportion of people randomised to CBA reported having sick days at 3 and 6 months, but this difference was small and not statistically significant. Overall there were no differences in the numbers of days off sick. Data on sickness, benefit claim and employment are given in Appendix 13.

Global indicators of change, benefit and satisfaction with treatment

Self-rated global assessments of change in back pain and satisfaction with treatment favoured CBA over AM (data shown in Table 27). Differences were statistically significant for all comparisons regardless of the time point of follow-up.

TABLE 27 - Change in back pain and treatment satisfaction (n, %)

p-value used chi-squared test from trend.

Month		AM	AM+CBA	p-value
Change in back pain
3	Completely recovered	1 (0.5)	3 (1)	< 0.0001
	Much improved	8 (4)	95 (27)
	Slightly improved	45 (24)	98 (28)
	No change	93 (50)	124 (35)
	Slightly worsened	29 (16)	20 (6)
	Much worsened	8 (4)	12 (3)
	Vastly worsened	3 (2)	2 (1)
6	Completely recovered	2 (1)	31 (12)	< 0.0001
	Much improved	19 (10)	105 (41)
	Slightly improved	35 (19)	82 (32)
	No change	81 (44)	133 (11)
	Slightly worsened	27 (15)	14 (1)
	Much worsened	18 (10)	20 (2)
	Vastly worsened	4 (2)	1 (1)
12	Completely recovered	6 (3)	44 (11)	< 0.0001
	Much improved	24 (12)	106 (27)
	Slightly improved	32 (16)	85 (22)
	No change	89 (45)	122 (31)
	Slightly worsened	24 (12)	20 (5)
	Much worsened	17 (9)	9 (2)
	Vastly worsened	5 (3)	9 (2)
Satisfaction
3	Very dissatisfied	19 (11)	20 (6)	< 0.0001
	Somewhat dissatisfied	23 (13)	11 (3)
	Neither satisfied nor dissatisfied	75 (44)	46 (14)
	Somewhat satisfied	38 (22)	117 (35)
	Very satisfied	17 (10)	137 (41)
6	Very dissatisfied	20 (12)	28 (8)	< 0.0001
	Somewhat dissatisfied	20 (12)	23 (9)
	Neither satisfied nor dissatisfied	84 (52)	71 (20)
	Somewhat satisfied	24 (15)	111 (31)
	Very satisfied	15 (9)	110 (31)
12	Very dissatisfied	14 (9)	19 (6)	< 0.0001
	Somewhat dissatisfied	22 (15)	21 (6)
	Neither satisfied nor dissatisfied	72 (48)	76 (23)
	Somewhat satisfied	30 (20)	104 (32)
	Very satisfied	13 (9)	108 (33)

By 12 months, 60% of participants in the CBA arm rated that they had improvements, whereas 31% in the AM arm reported improvement. Less than 25% of people randomised to AM reported a decline at 12 months, in comparison with 9% in the CBA arm. Similar trends were noted in the benefit question. Levels of satisfaction were greater in the CBA arm in comparison with AM.

Numbers needed to treat

Using a threshold of change of 30% to determine an MCID approximately seven participants would be needed on the AM+CBA arm to see one participant improve. The same estimate was gained using the MVK scales.

The numbers needed to treat using the global indicator of change scale was 7 (improvement versus stability or decline) or 3 (improvement and stability versus decline).

Estimates of the standardised difference (effect size)

The effect sizes for each of the outcomes are derived from the adjusted treatment estimates (Table 28).

TABLE 28 - Effect sizes of the clinical outcome measures at 12 months based on unadjusted differences estimated in regression modelling

FABQ, Fear Avoidance Beliefs Questionnaire; MVK, Modified Von Korff Scale; RMQ, Roland Morris Disability Questionnaire; SF-12, Short Form 12-item health survey.

	Effect size
RMQ (absolute)	0.31
MVK (disability)	0.41
MVK (pain)	0.37
Pain self-efficacy	0.35
FABQ	0.47
SF-12 (physical)	0.45
SF-12 (mental)	0.07

Adverse events

Only one adverse event occurred during the study, and this involved a participant receiving AM. An adverse event notification (Appendix 10) was submitted to West Midlands MREC on 3 August 2006. This patient was admitted to hospital on 11 July 2006 with acute spinal cord compression, which was subsequently decompressed operatively. Following discharge, the participant developed a pulmonary embolus as a complication, which required readmission. Final discharge was on 27 July 2006. As part of the ‘control intervention’, the patient received TheBack Book and a single session of advice. There is no indication that advice to stay active increases the risk of spinal cord compression, which affects a small proportion of back pain patients. The Back Book gives details of warning signs and advises immediate consultation with a doctor. The patient followed this advice and therefore the emergency was dealt with appropriately without delay.

Subgroup analysis

We undertook three prespecified subgroup analyses and one additional subgroup analysis based on the RMQ baseline score. For each of the three subgroups, no statistical interactions were found between the subgroups and the treatment assigned.

Sensitivity analysis

BeST interview study

Interviewee recruitment

Participants were eligible to be contacted for interview if they had agreed to this on their initial consent form. A total of 40 of the 701 BeST trial participants were invited for interview by a phone call from a researcher (V.N.) who outlined that the face-to-face interview would last approximately 45 minutes and be held at a venue of their choice (Figure 15). Potential participants were informed that the interviews would be about their experience of back pain and their views of the interventions tested in the trial, and were briefed that the interviews would be recorded if they were in agreement. Four interviews were undertaken as a pilot and are not included in the final analysis. Two potential participants declined to be interviewed: one because of an operation; the other did not give a reason.

FIGURE 15.

CONSORT diagram of interviews.

