Stratified primary care for adults with musculoskeletal pain: the STarT MSK research programme including RCTs

Article history

The research reported in this issue of the journal was funded by PGfAR as project number RP-PG-1211-20010. The contractual start date was in June 2014. The final report began editorial review in February 2021 and was accepted for publication in June 2022. As the funder, the PGfAR programme agreed the research questions and study designs in advance with the investigators. The authors have been wholly responsible for all data collection, analysis and interpretation, and for writing up their work. The PGfAR editors and production house have tried to ensure the accuracy of the authors’ report and would like to thank the reviewers for their constructive comments on the final report document. However, they do not accept liability for damages or losses arising from material published in this report.

Copyright statement

Copyright © 2023 Foster et al. This work was produced by Foster et al. under the terms of a commissioning contract issued by the Secretary of State for Health and Social Care. This is an Open Access publication distributed under the terms of the Creative Commons Attribution CC BY 4.0 licence, which permits unrestricted use, distribution, reproduction and adaption in any medium and for any purpose provided that it is properly attributed. See: https://creativecommons.org/licenses/by/4.0/. For attribution the title, original author(s), the publication source – NIHR Journals Library, and the DOI of the publication must be cited.

2023 Foster et al.

Background and rationale for the programme

Musculoskeletal pain such as back pain, neck, shoulder, knee and multisite pain is common and costly in terms of the burden on individuals, the NHS and society. 1 For some people these musculoskeletal problems are painful but short-lived; however, for others the painful episode fails to resolve or recurs, impacting day-to-day function including ability to work and leading to extensive NHS and societal costs.

Most patients with musculoskeletal problems are managed in primary care, where 20% of the registered practice population will consult their general practitioner (GP) annually with a musculoskeletal problem, accounting for 1 in 6 GP consultations. 2 There is limited evidence to guide how best to direct the right patient to the right treatment in primary care in ways that improve patient outcomes such as pain and function, and make the best use of health-care resource. The sheer number of patients with musculoskeletal pain and the wide variation in their prognosis means that it not appropriate to offer intensive or expensive treatments to all. 3

Stratified care involves targeting treatments according to patient subgroups, in the hope of maximising treatment benefit and reducing potential harm or unnecessary interventions. 4 Building on a previously successful model of prognostic stratified care for patients in primary care with low back pain,5–7 the aims of this programme were to adapt, finalise and test a prognostic stratified primary care model for a much larger group of patients with the five most common musculoskeletal pain presentations.

Aims of the research programme

The aims of this programme were to adapt, finalise and test a prognostic stratified primary care model for primary care patients consulting with the five most common pain presentations (back, neck, shoulder, knee or multisite pain), comparing it with usual primary care. A series of studies involving different methods (a prospective longitudinal cohort study, interviews with patients and clinicians, an evidence synthesis, consensus group workshops, the development of tools and training to support GPs to use the stratified care approach, and clinical trials including analyses of general practice medical record data and health economic analyses) were carried out to address the following research questions:

1. Can one prognostic tool [the Keele STarT MSK Tool (Subgrouping for Targeted Treatment for Musculoskeletal pain)] identify the risk of poor outcome for a wide range of patients with the most common musculoskeletal pain presentations in primary care, and does it discriminate subgroups at low, medium and high risk of persistent disabling pain?

2. What are the most appropriate treatment options that should be recommended for patients in each risk subgroup?

3. What are the views and experiences of patients and clinicians about using a prognostic stratified care approach in the management of musculoskeletal pain?

4. What is the feasibility of (1) delivering the stratified care intervention in primary care and (2) conducting a large randomised controlled trial (RCT) to test this new model of care?

5. In adults with the most common musculoskeletal pain presentations in primary care, does prognostic stratified care (involving use of the Keele STarT MSK Tool and recommended matched treatments) lead to better patient outcomes, greater cost-effectiveness and different clinical decision-making than usual non-stratified primary care?

In addition, a stratified care clinician training and support package comprising electronic templates, a stratified care tool and recommended treatment options to support delivery of stratified primary care was developed.

The research was carried out between June 2014 and September 2020.

The inter-relationships between the four work packages are summarised in Figure 1 (below). The prospective cohort study in work package 1 provided the sampling frame for the qualitative interviews in work package 2. The feasibility and pilot RCT in work package 3 provided the data set for the external validation of the Keele STarT MSK Tool in work package 1.

FIGURE 1.

The Keele STarT MSK research programme pathway diagram. KAPS, Keele Aches and Pains Study.

Research aim and objectives

The aim of this work package was to refine and validate an instrument, the Keele STarT MSK Tool, designed to identify risk of poor outcome in primary care patients with the most common musculoskeletal pain presentations.

The four specific objectives were as follows:

1. confirm the validity and optimise the predictive performance of the Keele STarT MSK Tool

2. determine the screening tool risk strata cut-off points based on optimal predictive values and suitability for matched treatment options

3. estimate the proportions of patients classified at low, medium and high risk of poor outcome and describe their characteristics

4. describe current health-care resource use by all patients and in each risk stratum.

Methods

We carried out a prospective cohort study of consecutive consulters in UK general practice [named the Keele Aches and Pains Study (KAPS)]. The protocol for the study has been published. 9

Patients aged ≥ 18 years presenting with one or more of the five most common musculoskeletal pain presentations (back, neck, shoulder, knee or multisite pain) were identified at 12 general practices in the West Midlands in England. Participants were recruited between July 2014 and February 2015. Patients were excluded if there were indications of potentially serious underlying pathology (such as cancer or infection), they had urgent care needs, they were vulnerable or if they were unable to communicate in English. We mailed out information about the study and baseline questionnaire and, for those providing consent, we sent further questionnaires at 2 and 6 months’ follow-up and analysed their primary care electronic medical records (EMRs). We estimated that we would need to identify approximately 3000 eligible patients to recruit 1800 participants at baseline (based on a 60% response rate). We anticipated that this would provide 1250 patients at 6 months follow-up, including 125 patients at high risk of poor outcome (the smallest subgroup), which would provide adequate power for validation of the draft Keele STarT MSK Tool.

Our main outcome measures were physical function [Short Form questionnaire-36 items (SF-36) physical component score (PCS)], pain intensity [0–10 numerical rating scale (NRS)], pain interference [Patient-Reported Outcomes Measurement Information System (PROMIS) pain interference scale] and health-related quality of life [EuroQol-5 Dimensions, five-level version (EQ-5D-5L)], measured at 2 and 6 months’ follow-up. For dichotomous outcomes, poor outcome on the PCS was defined as scores of < 37.17 at 2 months and < 39.61 at 6 months, based on lower tertiles from an independent study of UK primary care musculoskeletal pain patients. 10 Poor outcome on pain intensity was defined as NRS scores of ≥ 5 points. 11 Secondary outcomes are detailed in our protocol paper by Campbell et al. (2016). 9 The baseline questionnaires also included the draft Keele STarT MSK Tool (identical to the modified STarT Back tool) presented in our paper by Hill et al. (2016). 8 The responses to the KAPS baseline and follow-up questionnaires are provided in Appendix 1.

Analysis

Predictive performance was determined using linear regression of the association between baseline tool score and PCS and pain intensity at 2 and 6 months’ follow-up. Performance was assessed based on model fit (R²) and discrimination (c-statistic) and calibration (calibration slope and Hosmer–Lemeshow test).

Item redundancy and weighting was investigated within multiple linear regression models for estimating PCS score at 2 and 6 months, and pain intensity at 2 and 6 months. If items did not add significant predictive performance and/or if average standardised beta weight was small (i.e. < 0.05) in most analyses, then the item was deemed statistically redundant. The research team, in consultation with members of our patient and public involvement and engagement (PPIE) group, considered face validity in decisions about the removal of statistically redundant items.

The tool cut-off point for identifying the high-risk subgroup (compared with those at medium/low risk) was based on classification on the full score most likely to attain positive predictive values and specificity ≥ 0.8, and positive likelihood ratio ≥ 5, for predicting pain and function at 2 and 6 months’ follow-up. The tool cut-off point for categorising the low-risk subgroup (compared with those at medium/high risk) was based on the classification most likely to achieve negative predictive values and sensitivity ≥ 0.80, and negative likelihood ratio ≤ 0.2. 12–14 All decisions about tool cut-off points were based on statistical information in the sample overall and within pain sites, plus suitability for matched treatments.

Health-care utilisation data were collected from patient questionnaires and medical record review at 6 months’ follow-up. Information included primary and secondary care contacts, prescribed medications, treatments, tests and investigations. Unit costs were obtained primarily from standard national sources such as NHS Reference costs,15 Unit Costs of Health and Social Care16 and the British National Formulary (BNF),17 and applied to resource use data.

Descriptive statistics were used to summarise data on health-care utilisation and a total health-care cost per patient over 6 months was also estimated. Mean total costs for each patient risk subgroup were calculated and non-parametric bootstrapping (1000 replications) was used to estimate bias-corrected CIs around differences in mean costs.

External validity

Independent testing was carried out within the Keele STarT MSK feasibility and pilot trial data set in work package 3, with the Keele STarT MSK Tool completed during GP–patient consultations, and using pain intensity at 6 months’ follow-up as the outcome (see Work package 3 and the pilot trial findings publication by Hill et al. 18). Discriminant and predictive validity were investigated using model fit and discrimination as above. Descriptive analysis of outcomes within risk strata of the final Keele STarT MSK Tool were investigated.

Key findings

Overall, 4720 patients visited their GP about back, neck, knee, shoulder or multisite pain and were invited to participate in the cohort study. A total of 2057 patients responded (43.6% response rate), and 1890 consented to participate (40.2% adjusted response rate owing to incomplete/ineligible questionnaires and refusals). The mean age of participants was 58 years (range 18–96 years), and 61% were female. Response rates at 2 and 6 months were 75.8% (n = 1425) and 78.7% (n = 1452), respectively.

The mean baseline physical function (PCS) score was 36.2 [standard deviation (SD) 10.1], mean baseline pain intensity was 5.3 (SD 2.4) and 22% of the sample reported having pain for less than 3 months. Multisite pain was the most commonly reported reason for consulting the GP, followed by low back pain and knee pain (further baseline characteristics are shown in Appendix 1, Table 1). Response rates to the 2- and 6-month follow-up questionnaires were 75.8% and 78.7%, respectively. At 2 and 6 months, mean PCS scores rose to 38.1 and 38.6, respectively (indicating modest improvements in physical function); 47.7% (n = 560) at 2 months and 53.4% (n = 581) at 6 months were categorised as having a poor outcome on the PCS. Mean pain intensity reduced to 4.4 at 2 months and 4.1 at 6 months, with 45.6% (n = 582) and 42.3% (n = 482) categorised as having a poor pain outcome, respectively.

Alterations to initial plans and work package limitations

We anticipated 60% participation among those invited to participate in the KAPS cohort study, based on previous similar studies, but only 40% consented to participate. With approval from the Programme Steering Committee (PSC), we extended recruitment by 3 months to achieve the required sample, but the lower initial response rate may have introduced further bias into the characteristics of the study sample. This is most likely to be reflected in the proportions of participants in each subgroup. However, it is unlikely to strongly affect the data analyses and findings because these were internal comparisons and there was still sufficient variation within the sample, and sufficient numbers, to carry out the analyses.

All the key outcomes were available within our purposely designed development data set (the KAPS prospective cohort) but of those outcomes, only pain intensity was available in the independent validation data set (the feasibility and pilot trial in work package 3). As the tool appeared robust across outcomes in the cohort data set, it seems unlikely that it would show substantially different findings in the trial data set for outcomes of physical function (as measured by the PCS), but this cannot be empirically demonstrated.

The decision was made to examine the performance of the tool developed in the cohort study within the independent pilot trial data set; this took place after the publication of the KAPS cohort protocol paper but was agreed with the PSC. This decision was taken because the predictive performance of the initial refined tool was not as high as required. After examination of the tool in the external data set indicated sub-optimal performance, the requirement for candidate items to be potentially modifiable by treatment was removed because the team wanted to examine the potential for including non-treatment-modifiable factors in the tool such as ‘pain duration’, which was included in the final version to increase its predictive abilities.

A limitation of work package 1 as planned was that although we conducted the refinement and validation of the Keele STarT MSK Tool as a patient-completed set of questions, its use as completed by clinicians during a clinical consultation was not investigated.

