Matching treatment options for risk sub-groups in musculoskeletal pain: a consensus groups study

This study uses a consensus groups method; an approach commonly employed in the development of healthcare interventions, due principally to their capacity to ‘harness the insights of appropriate experts to enable decisions to be made’ [17]. The consensus group method used in this study is Nominal Group Technique (NGT). Originally developed by Delbecq, van de Ven and Gustafson [19], NGT is a systematic approach to building consensus using structured in-person meetings of experts. NGT follows a distinct set of stages that involves participants initially generating ideas on a particular issue or problem, which are then individually rated by participants. Following this, the results of these ratings are discussed within the group, and then re-rated individually by participants, who are given the opportunity to change their scores in light of group discussions. This process can be repeated several times until consensus is achieved. The threshold used to signal an acceptable level of consensus varies across NGT studies depending on the aims and design adopted; the threshold we adopted is discussed later; see ‘Assessing consensus’. NGT can be adapted, for instance through use of a ‘pre-elicitation technique’ [20], that is, providing participants with a summary of existing evidence to inform their decision-making. The pre-elicitation technique was used in this study due to the wide range of literature on MSK treatments (see below for further explanation). The stages of the NGT process are described in detail below.

We invited stakeholders from a range of clinical backgrounds, including general practice, physiotherapy, rheumatology, orthopaedics, clinical psychology and pain medicine (see Table 1 for participant characteristics). Participants were identified through existing clinical and research networks in the UK, and were invited via email. Dates for the meetings were arranged at the convenience of the practitioners. The same practitioners were invited to all three consensus group meetings (see Results below).

Ethical approval for the study was granted by the Keele University Ethical Review Panel (ref: ERP319). Participation in the study was entirely voluntary. Participants’ travel expenses were reimbursed, but beyond this they received no financial incentive for taking part. Written consent was obtained prior to each NGT meeting.

Three NGT meetings, each focusing on one of the three patient risk subgroups (low, medium, high risk of persistent disabling pain), were conducted between April–May 2015. Groups were convened at Keele University. Meeting 1 focused on agreeing matched treatment options for patients at low risk of persistent disabling pain, for all five MSK pain presentations. Meetings 2 and 3 focused on matched treatment options for patients at high and medium risk, respectively.

The meetings each followed the same set of stages (see Fig. 1). First a facilitator provided a brief study overview and outlined the aims and stages of the consensus process. Facilitators were members of the research team and did not participate in the group discussions. The same facilitators were used in each meeting. Participants were initially separated into small groups (4–6 participants with a spread of backgrounds and expertise so as to allow for multi-disciplinary discussions) each focusing on one or two MSK pain presentations. Each group was tasked to agree on the most appropriate treatment options for the MSK pain presentation and for the risk subgroup being considered (low, medium or high). Participants were asked to decide on treatment options for a general adult population based on risk of poor outcome, i.e. the patient’s likelihood of experiencing persistent disabling pain. To support this, participants were provided with: a) lists of UK NHS evidence based treatment options, developed for each risk subgroup across each of the five pain presentations (Fig. 2 outlines the development of these lists), and b) patient vignettes exemplifying ‘typical’ patients falling within the relevant risk subgroup, (e.g. for meeting 1, vignettes of patients at low risk). The vignettes were developed using data from patients stratified as low, medium and high risk using the validated STarT MSK tool™ as part of the earlier observational cohort study [16]. Vignettes were used as illustrative examples to give participants information about key characteristics of each patient subgroup in order to inform their decision-making on appropriate treatment options (see Table 2).

Participants first silently reviewed the resources provided in order to identify relevant treatment options they felt were missing from the list of UK NHS evidence based treatment options. Participants were encouraged to focus on first-line actions that could be taken by a GP, rather than treatment decisions made further along the treatment pathway, e.g. specific surgical procedures. Participants took turns to share any additional treatment options within their small group, which were noted on a flipchart by the facilitator. Once all additional ideas had been recorded, each additional treatment option generated through this initial stage (i.e. not including the existing options on the evidence-based lists) was discussed in turn within the small groups. Those that were agreed upon after discussion – in terms of their appropriateness for the given pain presentation and risk subgroup, and those available on the NHS – were added to the treatment options list. Participants then silently and anonymously rated, on a 7-point Likert scale, the appropriateness of each treatment option on the group’s list for the MSK pain presentation and risk subgroup being considered (e.g. appropriateness for patients with shoulder pain classified at low risk). The mean appropriateness score was then calculated for the group for each treatment option on the list by members of the research team. Options were then organised in rank order, i.e. the option with the highest mean ranked number 1, and so on.

