Progressive exercise compared with best-practice advice, with or without corticosteroid injection, for rotator cuff disorders: the GRASP factorial RCT

Problem and diagnosis

Shoulder pain is common. Annually, around 1% of adults aged > 45 years in primary care present with a new episode of shoulder pain, accounting for 2.4% of all general practitioner (GP) consultations in the UK. 1 This is most commonly attributed to the rotator cuff, which causes around 70% of cases. 2 The rotator cuff is a group of four muscles and their tendons/attachments. The rotator cuff actively moves and stabilises the shoulder joint, enabling a wide range of efficient movements at the shoulder. Disorders of the rotator cuff can be associated with substantial, persistent disability (e.g. being unable to dress independently) and pain. Rotator cuff disorders can persist for long periods. Up to half of those who present for treatment, particularly older people, continue to have pain and/or functional disturbance for up to 2 years. 3

The majority of shoulder pain is managed in primary care or in musculoskeletal interface services by physiotherapists and GPs. Musculoskeletal interface services are led by specialist practitioners who manage patients with musculoskeletal disorders through assessment, treatment, investigation and by referring to appropriate health-care professionals. Musculoskeletal interface services aim to promote more community-based management options for patients, rather than traditional hospital-based secondary care, and provide a more efficient, cost-effective and sustainable model for dealing with high-volume conditions. Treatments for rotator cuff disorders aim to improve pain and function. Standard primary care options include rest, advice, analgesia, non-steroidal anti-inflammatory drugs, physiotherapy and corticosteroid injections. 4,5 However, usual care can be highly variable and there are no National Institute for Health and Care Excellence (NICE) clinical guidelines.

A diagnostic algorithm2 has been developed as part of the NICE-accredited standards developed by the British Elbow & Shoulder Society (BESS) and other professional bodies (e.g. the Royal College of Surgeons, the Chartered Society of Physiotherapy and the British Orthopaedic Association) to confirm when a diagnosis of rotator cuff disorder is highly likely, based on a patient’s history and simple shoulder tests4 (see Appendix 1, Figure 13). The recommended tests have been selected with primary care application in mind,6 although they do require a reasonable degree of clinical skill. Imaging is not recommended in primary care because of the poor fit between structural change and symptomatic presentation. 7 We have used the BESS algorithm (see Appendix 1, Figure 13) to define the entry criteria for the GRASP (Getting it Right: Addressing Shoulder Pain) trial, thereby ensuring that the trial is consistent with national guidance.

Exercise interventions

In designing the trial, we looked at existing evidence regarding the choice of comparator interventions. There is promising evidence from small, short-term trials that physiotherapist-prescribed exercise is effective. 12–14 However, there is a lack of evidence regarding its long-term clinical effectiveness and cost-effectiveness,12–14 despite the widespread provision of physiotherapy for these conditions. There is also uncertainty about which types of exercise and delivery mechanisms (e.g. supervised or home based) are associated with the best outcomes. 12,13,15–17 This evidence is limited by problems in study design and choice of comparators. 13 There are also competing ideologies around which exercise programmes should be considered to ensure a worthwhile trial. Resistance training to improve muscular strength, whether supervised or home based, has been identified as a core component of exercise for rotator cuff disorders, although there is no evidence that any specific programme is superior. 18,19 Manipulation of the exercise volume and intensity is achieved by varying the frequency, load, number of sets, repetitions and rest intervals. 20 A trial of strength training found that duration, specificity of exercises, progression criteria and individualisation (i.e. adjusting the programme to suit each participant) were also important. 21 We did not consider other forms of physiotherapy-led interventions, such as electrotherapy, acupuncture, soft tissue mobilisation, manipulation or stratified care, because of the lack of evidence of their efficacy. 22,23

Little attention has been paid to the need for behavioural frameworks to enhance adherence to advice and exercise programmes and to tackle pain beliefs and behaviour in this context. 24 Non-adherence to physiotherapy treatment is estimated to be up to 70%. 25 In a large trial of exercise for lower back pain that did not include a behavioural component to increase exercise adherence, only half of the participants attended the minimum number of treatment sessions. 26 Risk factors for low adherence include low levels of physical activity, low self-efficacy, depression, anxiety, poor social support and greater perceived barriers to exercise. 24 Some of these risk factors are modifiable in the context of a physiotherapy intervention. We have previous expertise in this area27 and planned to include a behavioural component as part of the progressive exercise intervention.