Conclusions

A new tool was refined and validated, the Keele STarT MSK Tool, with which to subgroup adults consulting with back, neck, shoulder, knee or multisite pain in primary care into those at low, medium and high risk of poor outcome. The tool is available on request from www.keele.ac.uk/startmsk/startmskresearch/. It clearly and simply allocates patients to subgroups with distinct characteristics, different health-care usage and different prognoses, and its performance is acceptable upon independent validation. This study confirms that generic prognostic factors can be combined in one simple stratification tool and that this tool can be used to identify the risk of poor outcome (low, medium or high) in a wide range of patients consulting with different musculoskeletal pain presentations in primary care.

Interrelationship with other parts of the programme

The Keele STarT MSK Tool was refined and validated in this work package, ready for use in work package 3 (feasibility and pilot RCT) as one of two components of the new stratified care intervention. The data from work package 3 was used to examine the external validity of the tool, prior to work package 4 (Keele STarT MSK trial). Participants in the work package 1 KAPS cohort formed the sampling frame for (1) invitation to participate in the qualitative interviews in work package 2 and (2) selection of example patient case vignettes to aid the development of consensus on matched treatments in work package 2. In addition, the recommended matched treatment options generated in work package 2 contributed to decisions about the cut-off points on the tool used to allocate patients to low-, medium- or high-risk subgroups.

Research aims and objectives

The aim of this work package was to define and agree the matched treatment options for patients in each of the three risk subgroups and develop a training and support package for the delivery of stratified primary care.

The specific objectives were as follows:

1. summarise current best evidence on available treatments for the five most common musculoskeletal pain presentations in primary care

2. explore patients’ and GPs’ views on the acceptability of prognostic stratified care, and the anticipated barriers to and facilitators of its use in clinical practice

3. agree, through expert consensus, the most appropriate matched treatment options that should be recommended for patients in each risk subgroup

4. develop and specify a training and support package to support the delivery of stratified primary care.

Methods and key findings for each study

Alterations to initial plans and work package limitations

We conducted more focus groups with patients than planned (four rather than two) to ensure that larger numbers of patients’ views were included. Based on pragmatic considerations, only one physiotherapist focus group was conducted (rather than the two planned). In the original plans for the consensus group study, 70% or more was the intended cut-off point to recommend treatment options, but this was reduced to 50% or more given the challenge in gaining higher levels of consensus with our highly multidisciplinary group of clinicians. We may have reached higher levels of consensus about recommended matched treatments, or a smaller set of recommended matched treatment options, had we included a more homogenous group of clinicians.

In the evidence synthesis, despite an extensive search, we found a paucity of evidence about treatments for multisite musculoskeletal pain. The control interventions were often not well described, and it is possible this led to overestimation of the effectiveness of some treatments in some included studies. In the qualitative study, patients’ and clinicians’ views were sought prior to finalisation of the stratification tool and recommended matched treatments. Discussions were deliberately based on the general principles of prognostic stratified care so that the views of patients and clinicians could inform the specific stratified care approach. Experiences of clinicians delivering, and patients receiving, the stratified care intervention in this programme were sought in work packages 3 and 4.

Interrelationship with other parts of the programme

The qualitative focus group and interview study and the consensus group study both drew on participants from the KAPS cohort in work package 1. The agreed recommended matched treatment options from this work package informed the cut-off points of the Keele STarT MSK Tool in work package 1 and along with the electronic template and GP training/support package, were tested for feasibility of delivery in work package 3 (feasibility and pilot trial).

Research aims and objectives

The aims of this work package were to test the feasibility of a future main cluster RCT to compare stratified care with usual primary care, and to test the feasibility of delivery of the stratified care approach.

The four specific objectives were as follows:

1. estimate participant recruitment and follow-up rates for the main cluster RCT

2. examine evidence of selection bias between trial arms and between participants and non-participants

3. assess GP fidelity to the stratified care intervention (i.e. use of the stratification tool and matched treatment options)

4. conduct secondary descriptive analyses of GP decision-making and patient outcomes.

Methods

Full details of methods and results have been published in two papers: Hill et al. 18 and Saunders et al. 31 The design was a pragmatic pilot, parallel two-arm (stratified vs. non-stratified care), cluster RCT in eight UK GP practices (four allocated to the intervention and four as control). Practices were randomised with stratification by practice size, and the trial statistician and outcome data collectors were blind to allocation. Participants were adult consulters with one of the five most common musculoskeletal pain presentations (back, neck, shoulder, knee or multisite pain) and without indicators of serious pathologies, urgent medical needs or vulnerabilities. Potentially eligible patient records were electronically tagged following consultation at participating practices and individual patients were sent postal invitations to participate in data collection. The target sample was 500 participants.

Delivery of the stratified care intervention by GPs was supported by the bespoke EHR template which included the Keele STarT MSK Tool validated in work package 1 to stratify patients into low, medium or high risk of poor outcome and the 17 recommended matched treatment options agreed in work package 2. Patients at low risk were matched to options that supported self-management, including over-the-counter (OTC) medication, and unnecessary investigations or referrals were discouraged. For those at medium risk, in addition to the low-risk treatment options, recommendations included referral to conservative non-pharmacological treatments (e.g. physiotherapist-led exercise therapy) and workplace assessment and advice. For those at high risk, in addition to the low- and medium-risk options, recommendations included referral for corticosteroid injections, specialist clinical services (including rheumatology, orthopaedics and pain clinics) and opioid medications. Full details of the matched treatment options are described in Appendix 3, Box 2, and also in figure 2 of our pilot trial publication by Hill et al. 18

General practitioners in the four practices randomised to deliver stratified care were invited to participate in the training/support package developed in work package 2, facilitated by an experienced GP trainer and the trial lead. The sessions lasted a total of 3–4 hours at each practice and covered the rationale for stratified care, how it differs from usual care, and familiarisation with the EHR template and its fit within the flow of a musculoskeletal consultation. The sessions also provided an opportunity for discussion and questions or concerns to be addressed. GPs also received a training update halfway through their practice’s recruitment period at which feedback data were shared about individual GP intervention fidelity, with peer-to-peer comparisons and discussion. Further details about the training and GP peer-to-peer comparisons are available in Appendix 3, Tables 5 and 6.

A set of pre-defined feasibility success criteria were used to evaluate recruitment, follow-up rates, selection bias and GP intervention fidelity. To determine whether or not these success criteria were met, four sources of data were used:

1. the general practices’ EHR participant identification screen, which captured point-of-consultation data in all 8 general practices, including patients’ pain intensity and location (back, neck, shoulder, knee or multisite pain)

2. an initial and 6-month postal questionnaire which collected outcomes such as physical function, risk subgroup, overall musculoskeletal health, fear avoidance beliefs, patient-perceived reassurance (from their GP), health-related quality of life, satisfaction with care received, provision of written educational material from their GP, global rating of change in the musculoskeletal problem, employment characteristics, work absence and productivity and patient demographic descriptors (see our published paper by Hill et al. 18 for full details, and also see Appendix 3, Table 7, for a summary of the participant self-reported measures)

3. monthly follow-up using either short message service (SMS) texts or brief paper questionnaires to capture pain intensity, musculoskeletal pain-related distress and self-efficacy

4. an anonymised general practice EHR audit to collect data to describe GP decision-making including prescriptions, referrals, imaging requests, sick certifications and repeat GP visits over a 6-month period following the index consultation when the stratified care template was first fired.

Nested qualitative study

Stimulated recall interviews were conducted with patients and GPs in the stratified care arm of the feasibility and pilot trial (n = 10 patients, n = 10 GPs), prompted by consultation recordings, in order to explore the feasibility of delivering stratified care in consultations. Data were analysed thematically and mapped onto the capability, opportunity, motivation and behaviour (COM-B) behaviour change model, exploring the capability, opportunity and motivation GPs and patients had to engage with stratified care (see our published paper by Saunders et al. 30 for full details of the methods and results of the nested qualitative study).

Key findings

Participants were recruited between October 2016 and May 2017 from eight general practices, during which GPs screened 3063 patients (stratified care n = 1591, usual care n = 1472) and completed the bespoke EHR template with 1237 eligible patients (stratified care n = 513, usual care n = 724). A total of 524 participants (42% of those who received the bespoke EHR template) consented to providing questionnaire outcome data (stratified care n = 231, usual care n = 293) over 6 months’ follow-up. Follow-up questionnaires and EHR audit data collection were completed by December 2017.

Alterations to initial plans and work package limitations

The learning from this work package led to important changes to the original plans, all of which were agreed with the PSC and funder, prior to the main Keele STarT MSK cluster trial in work package 4, which were as follows:

1. This feasibility and pilot trial changed from the intended internal pilot phase to an external pilot trial, given it took twice as long as expected (28 weeks rather than 12) to recruit participants and only 2 of the 4 pre-specified success criteria were met.

2. The main trial sample size was reduced (from n = 3600 to n = 1200) by focusing on the overall between-arm comparison rather than also powering the trial to detect between-arm differences at the level of patient risk subgroups (i.e. low, medium and high risk).

3. The items within the Keele STarT MSK Tool were amended following further statistical analysis using the point-of-consultation data from the pilot trial to identify items that improved the tool’s predictive validity, and a clinician-completed version (interview style rather than self-report style) was developed. A license to obtain both the self-report and clinician-completed versions of the tool is available on request at www.keele.ac.uk/startmsk. The clinician-completed version of the tool is also available in Appendix 3, Figure 2.

4. The number of recommended matched treatment options was reduced slightly from 17 in the pilot to 14 in the main trial and some options were recommended for additional patient subgroups (e.g. opioid medications). Following meetings with linked physiotherapy services, participating GPs were also reassured that these NHS physiotherapy services were informed about the trial and had capacity to receive referrals.

FIGURE 2.

EMIS templates for matched treatment options. (a) Multisite pain patients classified as low risk; (b) multisite pain patients classified as medium risk; (c) multisite pain patients classified as high risk; (d) shoulder pain patients classified as low risk; (e) shoulder pain patients classified as medium risk; (f) shoulder pain patients classified as high risk; (g) back pain patients classified as low risk; (h) back pain patients classified as medium risk; (i) back pain patients classified as high risk; (j) neck pain patients classified as low risk; (k) neck pain patients classified as medium risk; (l) neck pain patients classified as high risk; (m) knee pain patients classified as low risk; (n) knee pain patients classified as medium risk; and (o) knee pain patients classified as high risk.

There were several research limitations that need to be highlighted. First, there is an inherent potential risk of bias from those involved in developing a new clinical tool being the researchers who test it in practice. In addition, although the qualitative research revealed useful insights about why GPs found the tool difficult to use and explored the perceptions of patients regarding the services and treatment options available to them, more information about the reasons for both of these elements would have been useful. A further limitation is that there may have been a difference in the suitability of the approach according to the length of condition duration (i.e. acute vs. chronic pain) that we have not yet been able to fully explore.

Interrelationship with other parts of the programme

This work package evaluated the feasibility of using the Keele STarT MSK Tool and matched treatment options that were agreed in work packages 1 and 2 and comprised the external feasibility and pilot trial that informed the main trial in work package 4. In addition, the data from this pilot trial were used for the external validation of the Keele STarT MSK Tool at the point of consultation.

Research aims and objectives

The aim of this work package was to determine whether or not stratified care (i.e. the use of the Keele STarT MSK Tool to subgroup patients and matching subgroups to recommended treatment options) leads to superior clinical effectiveness and cost-effectiveness compared with usual primary care in patients with one of the five most common musculoskeletal pain presentations (back, neck, shoulder, knee or multisite pain).

The four specific objectives were as follows:

1. determine the comparative clinical effectiveness of stratified care compared with usual primary care for patient outcomes

2. investigate GP fidelity to delivery of stratified care and the impact on clinical decision-making

3. undertake an economic evaluation of stratified care versus usual care

4. conduct a nested qualitative study to understand how stratified care was perceived and operationalised by clinicians and patients.

Methods

Full details of the methods are presented in our publication by Hill et al. 32 The main trial was a pragmatic, parallel two-arm (stratified vs. non-stratified care), cluster RCT with a health economic analysis and mixed-methods process evaluation. The setting was UK primary care, comprising 24 general practices randomised (stratified by practice size) in a 1 : 1 ratio (12 per arm) and with the blinding of the trial statistician and outcome data collectors. Randomisation units were general practices and units of observation were adults consulting with one of the five most common musculoskeletal pain presentations without indicators of serious pathologies, urgent medical needs or vulnerabilities. Potentially eligible patient records were electronically tagged at participating practices and patients were sent postal invitations to participate in data collection. The target sample was 1200 participants with ≈ 2500 musculoskeletal consultations available for anonymised medical record data comparisons.

