Optimal primary care management of clinical osteoarthritis and joint pain in older people: a mixed-methods programme of systematic reviews, observational and qualitative studies, and randomised controlled trials

Design and study population

Data were used from the NorStOP study, a series of population-based prospective cohorts of adults aged ≥ 50 years who were registered with one of eight general practices in North Staffordshire, UK. 60 These cohorts were originally set up through Medical Research Council (MRC) programme funding and NIHR primary care centre funding, and continue to be a resource for researchers from inside and outside Keele University. A two-stage mailing strategy was used, comprising a health survey sent to all persons in the sampling frame and a subsequent regional pain questionnaire sent to those who had responded to the health survey, given consent to be contacted again and indicated they had experienced pain in the hip, knee, hand or foot in the previous year. Participants received similar follow-up questionnaires at 3 years. Ethics approval was obtained from the North Staffordshire Local Research Ethics Committee (NS-LREC 1351 and 1430). This analysis includes all responders who reported hand, hip, knee or foot joint pain at baseline that had lasted ≥ 3 months over the previous 12 months and who returned the 3-year follow-up survey.

Outcome measures

Outcome measures for the prognostic models were (1) moderate or more severe pain in at least one joint site and (2) limitations in physical function in accordance with the Outcome Measures in Rheumatoid Arthritis Clinical Trials (OMERACT) consensus statement on key outcome measures for OA. 61

The pain subscales of the Western Ontario and McMaster Universities Osteoarthritis Index (WOMAC) (range 0–20),62 Australian/Canadian Osteoarthritis Hand Index (AUSCAN) (range 0–20)63 and Foot Pain Disability Index (FPDI) scores (range –3.32 to 3.33)64,65 were used to measure pain related to the hip/knee, hand and foot, respectively. All three questionnaires have been validated in relevant populations. To define a dichotomous measure of at least moderate pain, cut-off points for increased scores on the WOMAC, AUSCAN or FPDI scores were selected using receiver operating characteristic (ROC) curve analysis with a 0- to 10-point numerical rating scale (NRS) for pain with a validated cut-off point of 5 as the anchor (0–4 mild pain, 5–10 moderate to severe pain). 66 Based on maximal area under the ROC curve, the cut-off points for at least moderate pain obtained for WOMAC pain were 6 and 5 for the hip and knee, respectively, 9 for AUSCAN hand pain, and –0.479 for FPDI foot pain. Participants with scores below any of the relevant cut-off points at the 3-year follow-up were classified as having ‘no or mild pain’, while those with scores equal to or above the cut-off point were classified as having ‘moderate to severe pain’.

Limitations in physical function were assessed with the 10-item physical functioning subscale of the Short Form questionnaire-36 items (SF-36) (range 0–100, higher scores indicating better physical functioning),67 which was completed by all participants regardless of the location of pain. The scale was converted into a binary variable based on its median score of 55, as its distribution was heavily skewed. Respondents with a score of less than the median value were classified as having important functional limitation.

Potential predictors

The baseline health survey included questions on physical functioning, joint pain, participation restriction, general health, sociodemographic factors, lifestyle characteristics, comorbidity, medication use and psychosocial factors (including illness perceptions, anxiety and depression). Full details of the variables and their descriptions have been reported elsewhere. 59 These variables were all considered as potential predictors in the prognostic analysis as they have been shown to be associated with symptoms of OA or are considered to be potential predictors of outcome in pain conditions more generally. All variables with prevalence of < 10% or > 90% or with fewer than 30 participants giving a positive response were excluded from the analysis to facilitate successful convergence of the models68 and optimal discrimination between people with favourable or unfavourable outcome.

Statistical analysis

Response

In the original NorStOP cohorts, a total of 26,705 people aged ≥ 50 years were identified from eight general practices and posted the health survey. A total of 18,474 participants responded (estimated response 71.8% after adjustment for age and sex). Among this group of responders were 10,057 persons who reported joint pain in the previous year and gave their consent for further contact. This group was sent a second questionnaire at baseline (the regional pain questionnaire) and a follow-up questionnaire 3 years later.

The cohort for the current analysis, which is described here and was supported by the NIHR programme grant, was the subgroup of 3563 responders at the 3-year follow-up (57.6% of all 3-year regional pain questionnaire responders) who responded at 3 years and had reported pain in at least one joint site at baseline for a duration of ≥ 3 months over the 12 months prior to the baseline survey.

Characteristics of participants

Table 1 presents demographic characteristics of the sample, stratified according to the presence of severe pain and functional limitation at the 3-year follow-up. The mean age of participants was 64 years (range 50–93 years) and 60% were female. Non-responders at baseline and at the 3-year follow-up were slightly younger and more often male. At the 3-year follow-up, more than two-thirds (71%) of the participants reported severe pain, while 43% had poor function. The distribution of pain, physical function, obesity, anxiety and depression scores were similar for individuals responding at follow-up and those who dropped out from the study at 3 years (data not shown).

Predictors of poor outcome at three years

Multivariable Poisson regression identified a large number of baseline predictors that were statistically significantly associated with increased risk of moderate to severe pain at 3 years, including variables related to pain severity, poor physical function, socioeconomic factors, obesity, anxiety, hypertension and little use of pain medication or other remedies. Baseline predictors associated with increased risk of functional limitation at 3 years included several measures of poor function, being unemployed or retired from work, lower levels of activity, fewer years in education, joint pain severity, obesity, hypertension and poor health perceptions. Predictive performance was good for both models with c-statistics adjusted for optimism of 0.83 [95% confidence interval (CI) 0.81 to 0.85] for predicting moderate to severe pain and 0.91 (95% CI 0.90 to 0.92) for predicting limitation in function at 3 years.

Selection of important predictors

Both models included a large number of predictors, which limited their clinical utility. In this study, the most relevant predictors for severe pain and poor functional limitation at 3 years were selected using the rule employed by Smit et al. 71 to select subgroups of participants at increased risk of developing anxiety in later life. The rule selects predictors based on large effect size (IRR), high PAR and low NNT, which have the ability to select factors for which the highest possible health benefit (IRR and PAR) and the lowest possible effort and cost (NNT) can be achieved if interventions are completely successful. In addition, the predictive performance of the set of predictors had to be comparable to that of the original Poisson regression models.

Based on high effect size (IRR), high PAR and low NNT, the following prognostic factors were selected as the most relevant predictors of moderate to severe pain at 3 years: presence of knee pain in the last year (IRR 1.31, 95% CI 1.19 to 1.43), high WOMAC knee pain score (IRR 1.15, 95% CI 1.06 to 1.24), poor physical function (IRR 1.14, 95% CI 1.07 to 1.22), hand pain in the last year (IRR 1.12, 95% CI 1.05 to 1.20), not attended further full-time education after secondary school education has been completed (IRR 1.12, 95% CI 1.01 to 1.23) and obesity (95% CI 1.11, 1.05 to 1.17). The predictive performance (apparent c-statistic) of this reduced set of predictors was 0.75 (95% CI 0.73 to 0.77), indicating some reduction in performance for the reduced pain model. The adjusted PAR estimates for these factors were 17%, 9%, 11%, 7%, 9% and 2%, respectively, while their (unadjusted) NNT estimates were n = 5, n = 6, n = 4, n = 7, n = 9 and n = 6, respectively (Figure 1).

FIGURE 1.

The PAR and NNT for the most important predictors of moderate to severe pain at 3 years.

Using the same procedure for selecting predictors, poor physical function (subscale SF-36, IRR 2.48, 95% CI 2.02 to 3.05), poor physical health [SF-12 physical component score (PCS), IRR 1.44, 95% CI 1.24 to 1.67], being retired from work (IRR 1.39, 95% CI 1.18 to 1.64), reporting reduction in time of or change in activities in the past year (IRR 1.31, 95% CI 1.01 to 1.70), with IRR estimates shown separately for ‘much’ and ‘little’ reduction in Figure 2 and not going out for long walks (> 2 miles on any day) (IRR 1.28, 95% CI 1.10 to 1.48) were identified as the most important predictors for functional limitation (see Figure 2). Their adjusted PAR estimates were 48%, 21%, 19%, 8%, 13% and 15%, respectively, with unadjusted NNT estimates of n = 2, n = 2, n = 3, n = 3, n = 2 and n = 3, respectively (see Figure 2). The predictive performance (apparent statistic) of this reduced model was 0.71 (95% CI 0.69 to 0.73), indicating a considerable reduction in predictive performance when compared with the full Poisson model.

FIGURE 2.

The PAR and NNT for the most important predictors of limitation in function at 3 years.

Outputs

1. These findings were used as the basis for the modelling of primary care strategies for population health gain in OA (see Study 3: estimating cost-effectiveness of delivering core primary care management for osteoarthritis).

2. One potential implication of the findings for the primary care of OA is that improving the most severely affected patients will be needed in order to shift population levels of pain and disability in the long term. However, the analysis presented here was based on a general population sample, which may over-represent prevalent cases when compared with a sample of primary care consulters, and included some characteristics that may not be modifiable, at least in the short term. Future work will need to take better account of the presence of generalised pain in people with joint pain or OA and investigate the specific predictive value of modifiable prognostic factors (including physical activity and obesity) to identify risk groups and individuals for targeting treatment, such as the interventions investigated in the Managing OSteoArthritis In ConsultationS (MOSAICS) studies and Benefits of Effective Exercise for knee Pain (BEEP) trials (see Chapters 3 and 4). The second output of this programme is, therefore, a development of the analysis based on a more selective category (‘joint pain that interferes with daily life’) as the primary outcome, designed to reflect a more severely affected population. Predictors will focus on modifiable prognostic factors for poor outcome.

Search strategy

Systematic reviews and meta-analyses were searched in MEDLINE and The Cochrane Library from January 1990 to August 2010. Subsequently, an additional search in MEDLINE, covering the period from the year 2000 to August 2010, was conducted to identify individual trials that were not yet included in reviews. The search terms were developed in consultation with an information specialist and were based on the following (exploded) medical subject heading terms: OA, family practice, general practice, primary health care, community health services, ambulatory care, non-steroidal anti-inflammatory agents, exercise therapy. Reference lists from all retrieved systematic reviews and randomised controlled trials (RCTs) were checked to identify additional potentially eligible RCTs.

Selection of trials

Included in the evidence synthesis were full reports of RCTs, published in English, that evaluated the effectiveness of advice or information regarding self-management approaches and of paracetamol, topical NSAIDs and exercise in patients diagnosed (radiographically or symptomatically) with OA at one or more joint sites (hand, hip, knee or foot). Given the focus of this study on primary care, RCTs were only selected if they had been conducted in a primary care or direct access setting. One reviewer (JW) scored eligibility of all identified publications, with an additional reviewer (DvdW, MB or SJ) judging eligibility of all potentially relevant studies.