All participants then joined together with others for a full group discussion of the treatment options for each MSK pain presentation. The full group was presented with the list of treatment options for each pain presentation in rank order. All participants then had the opportunity to ask questions and suggest further relevant treatments they felt were missing. All suggested options were written on a flipchart and discussed in turn by the full group. Participants then voted via a show of hands as to which treatment options should be added to the final treatment option list (those agreed by more than half of the full group members).

Using the final list, all participants silently and anonymously rated, again on a 7-point Likert scale, the appropriateness of each treatment option (for the MSK pain presentation and risk subgroup). The consensus process described above was repeated for all five MSK pain presentations and for each risk subgroup.

Following the three group meetings, the mean scores were calculated across all participants for each MSK pain presentation and each risk subgroup. A list of treatment options identified for each pain presentation in each risk subgroup was compiled and sent to participants via email to consider and resolve any inconsistencies and approve the finalised list of treatment options. Through discussions within the research team, as well as considering consensus thresholds used in previous NGT studies reported in the research literature, it was decided that a mean rating of > 50% signalled an acceptable level of consensus for a treatment option to be included in the final list. This cut-off point reflected the challenge in gaining consensus from a multi-disciplinary group compared to a group from a single health professional discipline [21‐23]. Additionally, we wanted the final treatment option lists to be inclusive so that when these are used by GPs in consultations, the range of available options can allow for patient choice and shared decision-making.

In total, 20 participants attended at least one of the three NGT meetings; however, not all participants were able to attend all three group meetings (group 1: n = 18 participants; group 2: n = 16; group 3: n = 12). Participants had a range of clinical backgrounds, with a mix of experience levels/ length of time in practice (2–27 years in current clinical role; mean = 12 years). They also came from a range of geographical areas within the UK, giving a breath of knowledge on the diversity of existing treatments available within the NHS nationally (see Table 3).

The following matched treatment options were agreed and are presented in full in Table 4.

The agreed treatment options for patients at low risk principally involve GP-supported self-management, and do not include referral for further treatment, reflecting the aim of stratified care to avoid over-treating patients with a good prognosis [24]. The additional treatment option agreed only for low risk multisite pain – “refer to supported self-management and locally available community resources e.g. walking group, exercise on prescription” – may reflect the additional complexity of managing pain at more than one body region site, with the participants believing that these patients would benefit from further support in addition to that provided by the GP.

As well as the treatment options agreed for low risk, medium risk included the addition of onward referral options across all five MSK pain sites: “consider referral to physiotherapy” and “consider referral to MSK interface clinic”. Including options for onward referral reflects the increased need of patients at medium risk, who may experience higher levels of pain and greater functional limitations compared to low risk patients. There was also some variation in agreed treatment options for patients with different MSK pain-site presentations who are at medium risk of poor outcome. For instance, “consider referral to psychosocial intervention or pain management service” was included for back and multisite pain, but not for the other three pain sites; this may reflect the higher prevalence of psycho-social issues associated with these pain sites, e.g. both are strongly associated with an increased risk of developing co-morbid depression [25, 26]. Other agreed treatment options that differed across pain sites included “atypical analgesia”, which reached consensus for patients at medium risk with back, multisite and neck pain, but not for those with knee or shoulder pain; which is likely to reflect the greater prevalence of radicular pain from these three sites. A number of guidelines suggest that atypical analgesia is more effective for neuropathic pain and following discussion amongst the participants this was further simplified to being recommended for patients with radicular pain. Two treatment options were agreed only in relation to one of the five pain sites at medium risk: “corticosteroid injection” (shoulder), and “refer to rheumatology” (multisite). Again these results appear clinically appropriate: administering an injection is a common first-line treatment for shoulder pain; and the option to refer to rheumatology for multisite pain patients takes into account the importance of identifying in primary care early indications of inflammatory arthritis, in order to reduce diagnostic delay [27].