Delivery of the stratified care intervention by GPs was supported by the bespoke EHR template, which included the clinician-completed version of the Keele STarT MSK Tool and the reduced set of recommended treatment options for patients at low, medium or high risk of poor outcome. Full details of the per-protocol matched treatment options for each patient subgroup are provided in Appendix 4, Box 3, and in figure 3 of our main trial protocol paper by Hill et al. ,32 and were used to support secondary analysis based on treatment per-protocol in the main trial. Overall, the intervention sought to influence GP clinical behaviours in order to reduce pharmacological treatment, referral to imaging, and secondary care referrals, and increase non-pharmacological treatments such as self-management advice and referral to physiotherapy. In addition, when GPs made a referral to physiotherapy they were asked to provide additional information about the patient’s tool scoring and risk subgroup. The GP training and support package was similar to that described for work package 3, except it was shortened to 2 hours. GPs received a 1-hour training-update after the first month of recruitment to share and discuss the first of their monthly feedback reports showing individual GP fidelity to the stratified care intervention, and engage in peer-to-peer comparisons.

Data collection processes

Key findings

Objective 1: determine the comparative clinical effectiveness of stratified care compared with usual primary care for patient outcomes

Twenty-four general practices [12 per arm; total adult practice size of 104 GPs and 185,088 patients (96,397 allocated to stratified care, 88,691 receiving usual care)] participated from the West Midlands region of England. A total of 1211 (49%) patients consented to data collection (534 allocated to stratified care and 677 receiving usual care) and responded to the initial questionnaire within 30 days. Mean age was 60 years (range 18–95 years) and 58.9% of patients were female. Mean pain intensity on a 0–10-point NRS scale at the point-of-consultation was 6.73 points (6.77 points in the stratified care arm and 6.70 points in the usual care arm). Over 6 months of follow-up, 1178 (97%) participants provided at least one pain intensity response [515 (96%) in the stratified care arm and 663 (98%) in the usual care arm] and 80.9% responded to the follow-up questionnaire at 6 months (77.9% in the stratified care arm and 83.3% in the usual care arm). Full details of patient recruitment and follow-up through the trial are described in Appendix 6, Figure 4.

FIGURE 4.

Cost Effectiveness Acceptability Curve (Base-case).

There were few differences between the characteristics of patients who agreed to participate in data collection and those that did not (see Appendix 6, Table 8). The population characteristics of general practices randomised to each arm were similar (see Appendix 6, Table 9). Most characteristics of participants in each arm of the trial were similar, including the proportions of patients in each risk subgroup, which for stratified care and usual care, respectively, were 19.5% and 20.1% for low risk, 47.3% and 45.6% for medium risk, and 33.2% and 34.3% for high risk. Although there were statistically significant differences in age, employment status and pain site (see Appendix 6, Table 18), overall, these differences were modest and did not amount to a concerning signal of selection bias.

TABLE 8 - Descriptive and incremental health outcomes over 6 months for the base-case analysis and the complete case analyses. Values are mean (SD) scores unless stated otherwise

QALYs=quality-adjusted-life-years.

Base-case imputed dataset

Difference=SC-UC by generalized linear latent and mixed (gllamm) models adjusng for clustering. Reported CIs were generated using generalised linear regression methods.

Adjusted for baseline utility

	Total quesonnaires returned	Total quesonnaires returned (Including MDC)
6 month follow-up	978 (80.8%)	1070 (88.4%)
Health outcomes	SC n =534	UC n =677	Mean differenceb (95% CI)
Primary (Imputed) analysisa
Baseline EQ-5D	0.5572 (0.2293)	0.5542 (0.2349)	(-0.04530.0011, 0.0476)
6-month EQ-5D	0.6715 (0.2192)	0.6512 (0.2308)	(-0.02330.0166, 0.05662)
Unadjusted QALYs	0.3072 (0.1012)	0.3014 (0.1042)	(-0.01340.0054, 0.0242)
Adjusted QALYsc	0.3063	0.3021	0.0041 (-0.0013, 0.0095)
Complete-case analysisd	n = 445	n = 574
Unadjusted QALYs	0.3095 (0.1034)	0.3032 (0.1063)	(-0.01410.0068, 0.0277)
Adjusted QALYsc	0.3080	0.3027	0.0053 (-0.0005, 0.0112)

TABLE 9 - Resource components analysed in the trial. Values are mean (SD) resource per patient, by treatment group, for patients providing health care utilisation data at 6 months (n=978), unless stated otherwise

SC=Stratified Care; UC=Usual Care.

Resource use component	SC (n=415)	UC (n=563)
Primary care contacts:
Primary care general practitioner	0.96 (1.60)	1.15(2.11)
Primary care nurse	0.08 (0.57)	0.13 (0.65)
Primary care physiotherapist	0.35 (1.67)	0.29 (1.17)
Hospital-based care:
NHS consultant	0.33 (1.32)	0.37 (0.99)
NHS other physiotherapist	1.64 (2.64)	0.71 (1.87)
NHS acupuncturist	0.06 (0.59)	0.03 (0.44)
NHS osteopath	0.002 (0.05)	0.02 (0.22)
Private consultant	0.02 (0.23)	0.06 (0.40)
Private physiotherapist	0.21 (1.24)	0.37 (1.99)
Private acupuncturist	0.06 (0.55)	0.06 (0.42)
Private osteopath	0.29 (1.28)	0.26 (1.57)
Other Hospital-based care (n, %)
NHS surgery	10 (2.2%)	11 (2.0%)
Shoulder	2	1
Neck	0	1
Spine/Back	1	1
Hip	1	4
Knee	6	4
Private surgery	1(0.2%)	2 (0.4%)
Back	1	0
Knee	0	1
Shoulder	0	1
NHS Scans	13 (3.1%)	27 (4.8%)
NHS MRI investigations	64 (15.4%)	67 (11.9%)
NHS Blood tests	1 (0.2%)	5 (0.9%)
NHS injections	4 (1.0%)	24 (4.3%)
Private Scans	1 (0.2%)	1 (0.2%)
Private MRI investigation	12 (2.9%)	13 (2.3%)
Private Injections	1 (0.2%)	0

TABLE 18 - Baseline participant characteristics, per trial arm in the main trial

KOOS-PS, Knee Injury and Osteoarthritis Outcome Score Physical Function Short-Form; MSK-HQ, Musculoskeletal Health Questionnaire; NDI, Neck Disability Index; RQ, Reassurance Questionnaire; SF-12, Short Form questionnaire-12 items; SPADI, Shoulder Pain and Disability Index; TSK-11, Tampa Scale for Kinesiophobia-11 item version.

NRS-pain: 0 points = ‘no pain’, 10 points = ‘worst ever pain’.

Non-white group includes mixed, Asian, black and other.

Performance at work (0–10 NRS): 0 = problem ‘not at all’ affected performance over last 6 months, 10 = ‘so bad I am unable to do my job’.

NRS-distress: 0 points = ‘no distress’, 10 points = ‘extreme distress’.

NRS-confidence to manage: 0 points = ‘not at all confident’, 10 points = ‘extremely confident’.

Pain change on 11-point NRS scale (–5 to +5 points): –5 points = very much worse, 0 points = unchanged, +5 points = completely recovered (change from clinic appointment to time of self-report baseline completion).

RMDQ (0–24 scale): 0 = no low back pain/disability, 24 = maximum low back/pain disability.

NDI (0–50 scale): 0 = no disability, 50 = maximum disability.

SPADI– Disability subscale (0–100): 0 = no disability, 100 = maximum disability.

KOOS-PS (0–100): 0 = extreme disability, 100 = no disability.

SF-12 PCS (0–100): 0 = worst physical health score, 100 = best physical health score.

MSK-HQ (0–56 scale) based on summation of 14-items on a 0–4 scale and where 0 = worst musculoskeletal healthstatus and 56 = best musculoskeletal health-status.

MSK-HQ (0–56 scale) based on summation of 14-items on a 0–4 scale and where 0 = worst musculoskeletal health-status and 56 = best musculoskeletal health-status.

The Keele STarT MSK Tool score (0–12): 0 = lowest risk, 12 = highest risk.

EQ-5D-5L (–0.59 to 1.00):–0.59 = worst health status, 1.00 = best health status.

TSK-11 (11–44): 11 = minimum fear avoidance, 44 = maximum fear avoidance.

RQ (12–84 scale): 12 = no reassurance, 84 = high reassurance (subscales all recorded on 3–21 scale: 3 = no reassurance, 21 = high reassurance).

Note

p-values derived through linear or generalised mixed model accounting for GP–practice clustering (random factor). Bold indicates where there is a difference between the two groups for that variable.

Key characteristics		Stratified care arm (N = 534)	Usual care arm (N = 677)	p-value
Point of consultation with GP
Pain intensity (0–10-point NRS),a mean (SD)		6.8 (1.9)	6.7 (2.0)	0.726
Back		7.1 (1.8)	6.9 (1.9)	0.309
Neck		6.9 (1.6)	6.7 (2.2)	0.555
Shoulder		6.7 (1.8)	6.8 (1.9)	0.697
Knee		6.2 (2.1)	6.4 (2.3)	0.578
Multisite		7.2 (1.9)	7.0 (1.5)	0.524
Baseline questionnaire
Age (years), mean (SD)		57.8 (15.3)	61.8 (15.0)	0.004
Sex (female), n (%)		313 (58.6)	400 (59.1)	0.869
Days between consultation and returning questionnaire, mean (SD)		16.6 (27.3)	16.8 (28.0)	0.896
Ethnicity (white), n (%)b		513 (96.8)	659 (97.6)	0.400
Lives alone (yes), n (%)		91 (17.2)	121 (18.0)	0.936
Currently employed (yes), n (%)		275 (53.8)	286 (43.7)	0.001
Performance at work in past 6 months (current workers only, 0–10-point NRS), mean (SD)c		4.9 (3.1)	4.8 (3.1)	0.725
Performance at work in past 6 months (total study population, 0–10-point NRS), mean (SD)c		4.8 (3.2)	4.6 (3.2)	0.529
Time off work (yes), n (%)		99 (32.7)	203 (31.0)	0.803
Number of days off (from yes subgroup), median (IQR)		5 (2.5–15)	10 (4–20)	0.341
Health literacy (need help), n (%)				0.580
Never		434 (82.5)	539 (81.1)
Rarely		44 (8.4)	58 (8.7)
Sometimes		31 (5.9)	42 (6.3)
Often		14 (2.7)	18 (2.7)
Always		3 (0.6)	8 (1.2)
Pain area affected, n (%)				0.007
Back		214 (40.1)	243 (35.9)
Neck		61 (11.4)	68 (10.0)
Shoulder		71 (13.3)	59 (8.7)
Knee		157 (29.4)	222 (32.8)
Multisite pain		31 (5.8)	85 (12.6)
Pain intensity (0–10-point NRS),a by pain area, mean (SD)		6.3 (2.2)	6.4 (2.2)	0.775
Back		6.6 (2.2)	6.5 (2.2)	0.876
Neck		6.3 (2.0)	5.9 (2.2)	0.325
Shoulder		6.6 (1.9)	6.8 (2.2)	0.905
Knee		5.9 (2.3)	6.1 (2.4)	0.412
Multisite		6.5 (2.5)	6.8 (1.7)	0.600
Distress (0–10-point NRS), mean (SD)d		5.9 (2.6)	5.8 (2.6)	0.914
Confidence to manage (0–10-point NRS), mean (SD)e		5.1 (2.5)	5.3 (2.6)	0.186
Pain duration, n (%)				0.674
< 3 months		126 (23.9)	180 (26.7)
3–6 months		106 (20.1)	101 (15.0)
7–12 months		65 (12.3)	86 (12.8)
1–2 years		64 (12.1)	83 (12.3)
3–5 years		74 (14.0)	88 (13.1)
6–10 years		34 (6.4)	49 (7.3)
> 10 years		59 (11.2)	87 (12.9)
Overall pain change (–5 to +5 points), mean (SD)f		0.3 (2.1)	0.3 (2.0)	0.833
Previous episodes in last 3 years, n (%)				0.052
0	135 (25.5)		125 (18.5)
1	69 (13.0)		75 (11.1)
2–3	98 (18.5)		133 (19.7)
4–9	70 (13.2)		115 (17.0)
≥ 10	158 (29.8)		227 (33.6)
Previous surgery related to problem, n (%)				0.162
0	455 (89.6)		564 (85.5)
1	38 (7.5)		61 (9.2)
2	5 (1.0)		24 (3.6)
≥ 3	10 (2.0)		11 (1.7)
Days of moderate activity in last week, median (IQR)	2 (1–4)		2 (0–5)	0.716
Physical function, mean (SD)
Back (RMDQ)g	9.9 (5.8)		9.2 (5.6)	0.159
Neck (NDI)h	17.5 (8.7)		15.7 (8.7)	0.241
Shoulder (SPADI-Disability subscale)i	46.8 (24.1)		50.9 (26.1)	0.776
Knee (KOOS-PS)j	56.5 (15.2)		55.8 (17.9)	0.711
Multisite (SF-12 PCS)k	37.6 (8.8)		33.2 (10.3)	0.035
Standardised function scale (overall mean 0, SD 1)	0.01 (0.97)		0.00 (1.02)	0.844
MSK-HQ (0–56), mean (SD)l	28.7 (9.9)		29.5 (10.3)	0.604
The Keele STarT MSK Tool (clinical version), mean (SD)m	7.1 (2.7)		7.0 (2.9)	0.969
The Keele STarT MSK Tool (clinical version): risk subgroup, n (%)				0.964
Low risk (0–4 score)	98 (19.5)		126 (20.1)
Medium risk (5–8 score)	238 (47.3)		286 (45.6)
High risk (9–12 score)	167 (33.2)		215 (34.3)
Health-related quality of life (EQ-5D-5L), mean (SD)n	0.56 (0.23)		0.55 (0.24)	0.754
Fear avoidance beliefs (TSK-11), mean (SD)o	25.4 (6.4)		24.8 (6.5)	0.262
Listed long term conditions, n (%)				0.237
0	184 (34.5)		202 (29.8)
1	192 (36.0)		247 (36.5)
2	103 (19.3)		147 (21.7)
≥ 3	55 (10.3)		81 (12.0)
Perceived reassurance from GP consultation (RQ), mean (SD)p
Total	58.7 (16.0)		59.8 (16.6)	0.686
Data gathering	16.5 (4.3)		16.6 (4.6)	0.885
Relationship building	16.7 (4.3)		16.9 (4.5)	0.748
Generic	10.7 (5.0)		10.6 (5.2)	0.784
Cognitive	14.9 (5.1)		15.5 (5.1)	0.378
Satisfaction with GP care in last 6 months, n (%)				0.258
Very satisfied	123 (23.7)		190 (28.3)
Quite satisfied	173 (33.3)		244 (36.3)
No opinion	149 (28.7)		153 (22.8)
Not very satisfied	58 (11.2)		73 (10.9)
Not at all satisfied	16 (3.1)		12 (1.8)
Preferential mode of follow-up, n (%)				0.013
Text message	285 (53.4)		310 (45.8)
Post	249 (46.6)		367 (54.2