Data extraction

Data were extracted from RCTs included in this review to enable description of setting, study design, trial population, interventions and outcome measures. Estimates of absolute mean or changes in mean scores after treatment and their respective standard deviations (SD) were extracted and used to calculate standardised mean differences (SMDs) for outcomes of pain and functional limitation. 80 When relevant reports were available, effect estimates were calculated based on intention to treat analysis. Effect estimates of 0.2, 0.5 and 0.8 were described as small, moderate and large, respectively. 81

Assessment of risk of bias

Risk of bias of RCTs was assessed using The Cochrane Risk of Bias tool. 82

The risk of bias domains considered most important in this analysis were adequacy of randomisation, concealment allocation, blinding of outcome assessors and loss to follow-up. These were scored by JW as high risk, low risk or unclear risk of bias. Sensitivity analyses were performed, omitting trials with high risk of bias in at least one of the domains considered.

Meta-analysis

Meta-analysis focused on comparisons that were of interest to the design of the decision model (see Study 3: estimating cost-effectiveness of delivering core primary care management for osteoarthritis): (1) advice and information versus no treatment, (2) simple analgesics versus advice/placebo/no treatment, (3) topical NSAIDs versus advice/placebo/no treatment and (4) exercise versus advice/simple analgesics/no treatment. Pooled effect estimates for each comparison were calculated using a fixed- or random-effects model depending on the extent of heterogeneity present. Fixed-effects models were used when the I² estimate was ≤ 50%, otherwise random-effects models were used. Depending on the data presented in the original trial reports, SMDs for pain and functional limitation were derived using either mean change or the final outcome score at the end of treatment. This does not pose a problem in meta-analysis as both scores are considered to be addressing the same underlying intervention effect in RCTs. 83 Differences in mean final scores will be the same as differences in mean change scores if randomisation has been successful and baseline values of outcome measures are similar.

Funnel plots and Egger’s test84 were used to assess the risk of small-study bias for comparisons including an adequate number of studies.

Search results

A total of 41 RCTs (27 from Cochrane reviews, 11 from other reviews and three additional trials from MEDLINE) met the selection criteria and were included in the evidence synthesis. Four RCTs investigated the clinical effectiveness of advice and information, two investigated simple analgesics, four investigated topical NSAIDs and 31 examined exercise interventions. A full list of references and summary of the main characteristics of the included RCTs are available from the authors of this report.

The knee was the most commonly affected joint among the RCTs considered in this review, with 24 RCTs investigating the knee only and 17 enrolling participants with either knee or hip OA. No trials investigating hand or foot OA met the inclusion criteria for this evidence synthesis and none assessed treatment effects independently of the joint affected.

Risk of bias

In general, the RCTs included in this review appeared to have good methodological quality and none of the medication trials (analgesics or NSAIDs) showed a high risk of bias on any of the domains. All 41 trials were judged to be at low risk of bias in terms of random allocation of interventions. Among four trials investigating advice and information interventions, one was at a high risk of bias in terms of loss to follow-up. For exercise interventions, the proportion of RCTs at a high risk of bias was 7 out of 31 (23%) for blinding of outcome assessment and 5 out of 31 (16%) for loss to follow-up.

Clinical effectiveness of primary care interventions for osteoarthritis

The 41 RCTs included in this review provided data on 6715 subjects assessed for pain and 5322 subjects assessed for functional limitation. Table 2 summarises the results of the meta-analysis for each of the four comparisons.

Only one of the four RCTs investigating advice and information showed statistically significant improvement when compared with control. There was no evidence of heterogeneity across the studies for either pain (I² = 0.0%) or functional limitation (I² = 1.7%) outcomes. The pooled analysis showed a small, statistically significant, reduction in pain (SMD = –0.17, 95% CI –0.31 to –0.03, n = 771) as well as a small, statistically significant, improvement in function (SMD = –0.20, 95% CI –0.34 to –0.06, n = 771) in favour of advice and information (mean duration of treatment was 19 weeks) when compared with control.

The two RCTs investigating the effectiveness of analgesia (paracetamol, mean duration of treatment was 9 weeks) failed to demonstrate a significant difference compared with control interventions. The pooled effect estimates were small and not statistically significant: SMD = –0.11(95% CI –0.31 to 0.08, n = 400) for pain and –0.01 (95% CI –0.53 to 0.51, n = 57) for function. Given the small number of studies, these estimates have to be interpreted with caution.

Four RCTs compared the efficacy of topical diclofenac with placebo, with three RCTs showing beneficial effects. There was no evidence of heterogeneity of effects among the studies for either pain (I² = 0.0%; p = 0.985) or functional limitation (I² = 0.0%; p = 0.900). Pooled estimates of SMD showed moderate reduction in pain (–0.35, 95% CI –0.49 to –0.21, n = 790) and moderate improvement in function (–0.31, 95% CI –0.45 to –0.17, n = 789) of topical diclofenac compared with placebo after a mean duration of treatment of 5.5 weeks.

A wide range of exercise interventions was assessed among the 31 exercise trials, comprising both individual and group programmes, and with variable content focusing on strength, flexibility, balance and/or aerobic exercises. These references are included in Figure 3. Three RCTs showed a statistically significant reduction in pain (see Figure 3) and a statistically significant improvement in functional limitation. There was minimal statistical heterogeneity among the studies for both outcome measures (see Table 2). The pooled analyses showed a moderate reduction in pain (SMD –0.32, 95% CI –0.43 to –0.21, n = 1389) and moderate improvement in function (SMD –0.27, 95% CI –0.39 to –0.16, n = 1240) for exercise (mean duration of 19 weeks) compared with control interventions.

FIGURE 3.

Effect estimates (SMD) for pain outcomes of exercise interventions for OA. Reproduced with permission from Jerome Wulff. 59

The number of trials of exercise interventions was sufficient to reliably assess the risk of small-study bias. The funnel plots for both outcome measures appear to be symmetrically shaped (Figure 4) and Egger’s bias estimates confirm the lack of evidence for small-study bias for pain (bias estimate –0.49, 95% CI –1.41 to 0.44; p = 0.292) and function (bias estimate –1.14, 95% CI –2.47 to 0.18; p = 0.087).

FIGURE 4.

Funnel plot for trials investigating exercise interventions for OA (pain outcomes). Reproduced with permission from Jerome Wulff. 59

Sensitivity analyses were performed for advice and information, and exercise intervention studies, with results showing minimal differences in the pooled effect estimates after exclusion of studies with a high risk of bias.

Outputs

1. As stated above, estimates were provided for the modelling in the section Study 3: estimating cost-effectiveness of delivering core primary care management for osteoarthritis

2. The specific aim and content of the review presented here was extended into a full review and network meta-analysis of randomised trials of exercise interventions for OA as a linked output of workstreams and workstream 3. 122

Study population

The patient population was adults aged ≥ 50 years with symptomatic knee and/or hip pain or OA in a primary care setting in the UK, using data from the NorStOP cohort. 60 The original idea was to use a similar target population to that used for the prediction modelling study described above in study 1 (see Study 1: predicting long-term outcome in people with joint pain and osteoarthritis), in which participants had pain at baseline at one or more joint site (hand, hip, knee or foot). However, the proportion of participants with only hand or foot OA was low (< 5% each) and the evidence synthesis (described in study 2, see Study 2: summarising evidence regarding the effectiveness of core primary care interventions) provided effect estimates only for hip and/or knee OA. Therefore, it was decided to define this sample more narrowly than in study 1 of the workstream and the pain scores provided by the WOMAC questionnaire from study 1 were used to represent hip or knee pain status. If an individual had pain scores for both joints, the highest score was used to reflect the joint with greater pain.

Model structure

A Markov cohort model was built to reflect the clinical history of OA. The model considered four health states (no pain, mild, moderate and severe pain), defined using baseline WOMAC scores (range from 0 to 20). A score of zero was defined as no pain, 1–5 as mild pain, 6–10 as moderate pain and 11–20 as severe pain. 62 Everyone started in the model with some pain (mild, moderate or severe) and moved health state when their condition improved or worsened; alternatively, they could remain in the same health state. Therefore, the model includes no pain as a state through which people could pass during follow-up, even though persons in that state were not included as one of the start points. Movement could only be to the next better or worse health state. A time horizon of 3 years was used in the base-case analysis, reflecting the follow-up period of the NorStOP cohorts used in this model. Death was not included in the model as the time horizon was short and the condition does not directly lead to death. A 3-month time cycle was used, as it was considered to be a clinically meaningful time period for OA in terms of expected changes in the symptoms of OA, duration of treatment and timing of decision-making by a GP. The structure is presented in Figure 5.

FIGURE 5.

Markov model structure. Reproduced with permission from Jerome Wulff. 59

Interventions

Three packages of primary care were considered by the model: stepped care, ‘one-stop shop’ and current care. These hypothetical scenarios were discussed with clinicians to ensure assumptions were realistic.

Data inputs

Table 3 contains information on the baseline proportion of patients in each health state at baseline and provides estimates for the clinical effectiveness of individual treatments. The initial distributions in each health state at the start point of the model were estimated from NorStOP baseline data. Participants with no knee or hip pain at baseline were excluded because the model assumed only participants with pain would consult primary care and receive treatment. The numbers excluded from the sample used in study 1 above (i.e. those with only foot or hand pain) were small. Estimates of the SMD for each of the primary care interventions compared with their controls were obtained from the review in study 2.

Table 4 presents the transition probabilities for current care and all interventions. Current-care data on pain severity at baseline and 3 years from the NorStOP cohort was used to provide transition probabilities for the model. Matrix multiplication was utilised to transform actual transitions over 3 years into 3-monthly transitions. The treatment effect estimates were then applied to the current-care transition probabilities to obtain new transition probabilities. In the stepped-care intervention, only the effectiveness of the specific intervention at that step was applied, even though the patient was likely to be continuing to receive previous interventions. The estimate of effect (SMD) for the ‘one-stop shop’ was set at 0.5. This estimate was agreed during the consensus meeting of clinicians and OA researchers as it is larger than the strongest effect estimate of the individual primary care interventions. The clinicians in particular were clear that they often used the interventions as a package and agreed that a large additive effect between these concurrent interventions was unlikely.

Costs

Table 5 shows the unit cost data applied in this study. The cost of current care for each pain severity health state was estimated using patient-level data from the current-care arm of the exercise trial by Hurley et al. 124 The unit cost of drugs was updated from a price year of 2003/4 to 2010 using British National Formulary (BNF) costs126 and an average cost per 3 months of treatment calculated.

In stepped care, there was an initial cost for consultation with a GP, a nurse appointment for each new line of treatment and the cost of the intervention. The exercise intervention was six sessions of exercise at 30 minutes per session over a period of 6 weeks with an experienced physiotherapist. 128 It was assumed that when participants moved to the next line of treatment, they all continued to use the preceding interventions. The ‘one-stop-shop’ intervention cost included the cost of an initial GP consultation, ongoing costs for paracetamol and topical NSAIDs, a consultation with an experienced physiotherapist to introduce the exercise intervention at an initial consultation and the cost of exercise. Costs were accumulated over the 3-year time horizon to give total costs for each of the three modelled treatment strategies.