Description of intervention

The interview

A semi-structured interview schedule was designed and developed by undertaking four pilot interviews, which were not subsequently used in the analysis. In the final interview schedule, interviewees were asked about their lives in general and their back pain, as this formed the context for understanding their lived experience before and after the intervention.

The following questions regarding the interventions were asked of all interviewees:

• Which treatment did you have on this trial? Could you tell me what you thought of that?

• Were you given any written material to do with your back?

Questions asked only of the AM+CBA interviewees were:

• What do you think about physical activity for your back?

• How do you decide when to go back to work/normal activities?

• How do you feel about the whole process of back skills training?

• How did you feel about the logbooks and goal setting?

• How did you feel about being in a group?

• Any other comments about the sessions?

Additional questions were included in the interview schedule and are provided in Appendix 14. However, these were not part of the commissioned project and hence are not reported here.

Prompts were used to expand on questions to elicit personal experience and opinions surrounding their LBP and the interventions. For example, if a person mentioned that a component of the intervention was not useful they were asked why, what should be changed and whether it would be useful to others. At the end of the interview, participants were asked if there was anything in the interview they wished to have removed from the audio recording, and were reminded that the interview was confidential to the research team and would be made anonymous when transcribed.

One participant had not finished the group sessions by the time of the interview, so was contacted later for a short telephone interview to ask about the last group session. One participant asked to see a copy of the transcript which was sent with a prepaid envelope for any comments (this was duly returned with minor typographical changes only) and another asked for a copy of any excerpt which may be used in the final report, which was sent and elicited no response.

Data coding and analysis

All interviews were carried out by one researcher (V.N.). At the time of interview the interviewer was not aware of the detailed content of the course sessions, to reduce any tendency to ask leading questions. During analysis V.N. was briefed fully on the content to enable her to understand which part of the intervention interviewees were talking about.

Audio recordings of interviews were transcribed and checked by the interviewer. All transcripts were entered into NVivo, a computer software tool for managing qualitative data (NVivo, QSR, Portsmouth, UK). Thematic codes were developed based on the interview schedule, the components of the intervention and issues arising in the interviews. The research team agreed codes and their definition. All data were double coded and coding was compared. Where there were inconsistencies in coding these were discussed and agreed. The data were extracted from all interviews for each coded theme relevant to the experience of AM or AM+CBA. The number of participants mentioning a theme was counted, the content of the coded data was summarised and illustrative quotes were identified for presentation in the results.

Results

The results describe the interviewees’ experiences of back pain, being on the trial and their account of the interventions they received. The first section summarises the accounts of receiving AM only. We then draw on the accounts of those receiving AM+CBA and summarise their account of the AM component, the individual assessment and the group sessions. The results include numbers of interviewees who talked about themes. Each interviewee may have talked about one or more of these themes.

Conclusion

The CBA group intervention was added value for the trial participants over AM. Almost all participants were already familiar with the AM approach.

Most of the interviewees in the CBA intervention arm of the trial and who attended at least one session picked up key messages from the intervention and two-thirds talked about a reduction in fear avoidance and changes in their behaviour. Over half found the exercises helpful and many of these had incorporated exercise into their daily lives. Other aspects of the intervention were useful to some and not others, suggesting that offering the range of techniques is important.

Although participants of the BeST trial were considered to have received the intervention only if they had attended three or more sessions, the interview data suggest that fewer than three sessions may still be useful for some participants.

A minority of interviewees were unhappy with certain aspects of the group intervention and several of these felt they knew the content of the intervention already.

Issues that were raised by the interviews that should be considered in translating the intervention into practice were as follows:

• The individual assessment, although an unusual component of CBA, sets the agenda for the groups, begins the CBA and allows a relationship to be established with the therapist. It is an important element of the intervention.

• Exercises were considered an important component by most interviewees and demonstration and tailoring of exercises was valued. Great emphasis could be placed on tailoring and demonstrating exercises during the sessions.

• Participants can benefit even if they choose to use only certain aspects of the intervention or attend a limited number of groups. Translation into practice should consider giving participants more information about the sessions ahead of time, so that people are able to choose what to engage with and when. Although not mentioned specifically, a rolling programme of group meetings may mean that participants have a greater choice about being able to join when they choose.

• Group sessions where participants share experiences can provide them with reassurance, lessen isolation and enable participants to learn strategies from each other.

• Consider increasing training for therapists in group facilitation to increase sharing of experiences.

Introduction

We aimed to assess the cost–utility of adding a CBA programme to AM. Our primary outcome was cost–utility expressed as the incremental cost–utility ratio (ICUR),131 the shadow ‘price’ of CBA, which indicates the cost of gaining one additional QALY relative to the current therapy.

We assessed cost–utility over a 12-month time frame, and the base year was 2008. Costs were expressed in UK pounds (£) and actualised to 2008 using the Office of National Statistics Retail Price Index132 where required. Neither costs nor outcomes were discounted as the time frame for the analysis was 12 months.

We considered cost–utility from both the UK NHS and the general health-care perspective. 131 The NHS perspective included all health-care costs supported by the NHS to deliver AM or AM+CBA and to provide health care associated with back pain treatment and symptoms in the 12-month period after randomisation. The health-care perspective is broader and included both the total health-care costs to the NHS and the cost of privately purchased goods and services (medical and non-medical) related to LBP and private transport.

The base-case analysis was based on the within-trial resource consumption. To illustrate the generalisability of the cost-effectiveness analysis, we developed an alternative cost–utility scenario based on the potential resource consumption for the CBA programme delivered within NHS structures where therapists are almost entirely based at one site (and hence there is no need to allow for travel time and expenses). The within-trial analysis was the primary economic analysis and is presented in this chapter. The generalised NHS scenario was a secondary analysis and appears in Appendix 15.