In the primary analysis there were no statistically significant differences in pain intensity over months 1–6 between the two arms, with mean values of 4.4 points (SD 2.3 points) for stratified care and 4.6 points (SD 2.4 points) for usual care (see Appendix 6, Figure 5 and Table 19). The adjusted mean difference was –0.16 points [95% confidence interval (CI) –0.65 to 0.34 points; p = 0.535], translating to a SMD (effect size) of –0.08 (95% CI –0.33 to 0.17) (see Appendix 6, Table 19). Mean differences in pain intensity were consistently greater in the last 3 months than the first 3 months; however, the average mean difference between the stratified care arm and usual care arm over months 4–6 was not significant (0.33 points, 95% CI –0.84 to 0.19 points; p = 0.211). Most sensitivity analyses showed no statistically significant between-arm differences, despite showing consistent slightly favourable results for stratified care (see Appendix 6, Figures 10–17). Analysis of the MCIC of a > 1-point difference in pain outcome gave an overall odds ratio (OR) of 1.26 (95% CI 0.89 to 1.78; p = 0.188); however, there was a significant OR (1.54, 95% CI 1.09 to 2.17; p = 0.013) for the 4–6 months comparison (see Appendix 6, Table 20).

FIGURE 10.

Pain intensity (NRS) scores per trial arm for patients at low risk of poor outcome, over time. IQ, initial questionnaire; M1–6, months 1–6 SMS or brief questionnaire scores; M6FQ, month 6 follow-up questionnaire; PoC, point of consultation.

FIGURE 11.

Pain intensity (NRS) scores per trial arm for patients at medium risk of poor outcome, over time. IQ, initial questionnaire; M1–6, months 1–6 SMS or brief questionnaire scores; M6FQ, month 6 follow-up questionnaire; PoC, point of consultation.

FIGURE 12.

Pain intensity (NRS) scores per trial arm for patients at high risk of poor outcome, over time. IQ, initial questionnaire; M1–6, months 1–6 SMS or brief questionnaire scores; M6FQ, month 6 follow-up questionnaire; PoC, point of consultation.

FIGURE 13.

Pain intensity (NRS) scores per trial arm for patients with back pain. IQ, initial questionnaire; M1–6, months 1–6 SMS or brief questionnaire scores; M6FQ, month 6 follow-up questionnaire; PoC, point of consultation.

FIGURE 14.

Pain intensity (NRS) scores per trial arm for patients with neck pain. IQ, initial questionnaire; M1–6, months 1–6 SMS or brief questionnaire scores; M6FQ, month 6 follow-up questionnaire; PoC, point of consultation.

FIGURE 15.

Pain intensity (NRS) scores per trial arm for patients with shoulder pain. IQ, initial questionnaire; M1–6, months 1–6 SMS or brief questionnaire scores; M6FQ, month 6 follow-up questionnaire; PoC, point of consultation.

FIGURE 16.

Pain intensity (NRS) scores per trial arm for patients with knee pain. IQ, initial questionnaire; M1–6, months 1–6 SMS or brief questionnaire scores; M6FQ, month 6 follow-up questionnaire;, point of consultation.

FIGURE 17.

Pain intensity (NRS) scores per trial arm for patients with multisite pain. IQ, initial questionnaire; M1–6, months 1–6 SMS or brief questionnaire scores; M6FQ, month 6 follow-up questionnaire; PoC, point of consultation.

TABLE 19 - Pain intensity scores (0–10-point NRS) and between-arm differences in the main trial

The between-group differences were calculated through use of linear mixed models with practice and participants specified as random factors, and practice size, participants’ age, sex and point-of-consultation pain score specified as fixed factors.

If the last monthly SMS/brief questionnaire response was missing it was imputed using the corresponding pain response from the returned 6-month questionnaire (if completed within 20 days of the date of issue).

Primary end point (average of available data for 1–6 months follow-up). Average summary mean (SD) relates to the mean of available 1–6-month follow-up data.

Sensitivity analysis (1–6 months).

Sensitivity analysis comparing those with/without complete monthly follow-up.

Kenward-Roger/Satterthwaite adjustment to CI: average 1–6 months –0.68 to 0.37; average 4–6 months –0.87 to 0.22.

Post hoc analysis (not pre-specified).

Sensitivity analysis (4–6 months): post hoc analysis due to patients not receiving their intervention immediately owing to waiting lists.

Note

Contrast arm × time: chi-squared test 19.8, 5 degrees of freedom; p = 0.0014.

SMD (effect size) relative to NRS-pain point-of-consultation SD (1.95).

(i) Constrained baseline model –0.18, 95% CI –0.53 to 0.16; p = 0.302.

(ii) Per-Protocol analysis (based on consultation template) –0.11, 95% CI –0.62 to 0.41; p = 0.681.

(iii) Complier-average causal effect analysis (compliance based on consultation template) 0.47, 95% CI –2.16 to 3.10; p = 0.725.

(iv) MI inclusive of additional baseline covariates –0.03, 95% CI –0.48 to 0.42; p = 0.900.

(v) Practice-level analysis –0.29, 95% CI –0.95 to 0.36; p = 0.358.

(i) Constrained baseline model –0.35, 95% CI –0.74 to 0.04; p = 0.082.

(ii) Per-protocol analysis (based on consultation template) –0.30, 95% CI –0.84 to 0.23; p = 0.270.

(iii) Complier-average causal effect analysis (compliance based on consultation template) 0.93, 95% CI –1.98 to 3.83; p = 0.532.

(iv) MI inclusive of additional baseline covariates –0.15, 95% CI –0.64 to 0.33; p = 0.536.

(v) Practice-level analysis –0.45, 95% CI –1.16 to 0.27; p = 0.208.

(i) Average between-group difference in mean scores for participants with 6/6 (100%) monthly responses: –0.24, 95% CI –0.76 to 0.39; p = 0.389.

(ii) Average between-group difference in mean scores for participants with incomplete (< 6) monthly responses: 0.01, 95% CI 0.68 to –0.65; p = 0.971.

The additional analyses presented were additional sensitivity analyses carried out across 1–6 months, and 4–6 months, based on recommendations from the DMC. The first five are 1–6 months, the next five are 4-6 months (due to potential delayed impact of the intervention), and the final two were sensitivity analyses for related to participants who did/did not complete all follow-up time points.

Time point	Stratified care, n	Usual care, n	Stratified care mean (SD)	Usual care mean (SD)	Mean difference (95% CI)a	SMD (95% CI)	p-value
Point of consultation with GP	534	675	6.8 (1.9)	6.7 (2.0)	–	–	–
Baseline	533	675	6.3 (2.2)	6.4 (2.2)	–	–	–
Month 1	491	632	5.1 (2.5)	5.0 (2.6)	0.17 (–0.34 to 0.68)	0.09 (–0.17 to 0.35)	0.51
Month 2	470	623	4.7 (2.6)	4.6 (2.7)	0.09 (–0.43 to 0.60)	0.05 (–0.22 to 0.31)	0.74
Month 3	478	615	4.2 (2.7)	4.5 (2.8)	–0.21 (–0.73 to 0.30)	–0.11 (–0.37 to 0.15)	0.42
Month 4	455	604	4.0 (2.7)	4.5 (2.9)	–0.40 (–0.92 to 0.13)	–0.21 (–0.47 to 0.07)	0.14
Month 5	451	601	3.9 (2.8)	4.4 (2.9)	–0.35 (–0.88 to 0.18)	–0.18 (–0.45 to 0.09)	0.2
Month 6b	446	593	3.8 (2.8)	4.2 (2.9)	–0.24 (–0.78 to 0.30)	–0.12 (–0.40 to 0.15)	0.39
Average (1–6 months)c,d,e	515	663	4.4 (2.3)	4.6 (2.5)	–0.16 (–0.65 to 0.34)f	–0.08 (–0.33 to 0.17)	0.54
Average (4–6 months)g,h	486	642	4.0 (2.6)	4.3 (2.7)	–0.33 (–0.84 to 0.19)f	–0.17 (–0.43 to 0.10)	0.21

TABLE 20 - Number (%) achieving the MCIC for pain intensity NRS (at least 1-point change)

NNT, number needed to treat.

The between-group differences were calculated through use of linear mixed models with practice and participants specified as random factors, and practice size, participants’ age, sex and point-of-consultation pain score specified as fixed factors.

Absolute percentage difference derived from the odds ratio using the observed proportion in the control arm as the reference.

NNT (derived as the reciprocal of the percentage difference) where SC superscript denotes the NNT in respect of favourable stratified care arm and UC superscript denotes the NNT in respect of favourable usual care arm.

If the last monthly SMS/brief questionnaire response was missing it was imputed using the corresponding pain response from the returned 6-month questionnaire (if completed within 20 days of the date of issue).

Sensitivity analysis using different cut-off points (1–6 months).

Post hoc analysis (not pre-specified).

Sensitivity analysis using different cut-off points (4–6 months).

Note

Contrast arm × time: chi-squared test 20.3, 5 degrees of freedom; p = 0.0011.

(i) MCIC 2-point change OR = 1.34, 0.75 to 2.42; p = 0.322.