Corticosteroid injection

There is systematic review evidence that, in comparison with placebo, corticosteroid injections have a short-term benefit in the shoulder, as in other areas of the body. However, there are some concerns about the longer-term benefits and harms of corticosteroid injections. 28–30 The combination of injection and physiotherapy has intuitive appeal, with some evidence of an additive, but not interactive, effect in the short term (3–4 months). 30–33 The longer-term benefits of injections require further study. We planned to use a no-injection comparison, as finding an inert robust placebo is challenging and, given the existing evidence,28–30 we believed that it was unethical and undesirable to progress a placebo arm in a large Phase III trial. In our study based in NHS musculoskeletal services, extended-scope physiotherapists typically deliver the corticosteroid injections. This is increasingly common practice in the NHS, where therapists undertake additional post-registration training to deliver injections, working within a local Patient Group Direction and/or becoming qualified non-medical independent prescribers. 34 Although the use of ultrasound to guide injections in primary care has become increasingly common, evidence from the SUPPORT (SUbacromial imPingement syndrome and Pain: a randomised controlled trial Of exeRcise and injecTion) trial35 and other trials3 have demonstrated that it is no more effective than standard injection practice. Ultrasound guidance also substantially increases the cost and reduces the practicality of injection therapy. Therefore, we planned to deliver injections without the use of ultrasound guidance.

Internal pilot

An internal pilot was included as an integral part of the GRASP trial design, which mirrored the procedures and logistics undertaken in the main GRASP trial. The purpose of the internal pilot was to test and refine the recruitment process and explore treatment acceptability. The decision to progress to the main trial was made in collaboration with the Trial Steering Committee (TSC) and the National Institute for Health Research (NIHR) Health Technology Assessment (HTA) programme based on a predefined progression criterion: reaching the target recruitment rate (42 participants) within the specified time frame (4 months). Data from the internal pilot trial contributed to the final analysis, as there were no substantive changes in design or delivery of the trial interventions.

Study setting

The GRASP trial was conducted across 20 primary care-based musculoskeletal services and their related physiotherapy services in the NHS. These services treat people with a range of musculoskeletal conditions and are run by specialist practitioners, including extended-scope physiotherapists, GPs with a specialist interest in musculoskeletal conditions, clinical nurse specialists and, in some instances, rheumatologists and orthopaedic consultants. Sites were chosen so that they reflected a range of settings (urban and rural) and were able to deliver the trial interventions. The local principal investigator was responsible for the conduct of the research at their site.

Participants

Participants were recruited if referred by their GP or physiotherapist for treatment of a new, but not necessarily first, episode of shoulder pain attributable to a rotator cuff disorder. Participants were predominantly seeking treatment for one shoulder. People who self-referred directly to the musculoskeletal service were also assessed for eligibility, as the typical route of referral varied across services. The participants did not routinely undergo diagnostic imaging, such as magnetic resonance imaging or ultrasound, as a requirement of the trial, as this is generally not recommended in primary care.

Inclusion criteria

• Men and women aged ≥ 18 years.

• A new episode of shoulder pain (i.e. within the previous 6 months) attributable to a rotator cuff disorder (e.g. cuff tendonitis, impingement syndrome, tendinopathy or rotator cuff tear), using the diagnostic criteria set out in the BESS guidelines4 (see Appendix 1, Figure 13).

• Not currently receiving physiotherapy.

• Not being considered for surgery.

• Able to understand spoken and written English.

Exclusion criteria

• Participants with a history of recent significant shoulder trauma (e.g. dislocation, fracture or full-thickness tear requiring surgery).

• Those with a neurological disease affecting the shoulder.

• Those with other shoulder disorders (e.g. inflammatory arthritis, frozen shoulder or glenohumeral joint instability) or with red flags consistent with the criteria set out in the BESS guidelines. 4

• Those who had received corticosteroid injection or physiotherapy for shoulder pain in the previous 6 months.

• Those with contraindications to corticosteroid injections.

Recruitment of participants, screening and eligibility assessment

Potentially eligible participants were identified by clinicians in NHS musculoskeletal services. People attended their clinic appointments in accordance with standard NHS procedures. The treating practitioner in the musculoskeletal services undertook a clinical assessment according to their usual practice. If a patient fulfilled the criteria for a rotator cuff disorder, they were assessed to see whether or not they met the GRASP trial eligibility criteria. Patients were provided with a copy of the participant information sheet and asked if they wished to be considered for the trial. Those who met the eligibility criteria and wanted to participate were approached for informed consent. Participants who did not meet the eligibility criteria or who did not wish to participate received standard NHS treatment. We recorded anonymous information on the age and sex of those who declined to participate so that we could assess the generalisability of those recruited. Reasons for declining were also recorded.