Outcomes

Baseline Short Form questionnaire-6 Dimensions (SF-6D) component scores of the NorStOP participants with hip and/or knee pain were used to derive utility values for each health state, using the algorithms developed by Brazier and Roberts. 128 Table 6 shows the mean utility scores for the four health states used in this study. Utilities were accumulated over the 3-year time horizon to give total quality-adjusted life-years (QALYs) for each strategy in the model.

Base case

An incremental cost–utility analysis was undertaken to compare the cost-effectiveness of the two proposed primary care strategies with current care, from a health-care perspective. The interventions were ordered in descending order according to cost. Costs and QALYs were discounted at an annual rate of 3.5% in accordance with the current UK treasury guidelines. 129 Costs were expressed in Great British pounds using 2010/11 as the price year. A threshold of cost-effectiveness of £20,000 per QALY gained was adopted for this study. 15

Deterministic sensitivity analyses

Sensitivity analyses were carried out to test the robustness of the primary results by changing some of the most important assumptions used in the model construction.

The following sensitivity analyses were performed:

• Extension of the time horizon of the model from 3 years to 5, 10 and 20 years.

• Application of GP costs instead of nurse costs for subsequent consultations in stepped care.

• Selection of only those subgroups with moderate or severe pain categories (excluding mild pain patients) to reflect a health-care seeking population.

• Varying the effect size (SMD) of exercise using 95% CIs.

• Varying the effect size (SMD) for ‘one-stop shop’ between –0.4 and –0.6.

Outputs

1. The template developed, described and tested here is a stand-alone resource for future modelling exercises.

2. The practical application and content of the template described here was constrained by available data at this stage of the programme (population data and secondary care trials). A second application of the model will use the output of the primary care trial data, such as that which has now emerged from the later stages of the programme in workstreams 2–4 for new analyses in the future. This second application will also further develop the template through:

   1. probabilistic sensitivity analysis, taking into account all parameter uncertainty simultaneously, which was not undertaken here in this first application

   2. inclusion of data on the course of OA pain over time, including short-term fluctuations, to provide more appropriate transition probabilities

   3. adaptation to look at a broader range of OA pain sites.

The study populations

The registered populations of the eight participating general practices who were aged ≥ 45 years (n = 33,726) provided the population base for the MOSAICS studies. The eligibility for the participation of these practices, their staff and their patients is listed in Appendix 2 (see MOSAICS eligibility criteria).

There were three populations selected from this base in the pre-randomisation phase.

• Population survey responders. All persons aged ≥ 45 years registered with one of eight participating practices were sent a general health questionnaire to establish baseline pre-randomisation patterns of joint pain and OA in the population. A series of baseline analyses of all ‘survey responders’ were performed to determine the background frequency of multisite pain, self-reported uptake of NICE recommended treatments and prevalence of physical activity among persons with joint pain in this population (MOSAICS study 1.1, see Study 1.1: self-reported prevalence and patterns of osteoarthritis and its treatment in the registered population of primary care).

• Total OA consulting population. All consulters aged ≥ 45 years who were coded on the practice medical records system as having consulted about joint pain or OA, regardless of whether or not they had responded to the baseline population survey, formed the ‘total OA consulter population’ (Figure 7 and Box 2). This anonymised population provided occurrence estimates of joint pain and OA in the participating practices, using the total registered population aged ≥ 45 years of the participating practices as the denominator (n = 33,726) (see Figure 7).

• Consenter population. Those who responded to the survey, who consented to follow-up and a recorded review, and who had reported pain in at least one OA joint site (hand, hip, knee or foot) at baseline were eligible for individual follow-up in the cluster trial. This population was a subgroup of the second population above, drawn from the baseline survey responders in the first population. A consent form to further contact and to use of individual medical records was included in the survey. Survey responders with pain in at least one joint who gave their consent formed the ‘consenter population’. This identified, pre-randomisation, the pool of patients in all practices who would be willing to be recruited and followed up individually in the main cluster trial if they consulted their GP with joint pain during the randomisation phase. Such pre-randomisation of a consenting population of potential future eligible trial participants guards against bias in recruiting individuals within a cluster design post randomisation. 109 This subgroup, the self-reported questionnaires they completed during the follow-up phase and details from their medical records created the basis for the ‘individual-level’ analysis in the cluster RCT of clinical effectiveness and cost-effectiveness of the implementation of the new model of care (MOSAICS studies 4.2–4.3, see Study 4.2: effect of the MOSAICS programme on patient-reported outcomes and Study 4.3: cost-effectiveness of the MOSAICS intervention: health economic evaluation). The subgroup also provided the sampling framework for additional ‘individual-level’ quantitative and qualitative studies of implementation (MOSAICS studies 5.1–5.4, see Study 5.1: implementing a complex intervention in practice – an evaluation using a theoretical framework, Study 5.2: experiences of model consultations – patients, Study 5.3: experiences of model consultations – general practitioners, and Study 5.4: experiences of the model consultation – nurses).

A practice-level intervention (the quality-of-care e-template, which prompted OA assessment and NICE core OA interventions by the GP) was developed and introduced into all eight practices. This template was a pop-up computer screen that appeared whenever joint pain and OA consulting codes were entered. The effect of this intervention on routinely recorded items of care (such as prescriptions for OA) was investigated by a before-and-after comparison of downloaded anonymised practice record data 12 months before and 6 months after installation of the template in all eight practices [MOSAICS study 3.1, see Study 3.1: the osteoarthritis quality-of-care template (the ‘e-template’)]. In the subsequent cluster RCT of the MOSAICS intervention, this population (all patients consulting with OA symptoms or diagnosis in the participating practices) and the anonymised measures of quality of care provided by the template generated the level and method, respectively, of outcome analysis for the ‘practice-level’ implementation study. This compared anonymised downloaded consultation data between intervention and control populations (MOSAICS study 4.1, see Study 4.1: implementation of the MOSAICS intervention – practice-level effect on quality of care for patients with osteoarthritis).

Development studies

The development and testing of components of the model OA consultations and of the e-template were conducted in the pre-randomisation phase [MOSAICS studies 1.2, 2.1–2.5, 3.1 and 3.2, see Study 1.2: establishing the current evidence base for multidisciplinary osteoarthritis interventions in primary care settings; Study 2.1: development of the content of a ‘model’ general practitioner consultation and the training intervention to deliver it; Study 2.2: evaluation of intervention workshops; Study 2.3: development of the content of a model osteoarthritis consultation with a practice nurse and the training package for nurses to deliver it; Study 2.4: evaluation of practice nurse training to support osteoarthritis self-management; Study 2.5: development of the opportunistic osteoarthritis consultation with health-care professionals; Study 3.1: the osteoarthritis quality-of-care template (the ‘e-template’); and Study 3.2: developing patient-reported outcomes with the osteoarthritis research user group).

The main trial process

The eight practices were randomly allocated to intervention practices and control practices to form the cluster trial design. All practices continued to use the e-template to record care. Following training of GPs and practice nurses, consulters with joint pain or OA in the intervention practices received the MOSAICS model of care, whereas consulters with joint pain or OA in the control practices received usual care.

Patient and public involvement in the MOSAICS studies

The OA RUG at Keele University had provided the initial impetus and thread for the whole programme (see Appendix 1) by highlighting that a desire expressed by patients to try self-management approaches was not matched by sufficient active and informed professional NHS support to do so.

The RUG continued to play a substantial part in the creation, shaping, delivering and evaluation of many important components of the MOSAICS suite of studies. An overview of RUG activity in MOSAICS is as follows.

Summary

The MOSAICS is a complex suite of studies. This created challenges for the RUG members and for researchers. We have managed this by:

• generating of a glossary which is available to all RUG members

• providing support for RUG members at meetings via a PPI co-ordinator and user support worker

• producing lay summaries in advance of meetings

• continuing to provide feedback to patients who have been involved

• offering training (e.g. Contributing Assertively in Meetings, a regular workshop run for staff by Keele University’s Learning and Professional Development Centre).

Study 1.1: self-reported prevalence and patterns of osteoarthritis and its treatment in the registered population of primary care

This study incorporated three analyses of the baseline survey of the registered populations of the eight practices participating in the cluster trial. The survey was both a vehicle to identify a pool of consenters for potential future recruitment into the main trial if they consulted about joint pain or OA, and a basis for background studies of OA and its treatment in the general population.

The latter was the purpose of study 1.1 and is based on the fact that > 95% of people in the UK are registered with a general practice and so a survey of a registered practice population is a suitable vehicle for assessing a local general population sample.

Study 1.2: establishing the current evidence base for multidisciplinary osteoarthritis interventions in primary care settings

Parts of this section have been reproduced from Finney et al. 156 ©The Author(s). 2016. This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.

Study 2.1: development of the content of a ‘model’ general practitioner consultation and the training intervention to deliver it

Parts of this section have been reproduced from Porcheret et al. 134 © 2014 Porcheret et al. ; licensee BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly credited. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated; and from Porcheret et al. 143 © 2013 Porcheret et al. ; licensee BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Study 2.2: evaluation of intervention workshops

Parts of this text have been reproduced with permission from Porcheret et al. 159 This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.

Study 2.3: development of the content of a model osteoarthritis consultation with a practice nurse and the training package for nurses to deliver it

Parts of this section have been reproduced from Healey et al. 160 © The Author(s). 2016 This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.

Study 2.4: evaluation of practice nurse training to support osteoarthritis self-management

Parts of this section have been reproduced from Healey et al. 160 © The Author(s). 2016. This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.

Study 2.5: development of the opportunistic osteoarthritis consultation with health-care professionals

Parts of this section have been reproduced with permission from Finney A, Porcheret M, Grime J, Jordan KP, Handy J, Healey E, et al. 144 Defining the content of an opportunistic osteoarthritis consultation with primary health care professionals: a Delphi consensus study. Arthritis Care Res 2013;65:962–968. © 2013 by the American College of Rheumatology.

Study 3.1: the osteoarthritis quality-of-care template (the ‘e-template’)

Parts of this section have been reproduced from Edwards et al. 132 © The Author 2014. Published by Oxford University Press on behalf of the British Society for Rheumatology. This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (http://creativecommons.org/licenses/by-nc/4.0/), which permits non-commercial re-use, distribution, and reproduction in any medium, provided the original work is properly cited. For commercial re-use, please contact journals.permissions@oup.com; and from Edwards et al. 145 This is an Open Access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 3.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited and the use is non-commercial. See: http://creativecommons.org/licenses/by-nc/3.0/.

Study 3.2: developing patient-reported outcomes with the osteoarthritis research user group

Parts of this section have been reproduced from Blackburn et al. 166 © 2016 Blackburn et al. Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.