Data

The cost of cognitive behavioural therapy in the randomised controlled study (within-trial analysis)

Unit costs for the AM+CBA intervention were estimated from the trial data, using resource use described above and other staff cost information retrieved from published sources. 134 The unit costs for the components of the CBA programme were staff salary costs (including overheads and capital costs), training costs, consumables and equipment and travel costs. The total cost per CBA cycle was then computed for each group and allocated to each participant in relation to the number of sessions attended and the number of participants per session.

In the cost–utility analysis, CBA was estimated using the actual resource used to deliver the programme, i.e. the total working time provided within the context of the trial. Staff salary costs of one CBA cycle were then calculated as the cost of the total number of contact and non-contact hours spent by a therapist to deliver CBA multiplied by the normal salaried staff hourly rate.

The effective staff hourly rates paid in the trial were not likely to represent the normal salaried cost of staff. As with many trials, some of the activities were paid for under casual working arrangements. Although these include a holiday enhancement, they do not account for the broader costs of employment if undertaken on a salaried basis. Therefore the total normal staff hourly cost was computed, inflating the effective salary rates paid in the trial to include gross salary, employers’ and overhead costs, capital costs and capital overheads. Gross salaries were calculated based on effective seniority for each therapist combined with published NHS pay scales. 138 Salaries were inflated with 8% employers’ National Insurance costs and 14% for employers’ contribution to superannuation pay. 134 Salaries were then inflated at the year 2008 using percentage pay increases for the year 2006–7 (5.5%) reported in Curtis. 134 As salary increase data for the year 2007–8 are not available the same percentage increase was compounded for that period.

Overheads were applied to total staff time in a proportion of 5% of the salary for direct revenue overheads and £2904 per year for indirect overheads. 134 Capital costs and overheads were valued at £2752 per year. 134 Hourly rates were then computed from the total salary costs, assuming a total working time of 5 days and 52 weeks, less 27–33 days of holiday per year, depending on the therapist’s salary band, 8 days for statutory holidays and an average of 12 days’ sick leave and 3 days’ training. 138 Hourly rates used in the cost-effectiveness are summarised in Table 38.

Finally, the total salary cost of one cycle of CBA was computed by multiplying the number of worked hours per cycle by the therapists’ normal hourly salary rates. The total working time was the sum of contact and non-contact time observed for each therapist in the trial. Therapists’ travel time was included in non-contact time, as the intervention was delivered by therapists seconded to the trial from other institutions where they had a permanent position.

The total cost of training, supervision, consumables and equipment was allocated top-down to each CBA cycle, based on the number of groups in the CBA programme.

The cost of consumables and equipment was computed by multiplying the number of items bought for the trial by their acquisition costs.

Training costs were estimated by summing the cost of external lecturing fees, intervention materials, lecture room fees, and residential expenses, including accommodation, travel and subsistence costs. Salary costs of therapists attending training were not added, to avoid double counting as working time spent on training was already included in the computation of the therapists’ hourly rate. Training costs were allocated by CB cycle rather than as a component of salaries, as we requested that the staff did not deliver CB therapy outside the trial context. The CBA training course lasted 2 days. We assumed that therapists would have refresher training/mentoring for an average of 2 days a year, therefore we assumed that the costs of training would be repeated yearly. This seemed a reasonable approximation which accounts for the cost of continuous training for specialist groups, but excludes other types of training that may be unrelated to the CBA programme. 139

Finally, therapists’ travel expenses were computed from actual travel mileage declared by therapists and paid at £0.40/mile. Travel costs were allocated based on the actual travelled mileage for each group.

We did not incorporate a separate cost for space used within NHS physiotherapy departments. The space used for physiotherapy conducted on NHS premises (community and secondary care) comprises treatment cubicles, office space and a gym space with the space being used flexibly depending on the need for privacy, equipment, etc. Some patients may be treated in the gym and others in a treatment cubicle. We considered that the overheads in the Personal Social Services Research Unit costs would include the cost of accessing these spaces (on average). In the case of CBA, sessions were run for small groups (up to eight people) in spaces comparable to the gym or office used in normal NHS physiotherapy practice. From our experience, these two spaces are likely to be used for the majority of the patients, therefore we chose to use the estimated overheads of the Personal Social Services Research Unit as they incorporate space requirements. Although it could be argued that revising the cost of capital upwards to exclude the use of cubicles should be undertaken, we believe this is likely to have negligible impact on the cost of the CBA.

The total cost of one session was then obtained as a proportion (one-sixth) of the total cost of one CBA cycle, including supervision, training, consumables, equipment and travel time. The cost of one session was allocated to participants in proportion to session attendance, i.e. dividing the cost of one session by the number of participants in each group and session. Finally, the cost of CBA was computed for each participant multiplying the cost per patient of each session by the number of sessions attended.

The cost of the assessment visit was the cost of 1.5 hours’ contact time at each therapist’s hourly rate, and the cost of consumables and of CBA materials given to the participant (£2.60 in total per participant assessed). The cost of assessment was attributed to those participants who were assessed.

Methods of analysis

The base-case economic evaluation was conducted using the sample of participants who had completed questionnaires at baseline and at least at 3 months.

Results

Cases included in the economic evaluation

The proportion of returned questionnaires valid for the economic evaluation at 3, 6 and 12 months of follow-up was between 70% and 77% (Figure 16).

FIGURE 16.

Flow chart of loss to follow-up – economic data.

The cost–utility analysis was conducted using the cases with complete follow-up at least for 3 months, n = 528. This was the majority of cases in the trial (75%). Individuals who completed questionnaires at 6 or 12 months but not at 3 months were excluded from the cost-effectiveness analysis.