Informed consent and baseline assessment

After participants had been assessed for eligibility, informed consent for participation in the GRASP trial was obtained by a research facilitator at the site who was trained in Good Clinical Practice (GCP). The following was explained to the participant: the exact nature of the study; what it would involve for the participant, including expectations that the participant would be willing and able to attend sessions to receive the study intervention; and any risks involved. The potential participant was provided with a patient information sheet and was given the opportunity to discuss issues and ask questions. In most cases, the process of obtaining informed consent took place during the initial musculoskeletal clinic appointment. Some participants required a second research appointment because they required more time to consider the study or because of local resources at site. Participants were then asked to complete a baseline assessment questionnaire that recorded simple demographic information and baseline measurements for the primary and secondary outcomes (Table 1).

Subacromial corticosteroid injection

The subacromial corticosteroid injection was delivered prior to the progressive-exercise or best-practice advice intervention. The injections were predominantly carried out by extended-scope physiotherapists with appropriate post-registration qualifications in injection therapy who worked within a local patient group directive or as non-medical independent prescribers. 34 This reflects an increasingly common practice in the NHS and ensured that the injections were delivered in the most cost-effective manner possible. The corticosteroid injection was given as per its marketing authorisation and in accordance with its normal indication and therapeutic dosage. 44

The corticosteroid and local anaesthetic were given together in one injection or separately in two injections, depending on local treatment protocols at sites. The corticosteroid injected was either methylprednisolone acetate (Depo-Medrone^®, Pfizer Ltd, Walton Oaks, UK; up to 40 mg) or triamcinolone acetonide (Kenalog™, Bristol-Myers Squibb, Mulhuddart, Ireland; up to 40 mg), as per local treatment protocols. These are the two routinely injected corticosteroids for shoulder pain. There is no clear evidence that either corticosteroid is more effective than the other. 30 The local anaesthetic was either 1% lidocaine (up to 5 ml) or 0.5% bupivacaine hydrochloride (up to 10 ml). We selected sites that adhered to these prescribing boundaries. The choice and dose of corticosteroid, local anaesthetic (including volume) and the injection site were recorded for each participant on a trial injection data collection form (see Appendix 2, Figure 14).

Participants were advised to take care and avoid heavy lifting for 24–48 hours after the injection. Appointments were co-ordinated so that participants typically received their injection within 10 days of randomisation. Very occasionally, a second injection was given after 6 weeks in accordance with the trial protocol (but within 16 weeks of the patient being randomised), but this injection was administered to only those patients who received good initial benefit from their first injection and who requested further pain relief to facilitate their exercises. Any participants who received a second injection had the dose, drug and date of administration recorded on a trial injection data collection form.

Progressive-exercise intervention

Participants randomised to the progressive-exercise intervention received up to six individual face-to-face sessions with a physiotherapist over 16 weeks. These sessions included a behavioural component to encourage adherence to the exercises. A similar rationale has been used to good effect in other trials. 21,45 We chose the number of sessions, spread over this time, to enable progression of the intensity of exercise and provide a sufficient amount of time for a physiological response in the neuromuscular system. 46 Appointments were co-ordinated so that participants typically started their first exercise session within 14–28 days of randomisation, as per local appointment availability. The initial session lasted up to 60 minutes for assessment and setting up the home exercise programme, followed by up to five 20- to 30-minute follow-up sessions. Participants were provided with a folder containing an advice booklet, an exercise action planner and diary, and instructions on their exercise programme set up in collaboration with their physiotherapist. A resistance band was issued as required. The physiotherapists recorded the number of treatment sessions attended by each participant. The intervention was designed to support participants through a progressive dose of exercises and optimise adherence to the home exercise plan. The progressive-exercise programme was highly structured, but could be tailored to the needs and preferences of participants, with the aim of helping them achieve their rehabilitation goals. Importantly, the intervention could be delivered within the current NHS commissioning paradigm. 47