Study 4.1: implementation of the MOSAICS intervention – practice-level effect on quality of care for patients with osteoarthritis

Parts of this section have been reproduced from Jordan et al. 167 © 2017 The Authors. Published by Elsevier Ltd on behalf of Osteoarthritis Research Society International. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).

Study 4.2: effect of the MOSAICS programme on patient-reported outcomes

Parts of this section have been reproduced from Dziedzic et al. 168 © 2018 The Authors. Published by Elsevier Ltd on behalf of Osteoarthritis Research Society International. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).

Study 4.3: cost-effectiveness of the MOSAICS intervention: health economic evaluation

Parts of this section are summaries of material in Oppong et al. 172 © The Author(s) 2018. Published by Oxford University Press on behalf of the British Society for Rheumatology. This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (http://creativecommons.org/licenses/by-nc/4.0/), which permits non-commercial re-use, distribution, and reproduction in any medium, provided the original work is properly cited. For commercial re-use, please contact journals.permissions@oup.com.

Study 5.1: implementing a complex intervention in practice – an evaluation using a theoretical framework

Study 5.2: experiences of model consultations – patients

Study 5.3: experiences of model consultations – general practitioners

Study 5.4: experiences of the model consultation – nurses

Study 6.1: revising the training programme – engaging control practices post trial

Study 6.2: extending the MOSAICS intervention to new practices via the new NHS

Limitations of implementation

Implementation of a complex intervention designed to provide an integrated primary care service for patients with OA that was focused on model GP and practice nurse consultations did prove a challenge. Notably, only a minority of patients saw the nurse and so not all eligible patients were offered exercise, which is known to be effective in improving pain and increasing function in knee OA. 122 In addition, those that did see the nurse appeared not to have adopted the optimal change in exercise. The qualitative studies provided further insight into this to show that the patients understood the importance of exercise, but that barriers persisted for aerobic walking exercise because of comorbidities and lack of a suitable environment to encourage exercise.

It is possible that, despite the success of the training programme for GPs in changing some of their behaviours (as shown by changes in the content of their consultations with simulated patients), there has been less of an impact in practice and the impact has been insufficient to influence patients. However, the patient guidebooks, which met criteria for strong patient engagement in guidebook development and incorporated local information, were successfully delivered as part of the model OA consultation and, in the qualitative studies, patients reported relevant GP behaviours in the intervention arm and evaluated the guidebook as helpful. The e-template study provided evidence that prescribing did change and that relevant advice was recorded as being given. However, there were limitations. Only one-third of patients attended the nurse follow-up clinic in the intervention arm. There may be a number of reasons for this, including explanations for low referrals such as:

• non-referred patients did not have OA (this would happen if the GP fired the template, then considered later on in the consultation that the patient’s problem was not due to OA and decided not to refer the patient to the nurse)

• some GPs were untrained because of staff turnover rates

• behavioural change in GP competencies did not translate into changed performance.

The low referral rate to nurse-led clinics may be one explanation for the lack of clinical improvement. It meant that the full integrated MOSAICS model of care, incorporating improved consultations with the GP and referral to practice nurse-led OA clinics were not delivered as intended. Evidence has been provided that seeing the nurse was linked to full implementation of the MOSAICS method of delivering NICE core guidance, including the provision of written information and advice, and opportunities to discuss exercise and physiotherapy referral. This appeared sufficient to explain differences between control and intervention practices in meeting quality-of-care criteria for provision of written information.

However, the fact that only a minority of patients saw the practice nurse cannot finally explain the lack of improvement in pain and disability outcomes in the trial, as even in the subgroup who saw the nurse there was no difference between the intervention and control groups in these outcomes, although it may have affected power to show a difference. There was clear evidence in the practice-level analysis that the combination of GP consultation and nurse referral was achieving the highest level of implementation of NICE core guidance and quality of care. The qualitative studies confirmed the nurses’ enthusiasm for the new model of care and provided evidence of patient engagement being driven by the nurse-led clinics, but they also raised some questions about the clinics, based on patients’ judgement of the nurse intervention and from direct observation and audio-recording of the nurse-led clinics. In addition, the GPs’ use of the template did reduce with time and a similar drift may have occurred in the content of their consultations. Maintenance of behaviour change in clinicians and patients is a challenge for the primary care of OA patients.

Limitations of the intervention

It is possible that crucial elements of an effective intervention were lacking. It is notable that OA treatment remains focused on single-site disease and yet the evidence from this workstream emphasises that it is the person with multiple joint pain who is most commonly seeking help. Elements of primary care may still not be properly targeted or shaped for the majority of patients with multiple joint pain. The emphasis in the NICE guidelines on the importance of advice about exercise and weight loss may be an insufficient intervention in primary care; the guidelines recommend a whole range of further adjunctive treatments should the core intervention be ineffective. 15 The evidence about effectiveness of physiotherapy-led exercise interventions in reducing pain and disability in the short term among OA patients (reviewed and emphasised in Chapter 4, Introduction, describing workstream 3 of the programme) may mean current GP or nurse advice alone is insufficient to improve activity to levels which will relieve pain and disability. The finding about the shift in balance towards muscle strengthening and away from aerobic exercise in the intervention practices supports this possibility. Increasing the availability and use of physiotherapist-led exercise and activity as part of a revised MOSAICS intervention is one potential way forward.

However, this raises issues of resource. One question for future research is whether or not there are subgroups of patients in primary care for whom the NICE core interventions are both sufficient and effective. One important issue is the context of multimorbidity, as the baseline population survey confirmed that this is the most frequent state among older people in the community reporting joint pain. There is a continuing need to investigate the role of multimorbidity in determining differential outcomes in primary care patients with OA and how best to manage this.

Limitations of expectation

It is important to consider if process measures of successful delivery of care might be valued as important in their own right, given that there is evidence from elsewhere that the treatments and interventions are worth doing and that feedback from patients and clinicians indicate that the interventions produce a clearer and more positive model of care for patients with OA. Therefore, a second question for future research arising from this workstream is how far process measures of successful implementation of treatments of known efficacy should be used to drive policy in the absence of strong evidence that they are shifting clinical outcomes, and in the presence of evidence that some patient goals (better information, satisfaction with care) are being met by the model of care. One important issue when considering, for example, the application of evidence, to everyday primary care, that physiotherapy-led exercise and activity is efficacious and effective in reducing pain and disability, is that the evidence may have been gained outside the context in which it would be applied. One of the arguments for complex trials is that the evidence of efficacy is being examined in its practical context.

Limitations of resource

It must also be accepted that support for self-management provided in the GP consultation and the nurse-led clinic may be limited in its potential for impact on long-term clinical outcomes by the limited resources available to the patient outside the primary care setting. The qualitative study of patients highlighted challenges in the outside world, such as limited access to places to exercise, for patients continuing on a self-management path. Such external barriers will undermine the immediately beneficial effects of advice and information also observed in MOSAICS (e.g. shifts in knowledge, initial engagement in exercise). Similar challenges were highlighted in a German study of why GPs felt that implementation of guidelines for back pain would fail – lack of an external environment to meet the requirements of the guideline. 187 Future research could investigate integration between primary care, public health and community resources to support sustainability of OA self-management.

The last two points lead to the question of how implementation of the MOSAICS model might now proceed. A local CCG has engaged in the first stage of planned dissemination and views the evidence of effective processes of GP training, quality templates, nurse-led clinics and patient-generated information (the ‘process outcomes’ of the MOSAICS studies) as sufficient basis for the new approach. The finding of cost neutrality provides further reassurance for such implementation. Delivering high-quality care with a reduction in orthopaedic visits and use of imaging are desirable outcomes for health-care commissioners. The NPT model, which has framed the work on implementation throughout this workstream, highlights that there remain barriers to implementation, including clinician variation and lack of resource to deal with the dominant problems of low physical activity and obesity.

The lack of evidence demonstrating clinical effectiveness of the current model of optimising care is an important issue to tackle, as clinicians and patients will need evidence that new models of care do ultimately benefit patients if resources to implement them fully are available. However, the evidence that implementation of the model has achieved a higher quality of care, as judged by current standards, is relevant to policy decisions at the same time as future research tackles some of the questions about how those current standards might need to change in order to improve clinical outcomes.

It follows from this that the alternative explanation is that the MOSAICS model of care was not a sufficient intervention to shift pain and disability, and that its focus on core interventions limited the capacity of the individual-level analysis to assess the effect of the full spectrum of adjunctive treatments recommended by NICE if core interventions fail to benefit. Successful implementation of NICE core guidance itself, as demonstrated by both practice- and individual-level analysis of the MOSAICS model of care, may not lessen pain and disability. Optimal and accessible information for patients may be desirable for many reasons but may not, in itself, result in lessened pain and disability and, using the GP and practice nurse consultation as the means to deliver exercise and weight loss advice (although, again, desirable as a first step), may not deliver the full ‘dose’ of support and intervention needed to shift patient behaviour sufficiently to achieve the reductions in pain and disability that efficacy trials of, for example, physiotherapy-led exercise and physical activity have demonstrated in patients with OA.

One further question that arises is the conceptual one of the extent to which NICE guidelines will be effective when put into practice. One clear novelty of our study is that it has attempted to test the clinical effectiveness and cost-effectiveness of guideline implementation in practice. We have provided process evidence that health-care professionals and health-care systems can change sufficiently to implement components of guidelines, but patient benefits remain elusive to demonstrate.

The implementation study

• This focused on improved resources/technologies for primary care (including a patient-developed guidebook, computerised consultation template, nurse-led OA clinic, training for GPs and nurses) designed to promote behavioural change in GPs and nurses, and (via guidebook and advice) self-management behaviours in patients.

• The MOSAICS intervention demonstrated that a ‘whole-system’ implementation of NICE guidance can be achieved in primary care and can change clinician behaviour and support patient self-management. The study produced evidence of improved capacity for NICE implementation at the:

  • GP level (the training delivered behavioural change, guidebooks were handed out to patients, items on the template were carried out, prescribing changed and nurse referrals occurred)

  • nurse level (clinics were carried out and template items were reinforced)

  • patient level (patients received guidebooks, attended nurse-led clinics and improved their physical activity).

• On the basis of these findings, a wider implementation exercise is taking place in collaboration with commissioners, which incorporates changes and refinements to the MOSAICS model to tackle problems of generalisability (e.g. by using briefer, more focused health professional training).

The clinical effectiveness study

This did not show evidence of improved levels of pain and disability. Potential explanations include:

• The evidence base of clinical effectiveness in NICE guidance may not be strong enough to achieve change in clinical end points in practice, or some of the aims of the guidance (e.g. provision of satisfactory information for patients) are worthwhile and necessary in their own right regardless of effect on clinical measures of outcome.