For presentation purposes only, the descriptive analysis of mean resource use, costs and quality of life scores over a 1-year period was conducted on cases with complete follow-up at 12 months, 70% of the participants in the trial (n = 490).

There was no difference in length of follow-up between the AM+CBA and the AM groups for cases included in the economic evaluation. Mean follow-up for cases included was 360 days (SD 62) in the AM group and 365 days (SD 52) in the AM+CBA group. Cases excluded had shorter follow-up: 85 days (SD 124) and 68 days (SD 88) in the AM and AM+CBA groups respectively.

Utility scores

There were some differences in mean baseline utility score between AM (0.606) and the AM+CBA arm of the trial (0.574), although these differences were not statistically significant (data shown in Table 40). At baseline women expressed worse utility (0.569) than men (0.607), although differences were not statistically significant (p = 0.11). As would be expected, baseline utility differed according to baseline severity. Participants with RMQ scores ≥ 4 or more had a mean baseline utility of 0.552 whereas those with RMQ scores < 4 had a baseline utility of 0.750 (mean difference 0.198; p < 0.001).

TABLE 40 - Mean utility scores at baseline, 3, 6 and 12 months

SE, standard error.

For numbers of participants in sample, see Figure 16.

	AMa		AM+CBAa		Mean difference
	Mean	SE	Mean	SE
Baseline (n = 528)	0.606	0.019	0.574	0.014	– 0.032
3 months (n = 528)	0.567	0.022	0.628	0.014	0.061
6 months (n = 527)	0.593	0.021	0.630	0.016	0.037
12 months (n = 490)	0.592	0.023	0.640	0.016	0.048

Figure 17 shows the mean utility scores for AM+CBA at baseline and follow-up time points. Utility scores improved during follow-up for the AM+CBA group, with changes evident at 3 months and sustained to 12 months. By the end of follow-up, the mean utility scores had risen by 0.066, whereas in the AM group mean utility dropped initially at 3 months but improved in the follow-up, and remained unchanged over the 12 months, with a drop of – 0.014 at 12 months in comparison with baseline.

FIGURE 17.

Mean utility scores at baseline, 3, 6 and 12 months. Vertical bars show 95% confidence intervals; data points are offset to improve visibility. *Statistically significant difference between AM and AM+CBA, p = 0.001.

Quality-adjusted life-years

We calculated the QALYs as the weighted mean of utility weights between two measurements and time spent in the study. Figure 18 gives the cumulative QALYs over 3, 6 and 12 months. These are the total QALYs from baseline up to each time point. AM+CBA accumulated 0.662 mean QALYs and AM accumulated 0.624 mean QALYs over the 12-month follow-up, with an unadjusted difference of 0.038 QALYs at 12 months (Figure 18). These are crude estimates and are the mean of the area under the curve, obtained from the crude QALY scores multiplied by the length of follow-up. These estimates are illustrated here only for reference because they are not the final estimates of incremental QALYs based on the regression approach used to estimate the cost-effectiveness of CBA.

FIGURE 18.

Cumulative quality-adjusted life-years (QALYs) – unadjusted.

Cost-effectiveness

Separate models are presented for the NHS and the general health-care perspectives. Expected incremental costs and incremental QALYs are reported in Tables 50 and 51.

TABLE 50 - Incremental costs and QALYs, AM+CBA compared with AM – NHS perspective

ICUR, incremental cost–utility ratio.

	AM	AM+CBA	Difference
Total NHS costs	£278.99	£457.05	+ £178.06
Total QALYs	0.604	0.703	+ 0.099
ICUR			£1786

TABLE 51 - Incremental costs and QALYs, AM+CBA compared with AM – NHS perspective, by subgroup

ICER, incremental cost-effectiveness ratio; QALY, quality-adjusted life-year; RMQ, Roland Morris Disability Questionnaire.

	AM		AM+CBA		Incremental cost (£)	Incremental QALYs	ICER (£)
	Cost (£)	QALY	Cost (£)	QALY
Males (n = 214)	186.72	0.640	397.47	0.727	210.75	0.087	2422
Females (n = 314)	251.34	0.583	413.56	0.694	162.22	0.111	1461
Older than 60 years (n = 220)	218.53	0.563	454.15	0.690	235.62	0.127	1855
Younger than 60 years (n = 308)	228.11	0.620	372.65	0.714	144.54	0.094	1538
Duration of pain ≤ 3 years (n = 154)	259.35	0.681	383.74	0.749	124.39	0.068	1829
Duration of pain > 3 years (n = 374)	210.74	0.559	415.23	0.688	204.49	0.129	1585
RMQ score at baseline ≥ 4 (n = 442)	236.76	0.583	427.20	0.708	190.44	0.125	1524
RMQ score at baseline < 4 (n = 86)	151.53	0.729	302.18	0.722	150.65	–0.007	AM+CBA dominated

NHS perspective

From the NHS perspective, CBA was associated with an increase in health-care costs, + £178.06, and an improvement in QALYs, + 0.099. The ICUR was £1786, indicating that CBA is cost-effective at the currently accepted decision-maker’s cost-effectiveness threshold.

The cost–utility was not sensitive to cost outliers. The ICUR obtained excluding the cases with costs higher than the 90th percentile was £1797, with an incremental cost of £162.44.

The cost–utility was sensitive to the inverse weighting method. Excluding inverse weighting, the model estimated + 0.061 incremental QALYs and + £177.43 incremental costs (from the bootstrap procedure), yielding an ICUR of £2909. This ICUR remains nevertheless well below the recognised decision-maker’s cost-effectiveness threshold.

