From the selection of example trials presented, several observations were made and discussed at length within the workshop. Firstly, when common non-pharmacological interventions were compared with other primary care interventions, such as care from general practitioners (GP), then trials generally showed significant differences between groups. For example, a pragmatic RCT by Hay and colleagues 22 compared usual primary care with community physiotherapy for older adults with chronic knee pain, and showed short-term benefits of physiotherapy over usual GP care. Secondly, when two or more different non-pharmacological interventions in primary care were compared, then trials appeared most likely to show no significant differences between groups. For example, no difference between two or more physiotherapy approaches for low back pain patients 232529 or neck pain patients 26. Thirdly, when the primary outcome data were plotted on a graph over time, the similarity of findings, in terms of changes in pain and function, across trials of low back pain, neck pain, knee pain and shoulder pain was striking, irrespective of the specific interventions being tested.

Several explanations for the trends in the findings of recent trials were discussed. Given the similarity in trial findings, one key explanation was that non-pharmacological interventions in primary care provide little benefit for patients with musculoskeletal conditions and do not change the natural course of these conditions over time. Patient selection was discussed in relation to the problem of heterogeneity of the samples within previous trials, such that the average treatment effect masks a wide range of individual responses to treatment, including for example patients who benefit a great deal along with those who benefit little or not at all. Participants suggested there has been inadequate identification of important patient subgroups in previous trials. Discussion highlighted that previous trials had been based on a traditional, but perhaps unrealistic, expectation of moderate to large effects of non-pharmacological, primary care-based interventions for patients with common musculoskeletal patients. In fact, the overall message from previous trials is one of small and often short-term effects, some of which may be clinically meaningful whilst others may not. If the results of multiple high quality clinical trials give the 'true' picture for the majority of patients, then dramatic improvements seen in individual patients are not representative of the impact of non-pharmacological therapies in general for populations with musculoskeletal conditions. Powering clinical trials to compare different interventions to show anything other than small differences between groups appeared particularly unrealistic.

Participants debated the likelihood that previous clinical trials' investigators had overestimated the specific treatment effects of interventions and underestimated the non-specific treatment effects, such as the attention, support and empathy provided by a health care practitioner and individual patients' and practitioners' preferences and expectations. This was felt to help explain the lack of significant differences between interventions seen in previous trials. A further explanation was, that despite many trials now incorporating recommended and validated outcome measures, these may still be failing to capture what is really important to patients. There was a general view that there continues to be inadequate measurement of outcomes that are important to patients, in terms of both the timing of outcome assessments and the constructs being captured by available validated outcome measures, such as pain-related disability. A final explanation debated to what degree the interventions within previous trials were effectively applied, with some participants of the view that there has been insufficient attention to the assessment of skills and competencies of clinicians providing the interventions and their adherence to the intervention protocols stipulated by the trial procedures. Following these discussions, participants generated their recommendations for future studies.

As a result of the independent suggestions from participants, a list of 22 recommendations was generated, presented in Table 1. Following clarification and discussion, these were then subjected to independent ranking.

From the original set of 22 recommendations, an agreed list of 12 priorities was generated as a result of the independent ranking, highlighted in Table 1. The 12 recommendations were ranked, again independently, at the WCPT international symposium with 133 participants from 24 different countries. Of the 12 recommendations, 7 reached our agreed level of consensus and are highlighted in Table 1. Those describing themselves as researchers and those describing themselves as practitioners selected the same top 7 recommendations in their ranking exercise. There was clear agreement on the need for future research to focus further on implementation (research into practice) for musculoskeletal conditions, the potential benefit of developing national musculoskeletal research networks through which support for trials could be provided, to develop more innovative trial designs such as those based on stepped care and subgrouping for targeted treatment approaches and for health economic analysis to become a common component of future clinical trials. In addition, the recommendations to include more patient-centred outcomes in trials, to develop a core set of outcomes for new trials to allow comparisons in the future and to conduct studies that advance methodological approaches for clinical trials also met our consensus criteria. The implications of these priority recommendations are discussed below.