(ii) MCIC 3-point change OR = 1.28, 0.71 to 2.32; p = 0.410.

(i) MCIC 2-point change OR = 1.87, 1.01 to 3.43; p = 0.045.

(ii) MCIC 3-point change OR = 1.88, 1.02 to 3.48; p = 0.044.

The additional analyses presented were additional sensitivity analyses carried out across 1–6 months, and 4–6 months, based on recommendations from the DMC. The first five are 1–6 months, the next five are 4-6 months (due to potential delayed impact of the intervention), and the final two were sensitivity analyses for related to participants who did/did not complete all follow-up time points.

Month	Stratified care, n/N (%)	Usual care, n/N (%)	OR (95% CI)a	p-value	Percentage difference (95% CI)b	NNT (95% CI)c
1	319/491 (65.0)	404/632 (63.9)	0.89 (0.48 to 1.66)	0.72	–2.7 (–18.0 to 10.7)	36.6UC (5.6UC to 9.3SC)
2	332/470 (70.6)	418/623 (67.1)	1.11 (0.59 to 2.10)	0.74	2.3 (–12.5 to 14.0)	44.2SC (8.0UC to 7.2SC)
3	361/478 (75.5)	414/615 (67.3)	1.92 (1.01 to 3.67)	0.047	12.5 (0.2 to 21.0)	8.0SC (457SC to 4.8SC)
4	346/455 (76.0)	397/604 (65.7)	2.47 (1.29 to 4.74)	0.007	16.9 (5.5 to 24.4)	5.9SC (18.2SC to 4.1SC)
5	349/451 (77.4)	408/601 (67.9)	2.17 (1.12 to 4.20)	0.021	14.2 (2.4 to 22.0)	7.0SC (41.3SC to 4.5SC)
6d	347/446 (77.8)	412/593 (69.5)	2.03 (1.04 to 3.94)	0.037	12.7 (0.8 to 20.5)	7.9SC (121SC to 4.9SC)
Mean 1–6 monthse	371/515 (72.0)	434/663 (65.5)	1.66 (0.98 to 2.82)	0.061	10.4 (–0.5 to 18.8)d	9.6SC (218UC to 5.3SC)d
Mean 4–6 monthsf,g	367/486 (75.5)	423/642 (65.9)	2.22 (1.26 to 3.89)	0.006	15.2 (5.0 to 22.4)d	6.6SC (20.0SC to 4.5SC)d

FIGURE 7.

Incremental net benefits (INB) of SC vs UC (95% CI) complete case.

FIGURE 8.

Main trial flowchart from work package 4. MDC, minimum data collection.

FIGURE 9.

Overall pain intensity (NRS) scores per trial arm, over time. IQ, initial questionnaire; M1–6, months 1–6 SMS or brief questionnaire scores; M6FQ, month 6 follow-up questionnaire; PoC, point of consultation.

Subgroup analyses showed some between-arm mean differences with a greater difference (although statistically non-significant) in patients at high risk than those at low risk, and in those with shoulder and knee pain than those with neck and back pain (see Appendix 6, Table 21–23). There were no statistically significant differences in secondary clinical outcomes at 6 months except for a significantly greater improvement in shoulder pain and function and a higher satisfaction with care in the stratified care arm than the usual care arm (see Appendix 6, Table 24).

TABLE 24 - Secondary outcome measures at 6 months, per trial arm in the main trial

KOOS-PS, Knee Injury and Osteoarthritis Outcome Score-Physical Function Short-Form; MSK-HQ, Musculoskeletal Health Questionnaire; NDI, Neck Disability Index; SF-12, Short Form questionnaire-12 items; SPADI, Shoulder Pain and Disability Index; TSK-11, Tampa Scale for Kinesiophobia-11 item version.

The between-group differences were calculated through use of linear mixed models with practice specified as a random factor, and practice size, participants’ age, sex, point-of-consultation pain score and corresponding baseline measure (if available) specified as fixed factors. Generalised mixed models were used for categorical outcomes, labelled as OR.

NRS-pain: 0 points = ‘no pain’, 10 points = ‘worst ever pain’.

Pain change on 11-point NRS scale (–5 to +5 points): –5 points = very much worse, 0 points = unchanged, +5 points = completely recovered (change from clinic appointment to time of self-report baseline completion).

RMDQ (0–24 scale): 0 = no low back pain/disability, 24 = maximum low back/pain disability.

NDI (0–50 scale): 0 = no disability, 50 = maximum disability.

SPADI– Disability subscale (0–100): 0 = no disability, 100 = maximum disability.

KOOS-PS (0–100): 0 = extreme disability, 100 = no disability.

SF-12 PCS (0–100): 0 = worst physical health score, 100 = best physical health score.

MSK-HQ (0–56 scale) based on summation of 14-items on a 0–4 scale and where 0 = worst musculoskeletal health-status and 56 = best musculoskeletal health-status.

The Keele STarT MSK Tool score (0–12): 0 = lowest risk, 12 = highest risk.

EQ-5D-5L (–0.59 to 1.00):–0.59 = worst health status, 1.00 = best health status.

TSK-11 (11–44): 11 = minimum fear avoidance, 44 = maximum fear avoidance.

Performance at work (0–10 NRS): 0 = problem ‘not at all’ affected performance over last 6 months, 10 = ‘so bad I am unable to do my job’.

Note

Number of days absent restricted to 120 days (over a 6-month period). Bold shows results that fall below the 0.05 significance level.

Key characteristics	Stratified care (n = 534)	Usual care (n = 677)	Mean difference (95% CI)a	p-value
Pain intensity reported in questionnaire (0–10-point NRS), mean (SD)b.1	3.5 (2.7)	4.1 (2.8)	–0.45 (–0.97 to 0.07)	0.088
Overall global change (–5 to +5), mean (SD)c.1	1.7 (2.4)	1.2 (2.6)	0.29 (–0.10 to 0.68)	0.143
Days of moderate activity in last week, mean (SD).1	3.2 (2.1)	3.2 (2.3)	–0.06 (–0.35 to 0.22)	0.656
Physical Function, mean (SD).1
Back (RMDQ)d	6.4 (5.7)	6.1 (5.5)	–0.30 (–1.30 to 0.70)	0.558
Neck (NDI)e	11.5 (8.9)	12.3 (9.1)	–1.01 (–4.81 to 2.80)	0.604
Shoulder (SPADI-Disability subscale)f	25.5 (27.3)	39.5 (31.4)	–11.1 (–19.8 to –2.3)	0.013
Knee (KOOS-PS)g	68.1 (14.7)	65.6 (20.0)	0.35 (–4.94 to 5.64)	0.896
Multisite (SF-12 PCS)h	40.4 (9.9)	35.8 (11.5)	0.31 (–4.40 to 5.01)	0.899
Standardised function score (0, 1)	–0.06 (0.94)	0.05 (1.04)	–0.07 (–0.22 to 0.08)	0.341
MSK-HQ (0–56), mean (SD)i.1	39.2 (11.0)	37.4 (12.1)	1.57 (–0.30 to 3.45)	0.100
The Keele STarT MSK tool (clinical version), mean (SD)j.1	4.8 (2.8)	5.1 (3.1)	–0.27 (–0.73 to 0.20)	0.265
The Keele STarT MSK (clinical version): risk subgroup, n (%)			OR 0.76 (0.51 to 1.13)	0.174
Low risk (0–4 score)	211 (53.0)	263 (48.6)
Medium risk (5–8 score)	136 (34.2)	180 (33.3)
High risk (9–12 score)	51 (12.8)	98 (18.1)
Health-related quality of life (EQ-5D-5L), mean (SD)k.1	0.67 (0.23)	0.65 (0.24)	0.022 (–0.003 to 0.048)	0.082
Fear avoidance beliefs (TSK-11), mean (SD)l.1	22.7 (7.0)	23.3 (7.4)	–0.64 (–1.70 to 0.42)	0.240
Currently employed (yes), n (%)	177 (44.1)	219 (39.8)	OR 0.83 (0.43 to 1.59)	0.574
Performance at work over last 6 months (0–10-point NRS), mean (SD)	3.4 (2.8)	3.5 (3.0)	–0.18 (–0.62 to 0.27)	0.434
Time off work in last 6 months (yes), n (%)	47 (23.6)	57 (23.2)	OR 0.97 (0.53 to 1.78)	0.919
Satisfaction with care, n (%)			OR 0.74 (0.57 to 0.98)	0.033
Very satisfied	125 (30.3)	138 (24.7)
Quite satisfied	141 (34.1)	205 (36.7)
No opinion	97 (23.5)	134 (24.0)
Not very satisfied	39 (9.4)	64 (11.5)
Not at all satisfied	11 (2.7)	18 (3.2)

Alterations to initial plans and work package limitations

There were no further alterations to the plans for the main cluster RCT during work package 4 (see Work package 3, Alterations to initial plans and work package limitations for the changes made to the main trial following the pilot RCT). It was not possible to limit the automatic firing of the stratified care EHR template within the EMIS system as much as would be desirable in the consultation. This meant it fired when the patient’s musculoskeletal treatment plan was not necessarily being considered in a consultation (e.g. when recording medication-only changes), or where musculoskeletal pain was not the primary reason for the consultation (e.g. in cases where musculoskeletal pain was a comorbidity to a different main clinical issue). In addition, to reduce GP burden, the stratified care EHR template only fired once per patient and not during their subsequent consultations. This meant that the opportunity to change GPs’ clinical decision-making for an individual patient according to stratified care was limited to the one consultation in which the stratified care template ‘fired’.

There are several limitations that are worth highlighting, particularly as the trial findings contrast with those of our STarT Back stratified care trial6 conducted among patients with low back pain. Possible explanations for this include the low GP fidelity to using the risk tool (29.76%) and the potential lack of effectiveness of the matched treatments options used. We saw no evidence that the low levels of GP tool use were due to difficulties in accessing or using the medical record interface (probably because our intervention template was embedded into their existing record system). We think it more likely that the low levels of use were due to the timing of template trigger, which often occurred after patients had left the room, and also due to the current time-pressured context of UK primary care. This was evidenced by GPs stating they ‘do not have time’, or ‘patient was not present’ in 49.80% of potentially suitable musculoskeletal consultations. Following the pilot trial findings (where GP tool fidelity was 40% of eligible consultations) we revised our GP fidelity expectations to 25% of musculoskeletal-coded consultations, primarily because GPs reported the intervention was only appropriate where musculoskeletal pain was the primary problem for the visit, which may be as low as only 50% of all GP musculoskeletal-coded consultations.

In relation to the specific limitations of the matched treatment options, it should be noted that in this trial we did not train physiotherapists in biopsychosocial approaches or optimise the musculoskeletal clinical pathways to deliver improved risk-matched treatments. In our previous STarT Back trial6 we included a 3-day physiotherapy training programme for physiotherapists delivering the medium risk treatment and a 6-day training programme (and ongoing regular mentoring) on psychologically-informed physiotherapy skills for those treating patients at high risk of poor outcome. In the STarT Back trial,6 mean change in Roland Morris Disability Questionnaire (RMDQ) score in the intervention arm at 4 months was 4.7 versus 3.0 points in the control arm. In the Keele STarT MSK trial,6 back pain function improved by much less at 6 months (RMDQ score: intervention = 3.5 points versus control = 3.1 points). To date, the authors are not aware of any other stratified care trials for low back pain providing as intense a training and mentoring programme as the STarT Back Trial. 6 Another consideration is that other successful primary care risk-prediction tools (such as QRISK, a risk prediction algorithm to help enable doctors to identify patients at most risk of heart disease and stroke,44 for estimating cardiovascular risk) have the benefit of highly effective pharmaceutical treatment for those identified at increased risk. Therefore, it seems that a key challenge for future trials of musculoskeletal risk-based stratified care will be to provide more effective treatments for those at increased risk of poor outcome.

Conclusions

The results of this cluster RCT demonstrated that the model of stratified care tested in this programme and delivered by GPs for patients consulting with one of the five most common musculoskeletal pain presentations did not lead to superior clinical outcomes compared with usual non-stratified care. However, data on clinical decision-making showed that the stratified care EHR template and clinician training led to positive changes in GP clinical decision-making, with fewer prescriptions for muscle relaxants, better provision of written self-management information and many more referrals to physiotherapy. The health economic analysis showed that although the costs of stratified care were very similar to usual care, there were small benefits from the stratified care intervention. Although the findings suggest that risk-based stratified care is potentially a cost-effective use of health-care resources when applying conventional rules of cost-effectiveness, the minimal differences suggest caution in the interpretation of this result. Qualitative data from GPs, physiotherapists and patients involved in the main trial about their experiences of stratified care were generally positive.