Uncertainty

To assess the variability of this estimate, we plotted the incremental costs and incremental QALYs on the cost-effectiveness plane (Figure 19) calculated for each of the 1000 bootstrap estimates upon which the estimation of the ICUR was made. Figure 19 shows that the incremental cost and QALY of CBA are both always positive. Therefore CBA is never dominant (i.e. less costly and more effective).

FIGURE 19.

Cost-effectiveness plane, incremental costs and incremental QALYs, AM+CBA versus AM – NHS perspective.

To visually assess the range of variation for the ICER, we also plotted the cost-effectiveness thresholds at £500 and £4000. These are the dotted and dashed lines in Figure 19.

As the ICUR is the ratio of incremental costs and QALYs, the slope of the line joining any of the points with the origin is a graphical representation of the ICER. Therefore all points above the £500 line and below the £4000 line are points associated with an ICER of between £500 and £4000.

We summarised the variations of the ICUR in a cost-effectiveness acceptability curve (Figure 20). This is the probability that CBA is cost-effective at the decision maker’s cost-effectiveness threshold λ, the willingness to pay for one QALY, plotted here for the range £0–£35,000. The probability of CBA being cost-effective reaches 90% at about £3000 and remains at that level or higher above that threshold. At £30,000 CBA has an almost 100% probability of being considered cost-effective.

FIGURE 20.

Cost-effectiveness acceptability curves, AM+CBA versus AM – NHS perspective.

Subgroup analysis

We estimated the cost–utility of CBA by age and sex to test the robustness of recommendations within the general population with LBP. The descriptive analysis of costs and QALYs (see Costs) showed that costs and QALYs are not likely to vary by sex or age. However, back pain severity (RMQ scores) was shown to be associated with higher costs. The subgroup analysis reported here (Table 51) shows confirmatory evidence that age, sex and duration of back pain do not have a large impact on the cost-effectiveness of CBA, which remains well below currently recognised cost-effectiveness thresholds. However, in the subgroup analysis for RMQ scores we found that CBA was associated with positive incremental costs whilst incremental QALYs were negative in the low-score range (RMQ < 4); therefore CBA was ‘dominated’ by AM in this group, suggesting the CBA may not be cost-effective in less severe back pain.

The cost-effectiveness acceptability curves by subgroup (Figure 21) show that the probability of CBA being cost-effective is between 80% and 99% for age and sex at cost-effectiveness thresholds of about £5000. Over this threshold, the probability of CBA being cost-effective remains constant. For individuals with less severe problems from their back pain (RMQ scores < 4) the probability that CBA is cost-effective is less than 40% over the customary range of decision-makers’ willingness to pay.

FIGURE 21.

Cost-effectiveness acceptability curves, AM+CBA versus AM – NHS perspective, by subgroup.

General health-care perspective

From the health-care perspective, CBA was more expensive by approximately £314, when calculated from within-trial costs. With an improvement in QALYs of + 0.099, the ICUR was £3093, with CBA still cost-effective at the currently accepted decision-makers’ cost-effectiveness threshold (Table 52).

TABLE 52 - Incremental costs and QALYs, AM+CBA compared with AM – general health-care perspective

ICUR, incremental cost–utility ratio; QALY, quality-adjusted life-year.

	AM	AM+CBA	Difference
Total health-care costs	£723.41	£1037.78	+ £314.37
Total QALYs	0.604	0.703	+ 0.099
ICUR			£3093

The cost–utility analysis was not sensitive to cost outliers. The ICUR computed using cases with costs lower than the 90th percentile was £2098, with an incremental cost of £189.62.

Uncertainty

The incremental costs and QALYs of CBA under the general health-care perspective remain positive, indicating no dominance for CBA (Figure 22).

FIGURE 22.

Cost-effectiveness plane, AM+CBA versus AM – general health-care perspective.

The probability that the ICUR is below the accepted cost-effectiveness threshold is lower than that under the health-care perspective for low levels of willingness to pay.

At a threshold of £5000, the probability that CBA is cost-effective is 89% and this increases to 99% at about £10,000, after which it remains constant (Figure 23).

FIGURE 23.

Cost-effectiveness acceptability curves, AM+CBA versus AM – general health-care perspective.

Subgroup analysis

As for the NHS perspective, the subgroup analysis (Table 53) shows that age, sex and duration of pain do not have a large impact on the cost-effectiveness of CBA, well below currently recognised cost-effectiveness thresholds; however for individuals with low back pain severity, CBA is dominated, i.e. more costly and less effective than current practice.

TABLE 53 - Incremental costs and QALYs, AM+CBA compared with AM – general health-care perspective, by subgroup.

ICER, incremental cost-effectiveness ratio; QALY, quality-adjusted life-year; RMQ, Roland Morris Disability Questionnaire.

	AM		AM+CBA		Incremental cost (£)	Incremental QALYs	ICER (£)
	Cost (£)	QALY	Cost (£)	QALY
Males (n = 214)	433.89	0.640	774.22	0.727	340.33	0.087	3912
Females (n = 314)	522.26	0.583	817.20	0.694	294.94	0.111	2657
Older than 60 years (n = 220)	469.61	0.563	938.46	0.690	468.85	0.127	3692
Younger than 60 years (n = 308)	499.96	0.620	705.33	0.714	205.37	0.094	2185
Duration of pain ≤ 3 years (n = 154)	424.64	0.681	653.55	0.749	228.91	0.068	3366
Duration of pain > 3 years (n = 374)	512.74	0.559	866.76	0.688	354.02	0.129	2744
RMQ score at baseline ≥ 4 (n = 442)	499.10	0.583	861.95	0.708	362.85	0.125	2903
RMQ score at baseline < 4 (n = 86)	405.15	0.729	506.48	0.722	101.34	– 0.007	AM+CBA dominated

At the £10,000 cost-effectiveness threshold, CBA is cost-effective with a probability of at least 80% for all age and sex subgroups, and remains constant above that threshold (Figure 24). However, as for the NHS perspective, the probability that CBA is cost-effective in individuals with RMQ scores < 4 is low, below 40% at all accepted willingness-to-pay thresholds.