• The implementation was not complete enough to achieve changes in clinical outcomes. There was evidence (quantitative and qualitative) for this in terms of:

  • content of consultations (e.g. exercise advice may have encouraged muscle strengthening at the expense of aerobic exercise such as walking)

  • numbers of patients engaged in full implementation – only a proportion of patients were engaged with the full template or referred on to the nurse and the trial may have lacked power to show an effect at the group level

  • heterogeneity of this patient group

  • resources – the primary care package may be insufficient to achieve patient behavioural change (exercise and weight reduction) without more structured or specialised input to support it (e.g. physiotherapists).

• It may be inappropriate to seek more than minor clinical effect sizes in pain and disability in practice. Maintenance of activity and participation, despite continuing pain, may, for example, be a better long-term measure of effect.

The future

• Evidence from MOSAICS of how to implement the NICE OA guidance in practice is now driving wider dissemination projects (e.g. Chapter 3, Study 6.2: extending the MOSAICS intervention to new practices via the new NHS). These provide an opportunity to adapt MOSAICS to generalisable formats and to improve on those areas listed above which limited its implementation.

• Given the findings that patients with an OA diagnosis had better indicator outcomes and that the GP task of giving a diagnosis of OA did not improve after training suggests future implementation and research should address this component of the consultation.

• There is also a need in future research to reflect critically on whether or not:

  • pragmatic trials of clinical effectiveness of NICE guideline implementation are appropriate compared with a combination of implementation studies and new efficacy studies to strengthen the evidence base for the guidelines

  • OA outcomes in practice need to be concerned with aspects over and above improving short-term levels of pain and disability.

Intervention 1: usual physiotherapy care

Usual physiotherapy care consisting of advice and exercise was the most appropriate control group for the BEEP trial given that randomised trials34 and systematic reviews33 consistently show that interventions that include exercise are superior to those that do not. We did not use an attention control group as this trial was designed explicitly as a pragmatic trial, building on evidence about the effectiveness of exercise interventions. Clinical practice guidelines recommend exercise for all patients with OA. 15 In current UK clinical practice, patients are seen in relatively few treatment sessions and are provided with brief and relatively standardised exercise programmes that do not take into account differences between individual patients. Our ABC knee study described usual UK physiotherapy practice and formed the basis for the BEEP trial protocol for UC. 199

The BEEP trial protocol for UC consisted of advice and lower limb exercise previously tested and shown to be more effective in the short-term than GP-led care and advice alone. 34 Exercises were selected from an agreed template of commonly prescribed exercises {printed from the commonly used PhysioTools computer software [PhysioTools©, Product ID RG-PT1ENG, General Exercises Second Edition (English), Tampere, Finland]}, including specific lower limb muscle strengthening (non-weight-bearing and weight-bearing) and range of movement or stretching exercises. Patients were to receive up to four one-to-one treatment sessions with a physiotherapist over a period of 12 weeks, during which advice to continue to exercise was provided but individualisation, progression and supervision of the exercise programme was minimal, limited to up to four sessions as in UC. Other interventions used frequently by physiotherapists such as ice therapy, manual therapy and electrotherapy were permitted as per UC and they were recorded on CRFs, but the emphasis of the intervention was on supporting the patient to self-care and to follow the advice and exercise programme at home. This UC protocol matched usual UK practice in that it:

• focused on lower limb-strengthening exercises

• relied on self-report rather than the use of exercise diaries to monitor adherence and progress

• was delivered over few treatment sessions and, thus, had limited opportunity for good individualisation, supervision or exercise progression

• provided advice about general physical activity only

• did not offer refresher or booster sessions following the end of the episode of care. 199

Intervention 2: individually tailored exercise

Standardised exercise programmes such as those used in the UC group in the BEEP trial do not take into account differences between individual patients. Consequently, individuals will be working at relatively different intensities, which may be too little for some to get a training effect and may be too difficult for others. To be optimally beneficial, exercise should be progressed so that appropriate physical stress is placed on the individual for further improvements to be obtained. 213 Exercise self-efficacy (confidence to exercise despite the knee pain) has been identified as a predictor of exercise behaviour in many populations. 214,215 Supervision of exercise can enhance patients’ exercise self-efficacy and self-regulatory skills and reassure them that they can perform the exercises. It also provides good opportunity to reassure the patient about pain responses to specific exercises and to change the exercise prescription to ensure optimal performance or more tolerable pain response. Lack of adequate individual tailoring, supervision and progression of the lower limb exercise programme may, in part, explain the small benefits seen in some previous exercise trials. Our previous trial results suggested that a lower limb exercise programme that was supervised and progressed over six treatment sessions resulted in greater improvements in pain than one that was provided over an average of four sessions. 34,203 Therefore, the protocol for ITE was developed to ensure the prescription of an individualised, supervised and progressed lower limb exercise programme.

The BEEP trial protocol for ITE consisted of a supervised, individually tailored and progressed exercise programme. The aim of the intervention was to initiate and progress an ITE programme, which was supervised in clinic, practised at home and progressed in terms of intensity over 12 weeks. The intervention was modelled on a previously successful exercise intervention from one of our previous trials203 and focused on individualised lower limb muscle strengthening (non-weight-bearing and weight-bearing), range of movement or stretching exercise and balance exercise. The patient and physiotherapist were expected to agree a lower limb exercise programme and targets were to be reviewed and progressed. Individualisation was based on the findings of the physiotherapy assessment of each individual, including biomechanical and physiological observations, pain responses to specific exercises, starting levels of strength, range of movement and balance. Exercises were prescribed for each individual and participants were given their own individual print out of their specific exercise prescription (selected and printed from PhysioTools computer software), and these exercise prescriptions (and print-out instructions) were expected to change over time as the exercise programme was progressed. Physiotherapists were to encourage exercise behaviour change using self-monitoring through use of the BEEP trial lower limb exercise diary to record their adherence with their lower limb exercise prescription (see trial protocol188). In order to provide greater opportunity for individualisation, supervision and progression of exercise, patients were to receive between six and eight one-to-one treatment sessions with their physiotherapist. Physiotherapists were to provide advice about general physical activities and there were no scheduled follow-ups (refresher or booster sessions) with the physiotherapist beyond 12 weeks.

Intervention 3: targeted exercise adherence

Adherence to long-term treatment regimes, particularly those involving behavioural components such as exercise, is consistently lower than adherence to medication. 15 Exercise adherence, irrespective of exercise type, may be a key factor determining the success of exercise. In a previous trial of exercise for this patient population, we observed high exercise adherence rates in the short term through to the end of treatment at 12 weeks, which fell to just over 50% at 12 months. 203 Long-term adherence to lower limb exercise is perhaps unrealistic, as these can be challenging to incorporate into daily routines and individual lifestyles, and many patients do not consider them enjoyable, stopping them once symptoms reduce or resolve. In order to support engagement in and adherence to exercise, it may be more realistic to target general physical activities that individuals have previously engaged in and enjoyed (as previous exercise behaviour is, theoretically, the strongest source of self-efficacy information),215 or physical activities that individuals have positive expectations about (as these positive expectations may increase exercise intentions and ultimately exercise behaviour). Our previous ABC knee study216 highlighted the many different barriers to and facilitators of exercise and physical activity, and that no single exercise type or exercise setting is acceptable to all. Thus, we designed this TEA intervention to include an adherence-enhancing ‘toolkit’ of optional tools and techniques for physiotherapists to use with different participants, based on their assessment of individual participants and early feedback from participants (see trial protocol188 for a summary of the contents of the toolkit). The content of the intervention was directly informed by the results of our Cochrane systematic review208 and a networking meeting supported by Arthritis Research UK, during which national experts and patient representatives agreed the intervention. Although the general physical activity that was identified and encouraged by physiotherapists was individualised for participants, we anticipated that many may choose walking as it is seen as inexpensive and accessible. Pedometers have been shown to increase step counts in older people217 and to generally increase physical activity218 and, therefore, we included pedometers within the suite of options for physiotherapists to give participants who wished to target increases in walking activity.

The BEEP trial protocol for TEA began with a focus on the lower limb (as in the ITE group) but transitioned to focus increasingly on general physical activity adherence over time. In addition to prescribing an individualised, progressed and supervised lower limb exercise programme, physiotherapists were trained to assess patients’ current general physical activity levels, their intentions to increase their physical activity levels and their attitudes to exercise for knee pain and general health, as well as exploring their individual barriers to and potential facilitators of exercise. This group was to receive four treatments up to week 12 and a further four to six contacts from week 12 through to 6 months (a total of 8–10 treatment contacts). In the first four treatments, the aim was for participants to initiate and progress an exercise programme with supervision that would include lower limb and general exercise, and to identify (with the support of the physiotherapist) general physical activity opportunities within the local community that were suitable for, and of interest to, the individual. Proactive follow-up from the physiotherapist from week 12 through to 6 months was expected, using choices of telephone and face-to-face contact, providing an additional four to six contacts with each participant. The aim was to enhance long-term exercise adherence, promote increased general physical activity and encourage participants to develop an exercise and physical activity ‘habit’, shifting the focus away from lower limb exercise in the earlier treatment sessions and towards sustainable lifestyle changes in physical activity in later treatment sessions. The follow-up sessions were designed to ascertain and promote exercise adherence and increases in physical activity levels, support patients to integrate exercises into their activities of daily living, allow repetition and amendment of the exercise programme based on individuals’ experiences or concerns, and support progression at a pace suitable for the individual. The adherence enhancing ‘toolkit’ contained different educational, behavioural and cognitive–behavioural tools and techniques for facilitating physical activity behaviour change, selected for use based on an individualised assessment of each patient. As our Cochrane review identified that no one model of behaviour change was superior for facilitating adherence to exercise for chronic musculoskeletal pain, different theoretical models underpinned the development of the toolkit, including self-efficacy214 and self-regulation theory. 219 Educational tools included written education material on a range of relevant issues (including healthy eating and pain medication) and individualised PhysioTools exercise sheets. Behavioural strategies included ‘How exercise/physical activity feels’ and ‘How to monitor heart-rate’ guides to facilitate physical activity intensities at moderate levels, self-monitoring through use of physical activity diaries,188 heart-rate diaries, a visual feedback chart, and graded activity sheet, reminder post-cards for physiotherapists to post to patients, and pedometers to support increases in walking. Cognitive–behavioural strategies included guides to questions to elicit patients’ health-related beliefs; assessment of patient barriers and motivators; several ‘rulers’ to use with patients to assess their readiness; perceived importance and confidence to engage in physical activity; decisional balance sheets; specific, measurable, achievable, realistic, time-related (SMART) goal-setting and behavioural contracting for physical activity to facilitate the translation of intention into physical activity behaviour;212 and example templates to discuss and generate an individual exercise set-back plan. Finally, physiotherapist-generated information about local physical activity opportunities and facilities in the community were developed for each treatment site, with a focus on easy to access and inexpensive options. The target by the end of the 6-month period was that participants would be engaged in physical activity opportunities within their locality and have had support from the physiotherapist to overcome initial problems or barriers in engaging in these activities. Therefore, the emphasis was on maintenance of physical activity beyond the period of support from a health professional and NHS-based programme.