Interrelationship with other parts of the programme

This work package followed work packages 1–3 and tested the clinical effectiveness and cost-effectiveness of the prognostic stratified care approach for patients consulting in primary care with one of the five most common musculoskeletal pain presentations in primary care.

Work package 1

During development of the plans for this work package, a PPIE meeting was held to discuss and feedback on patient-facing materials (e.g. patient information leaflets, consent forms and questionnaires). PPIE members provided feedback on the medical record review procedures as well as the novel patient reminder systems (via text and e-mail reminders). Members reviewed the draft stratification tool (the Keele STarT MSK Tool) for face validity, and gave important feedback on the potential replacement candidate items for the revised tool. Once data collection and analysis were complete, a PPIE meeting (attended by 9 patients/carers) discussed the findings and suggested how best to share results with patients and general practices. This resulted in results being displayed on posters in the practices rather than distributed through leaflets. PPIE members suggested simplification of the content shared with participants, with an emphasis on thanking them for their involvement rather than sharing complex diagrams of the results. These suggestions were implemented.

Work package 2

The evidence about the effectiveness of available treatments for the five most common musculoskeletal pain presentations was summarised and presented to stakeholders, including our programme PPIE members. A PPIE member was invited to the meeting of the wider stakeholder group for work package 2 to assist the team with decisions about which existing patient-facing websites and patient-written self-management information leaflets GPs should use as part of the suite of matched treatment options.

Work package 3

Our PPIE co-applicant (Mr John Murphy) attended a protocol development day for work packages 3 (feasibility and pilot trial) and 4 (main trial) to advise on the protocol from a patient’s perspective. Subsequently, a PPIE day (attended by 12 PPIE members with musculoskeletal pain conditions) was held to review the protocol and assist with development of documentation for the trials. Those who attended were later involved in follow-up activities via post and e-mail to review further amended documents, including documents supporting the qualitative research in the trial and the postal questionnaires that were sent to patients.

Once the feasibility and pilot trial was under way, nine patients/carers took part in a further PPIE meeting to discuss progress and the emerging plans for the main trial, including the patient-facing documentation. PPIE members made suggestions on ways to improve the uptake of patient interviews for the qualitative component. At the end of the pilot trial, seven PPIE members met to discuss plans for the main trial, provided valuable feedback on the pilot trial results and helped finalise the wording of the clinician-completed version of the Keele STarT MSK Tool, pointing out that asking about the last 2 weeks for patients with a long-term condition may not seem relevant and suggesting that the GP needed to provide context to help the patient understand why they were asking the questions on the tool. They also suggested re-ordering the questions to improve flow and understanding. Amendments were made as a result and the decision was made to use the clinician version of the tool at the point-of-consultation in the main trial.

The qualitative researchers met with four PPIE members to discuss emerging themes from the qualitative research in the pilot trial, as well as data extracts from patient interviews. PPIE members provided their interpretation of the key issues in the data. This informed analysis and fed into the resulting paper from the pilot trial.

Work package 4

The study PPIE members met to review the near-final versions of the main trial patient-facing documentation (e.g. the invitation letter, information leaflet, questionnaire and consent form) prior to submission for ethics approval. They suggested a number of improvements to the documents that were implemented, including improved consistency of the wording used, use of bullet points rather than lengthy sentences, use of an A4 booklet for the information leaflet and more use of colour. Plans for the main trial were also adapted based on discussions held during the meeting. The patient prize draw used in the pilot trial to incentivise patients to return their questionnaires was not taken forward into the main trial plans. PPIE members were very supportive of the plan to ask GPs to seek verbal consent from patients in the consultation to share their contact details with the research team in order to send out information about the study.

A further PPIE meeting was held to discuss the main trial with the PPIE group and to share available quantitative and qualitative findings in order to seek PPIE members’ views on the interpretation of the data. PPIE members contributed by reviewing some of the qualitative research data from the main trial, reading through direct quotes from trial participants in pairs and noting their thoughts before joining a wider group discussion about the interpretations of the findings. PPIE members helped to influence the plans for patient-focused dissemination of the results from the main trial. Beyond this programme, our PPIE members continue to be involved in other research studies (e.g. as steering group members) and have contributed to other bids and funded research led by Keele University, sustaining and further developing PPIE partnership working and supported by the broader RUG at Keele.

Implications for practice

1. The Keele STarT MSK Tool is a valid tool with which to identify patients with different levels of risk of persistent pain, and therefore provides additional systematic information about an individual patient’s prognosis that can help clinicians direct patients to the most appropriate treatments. The approach of using one tool for this wide range of patients has the major benefit of simplicity for clinical practice, removing the complexity that would result from multiple, pain site-specific stratification tools.

2. Given the high prevalence of musculoskeletal pain and the variability in clinical decision-making in primary care, the positive impacts on some aspects of clinical decision-making we observed together with the finding that this new way of working does not negatively affect patients’ outcomes and is likely to be cost neutral suggests that this model of stratified care may bring benefits for the NHS.

3. Supporting GPs with bespoke computerised EHR templates that fire within the consultation with a patient and support clinical decision-making can help change specific behaviours in ways that reduce low-value care. Specifically, we showed that our EHR template increased the provision of written self-management information and referral for non-pharmacological treatments for patients with musculoskeletal pain.

4. Key challenges to overcome, however, include how to support GPs in busy clinical consultations to use stratification tools with and offer recommended treatments to as many appropriate patients as possible, and how to improve primary care treatments in ways that lead to better patient outcomes.

Recommendations for future research

1. There are challenges in designing and testing stratified care interventions that are sufficiently potent to bring about change in the face of challenging clinical contexts and likely losses to fidelity. We recommend conducting careful feasibility work prior to testing stratified care interventions in definitive pragmatic trials.

2. Our stratified care intervention had two components: the use of a stratification tool and then the matching of patient subgroups to recommended treatment options. The first of these worked well to identify patient subgroups and the tool has already been shared with over 1000 tool license requestees, leading to other research. We found that GPs struggled to incorporate use of the tool in their short consultations with patients, and losses in fidelity to the matched treatment options are key potential explanations for the trial results. Future research to test ways to better support clinicians to use stratification tools (including qualitative research) and to better match patient subgroups to appropriate treatments is needed.

3. To help address GPs’ reluctance to use stratification tools. In particular, testing the use of an EHR template that fires at every patient consultation and/or testing a simpler model of stratified care that may be easier to deliver would be valuable.

4. The stratification tool relied on self-reported data collected by GPs from patients within short consultations, since routinely collected primary care data does not contain many variables that are known to predict outcomes specifically in musculoskeletal pain. Future research that identifies ways to ensure such prognostic factors are routinely collected and can be used to provide clinicians with prognostic information in ways that do not add time to short consultations would be helpful. This approach would also overcome difficulties for patients who may have low health literacy levels or learning needs.

5. A key next step in research is to use prognostic information not only to stratify patients into subgroups as part of stratified health care, but to provide personalised outcome predictions as part of personalised health care. This could include predicting an individual’s likely future pain, function scores or their probability of being absent from work to achieve greater treatment tailoring. The European Union-funded Back-UP research programme (https://backup-project.eu/) developed personalised prognostic models for back and neck pain patients using the data sets from this Keele STarT MSK programme.

6. Clearly it would be useful for the prognostic tool to be available in more languages than English. We therefore recommend researchers help to culturally translate the tool and validate it in different languages. To date, there have been several translations of the STarT MSK Tool, including Hebrew, German, Norwegian, Persian, and Dutch versions. However, there are around a further 15 language versions in development. This information will be available in due course online (https://www.keele.ac.uk/startmsk/).

7. We recommend that further research is conducted to better understand how the new prognostic tool relates to the secondary care context. There may also be possibilities to use the tool and evaluate its role in initiatives focused on reducing current long NHS waiting lists (e.g. in supporting decisions about which patients might be appropriate for sign-posting to self-management resources).

Non-author contributions

Ying Chen (Associate Professor in Biostatistics & Epidemiology) served as the junior statistician for work package 3. Steff Garvin (Trial manager) served as the trial manager for work packages 3 and 4 and supported their operational management for the Keele Clinical Trials Unit. Vince Cooper (Senior Lecturer in General Practice) served to support work packages 2, 3 and 4 by providing specialist GP clinical knowledge to help design the intervention, recruit practices, design and deliver the GP training, interpret the pilot and main trial findings and support the publications for these work packages. Christian Mallen (Professor of General Practice) served as a co-applicant for the programme, and supported the interpretation of the main trial results and its publication. Danielle van der Windt (Professor of Primary Care Epidemiology) served as a co-applicant for the programme, and supported the interpretation of the main trial results and its publication. Elaine Hay (Professor of Community Rheumatology) served as a co-applicant for the programme, and supported the interpretation of the main trial results and its publication.

Contribution of authors

Nadine E Foster (https://orcid.org/0000-0003-4429-9756) (Professor of Musculoskeletal Health) served as overall chief investigator for the whole programme of research, led the funding application, provided senior support on the detailed protocol development and delivery of each work package, chaired the Programme Management Group, and was the senior author on the pilot and main randomised trials.

Kate M Dunn (https://orcid.org/0000-0002-6202-2606) (Professor of Epidemiology) served as the principal investigator for work package 1 and led the conception and writing of this work package for funding; led the design and development of the STarT MSK tool and its validation; and led or contributed to the publications from this and other work packages.

Joanne Protheroe (https://orcid.org/0000-0002-6202-2606) (Professor of General Practice) served as the principal investigator for work package 2; led the conception and writing of this work package for funding; led the design, development of the matched treatment options and intervention format and content of GP support packages; and led or contributed to publications from this and other work packages.

Jonathan C Hill (https://orcid.org/0000-0001-6246-1409) (Professor of Physiotherapy) served as the principal investigator for work packages 3 and 4; led the conception and writing of these work packages for funding; led the design, delivery, analysis and write-up of these work packages; and led first draft and had overall responsibility for submitting this overall report for publication. He is the corresponding author for the report.

Martyn Lewis (https://orcid.org/0000-0001-5290-7833) (Reader in Medical Statistics) served as the senior statistician for the trial. He had full access to the data and took responsibility for data integrity and accuracy of data analysis. He participated in the design, conduct, statistical analysis and publication writing of all work packages.

Ben Saunders (https://orcid.org/0000-0002-0856-1596) (Lecturer in Applied Health Research) served as qualitative research associate in work packages 2 and 3, and social science lead in work package 4; and led the design, data collection, analysis and publication of the qualitative research informing the intervention development and the nested qualitative studies within the feasibility trial and main trial.

Sue Jowett (https://orcid.org/0000-0001-8936-3745) (Professor of Health Economics) served as the lead health economist, leading on the design, data collection, analysis and publication of the health economics presented in work packages 1 and 4.

Susie Hennings (https://orcid.org/0000-0002-8160-6658) (Senior Trials Manager for the Keele Clinical Trials Unit) supported the operational management of the programme of research for the final two work packages.

Paul Campbell (https://orcid.org/0000-0001-9148-882X) (Senior Research Associate) served as research fellow for work package 1 of the programme of research; contributed to the development of the Keele STarT MSK Tool; led the work package 1 protocol publication, co-ordinating work package 1 cohort study, supporting data analysis; and was co-author on the Keele STarT MSK Tool and health economic publications.

Kieran Bromley (https://orcid.org/0000-0002-4129-2519) (Research Associate) was the junior statistician who supported the analyses of the main trial in work package 4.

Bernadette Bartlam (https://orcid.org/0000-0002-8557-6222) (Senior Social Scientist) served as senior qualitative lead for work package 2, where she led the design, data collection, analysis and publication of the qualitative aspects of work package 2. She also supported Ben Saunders in the qualitative aspects of work packages 3 and 4 in relation to the design, data collection and publications.

Opeyemi Babatunde (https://orcid.org/0000-0002-5064-6446) (Lecturer: Evidence Synthesis and Applied Health) served as the research associate (systematic review) in work package 2, leading on the design, data collection, analysis and publication of the evidence synthesis informing the intervention development.