FIGURE 24.

Cost-effectiveness acceptability curves, AM+CBA versus AM – general health-care perspective, by subgroup.

Aims and overview of findings

The discussion focuses on the interpretation of the findings, internal and external validity of the trial and associated analyses, comparison with other trial data, implications for clinical practice in the NHS and further research.

Low back pain is a common problem. Our aim was to optimise the design of a CBA package that could be delivered within the NHS, and to estimate the effectiveness, cost-effectiveness and acceptability of CBA in addition to best practice advice in primary care using a pragmatic trial design.

This was a definitive large-scale randomised controlled trial, which has demonstrated the long-term effectiveness and cost-effectiveness of CBA in treating subacute and chronic LBP. The clinical and cost-effectiveness outcomes are likely to make this intervention attractive to patients, clinicians and purchasers. Our short-term (3-month) clinical effects are similar to those found in high-quality studies of other therapies including manipulation, exercise and acupuncture. 144–146 Strikingly, and in contrast to many previous studies, the benefits we observed were maintained over the long term (12 months). The intervention is extremely cost-effective from an NHS and a societal perspective; cost per QALY is less than half that of competing interventions for LBP. Finally, since the intervention can be delivered by existing NHS staff following a brief, 2-day training session, the back skills training programme could be implemented into the NHS with relative ease.

External validity and generalisability of findings

The external validity of the trial is good. The trial sample was recruited from a range of general practices, including single-handed practices. The regions we recruited ranged from inner city to rural areas, and included areas of significant deprivation. The intervention was delivered by NHS staff and as such, the trial achieved the aim of testing the intervention in an everyday setting and in a sample that was representative of a range of NHS users and providers.

The average age of participants was somewhat older than seen in previous trials. This may be attributable to the ageing of the population. However, in contrast to many previous trials of LBP we did not specify an upper age limit to our entry criteria. Concerns about osteoporosis often predicate the exclusion of older people from trials of spinal manipulation.

The sample is representative of people who will accept the invitation to participate in CBA. This cannot be disentangled from the invitation to participate in the trial. People who were potentially eligible but declined were somewhat younger, but no different in other respects. The most common reason for declining was that people did not want to complete questionnaires during follow-up, and that despite reporting troublesome LBP, people did not consider it bad enough to engage in treatment. Neither of these factors has a significant impact on the interpretation of the findings. There is no suggestion of bias arising from selective loss to follow-up or withdrawal.

Overall, the trial sample was broadly representative of the ethnic mix of the UK. 147 Four per cent of people recruited were of Asian origin, although this ethnic group was predominantly recruited from the Birmingham cluster.

Critique of methods and threats to internal validity

We raised concerns with regard to the psychometric properties of the RMQ before starting the trial, and chose to adopt two primary outcome measures at the outset (the RMQ and the MVK scores). Difficulties relating to the ceiling effect of the RMQ have been observed in several studies, and more recent trials of back pain have chosen to define eligibility criteria of having an RMQ score of > 4 to avert some of these problems. 94,148,149 However, this approach reduces external validity – the RMQ is not used in clinical practice to triage or characterise LBP. Any estimates of the cost-effectiveness using a higher RMQ score as an entry criterion should also incorporate an estimate of the costs of screening all people on the RMQ and the quality of life consequences of misclassification with the RMQ. We are not aware of any studies that do. We have presented an analysis of the subgrouping by RMQ score to enable closer comparison with other studies that have used the threshold approach. Interestingly, we report that the CBA intervention is unlikely to be cost-effective in people with low RMQ scores at baseline. This analysis should be treated as exploratory – the difference is almost entirely on differences in cost-expenditure and not associated with clinical outcomes.

Instead of using an RMQ cut-off point, we chose to include people who reported their back pain as at least moderately troublesome. The simplicity and clinical relevance of the troublesome question is widely accepted. 150,151 The BeST trial included people who have troublesome back pain but who scored < 4 points on the RMQ. Hence, we had greater potential to encounter analytical problems relating to ceiling effects in the RMQ. We provided strong justification for the between-group difference that we wished to detect at the trial protocol stage because we anticipated that the difference between groups would be smaller than in previous trials because of the score range within which we were operating. This point has been supported by Lauridsen et al. ,125 who observed that for individuals who scored < 6 on the RMQ, an MCID was estimated to be 2, but for those individuals with scores between 6 and 12, the MCID was in the order of 6. There have been a number of data-driven attempts to determine the MCID for the RMQ, although these should be interpreted with caution because many are based on samples that have used a threshold of > 4 RMQ points to determine eligibility.

The scaling properties of the RMQ also make calculation of numbers needed to treat difficult. Recommendations suggest that a threshold of 30% change should be used to determine a minimally important change on the measure,124 but because the RMQ is a categorical scale, substantial errors are incurred in the calculation (because of rounding errors). We also calculated the numbers needed to treat for the MVK. There is no consensus on the MCID for the MVK, and we used guidance in the chronic pain literature that a 30% change in a 10-point numeric rating scale is clinically significant. 152–154 All estimates of numbers needed to treat have to be interpreted with caution.

We specified in the original protocol that we would undertake an interim analysis to compare the psychometric properties of the two primary outcome measures. Standard psychometric approaches did not, however, yield any useful data on which to base a choice – the main difficulty with the RMQ is the skewness in the distribution of raw scores, which severely limits what type of statistical methods can be applied to investigate between-group differences, particularly where these involve multilevel modelling. We opted to retain two primary outcomes and, reassuringly, the findings are consistent across both.