Content of the training programme

The content of the BEEP trial training programme was informed by our previous national surveys of physiotherapists and of older adults in the community with and without knee pain, and our Cochrane review (described in Development). It primarily focused on the rationale and specific content of each intervention, but also included practical information for physiotherapists about treating patients within RCTs [e.g. the importance of blinding, working to intervention protocols, prevention of contamination and summaries of procedures for serious adverse event (SAE) reporting].

The training programme was ‘stepped’, in that all physiotherapists attended the first day and received an update about OA, based on the NICE OA guidelines. 15 Key components of the first day included the primary care (clinical) diagnosis of knee OA, the central role of exercise as a ‘core’ treatment for older adults with knee pain, current physiotherapy practice for knee pain in older adults and OA, and the comparison between current practice and guidelines. 15,32,199,220

The second and third days were attended by physiotherapists delivering the ITE and TEA interventions and focused on how to improve outcomes from exercise for older adults with knee pain. The importance of individualisation, progression and supervision of lower limb exercise was highlighted from both physiological and psychological perspectives. Exercise self-efficacy was discussed as an important predictor of exercise behaviour214,215 and emphasis was placed on the importance of ‘selling’ exercise to patients. Tools to facilitate physiotherapists to individualise, supervise and progress exercise were provided and practised, including developing written individualised exercise programmes using PhysioTools computer software and use of lower limb exercise diaries.

The fourth and fifth (final) days were only attended by physiotherapists delivering the TEA intervention and focused on the importance of exercise adherence, the physiotherapist’s role in facilitating exercise and physical activity behaviour change, and shifting from a lower limb exercise programme to physical activity that might be more likely to be sustained in the long term. A number of behavioural models were drawn on within the TEA training programme, including self-efficacy214 and self-regulation theory. 219 Each physiotherapist was provided with their own ‘adherence enhancing toolkit’ containing multiple strategies to be used on an individualised basis with patients to facilitate and sustain physical activity behaviour change (described in Interventions development). Use of each strategy was practised through role play and participating physiotherapists were encouraged to practise using the tools with patients in their routine clinical practice, prior to the commencement of BEEP participant recruitment.

The training programme included lectures, interactive workshops, role play, group discussion, problem-solving and case studies, with homework set to consolidate learning. In addition, approximately 10 months after the training programme, all physical therapists were invited to attend a half-day workshop to share best practice and discuss any challenges faced with other physical therapists delivering the same intervention. This was designed to help support physiotherapists to deliver the treatments in line with the intervention protocols. A total of 26 physiotherapists attended this additional workshop.

Evaluation of the training programme

We investigated whether or not taking part in the BEEP trial training programme increased physiotherapists’ confidence in managing older adults with knee pain and altered their intended clinical behaviour and their attitudes and beliefs about the role of exercise for chronic knee pain, and explored whether or not the more intensive training for physiotherapists delivering the ITE and TEA interventions led to greater changes in these variables than in those physiotherapists delivering UC.

All physiotherapists were asked to complete a questionnaire before (pre training), immediately afterwards (post training) and after delivering the BEEP trial exercise interventions (post intervention, approximately 12–18 months after their training). The content of the questionnaire was based on our previous national survey of physiotherapists. It included a vignette describing a ‘typical’ patient ≥ 45 years with knee pain, associated clinical management questions, attitude statements about the potential benefits of exercise against which physiotherapists rated their level of agreement, and a measure of PSC in diagnosing and managing older adults with knee pain (adapted from a scale developed for back pain161). In order to fully describe participating physiotherapists, the pre-training questionnaire also included questions on their clinical characteristics (e.g. number of years in practice and previous training undertaken on OA and exercise therapy).

Analysis

Descriptive statistics were used to describe the sample as a whole, compare results over time (pre training, post training and post intervention) and between the groups of physiotherapists trained to deliver each BEEP trial intervention. Level of agreement with each attitude statement about the role of exercise for knee pain was determined by the percentage of respondents who ‘largely’ or ‘totally agreed’. We defined consensus as follows: 100% = unanimity, 75–99% = consensus, 51–74% = majority view, and 0–50% = no consensus, in line with our previous research. 216,220 Median scores were provided for the measure of physiotherapists’ self-confidence in diagnosing and managing knee pain in older adults, as data were positively skewed. Changes over time and differences between groups of physiotherapists (i.e. those trained to deliver each BEEP trial intervention) were not tested for statistical significance owing to the relatively small number of participating physiotherapists overall.

Results

In total, 52 out of the 53 physiotherapists returned the pre-training questionnaire, 44 (83%) returned the post-training questionnaire and 39 (74%) returned the post-intervention questionnaire. The majority of physiotherapists who took part in the training programme were female (63%), worked exclusively within the NHS (67%), had treated at least one patient ≥ 45 years with knee pain per week (62%), and had not received previous post-graduate training in the field of chronic knee pain (73%) or exercise therapy (62%). The mean number of years of clinical experience was 14 (SD 11 years), but ranged from 1 to 42 years (Table 22).

Physiotherapists who did not return their questionnaires post training and post intervention were similar in terms of the characteristics that were assessed to those who did complete and return their follow-up questionnaires. Several had moved jobs or had maternity leave during the lifetime of the BEEP trial.

Discussion

We aimed to determine whether or not taking part in the BEEP trial training programme increased physiotherapists’ confidence in managing older adults with knee pain and whether or not it altered their intended clinical behaviours, attitudes and beliefs about the role of exercise for chronic knee pain in line with the specific training package that they received.

Prior to the BEEP training programme, participating physiotherapists were already quite confident in their ability to diagnose and manage older adults with knee pain. However, as seen in our previous national survey of UK-based physiotherapists,201 there were some disparities between their intended clinical behaviour and current clinical guidelines and exercise recommendations for knee OA. 15,32 For example, although nearly all physiotherapists reported that they would use therapeutic exercise and provide advice as part of their treatment package, exercise was provided over relatively few treatment sessions, local (lower limb) muscle-strengthening training was favoured over general physical activity, there was limited supervision of exercise and there was a lack of focus on exercise adherence. There was also uncertainty about the potential benefits of exercise for knee pain.

Directly after the training programme, there were some key changes in the intended clinical behaviours of all groups of physiotherapists, including those who participated in only 1 day of training to support the delivery of UC in the trial. Even within physiotherapists who participated in the UC training programme, use of additional interventions (such as electrotherapy and manual therapy) alongside the exercise programme reduced, there was a greater focus on general physical activity and functional task training, supervision of exercise increased, and level of agreement with the attitude statements about the role of exercise for knee pain also increased. Therefore, these results demonstrate that taking part in the BEEP trial training programme was associated with changes in physiotherapists’ attitudes, beliefs and intended clinical behaviours with regard to exercise for a typical patient case (an older adult with knee pain).

The results showed that changes in attitudes, beliefs and intended behaviours were more pronounced in physiotherapists trained to deliver the ITE and TEA interventions. Within these groups, in addition to the changes described above, the number of treatment sessions that physiotherapists reported that they would provide also increased, more physiotherapists said they would monitor exercise adherence using an exercise diary, and there was a higher level of agreement with more of the attitude statements about the role of exercise for knee pain. In addition, their confidence in managing older adults with knee pain also increased. This may be linked to the more specific, targeted content in the training programme about how to individualise, supervise and progress exercise, and how to support longer-term exercise adherence. It may also relate to the intensity and type of content within the training programme for each group of physiotherapists. For example, the ITE and TEA training programmes included more time (in days) and more interactive sessions than the single-day training programme underpinning UC. Previous research has identified that interactive sessions are more effective than didactic educational techniques at facilitating behaviour change within health-care professionals, including physiotherapists. 221–224 However, even in physiotherapists who had undertaken the training to deliver the ITE and TEA interventions, not all changes in intended clinical behaviour and attitudes and beliefs were maintained over the longer term (12–18 months after completing the training programme), a pattern seen in professional groups within other areas of health care. 225 There may be several reasons for this, including barriers at the level of the patient, the professional and the wider health-care organisation. Multimodal approaches that not only address the exercise attitudes and beliefs and intended behaviours of physiotherapists, but also identify and address other potential barriers to long-term behaviour change in physiotherapists, are likely to be required to achieve greater or more sustained changes in these variables over time. 224,226,227

Although a large number of physiotherapists participated in the BEEP trial training programme (n = 53) and we achieved good follow-up rates on their self-completed postal questionnaires after training (83%) and 12–18 months later (74% response rate), the numbers were too small for robust statistical testing to compare changes in beliefs, attitudes or intended behaviours over time and between groups of physiotherapists delivering the three interventions. In addition, as with all survey research that utilises patient case vignettes, it is possible that intended practice behaviour in response to case vignettes may be different from actual clinical practice.

Conclusions

Physiotherapists who participated in the BEEP trial training programme to deliver the ITE and TEA interventions reported greater confidence in their ability to diagnose and manage older adults with knee pain. For all groups of physiotherapists, the reported attitudes, beliefs and clinical behaviours about exercise for chronic knee pain changed in line with the specific training package that they received. Some of these changes appeared to be sustained through to long-term follow-up at 12–18 months following the training programme and after the end of the period of treatment for BEEP trial participants, while others reverted back towards pre-training levels. The intensity and content of the training programme delivered seemed to be important, as changes in the survey variables among those physiotherapists delivering the ITE and TEA interventions, who attended longer and more interactive training programmes, were greater and more likely to be sustained. Whether or not these changes in physiotherapists’ self-reported behaviour translated into changes in their actual behaviour for BEEP trial participants is determined by analysis of the physiotherapy CRFs in the BEEP trial (see Interventions delivered and adverse events).

Objective(s) of the pilot study

The primary objectives of the pilot study were to assess the feasibility and acceptability of the ITE and TEA interventions.

The secondary objectives were to:

• test the training programme for participating physiotherapists and inform decisions about resource allocation of staff to the three arms of the trial

• test ways to assess whether or not the interventions are delivered per protocol

• test ways to assess if the exercise programme is individualised and adherence achieved, using physiotherapy CRFs, attendance at clinic sessions, physical activity questionnaires, exercise diaries and accelerometers

• determine the most appropriate tool for measuring self-reported physical activity in the trial population by comparing the PASE and IPAQ questionnaires

• test the recruitment processes, data collection methods and outcome measures to be used in the trial to ensure they operate as expected and they record sufficient information about the detail and fidelity of the interventions.

Methods

Findings

Key changes from the BEEP pilot study

The key learning from the pilot study was related to challenges in participant recruitment using only one recruitment method (from patients already referred to NHS physiotherapy services with knee pain). As a result, two additional recruitment methods were added to the protocol for the main BEEP trial. The recruitment methods for the main trial are fully described in Invitation, recruitment, consent and randomisation.