Simon Wathall (https://orcid.org/0000-0002-7107-5785) (Health Informatics Specialist) served by building the intervention and recruitment computer templates used inn work packages 3 and 4, and developing the data extraction searches used to extract patient data from GP electronic health records across the programme work packages.

Raymond Oppong (https://orcid.org/0000-0002-0815-4616) (Associate Professor in Health Economics) served as a health economist and was involved in design, data collection, analysis and publication of health economics aspects in work package 1.

Jesse Kigozi (https://orcid.org/0000-0001-7608-4923) (Assistant Professor in Health Economics) served in a health economics analysis and was involved in design, data collection, analysis and publication of health economics work presented in work package 4.

Adrian Chudyk (https://orcid.org/0000-0002-2990-9651) (Clinical Lecturer in General Practice) served to support work packages 3 and 4 by providing specialist GP clinical knowledge to help design the intervention, recruit and train general practices, interpret the pilot and main trial findings and the publications for these work packages.

Publications

Campbell P, Hill JC, Protheroe J, Afolabi EK, Lewis M, Beardmore R, et al. Keele Aches and Pains Study protocol: validity, acceptability, and feasibility of the Keele STarT MSK Tool for subgrouping musculoskeletal patients in primary care. J Pain Res 2016;9:807–18. https://doi.org/10.2147/JPR.S116614

Dunn KM, Campbell P, Lewis M, Hill JC, van der Windt DA, Afolabi E, et al. Refinement and validation of a tool for stratifying patients with musculoskeletal pain. Eur J Pain 2021;25:2081–93. https://doi.org/10.1002/ejp.1821

Oppong R, Lewis M, Campbell P, Dunn KM, Foster NE, Hill JC, Jowett S. Comparison of healthcare utilisation, costs and health-related quality of life across the subgroups defined by the Keele STarT MSK Tool. Rheumatol 2022;3:keac560.

Campbell P, Lewis M, Chen Y, Lacey RJ, Rowlands G, Protheroe J. Can patients with low health literacy be identified from routine primary care health records? A cross-sectional and prospective analysis. BMC Fam Pract 2019;20:101. https://doi.org/10.1186/s12875-019-0994-8

Saunders B, Bartlam B, Foster NE, Hill JC, Cooper V, Protheroe J. General practitioners’ and patients’ perceptions towards stratified care: a theory informed investigation. BMC Fam Pract 2016;17:125. https://doi.org/10.1186/s12875-016-0511-2

Babatunde OO, Jordan JL, Van der Windt DA, Hill JC, Foster NE, Protheroe J. Effective treatment options for musculoskeletal pain in primary care: a systematic overview of current evidence. PLoS One 2017;12:e0178621. https://doi.org/10.1371/journal.pone.0178621

Protheroe J, Saunders B, Bartlam B, Dunn KM, Cooper V, Campbell P, et al. Matching treatment options for risk sub-groups in musculoskeletal pain: a consensus groups study. BMC Musculoskelet Disord 2019;20:271. https://doi.org/10.1186/s12891-019-2587-z

Hill JC, Garvin S, Chen Y, Cooper V,Wathall S, Saunders B, et al. Stratified primary care versus non-stratified care for musculoskeletal pain: findings from the STarT MSK feasibility and pilot cluster randomised controlled trial. BMC Fam Pract 2020;21:30. https://doi.org/10.1186/s12875-019-1074-9

Saunders B, Hill JC, Foster NE, Cooper V, Protheroe J, Chudyk A, et al. Stratified primary care versus non-stratified care for musculoskeletal pain: qualitative findings from the STarT MSK feasibility and pilot cluster randomised controlled trial. BMC Fam Pract 2020;21:31. https://doi.org/10.1186/s12875-020-1098-1

Hill J, Garvin S, Chen Y, Cooper V,Wathall S, Bartlam B, et al. Computer-based stratified primary care for musculoskeletal consultations compared with usual care: study protocol for the STarT MSK cluster randomised controlled trial. JMIR Res Protoc 2020;9:e17939.

Hill JC, Garvin S, Bromley K, Saunders B, Kigozi J, Cooper V, et al. Risk-based stratified primary care for common musculoskeletal pain presentations: results of the STarT MSK cluster randomised controlled trial. Lancet Rheumatol 2022;4:E591–E602. https://doi.org/10.1016/S2665-9913(22)00159-X

Data-sharing statement

Any requests for access to the anonymised data will follow our data-sharing procedure. Requests for anonymised data will be reviewed by our Data Custodian and Academic Proposals Committee. The full statement on data sharing is publicly available from www.keele.ac.uk/health/fmhsresearchthemes/. All information will be held securely and in strict confidence. Each person in this study will be given a study number so that data from the study will not have any identifiable information, such as names and addresses, and cannot be traced. On this basis, these anonymised data will be kept electronically and may be used in other research studies. Requests for access to the data should be addressed to the corresponding author or to the data custodian.

Patient data

This work uses data provided by patients and collected by the NHS as part of their care and support. Using patient data is vital to improve health and care for everyone. There is huge potential to make better use of information from people’s patient records, to understand more about disease, develop new treatments, monitor safety, and plan NHS services. Patient data should be kept safe and secure, to protect everyone’s privacy, and it’s important that there are safeguards to make sure that it is stored and used responsibly. Everyone should be able to find out about how patient data are used. #datasaveslives You can find out more about the background to this citation here: https://understandingpatientdata.org.uk/data-citation.

Disclaimers

This report presents independent research funded by the National Institute for Health and Care Research (NIHR). The views and opinions expressed by authors in this publication are those of the authors and do not necessarily reflect those of the NHS, the NIHR, CCF, PGfAR or the Department of Health and Social Care. If there are verbatim quotations included in this publication the views and opinions expressed by the interviewees are those of the interviewees and do not necessarily reflect those of the authors, those of the NHS, the NIHR, the PGfAR programme or the Department of Health and Social Care.

Detailed Health Economics Report for the main trial

Results

Response rates and data completion

Resource use and cost per participant are reported by category, for all those with complete data at 6 months. Estimates for the base-case NHS/PSS perspective were based on the imputed participants’ data for NHS cost and EQ-5D data (n= 1211). The proportion of returned questionnaires at 6 months follow-up is reported (See Table 17: Descriptive and incremental health outcomes over 6 months for the base-case analysis and the complete case analyses). Complete cost data were available for 978(81%) participants at 6 months, while outcome data were available for 1070 (88%) including MDC data at 6 months. Aer excluding missing EQ-5D items, 1019 (84%) participants had a valid EQ-5D-5L score at 6 months.

TABLE 17 - General practice characteristics per trial arm in the main trial

IMD, Index of Multiple Deprivation.

For the English indices of area deprivation, the 32,844 Lower layer Super Output Areas in England are ranked according to their deprivation score. For each of the neighbourhood-level indices, the most deprived Lower layer Super Output Area in England is given a rank of 1, and the least deprived a rank of 32,844 (IMD is a weighted aggregate of the seven sub-indices).

Key characteristics	Stratified care (n = 12)	Usual care (n = 12)
Registered population size
Mean (SD)	8033 (3214)	7391 (3574)
Median (IQR)	7361.5 (6277.0–9954.5)	6981.5 (5616.5–8628.0)
Minimum, maximum	1994, 13248	2031, 13894
English indices of area deprivation, median (IQR)a
Income deprivation	19,390.5 (14,171.0–25,490.5)	17,483.5 (10,423.5–25,405.5)
Employment deprivation	14,676.0 (11,254.5–24,009.5)	14,010.0 (9897.0–24,563.0)
Education, skills and training deprivation	20,409.0 (10,228.0–28,632.5)	18,106.0 (8492.5–25,586.5)
Health deprivation and disability	15,656.0 (9502.0–24,854.0)	17,587.5 (12,628.0–22,859.5)
Crime	15,095.0 (52,53.5–24,809.5)	19,234.5 (12,500.5–26,251.5)
Barriers to housing and services	23,271.5 (17,355.5–26,325.5)	26,846.5 (19,260.0–30,273.0)
Living environment deprivation	9448.0 (2717.0–25,406.5)	15,491.0 (7144.5–21,202.5)
IMD	17,995.0 (11,240.0–23,126.5)	15,626.0 (11,146.0–24,248.5)

Health-related quality of life outcomes

Health-related outcomes over 6 months are reported in Table 17. At baseline, participants in the SC arm had higher EQ-5D scores compared to those in UC (0.5572 and 0.5542 respectively) and a slightly higher score at 6 months than those in the UC group (0.6715 and 0.6512 respectively). At 6 months, the mean adjusted QALY difference between the two groups was 0.0041 (-0.0013, 0.0095) in favour of SC, and this difference was not significant. Adjustments were made for clustering, age, gender and baseline EQ-5D. There was overall improvement over the 6 months reflected by increased EQ-5D scores from baseline through to 6 months in both trial arms.

Primary care consultations and medications

There were minimal differences in primary resource use and costs between the two groups, but overall, the primary care costs for UC were slightly higher than SC (see Tables 8 and 9 Costs related to prescribed medication were obtained from medical records. The mean prescription costs of prescribed medications were slightly lower for SC than UC (£1.07(5.10), £2.44 (11.60)), respectively) (see Table 9).

Hospital-based care

Hospital-based care resource use and costs of health professional contacts were very similar between the two groups with the exception of NHS physiotherapy treatment in the SC arm which was higher than UC (£95.18 and £41.00 respectively) (see Tables 8 and 9).

Private healthcare use

The proportions of individuals reporting private healthcare use are given in Table 8 (Resource components analysed in the trial) and Table 9 (Cost components analysed in the trial). The biggest area of difference in resource use and cost was in the use of private physiotherapy care among those allocated to UC (£12.44 and £21.43 respectively).

Imaging and other tests

Information about resource use and costs associated with hospital tests and investigations is shown in Tables 8 and 9. Data from the procedures and investigations showed more UC participants received NHS scans and injections than SC, with costs of overall NHS investigations and treatments £31.61 an d £61.13 for SC and UC respectively.

Work-related outcomes

Results regarding paid employment, work status, MSK-related work absence and reduced productivity are reported in Table 10 (Productivity costs analysed in the trial). Overall, reduced productivity (presenteeism) was very similar between the two groups (3.41 and 3.57 for SC and UC respectively) and the costs of work absence were slightly less in SC compared to UC (£698 and £738 respectively).

TABLE 10 - Cost components analysed in the trial. Values are mean (SD) costs per patient, by treatment group, for patients providing health care utilisation data at 6 months (n=978), unless stated otherwise

Between-group difference in mean scores (SC– UC) by generalized linear latent and mixed (gllamm) models adjusting for clustering,

Includes investigations such as MRI scans, CT scans and ultrasound scans, and injections; †Includes shoulder, neck, spine, back, hip, knee injury. SC=Stratified Care; UC=Usual Care.

Cost component	SC (n=415)	UC (n=563)
Primary care contacts:
Primary care general practitioner	31.81 (52.81)	37.87 (69.58)
Primary care nurse	3.14 (23.77)	5.52 (27.11)
Primary care physiotherapist	20.40 (96.94)	17.10 (68.11)
Prescripcttions	1.07 (5.10)	2.44 (11.60)
Hospital-based care
NHS consultant	51.72 (209.25)	59.31 (158.67)
NHS other physiotherapist	95.18 (152.96)	41.00 (108.72)
NHS Acupuncturist	3.49 (34.15)	1.95 (25.702)
NHS Osteopath	0.14 (2.847)	0.93 (12.67)
NHS other professionals	3.54 (41.36)	5.25 (78.07)
Private consultant	3.83 (36.45)	9.32 (64.00)
Private Physiotherapist	12.44 (72.11)	21.43 (115.565)
Private Acupuncturist	3.63 (31.79)	3.50 (24.21)
Private Osteopath	16.63 (74.38)	15.25 (91.05)
NHS private other	2.65 (33.98)	2.37 (32.27)
Other hospital-based care
NHS surgery†	110.07 (728.99)	106.34 (808.99)
NHS investigations and treatments††	31.61(86.22)	61.13(254.14)
Private surgery†	2.80 (57.04)	16.38 (284.23)
Private investigations/treatments††	6.86 (63.76)	3.93 (31.63)

Incremental costs and cost-effectiveness analysis

Table 11 (Cost-Utility analysis using the net-benefit regression approach) also provides a summary of the incremental costs, outcomes, net monetary benefit results and cost-effectiveness analysis using data from the base-case NHS perspective. Results from the healthcare and societal perspective are also reported.