A potential criticism is that although we randomised individuals to the treatment options, we did not randomly allocate therapists to participants. It was impossible to achieve this logistically. Also, the method of allocation we studied is reflective of the intervention delivery mechanism if generalised into primary care.

How have we determined that this is a clinically important result?

Interpretation of any trial is complex. The first step is to rule out that the differences we observed between treatment groups were the result of chance. The study was powered to be definitive, and the statistical significance of the comparisons we drew across a range of measures supports rejecting the null hypothesis (there is no difference between the groups) at low levels of uncertainty. Steps were taken to minimise type I (false-positive) error by including prespecified analysis plans, setting the alpha to 0.01, and prespecification of primary outcome measures.

Determining clinical significance is more challenging. We specified a priori (protocol version 1) that we wished to detect a difference of 1.4 RMQ change score points between two treatment comparators (i.e. for the new treatment to be improved by about half as much again when compared with the control group). For the purposes of the sample size estimate, this approximated to a standardised difference of 0.35 (assuming data were normally distributed and an SD of 4.0) at 12 months. We modified the trial design before starting recruitment to an unbalanced randomisation, and chose to retain the sample size at the original 700 patients. One eventuality of this scenario was that we might have reduced power by a small amount, or that we would only be able to detect larger effect sizes. Ultimately, with retention remaining higher than anticipated and little or no clustering in the data, there was minimal loss of power to detect the differences that we specified originally. We were able to demonstrate a statistically significant effect for our primary outcomes within the range we prespecified as the MCID. Participants who were randomised to CBA received on average, twice as much improvement in outcome across a range of measures. As a consequence, we are confident that we have demonstrated a difference that is both clinically important and statistically significant. The effect size is similar in magnitude to those found in the more robust trials used to inform the draft National Institute for Health and Clinical Excellence (NICE) guidance on LBP. Estimates of the standardised difference for the RMQ have to be interpreted with caution because the distribution of scores was significantly skewed. When we looked at scores integrated over the entire 12 months, the treatment differences were a little smaller. This was because differences in the early phase of the follow-up were smaller than those being observed at the later stages.

Comments on best practice advice

Recommendations to support best practice advice are based largely on expert opinion supported by epidemiological data highlighting the importance of physical activity and a limited amount of trial evidence. 36,155 A trial comparing sham and biomedical advice demonstrated a small short-lived effect of advice on subacute LBP. 38

The quantitative and qualitative data collected in the best practice advice arm provide insight into the natural history of LBP when managed with a low-intensity intervention. The intervention we selected for the control arm was consistent with international LBP guidance for primary care, and allows comparison across a range of studies that have used a similar control. Back pain improved in the best practice arm. Just over one-quarter of the people who received best practice improved in their symptoms at 1 year (28%). This is broadly consistent with the natural history of back pain in people – approximately 27% will be pain-free or improved at 12 months. 156–158

Comparisons between epidemiological data and the BeST trial data are difficult to draw – we targeted people who had subacute and chronic LBP and experienced trouble with their backs and in whom the recovery rate might be expected to be lower than in the general population with back pain. Within our trial, The Back Book and brief session of advice had no lasting effect on beliefs about back pain or confidence to deal with the pain. This is disappointing; The Back Book was designed to target these beliefs and clearly fails to achieve any lasting change in mindset for the great majority of people. Reinforcing advice at regular intervals might improve the efficacy of the best practice, but evidence is needed to demonstrate this.

Comments on the cognitive behavioural approach tested in BeST

Cognitive behavioural interventions take a variety of guises, varying in formulation, intensity, depth and integration with other treatments. The CB method is based on accessing and modifying beliefs that are conceptualised at three levels – negative automatic thoughts, ‘assumptions’ or ‘rules’, and core beliefs. The latter are deeply held core beliefs. We included the negative automatic thoughts, but not the other belief levels, because their relationship to LBP is not known; and the management of complex beliefs requires specialist psychological expertise and was beyond the scope of group setting we proposed. Guided discovery is the key skill required to elicit beliefs, and is often accompanied by education in skills such as pacing and goal setting.

Just over half of people who were randomised to the CBA intervention reported improvements – this is substantially greater than would be expected from natural recovery or from the provision of AM alone. CBA is used to treat a range of conditions, and we tailored the approach to LBP by using CB principles to target health beliefs and behaviours associated with poor outcome in this condition. The targets were avoiding movement and activity through fear of pain and further damage, promoting steady build up in physical, leisure and occupational activity. The tailoring involved fusing knowledge on risk factors and effective treatments. It is generally accepted that increasing physical activity is an essential part of the pathway to recovery in LBP, and therefore, we ensured that therapists were trained in up to date knowledge about activity and back pain, as well as CBA. This was particularly important for therapists who were recruited from a psychological discipline. Given the importance of physical activity, we stipulated that one of the goals set at the initial assessment should be related to physical activity or exercise. This was well received by participants and in the pilot evaluations we responded to participant comments that they would like more demonstration and checking of exercise and activity. That said, only very limited amounts of time were spent on direct supervision of exercise, and this was very much up to individual choice. We estimate that on average direct supervision of exercise occurred in less than 20% of participants, for less than 5% of the intervention time.

Other literature: comparisons of cognitive behavioural therapy with advice interventions in primary care

Cognitive behavioural approaches have been applied in a variety of ways in back pain – in isolation, alongside as opposed to integrated with other interventions, integrated with other treatments (notably exercise), and in formats in which the educational skills are used but scant attention is paid to eliciting beliefs. The recent draft NICE guidance on LBP has attempted to classify interventions, distinguishing between psychological interventions and combined psychological/physical interventions. 159,160 As our primary intention was to challenge beliefs that would effect a change in a range of behaviours, and over 95% of the intervention contact time was directed toward psychological goals, we would label our intervention as psychological.