The treatment CRFs from the pilot study showed that the BEEP interventions were deemed feasible and acceptable to both participants and physiotherapists. UC and the ITE interventions were generally delivered in line with the intervention protocols in terms of the number of sessions delivered and the content of treatment. In contrast, the majority of patients allocated to receive the TEA intervention did not receive the number of treatment contacts initially protocolised (between 8 and 10 contacts, which could be a combination of face-to-face and telephone contacts). This issue was therefore particularly highlighted within the main BEEP physiotherapy training programme, for which the importance of treating patients in line with the trial intervention protocol was emphasised. Overall the training programme in the pilot study was well received and only minor changes were made to the programme for the main trial.

The qualitative interviews in the pilot study established that both patients and physiotherapists were generally positive about the BEEP trial and the interventions. However, it was recognised that there was a need for some minor amendments to the format of the interview paperwork and the procedures for telephone interviews, and more space for writing on the CRFs was also requested by the physiotherapists. These amendments were made for the main trial.

In terms of the outcome measures used within the pilot study, these were generally well completed. However, at 6 months the response rate to the follow-up questionnaire was lower than expected (57%). In response to this we ensured that, for the main trial, minimum data collection procedures were in place at the key timepoints (6, 18 and 36 months), as this has been shown to substantially improve response in our previous trials. 34,203 One of the objectives of the pilot study was to determine the most appropriate outcome measure to assess self-reported physical activity within the main trial. We had no previous experience of using the PASE and the IPAQ, so completion rates were particularly important to assess for these measures. Following the pilot study and consultation with the independent Trial Steering Committee, the PASE was chosen for use in the main trial for three reasons: (1) the scoring is simpler than the IPAQ, (2) it appears easier for respondents to complete (all questions are closed form rather than open) and (3) it correlated slightly better with the objective physical activity data from accelerometers (data not shown).

Identification of potentially eligible participants

The key learning point from the pilot study was the need to increase recruitment to identify all potentially eligible participants. Therefore, in the main BEEP trial, participants were identified in one of three ways: (1) from general practice computer record reviews to identify those who have consulted for knee pain in the last 12 months, (2) from a population survey of older adults registered with participating practices and (3) from patients referred from their general practice to physiotherapy services for knee pain. The above three recruitment methods proceeded in parallel until the sample size required was reached, although methods 1 and 2 were operationalised in different GP practices. The recruitment methods are described in more detail in numbers (1), (2) and (3) below. Duplication checks ensured that eligible participants were not invited to the BEEP trial more than once. Figure 8 shows a summary flow chart of BEEP trial recruitment and reasons for ineligibility.

FIGURE 8.

BEEP recruitment flow diagram. a, Based on the 2012 practice size data from the NHS Information Centre for Health and Social Care. 231

1. General practice record review

   Members of the PCRN informatics team, contracted to work in the participating general practices, screened computer records for adults aged ≥ 45 years who had consulted with knee pain in the last 12 months at the practice. The electronic screen identified patients based on the 14 most frequent knee pain-related Read codes informed by our previous research. 116 Read codes are the standard clinical terminology system that GPs enter onto their computer systems in the UK, identifying patients’ clinical symptoms and diagnosis. In addition, the electronic screening protocol excluded those with potentially serious pathology (e.g. inflammatory arthritis such as rheumatoid arthritis and malignancy) and those in nursing home accommodation. GPs were invited to screen the sample list and exclude those patients whom they considered inappropriate to be invited to participate in the trial. PCRN staff then administered the mailing of a short screening questionnaire to check eligibility and determine the patient’s Chronic Pain Grade classification. 201 Patients recruited through this method had a Chronic Pain Grade severity of between 2 and 4. The final section of the brief screening questionnaire asked patients if they would be happy to give their consent for further contact. Patients returned their screening questionnaires to the Arthritis Research UK Primary Care Centre at Keele University and only those who appeared to be eligible and consented to further contact formed the sample that was sent information about the BEEP trial.

2. Population survey

   Older adults (aged ≥ 45 years) registered with participating general practices were posted a short screening questionnaire to identify potentially eligible trial participants. GPs were invited to screen the mailing list and exclude those patients whom they considered inappropriate to be invited to participate in the trial. This screen was used to identify those in the community with knee pain of sufficient severity to be included in the BEEP trial229 and to exclude those who did not meet the eligibility criteria. Patients recruited through this method had a Chronic Pain Grade severity of between 2 and 4. The final section of the brief screening questionnaire asked older adults registered with the practice if they would be happy to give their consent for further contact. Those who were eligible and consented to further contact formed the sample that was sent information about the BEEP trial.

3. Physiotherapy service referrals

   Older adults (aged ≥ 45 years) with knee pain and/or stiffness referred by their GP or who self-referred to participating physiotherapy services were first screened by a member of the physiotherapy service team for key eligibility criteria. Those potentially eligible were contacted by a PCRN research nurse to find out if they were willing to receive further information about the BEEP trial.

Recruitment and consent of participants to the BEEP trial

From recruitment methods (1), (2) and (3) above, those who met the eligibility criteria and agreed to further contact were posted information about the BEEP trial [a cover letter, a participant information sheet (see trial protocol188), a baseline questionnaire, a consent form and a freepost return envelope]. No less than 48 hours after receiving the posted information, a PCRN nurse telephoned the person to further check and confirm eligibility. This check covered the full eligibility criteria in order to ensure that only those who met these criteria were recruited to the trial. In addition, the nurse screened out individuals with known unstable cardiovascular disorders and those who had such severely restricted mobility that they could not get to the physiotherapy clinics for treatment. All potentially eligible participants had the opportunity to discuss the trial with the research nurse prior to deciding whether or not to participate. Those who wished to take part in the trial were asked to sign and date the written consent form, supported by the research nurse over the telephone, and they returned it along with their completed baseline questionnaire to the nurse in a pre-paid envelope. Those who did not wish to take part were asked to indicate this on the consent form and to return it to the nurse in the pre-paid envelope provided. This consent process had been tested in the pilot study and regular audits of the nurse telephone calls formed part of the quality assurance procedures of the BEEP trial. It was possible, based on this consent method, that a very small number of participants in the BEEP trial would be found to be subsequently ineligible, as it was only after consent that participants had a detailed physical assessment by a BEEP trial physiotherapist. Examples of this were expected to include a very small number of participants who had radiating leg pain from a spinal problem or a hip joint problem with referred pain in the area of the knee.

Participating general practices were supported to assist with identification of potentially eligible participants for the BEEP trial through small practice payments to reimburse their time for screening patient lists. Physiotherapy services were supported to participate through financial reimbursement for the time taken out from service delivery for the training programme and additional time for BEEP treatments. Participants did not receive any financial incentives to return screening questionnaires, to take part in the trial or to return their follow-up questionnaires.

Randomisation and allocation concealment

Following receipt of a signed consent form and baseline questionnaire, the BEEP trial administrator randomised the participants using a computer-generated randomisation schedule provided by the musculoskeletal clinical trials unit (CTU) at Keele University, which was password-protected to ensure that research nurses and trial statisticians remained blind to treatment allocation. Participants were individually randomised to one of the three treatment groups in a 1 : 1 : 1 ratio using random permuted blocks of size 3. To ensure that patients at each physiotherapy clinic had a chance of receiving any of the interventions, randomisation was stratified by physiotherapy clinic. Following randomisation, the trial co-ordinator liaised with the appropriate physiotherapy clinic staff to arrange the first appointment for each trial participant. The patient was then informed in writing of the date, time and location of their first appointment in the physiotherapy clinic. The GP of each trial participant was sent a letter to confirm that their patient was taking part in the BEEP trial. Thus, our procedures ensured that baseline data were collected prior to randomisation, that the allocation was concealed until after the patient was recruited into the trial and until the moment of randomisation, and that the person assigning participants to intervention groups (study administrator) had no involvement in the eligibility screen, consent or treatment processes.

Primary outcomes

This trial had two primary outcomes, lower limb pain and function, measured using the WOMAC Index228 collected at baseline and all follow-up time points (3, 6, 9, 18 and 36 months). The primary time point was 6 months after randomisation. The primary comparisons were between each of the two new physiotherapist-led exercise interventions (ITE and TEA) versus UC. The psychometric properties of the WOMAC228 have been extensively studied in knee pain populations in clinical trials of different interventions including exercise89,232 and the WOMAC has been recently shown to be the most responsive of five pain measures. 233 The pain subscale ranges from 0 (no pain) to 20 (maximum pain) and the function subscale ranges from 0 (no disability) to 68 (maximum disability). It is particularly suitable for the BEEP trial as it specifically captures self-reported pain during activities and the degree of difficulty with everyday physical activities, both of which are key treatment targets of physiotherapy-led exercise.

Secondary outcomes

A range of secondary outcomes were collected: the proportion of treatment responders using the internationally agreed OMERACT clinical responder criteria200,201 that combine data on pain and function from the WOMAC228 with patient’s global assessment of change (recorded using a 6-point Likert scale); physical activity levels (PASE, which assesses physical activity levels over a 1-week period combining physical activity from several domains including household, occupational and leisure);169 self-reported BMI (calculated from self-reported height and weight); exercise adherence (attendance at treatment sessions, self-reported adherence to prescribed exercise programme); use of local physical activity facilities in the previous 7 days (single item); a modified version of a measure of treatment acceptability and credibility;234,235 a measure of illness perceptions (Brief Illness Perceptions Questionnaire236); confidence in ability to exercise [Self-Efficacy for Exercise (SEE) scale237]; outcome expectations from exercise [Outcome Expectations for Exercise (OEE) scale 2238]; anxiety [seven-item Generalised Anxiety Disorder Assessment (GAD-7)239]; depression [Patient Health Questionnaire (PHQ) depression scale240]; self-reported health-care resource use (both NHS and private health care); and overall health status (EQ-5D-3L241). Resource use and EQ-5D-3L data were used in the cost–utility analysis (further details are provided in Health economic analysis).

Accelerometry outcomes

Physical activity was also measured in a subsample of participants (n = 89 of the 90 initially planned) through snapshots of 7-day accelerometry at each follow-up time point. Accelerometers are motion sensors worn on the hip and were used to estimate physical activity as counts per minute, time spent in light, moderate and vigorous physical activity and proportions of participants who met guideline levels of physical activity. 242 Accelerometers were allocated at the point of randomisation. The accelerometer units were posted out to the participants with full instructions. Participants were asked to wear the unit during waking hours for 7 consecutive days and then to post the unit back to the research centre where the data collected were downloaded and analysed. During the trial, accelerometers were allocated at regular intervals (approximately monthly) to the next three participants randomised to each treatment group. A random allocation procedure was not used as not all participants were willing to wear them. A regular accelerometer allocation procedure was needed to enable accelerometers to be available to collect data (at baseline and follow-up) from a pool of 30 accelerometers available for the trial. The allocation procedure was phased in at the start of the trial to ensure that the system was running smoothly and that accelerometers were being returned in time to be reused by other participants. Those who did not return their accelerometers were posted a written reminder.