The desire to implement research findings into practice has increased in priority with the amount of evidence now generated in the musculoskeletal pain literature. Simultaneously there has been an acknowledgement that the evidence alone cannot be implemented without facilitation 42434445. However optimal methods for achieving this are still uncertain. Development of clinical guidelines has been one mechanism for enhancing uptake of research into practice 46. For example, exercise is recommended as a core treatment by the National Institute of Health and Clinical Excellence in the UK in their recent osteoarthritis guidelines 47 as a result of evidence from high quality RCTs of non-pharmacological interventions [e.g. 22283648]. Unfortunately implementation strategies for clinical guidelines are complex and studies in this field are challenging. It is not surprising therefore that this recommendation was highly rated by both national and international participants in this consensus study.

This priority is timely, and reflects a growing commitment of the research community to adopt a collaborative, rather than competitive, approach to clinical research. It is increasingly unrealistic for lone researchers, working in isolation, to produce high quality RCTs and other clinical studies required to answer important research questions. As an example of the increasing recognition of this, in October 2007 the Arthritis Research Campaign (arc), the fourth largest medical charity in the UK and the only one solely dedicated to research in musculoskeletal conditions, launched a new initiative. They formed seven Clinical Studies Groups, to support clinical trials and related research within the UK 49. In partnership with the UK Comprehensive Research Network, the Clinical Studies Groups will develop nationally agreed strategic plans for clinical research in arthritis and related musculoskeletal conditions. This initiative aims to be inclusive, with the expressed aim of encouraging and facilitating clinician engagement in research in a variety of ways; for example by helping develop research questions, by developing and delivering novel interventions, through to being Principal Investigators on named projects. Full details of the arc Clinical Studies Groups can be found on http://www.arc.org.uk but this provides one key example of how this recommendation can be taken forward at a national level.

Ideally, strictly designed RCTs are carried out among homogeneous groups of patients, in order to limit individual variation and optimize prognostic similarity of intervention groups. The drawback of this is the limited generalisability of findings. In recent years there has been a shift towards pragmatic RCTs, in which more heterogeneous patients group are recruited. In these trials the reported average effect of treatment may obscure a wide variation in individual responses, with some patients showing dramatic improvement while others hardly respond to the same intervention 50. One solution may be to offer more complex or more intensive interventions only to those patients who do not respond to initial treatment. The effectiveness of such a stepped care approach is currently being evaluated in a trial with patients with whiplash-related disorders 38. Treatment effectiveness may also be improved by closer matching of treatments to individual patients' characteristics. Secondary analyses of a trial with back pain patients comparing a psychosocial intervention to usual primary care showed that profiles of patients responding favorably to treatment differed between intervention groups even though the effects measured at group level showed no difference at all 50. The main challenge is how to select or subgroup patients that may benefit more from specific (or targeted) interventions. Prediction rules have been designed to identify patients responding particularly well to some non-pharmacological interventions 5152 and a community-based intervention has been developed to identify and target psychosocial risks in individual patients 53. Further research is needed to establish whether the use of prediction rules (prognostic stratification) indeed leads to better patient outcomes and more efficient care 39. As such trials address complex interventions and have more sophisticated designs, they require large sample sizes. They also tend to require several phases of developmental work, for example as described by the MRC framework for the development and evaluation of complex interventions 5455.

The economic burden of musculoskeletal conditions has been well documented 56. For low back pain in the United States alone these costs are over $100 billion per year 57 and much of these costs result from prolonged care for those patients who fail to recover and progress to chronic disability. An important goal of primary care interventions is thus to reduce the likelihood of, and costs associated with, chronicity and ongoing management 58. Including health economic alongside clinical outcomes in RCTs is needed to be able to fully evaluate interventions. This is particularly important when the differences in clinical outcomes between treatments are small. Cost-effectiveness analysis, which compares the relative clinical benefits of different treatments to their relative costs, can identify treatments that provide the best value. For example, a recent trial found no significant differences in clinical outcomes between two non-pharmacologic interventions for patients with low back or neck pain; an approach based on cognitive-behavioural principles versus one based on the McKenzie approach to mechanical diagnosis and therapy 59. The economic analysis included the clinical outcomes, the relative costs of each approach and of additional health care services, medication, lost wages and work productivity, and suggested the McKenzie approach was preferable 60. Thus including health economic analysis is an important priority for future research to assist policy makers and patients in making decisions about interventions.