TABLE 11 - Productivity costs analysed in the trial. Values are mean (SD) costs per patient, by treatment group, for patients providing work outcomes data at 6 months

The evaluation of work-related outcomes and the estimation of indirect costs focussed on the subsample of respondents in paid employment at 6 months (396/1211).

Mean performance at work on a scale of 0 to 10 where 0 indicates work performance not affected.

Mean difference of reported days off-work over 6 months and adjusted for paid employment at baseline and clustering.

Productivity costs obtained from days off-work over 6 months and adjusted for paid employment at baseline and clustering.

Work-related outcomesa	SC	UC
Baseline: working in paid employment (n, %)	275 (53.8)	286 (43.3)
Baseline: reported me off work during the last 6 months (n, %)	91 (30.6)	101 (30.2)
Working in paid employment at 6-months (n, %)	177 (44.1)	219 (39.8)
Performance at work at 6 months (mean, SD)b	3.41 (2.71)	3.57 (2.89)
Reported me off work at 6-months (n, %)	47 (23.6)	57 (23.2)
Days off-work during the last 6 months (mean, SD)a	6.13 (20.05)	6.48 (19.66)
Mean differencec (95% CI)	-1.09 (-5.87, 3.68)
Productivity costs during the last 6 months (mean, SD)a	698.81 (2285.24)	738.14 (2240.95)
Mean differenced (95% CI)	-124.89 (-669.73, 419.93)

The results show that, although the SC and UC costs to the NHS perspective were very similar between the groups, the cost of SC was slightly higher overall than the UC cost (SC: £356.36 and UC: £343.44 difference £6.85 (95% CI: (-107.82, 121.54)), and was associated with minimal health gains of 0.0041 QALYs (-0.0013, 0.0094). The net monetary benefit (NMB) was £132 if society’s willingness to pay for a QALY (λ) is valued at £20,000 (see Figure 4; Incremental net benefits (INB) of SC vs UC (95% CI) (Base-case) and Figure 6; Incremental net benefits (INB) of SC vs UC (95% CI) complete case). Uncertainty around the values is illustrated in Table 12 (Cost-effectiveness results for the within-trial economic analysis with 6-month horizon) for the NHS perspective. The uncertainty analysis showed that SC was more costly and slightly more effective, and is likely to be cost-effective at the £20,000 per QALY threshold range with a 0.73 probability for the NHS perspective. The CEACs show the probability that the SC intervention is cost-effective at different levels of willingness-to-pay for a QALY (See Figure 3; Cost Effectiveness Acceptability Curve (Base-case) and Figure 5; Cost Effectiveness Acceptability Curve (Complete-case) analysis). Uncertainty around the values is illustrated in Table 12 for the healthcare and societal perspective.

FIGURE 3.

The Keele STarT MSK Tool developed in work package 3 (clinician-completed version).

FIGURE 5.

Incremental net benefits (INB) of SC vs UC (95% CI) (Base-case).

FIGURE 6.

Cost Effectiveness Acceptability Curve (Complete-case) analysis.

TABLE 12 - Cost-Utility analysis using the net-benefit regression approach†

NHS, National Health Service, QALY, Quality-Adjusted Life Year.

Mean differences adjusted for clustering in cost and QALY outcomes^.

Complete-case dataset

	SC N=534	UC N=677	Mean differences or ICER (95% CI’s)
Imputed dataset analysis
Cost Analysis †
NHS cost (£)   Mean (Standard Deviation)	356.36 (864.01)	343.44 (942.92)	6.85 (-107.82, 121.54)
Healthcare cost (£)   Mean (Standard Deviation)	411.1 7(890.161)	417.11 (1035.28)	-5.92 (-116.3216, 104.46)
Societal cost (£)   Mean (Standard Deviation)	838.53 (2102.12)	762.05 (2028.20)	65.01 (-195.56, 325.59)
Effectiveness Analysis †
Unadjusted QALYs gained*   Mean (Standard Deviation)	0.3072 (0.1012)	0.3014 (0.1042)	0.0053 (-0.0134, 0.0241)
Adjusted QALYs gained*	0.3065	0.3018	0.0041 (-0.0013, 0.0094)
Complete-case analysis
Cost Analysis †
NHS cost (£)   Mean (Standard Deviation)	351.99 (910.58)	338.89 (988.83)	11.61 (-118.84, 142.08)
Healthcare cost (£)   Mean (Standard Deviation)	400.85 (940.16)	411.06 (1082.03)	-10.21 (-139.95, 119.53)
Societal cost (£)   Mean (Standard Deviation)	762.07 (2195.45)	745.16 (2111.96)	19.05 (-295.94, 334.04)
Effectiveness Analysis †
Unadjusted QALYs gained*   Mean (Standard Deviation)	0.3095 (0.1034)	0.3032 (0.1063)	0.0068 (-0.0141, 0.0277)
Adjusted QALYs gained*	0.3092	0.3033	(-0.00050.0053, 0.0112)

Sensitivity analyses

Healthcare and societal perspectives

Details of the sensitivity analysis from alternative perspectives are presented in Table 11 and 12. The results between the two groups showed that SC and UC were very similar, but overall SC was slightly more costly that UC from a societal perspective. The probability that SC was cost-effective at £20,000 per QALY was 0.76 and 0.61 from a healthcare and societal perspective respectively.

Complete-case analysis

Under this scenario from an NHS perspective, SC was slightly more costly (£11.61) but generated more QALYs on average (Diff, CI: 0.0053: -0.0005, 0.0112) when the analysis was restricted to participants with complete cost and outcome data (Table 12). The net monetary benefit value was £162 (λ=£20,000); and the ICER was £2,190 per QALY gained. The probability of SC being cost-effective at λ = £20,000 was 0.75. The details of the incremental cost-effectiveness and uncertainty analysis are reported in Table 12.

Subgroup analysis

Impact of stratified care on health-related quality of life by study group

Quality of life data (EQ-5D-5L and QALY scores) at baseline and 6 months follow-up by treatment arm and MSK subgroup are provided in Table 13 (Subgroup analysis: Costs and health outcomes mean (SD) scores by STarT MSK risk subgroups (Imputed analysis). At baseline and at the end of 6 month follow-up, participants in the SC arm had slightly lower EQ-5D scores compared to UC, in low risk and medium risk, but higher in high risk group. At 6 months, QALY outcomes in the low risk and medium risk groups slightly favoured UC: (difference SC-UC: -0.0015, -0.0001 respectively) and favoured SC for the high risk group (0.0117) with all CIs crossing zero.

TABLE 13 - Cost-effectiveness results for the within-trial economic analysis with 6-month horizon

Adjusted for age, sex, treatment allocation, study site, impingement type, baseline health-related quality of life and baseline costs.

Mean ICERs in north east quadrant of the cost-effectiveness plane where SC is more costly and more effective. NHS, National Health Service, QALY, Quality-Adjusted Life Year.

Mean differences adjusted for clustering in cost and QALY outcomes.

Complete-case dataset. ICER = Incremental cost-effectiveness ratio; CI = confidence interval

	Cost-effectiveness outcomes – NHS			Probability SC is cost- effective at cost-effectiveness threshold of
	Mean incremental costs (95% CI), £	Mean incremental QALYs(95% CI)	ICER	£20,00 0 per QALY	£30,000 per QALY	£50,000 per QALY
Base-case analysis	6.85 (-107.82, 121.54)	0.0041 (-0.0013, 0.0094)	£1,670	0.73	0.73	0.73
Sensitivity analysis 1: Alternative-perspectives
Healthcare perspective	-5.92 (-116.3216, 104.46)	0.0041 (-0.0013, 0.0094)	Dominant	0.75	0.75	0.74
Societal perspective	65.01 (-195.56, 325.59)	0.0041 (-0.0013, 0.0094)	£15,856	0.62	0.61	0.63
Sensitivity analysis 2: Complete-case analysis††
NHS cost (£) Mean (SD)	11.61 (-118.84, 142.08)	0.0053 (-0.0005, 0.0112)	£2,190	0.75	0.74	0.73
Healthcare cost (£) Mean (SD)	-10.21 (-139.95, 119.53)	0.0053 (-0.0005, 0.0112)	Dominant	0.78	0.77	0.75
Societal cost (£) Mean (SD)	19.05 (-295.94, 334.04)	0.0053 (-0.0005, 0.0112)	£3,594	0.74	0.72	0.71

Impact of stratified care on costs and cost-effectiveness results by study group

Point estimates of incremental NHS costs, QALYs and ICERs by risk group are reported in Table 13. Results for NHS perspective costs showed that SC was slightly more expensive for participants in the low risk and medium risk groups but slightly cheaper for those in the high-risk group, due to the higher number of spinal injections and scans in the UC arm. Relatively similar results were observed in the healthcare perspective estimates incorporating private healthcare resource use. From a societal perspective, SC was more costly than UC due to higher productivity costs in the medium and high-risk groups.

There was some uncertainty due to the reduced sample size in each group, but overall SC was associated with slightly fewer QALYs on average, and was slightly more expensive than UC for the low and medium risk group, and was therefore dominated by UC from an NHS perspective. For those in the high-risk group however, SC dominated UC over 6 month follow-up, that is, beer health benefits (0.0129 additional QALYs) and slightly reduced NHS costs (-£65.96) (NICE, 2013). 15a At a willingness-to-pay threshold of £20,000 per QALY, the probability that SC is cost-effective compared with UC from an NHS perspective was 0.23, 0.29 and 0.87 in the low, medium and high-risk groups, respectively (See Table 14; Cost-effectiveness outcomes for STarT MSK risk subgroups). From healthcare and societal perspectives the probability that SC is cost-effective compared with UC was 0.16, 0.40, and 0.89 and 0.02, 0.47, 0.69 respectively (See Table 14). Further details of the incremental cost-effectiveness analysis and uncertainty analysis of the subgroup analysis are reported in Table 14.

TABLE 14 - Subgroup analysis: Costs and health outcomes mean (SD) scores by STarT MSK risk subgroups (Imputed analysis)

SC = Stratified Care; UC = Usual Care

Incremental QALY estimates following multiple regression-based adjustment for age, gender and baseline EQ-5D

	Low risk		Medium risk		High risk
	SC	UC	SC	UC	SC	UC
Total NHS cost	250.62 (800.71)	173.11 (586.11)	359.85 (764.10)	324.274 911.25	429.87 (1077.50)	495.83 (1193.57)
Mean difference (95% CI)	77.50 (-103.34, 258.35)		35.57 (-109.90, 181.06)		-65.96 (-296.69, 164.77)
Total Health care cost	314.87 (846.92)	204.42 (600.38)	406.23 (793.68)	403.962 1041.25	493.95 (1092.04)	595.94 (1276.64)
Mean difference (95% CI)	110.45 (-78.38, 299.29)		2.27 (-158.55, 163.09)		-101.99 (-343.85, 139.87)
Total Societal cost	586.49 (1730.83)	250.44 (642.97)	682.59 (1333.82)	759.47 (2128.87)	1293.22 (3089.00)	1175.37 (2510.19)
Mean difference (95% CI)	336.05 (9.74, 662.36)		-76.87 (-387.82, 234.07)		105.97 (-498.20, 710.14)
Baseline EQ-5D	0.7446 (0.1076)	0.7603 (0.1146)	0.5895 (0.1726)	0.5986 (0.1628)	0.3979 (0.2473)	0.3885 (0.2432)
Month 6 EQ-5D	0.8113 (0.1442)	0.8177 (0.1260)	0.6983 (0.1778)	0.6986 (0.1609)	0.5595 (0.2487)	0.5088 (0.2619)
QALYs	0.3890 (0.0504)	0.3945 (0.0489)	0.3220 (0.0767)	0.3243 (0.0635)	0.2394 (0.1111)	0.2243 (0.1104)
Mean difference (95% CI)	-0.0068 (-0.0224,0.0088)		-0.0023 (-0.0143, 0.0097)		0.0129 (-0.0145,0.0404)
Adjusted QALYs *	0.3917	0.3924	0.3233	0.3232	0.2375	0.2258
Mean difference (95% CI)	-0.0015 (-0.0115, 0.0085)		-0.0001 (-0.0069,0.0066)		0.0117 (0.0009,0.0225)