Follow-up

Outcome measures were collected by self-report, postal questionnaire before randomisation and at 3, 6, 9 and 18 months (Table 31). They are also being collected at 36 months follow-up for further analysis in the future. Therefore, adherence to exercise was measured twice throughout the early phase (0–6 months) and three times during the maintenance phase of the exercise interventions (≥ 9 months). Non-responders were followed up using our standardised CTU follow-up procedures for trials. This comprised a questionnaire, a reminder postcard at 2 weeks and a further copy of the questionnaire at 4 weeks. For participants who did not respond to any of these reminders, we attempted to collect minimum outcome measure data via telephone (by research nurses who were blind to treatment allocation) and by post at key outcome time points (6, 18 and 36 months) in order to try to capture primary outcome data and to minimise missing data. Quality assurance processes within the CTU ensured that training and auditing of the research nurses conducting minimum data telephone calls. In order to reduce participant burden, the 3 and 9 month questionnaires were slightly shorter than those at 6, 18 and 36 months. As direct measures of physical activity may themselves increase physical activity, similar measures of physical activity were used in each intervention group. Table 31 includes a list of all measures and the time points at which they were collected.

Adverse events

The occurrence of adverse events from all interventions was monitored and assessed using CRFs. An expected minor and transient adverse event from unaccustomed exercise and physical activity is temporary, mild muscle soreness. Physiotherapists delivering the interventions were to advise participants about how to manage such symptoms. Each physiotherapy site was asked to report any SAE experienced by a trial participant immediately to the trial chief investigator that may possibly be related to either the interventions or the trial procedures.

Other data

Other variables collected in participant questionnaires were age, sex, marital status, comorbidities, pain location using a pain manikin, duration of the knee problem, experience of exercise and work status. The following process data were also recorded from the physiotherapy CRFs: number of treatment sessions attended, the main content of each treatment, physiotherapists contact time with participants, the number of treatment withdrawals and treatment non-attendances.

Recruitment and follow-up

The numbers of participants identified and recruited using each recruitment method are reported in a flow chart, along with the numbers of participants returning a questionnaire at each follow-up stage. The baseline characteristics of participants are reported for the three treatment groups (to explore the effectiveness of randomisation) and for those with and without data at each follow-up time point (to explore any selective loss to follow-up).

Primary and secondary trial analysis for clinical outcomes

The primary and secondary trial analyses were conducted blind to intervention group by the statistician and an independent statistician verified the analysis of the primary outcomes. The primary analysis compared each of the ITE and TEA interventions with UC on an intention-to-treat basis for the primary outcome at 6 months post randomisation. Secondary analysis included the analysis of the primary outcome at the secondary end points (3, 9 and 18 months) and analysis of the secondary clinical effectiveness outcomes at all follow-up time points. Estimates of clinical effect were derived using analysis of covariance (ANCOVA) for continuous outcome measures and logistic regression for categorical outcomes and are presented as mean or percentage differences (as appropriate) with 95% CIs after adjustment for covariates defined a priori as:

• baseline for the outcome of interest (baseline adjustment is not relevant for the OARSI responder criteria as it incorporates baseline levels of knee pain and function into the follow-up outcome)

• age

• sex

• duration of the knee problem

• physiotherapy treatment centre (as was used in the randomisation algorithm).

A secondary analysis modelled the longitudinal trajectory of WOMAC pain and function over time (the primary outcome) and the PASE physical activity measure using generalised estimating equations. Model predictors included the a priori covariates listed above (with the baseline for the outcome of interest modelled as a covariate rather than outcome as recommended by Peduzzi et al. 249), time, treatment and a time × treatment interaction. A linear model was fitted to the data (with time as a continuous measure) initially; however, if the trajectory over time was non-linear, quadratic or cubic terms were explored. We also explored whether or not conclusions differed if time was represented in the model as a categorical variable. All model estimates were presented with 95% CIs derived using robust standard errors. The primary and secondary trial analyses were conducted after imputation of missing data. Missing data was imputed using the multiple imputation routines in Stata version 12. All primary and secondary clinical effectiveness outcomes (excluding accelerometry) were included in the imputation model and had their missing data imputed. The intention-to-treat analysis included participants who were treatment protocol violators, but did not include participants who were post-randomisation exclusions. Information on any adverse events is reported.

Sensitivity analysis for clinical outcomes

The following sensitivity analyses were completed and are reported.

• Therapist effects: these were explored by adding a random-effect term (to represent the treating therapist) to the main treatment models. The models with and without the therapist effect were compared at each outcome time point to explore the impact of the treating therapist on the effect estimates obtained.

• Imputation of missing data: a complete case analysis was conducted and the results from this compared with those from imputed data.

• Model covariates: treatment models were run on an unadjusted basis to investigate if the inclusion of model covariates had an impact on the main trial findings.

• A per-protocol analysis: this was completed on the subsample of participants judged to have received treatment in line with the specified treatment protocols. Criteria for determining the per-protocol group assignment were established by the Trial Management Group and approved by the Trial Steering Committee before analysis began.

Exploratory subgroup analysis

A specific subgroup analysis with strong theoretical rationale compared the clinical outcomes of pain and function of those who reported high exercise adherence with those who reported lower exercise adherence. We hypothesised that those who reported higher adherence would have improved pain and function outcomes.

Analysis of adherence data

Descriptive statistics were used to examine exercise adherence and were reported as numbers and percentages, or means and SDs, as appropriate. Exercise adherence was measured by reporting the number of treatment sessions attended in each treatment group and by reporting the following measures at each follow-up time point:

• self-reported exercise adherence (measured by agreement with the statement ‘I have been doing my exercises as often as I was advised’)

• frequency and duration of physiotherapy exercise completed by the participant

• change (from baseline) in level of physical activity (measured by PASE and, in a subsample of participants, accelerometry)

• self-reported use of local exercise and physical activity facilities.

Analysis of accelerometer data

Accelerometry data from the subsample of BEEP trial participants was analysed using ActiGraph (version 6.6.3; Actigraph© Lifestyle Monitoring System, Pensacola, FL, USA)250 accelerometer software. Prior to analysis, data were cleaned by excluding time periods containing > 60 minutes of zero count (for which it is assumed the accelerometer was not worn) and by including only those participants who had worn the accelerometer for at least 5 days for ≥ 10 hours. Two sensitivity analyses were conducted by changing the threshold to exclude minutes of zero count at 30 and 90 minutes, respectively, with the later threshold recommended for participants with knee pain. 251 For each participant, we generated the following at each data collection point: number of valid minutes, counts per minute and proportion of time spent at each level of physical activity using the cut-off points from Freedson et al. 252 We also calculated the proportion of participants meeting exercise/physical activity recommendations. 242 Descriptive statistics are given for the accelerometer variables at baseline and by treatment arm. Change (between baseline and each follow-up time point) in the average count per minute and in the proportion meeting the exercise recommendations were analysed using ANCOVA and logistic regression, respectively, with analyses adjusted for the covariates defined for the clinical effectiveness analysis and by treatment arm.

Health economic analysis

A cost–utility analysis was undertaken to compare ITE with UC and TEA with UC, using QALYs as the measure of benefit. The base-case analysis for the evaluation adopted a NHS and Personal Social Services (PSS) perspective. A complete case analysis and a broader perspective incorporating NHS/PSS costs, patients’ personal expenditure and costs associated with work over an 18-month follow-up period was considered as the sensitivity analysis.

Responses from the EQ-5D-3L generic instrument were used to measure preference-based health-related QoL at baseline, 3, 6, 9 and 18 months. For each patient, EQ-5D-3L index scores from responses to the questionnaire were obtained using the UK value set. 253 Using the area-under-the-curve approach that links each patient’s utility scores at different time points,254 EQ-5D-3L responses were used to generate QALYs for each patient.

Knee-related health-care resource use data were collected by patient self-report questionnaires administered at 6 and 18 months. NHS resource use data included primary care contacts (GPs, practice nurses, community physiotherapy), contacts with other health-care professionals (e.g. hospital consultants, physiotherapists and acupuncturists), hospital-based investigations and procedures [e.g. radiography, magnetic resonance imaging scans, knee-related injections and surgical procedure]), and prescribed medications. Additional information, reported in other trial communication by participants, about any knee surgery undertaken or planned within the 18-month follow-up period was also recorded. Non-NHS (health-care) costs were obtained by asking patients about their use of private health care and purchase of over-the-counter medicines and use of local exercise facilities, treatments or appliances. In order to assess broader economic consequences of the interventions beyond health-care resources, self-reported data on occupation and time off work taken because of their knee pain over the 18-month period was also collected.

Details of the number of trial-related physiotherapy sessions attended by each participant were collected through the CRFs. The costs required to deliver the two new interventions included additional physiotherapy sessions, telephone contacts and pedometers supplied to participants in the TEA intervention. The cost of the interventions included an average 45-minute initial assessment and treatment session, followed by 20-minute face-to-face physiotherapy sessions and 7.5-minute telephone contacts in the TEA intervention. Unit costs were obtained from various sources, including the BNF,174 NHS Reference Costs175 and Unit Costs of Health and Social Care173 and applied to resource use items. All unit costs used in this analysis are reported in Table 32, using a common 2012/13 price year.

All analysis was based on the intention-to-treat principle. In the base-case analysis, multiple imputations were used to impute all missing values for the EQ-5D-3L and non-surgery NHS costs for non-responders to the 6- and 18-month questionnaires. The base-case analysis involved imputing only the non-surgery NHS costs, in order to account for the additional knee surgery data retrieved for the 18-month follow-up period that were recorded separately from postal questionnaires and trial communication. A complete case analysis was also undertaken and is presented as a sensitivity analysis. Mean differences are all presented with bootstrapped 95% CIs.

An incremental cost-effectiveness analysis was conducted to determine the difference in costs and outcomes between ITE, TEA and UC. The unit of outcome was the incremental cost per QALY gained (incremental costs divided by QALYs) associated with the ITE and TEA interventions. The estimation of cost-effectiveness within the trial was based on the principles of dominance. If an intervention was found to be less effective and more costly than at least one of its comparators, then it was not included further in the economic analysis. To account for uncertainty, non-parametric bootstrapping was used to derive 5000 paired estimates of mean differential cost and QALY scores. These were then presented graphically on a cost-effectiveness plane. 255,256 The analysis also controlled for any possible between-group imbalance in baseline utility using a regression-based adjustment in order to avoid biased estimates of QALY scores. 257 Although costs were collected over an 18-month period, discounting was not applied to the 18-month questionnaire data, as this contained resource use data from both the last 6 months of year 1 and the first 6 months of year 2 and could not be disaggregated. When productivity loss was reported, costs were assigned using the human capital approach; self-reported days of work absence were multiplied by respondent-specific wage estimates identified from annual earnings data and UK Standard Occupational Classification coding. 258,259 Productivity costs were not imputed and a complete case analysis was undertaken. Statistical analysis was performed using Stata v12.