The need to develop robust patient-based outcome measures and core sets of these for use within RCTs has been previously recognised as important within the general clinical trial literature 61 and within the musculoskeletal pain community. For example, the OMERACT initiative http://www.omeract.org/, an informal international network of health professionals that organizes meetings focused on outcome measurement across RCTs and observational studies 62 recognised the need for patients to actively participate in this activity. There are now excellent opportunities to use such core outcome sets within RCTs with musculoskeletal pain patients. In collaboration with OMERACT, disease specific outcome measures for RCTs have been recommended [e.g. 6364]. In the area of low back pain, an international group of researchers considered recommendations for standardised outcome measurement in clinical trials 65, whilst others have evaluated a core set of outcomes for whiplash-related disorders 66. Not every musculoskeletal condition is covered but, increasingly, consensus on the important core domains is being reached. The development and use of patient-individualised measures are relatively rare but of increasing interest, as they allow the respondent to select issues or domains that are of personal concern rather than those which are solely predetermined by the investigator's list of questionnaire items 61. Some such patient-individualised measures are available [e.g. 67] and the suitability of using traditional outcomes alone for musculoskeletal pain patients questioned 68. Alongside broadening the specific choice of outcomes is the possibility of revisiting the timing of outcome measurements in RCTs for common musculoskeletal conditions and considering the trajectory of the pain condition over time as an outcome rather than focusing on specified follow-up time points alone.

We consider key strengths of this consensus initiative to include the independent generation of recommendations for future research, their independent ranking by both national and international researchers, clinicians and user representatives and the level of consensus reached by the top seven priorities. Participation in the workshops was excellent, from a wide range of disciplines, an indication of the relative importance placed on this topic by the community of researchers. Overall, participants wanted to acknowledge the positive progress made over recent years, in providing many high quality RCTs of non-pharmacological interventions for musculoskeletal conditions. There are now more clinical trialists within disciplines such as physiotherapy, occupational therapy and podiatry, than ever before, who aim to provide patients, practitioners and health policy makers with reliable evidence about non-pharmacological therapies. The workshops provided opportunity for discussion, debate and sharing of ideas and there was a great deal of enthusiasm for future similar opportunities. Potential weaknesses, as with all efforts to generate consensus, are that different participants may have provided different recommendations or prioritised the recommendations differently. Inevitably the results of priority-setting initiatives are dependent on the composition of the group of participants and further consensus-based initiatives in other countries and with other clinical disciplines would be useful. In addition, our focus was specifically on recommendations for future studies testing the effectiveness of non-pharmacological interventions for common musculoskeletal pain. Clearly, other types of research are also needed, such as research that provides better understanding of aetiological factors and mechanisms of action of interventions.

As a whole, this study should be considered as a step towards improved quality and innovation in future intervention studies, testing the effectiveness of non-pharmacological, primary-case based treatment options for musculoskeletal pain. The lessons learnt from previous trials, and the priorities identified can guide the development of further work of those involved in designing RCTs within the field of musculoskeletal care. The value of the agreed priorities will ultimately be in their use, for example, in justifying the case for funding of new studies, in their uptake by funding agencies as part of their commissioning processes and strategic plans, in the incorporation of these recommendations within new RCTs of non-pharmacological therapies, and in an increase in translational research studies that improve research into practice for patients with musculoskeletal conditions 69. We hope these recommendations will be incorporated within new clinical trial proposals, help to develop future research collaborations to ensure standardisation of outcome measures that can facilitate future meta-analyses and secondary outcome syntheses, lead to an increase in the use of more innovative trial designs and encourage the publishing of clinical trial protocols in this field. We welcome suggestions of other national and international dissemination mechanisms for these research priorities, such as the involvement of relevant professional bodies, clinical interest groups and funding agencies.