During the lifespan of the BeST study, several variants of CB approaches have been tested in comparison with best practice advice. Two additional trials provide insight into the effectiveness of best practice advice as a comparator.

Johnson et al. 149 selected people if they had an RMQ > 5 and back pain of longer than 3 months’ duration, but excluded those who had consulted for symptom management in the last 6 months. The interventions tested were a booklet and audio advice about back pain (based on CB principles) in comparison with the advice plus group CBA sessions. The estimates of between-group differences were smaller than we observed in BeST although they bordered on statistical and clinical significance at some time points. The sample size was small, making interpretation very difficult. Many trials of LBP have chosen unrealistically large effect sizes, and have resulted in trials underpowered to detect smaller differences that are substantially important when the public-health impact of LBP is considered. In comparison with BeST, there are some differences in the intervention – the training and intervention was provided by physiotherapists, and the degree of psychological input and compliance with British Association for Counselling and Psychotherapy definitions of a CB intervention is unclear. The control advice intervention may have been of a greater intensity, sufficient to match the group treatments. The participants of the Johnson trial reported more chronic symptoms, although our subgroup data indicate that this should not be an important variant in the treatment response.

Jellema et al. 161 trained GPs to identify and intervene on psychosocial factors. The intervention was a 20-minute CB session, in addition to the original consultation. GPs were trained to use standardised questions to elicit information on fear avoidance, beliefs about the causes of LBP, worries, catastrophising and other behaviours. Treatment techniques included information giving, and goal setting, supplemented by The Back Book. A follow-up appointment was made if persistent LBP was anticipated. GPs were given 5 hours of training in the method. The control arm received the usual care of ‘wait and see’, graded uptake of activities and pain medication. There was no difference in the effectiveness of the strategies, and no change in either fear avoidance or coping. The intervention may have been too brief, and process evaluation suggests that the GPs found difficulty in identifying psychosocial risk factors. The intervention was also less intensive, and some may question whether it is a CB intervention at all.

Comparison of cognitive behavioural interventions with other physical interventions in primary care

The Hands on, Hands off trial of Hay et al. 148 describes a pain management programme for primary care in Staffordshire, UK, that was delivered on a one-to-one basis that included some of the types of education skills used in BeST. The overall duration of the programme was much shorter, suggesting that belief elicitation may have been a smaller component of the programme. The intervention was no more effective than the comparator, which was physiotherapy including manipulation. This makes interpretation of this trial difficult. Both treatments may have been equally effective, or equally ineffective. Frost et al. ’s Oxfordshire Low Back Pain Trial37 reported that physiotherapy delivered within the NHS is no more effective than a best practice advisory session. Comparison with BeST results is also complicated because Hay et al. 148 recruited people with back pain of short duration, excluding those with pain for longer than 12 weeks. The one-to-one CB intervention tested in the Hands on, Hands off trial was more expensive than routine physiotherapy. 162

Two studies from the Netherlands have tested interventions that are very intensive. Smeets et al. 47,162 recruiting people with chronic LBP reported that a highly intensive CB intervention was more effective than a waiting list control, but of similar effectiveness to an intensive exercise training protocol. An average of 22 group CB therapy sessions were delivered per participant. Van de Roer et al. 163 drew a comparison between physiotherapy and an intensive cognitive group programme (up to 30 sessions for the group-based intervention, and an average of nine sessions of physiotherapy). There was no difference in cost-effectiveness between the two approaches (clinical outcomes have not been reported to date).

Critchley et al. 86 investigated the effectiveness of a pain management programme that was predominantly supervised exercise, and additional sessions on goal setting and positive coping strategies. There is no mention of eliciting and challenging beliefs. There were no differences between individual sessions of physiotherapy, pain management groups, or spine stabilisation exercises, with all groups demonstrating improvement. Again it is difficult to draw strong comparisons with the BeST trial or conclusions on effectiveness, as there was no usual care or best practice arm.

Comparison of other physical treatments with advice only in primary care

Our short-term (3-month) clinical effects are similar to those found in high-quality studies of other therapies including manipulation exercise and acupuncture. 86 Strikingly, and in contrast to many previous studies, the benefits we observed were maintained over the long term (12 months).

The UK BEAM trial compared exercise and manipulation (alone and in combination) with best practice advice. 94 They found that a spinal manipulation package produced a small to moderate benefit at 3 months and by a smaller but still statistically significant margin at 1 year, that exercise produced a small benefit at 3 months but not at 1 year, and that manipulation followed by exercise produced a small to moderate benefit at 3 months and a smaller benefit at 1 year.

Unlike acupuncture, CBA had a broad range of effects. In a well-conducted definitive trial, acupuncture reduced pain, but did not impact on disability outcomes. 143 The CBA tested in BeST impacted on a range of outcomes including disability, pain, health-related quality of life and general health status. We detected a larger effect in global health measures and health-related quality of life than in other trials of physical interventions, suggesting that the CBA has effects outside the spine and so has potential for greater overall benefit than those more localised treatments.

Cost-effectiveness

Despite a reasonable number of reports on the cost of behavioural interventions in the literature, relatively few studies are of sufficient quality to guide policy decisions. In the absence of well-conducted comprehensive decision analyses, studies of cost–utility, conducted alongside randomised controlled trials provide the best evidence. 86,94,164–167 When drawing comparisons with trials that have tested interventions against a best practice advice or usual care arm, the BeST intervention is extremely cost-effective from both an NHS and a patient perspective; cost per QALY is less than half that of competing interventions for LBP (Table 54).