There was some overlap in the matched treatment options that reached consensus for patients at medium and high risk of poor outcome, although additional options were agreed for patients at high risk across all five pain presentations. Two treatment options reached consensus for all five MSK pain presentations for patients at high risk, but did not reach consensus for any of the MSK pain presentations for patients at medium risk: “opioids” and “refer to expert patient programme”. That opioids were agreed only for patients at high risk of poor outcome reflects current UK practice guidelines– primary care clinicians are encouraged to prescribe non-opioid based analgesia where possible, and only prescribe opioids for patients with high pain levels or who have an inadequate response to paracetamol and/or nonsteroidal anti-inflammatory drugs [14]. The expert patient programme places focus on self-management of long-term conditions and therefore may be appropriate for some high risk patients given their poorer prognosis, and subsequent higher probability of experiencing long-term disability. “Refer for surgical opinion” reached consensus only for patients at high risk, for all MSK pain sites except multisite pain; whilst “refer to rheumatology” was agreed for patients with multisite pain only. This again reflects the different perspectives from this group of practitioners about how best to manage multisite pain – and particularly widespread pain – in comparison to single site pain. “Corticosteroid injection” reached consensus for patients with knee and shoulder pain at high risk, but not for patients with MSK pain at the other three sites; again reflecting common practice of injecting peripheral joints [28]. “Refer for lifestyle intervention e.g. dietician, slimming world etc.” reached consensus only for patients at high risk with knee pain and not for any of the other four pain presentations. This reflects the evidence base regarding the association between obesity and knee joint pain [29] and the functional difficulties resulting from knee pain that can hinder patients’ efforts to engage in physical activity.

This study innovatively combined NGT with pre-elicitation techniques of patient vignettes derived from epidemiological data and supported by a rigorous review of the available best evidence [18] (see Fig. 2). The multidisciplinary nature of the sample of participants was also an important strength as this allowed access to a varied range of expertise in the field of MSK pain. It was felt that this would lead to less restricted results than might have been found had the participants been made up of GPs alone. Whilst this represents a strength of the study, including a wide range of clinicians made it necessary to use a lower threshold to determine consensus than might have been the case if the group comprised clinicians from a single discipline, who may have more homogenous views towards treatment [22]. The > 50% cut-off point used for signalling agreement may be viewed as a limitation of the study. However, our aim was not to identify the single ‘best’ treatments that would be appropriate for all patients in a given risk subgroup, but to identify a suite of treatment options that can be recommended to GPs, so that when these options are used in consultations as part of the new stratified care model, this can allow for professional and patient shared decision-making. Patients stratified into the same subgroup may have different treatment needs despite having similar risk profiles, and certain matched treatment options may be more or less appropriate for an individual patient; therefore, a high threshold for signalling agreement would not have been appropriate. Using a higher threshold, e.g. > 75%, would have resulted in fewer recommended treatment options. Taking high risk knee pain as an example using a > 75% cut off, in addition to the two generic treatments agreed across all three risk subgroups, only two further treatment options would have gained consensus: “consider referral to physiotherapy”, and “consider referral to MSK interface clinic”. The other eight treatment options would have been excluded; similar findings regarding the restrictive nature of using a higher consensus threshold are reported in previous NGT studies [30, 31]. The decision to use a > 50% cut-off was further reinforced through agreement from the study participants as to the suitability of the final treatment options which were sent to them via email following the consensus group meetings (see Methods, above).

The UK focus of the study represents both a strength and weakness. Whilst the geographical spread of participants enabled us to produce a list of recommended treatment options that is likely to have good representativeness across the UK, the applicability beyond the UK depends on treatment availability in other settings. However, the majority of treatment options that reached consensus are not UK-specific; therefore, findings can have wider applicability beyond the UK context and may be useful for informing research and practice in other settings.

A limitation is that the use of a consensus group methodology resulted in the exclusion of newer or less widely used treatment options. However, as previously mentioned, we included a wide range of clinicians to avoid missing any treatment options considered to be best practice, even if not always supported by available evidence. It was not considered to be appropriate to recommend untested treatment options; these treatment options would require further research and testing. A further limitation of this study is that we did not include patient representatives in our expert groups. Whilst we recognise that expert patients would have valuable input for decisions about treatment options, it was felt that this study was specifically about clinical expertise and therefore was limited to clinician input. Extensive patient representation has been included in the broader research programme [6], and the development and testing of the stratified primary care intervention.