Best-practice advice intervention

Participants randomised to the best-practice advice intervention received a single, individual face-to-face session with a physiotherapist, lasting up to 60 minutes. Again, appointments were co-ordinated so that participants typically started their exercise session within 14–28 days of randomisation, as per local appointment availability. After a comprehensive shoulder assessment, participants were given an advice booklet. The content of the advice in the booklet was the same as that provided for the progressive-exercise group, with the exception of the different exercise programme. An exercise diary was also provided, along with a simplified version of the exercise action planner (see Chapter 3). Tailored education, reassurance and self-management exercise advice, including advice on pain management and activity modification, was offered. Participants were also given a simple set of self-guided exercises, including at least one level of resistance band and an exercise video [available on a website and digital versatile disc (DVD)], which could be progressed and regressed, depending on their capability. The exercises were designed using similar concepts to those of the progressive-exercise intervention, such as increased resistance, but with a simpler range of exercise options that were not supervised.

The best-practice advice intervention was selected as the comparator because it is consistent with current clinical practice guidelines regarding the self-management advice that should be provided to people with rotator cuff disorders. 4,5 In addition, people may find a single advice session and DVD preferential to a course of face-to-face physiotherapy sessions, as they do not have to come back to the hospital or clinic, take time off work or make carer arrangements.

Concomitant care

All participants were advised that they could take over-the-counter analgesia as required (e.g. paracetamol with or without codeine, or an oral non-steroidal anti-inflammatory drug) in accordance with the BESS guidelines. 4 Participants could seek other forms of treatment during the follow-up period, but were informed that they should use usual routes (predominantly NHS referral) to do so. Additional treatments, including contact with their GP or other health professional, changes in medication, use of physical treatment and alternative therapies, were recorded as a treatment outcome through the patient questionnaires at 8 weeks, 6 months and 12 months post randomisation.

Training and monitoring of intervention delivery

All physiotherapists delivering study interventions, progressive exercise and best-practice advice had access to a comprehensive intervention manual and were required to have undertaken trial-specific training by a GRASP trial research physiotherapist. A rigorous quality control programme was also conducted to ensure intervention fidelity (see Chapter 3).

Primary outcome

The primary outcome was shoulder pain and function over the 12 months post randomisation measured using the SPADI,10,11 which was developed to measure current (i.e. in the last week) shoulder pain and disability in an outpatient setting. The SPADI scale is based on 13 questions, all scored on a 0–10 scale, on which 10 is the worst score. In addition, the SPADI scale has a five-item pain subscale and an eight-item disability subscale. The subscale items are summed and converted to a 0–100 scale, where a higher value denotes more pain and/or disability. A systematic review of outcome measurement sets for shoulder pain trials showed that SPADI is the most commonly used measure to assess pain and disability. 48 The SPADI scale has good psychometric properties, is used widely in the field and can be completed using a postal questionnaire.

Secondary outcomes

Secondary outcomes (see Table 1) were subdomains of the SPADI, which are pain measured using the SPADI five-item pain subscale10,11 and function measured using the SPADI eight-item disability subscale;10,11 health-related quality of life, measured using the well-validated EuroQol 5 Dimensions, five-level version (EQ-5D-5L) score;38 psychological factors, measured using the Fear Avoidance Belief Questionnaire – Physical Activity (FABQ-PA) five-item subscale39 and the Pain Self-Efficacy Questionnaire, two-item version (PSEQ-2);40 sleep disturbance, measured using the Insomnia Severity Index (ISI);41 patient global impression of change;42 RDA, including work, social life and sport activities; patient adherence to exercise; any serious adverse events (SAEs); health resource use, including consultation with primary and secondary care, prescribed and over-the-counter medication use, additional physiotherapy or injection use, and hospital admission; additional out-of-pocket expenses; and work absence (i.e. number of sickness days).

The EQ-5D-5L38 is a validated, generic health-related quality-of-life measure comprising five dimensions, each with a five-level answer possibility and a health thermometer scale. The EQ-5D-5L can be used to report health-related quality of life in each of the five dimensions and each combination of answers can be converted into a health utility score, where 1 represents perfect health and 0 indicates health states equal to death. The health thermometer scale (EuroQol visual analogue scale) takes values between 0 and 100, where 0 represents worst imaginable health and 100 best imaginable health. It has good test–retest reliability and gives a single preference-based index value for health status that can be used for broader cost-effectiveness comparative purposes.

The Fear Avoidance Belief Questionnaire (FABQ)39 is a validated measure of fear-avoidance behaviour. The FABQ-PA is a subscale of the FABQ that measures fear-avoidance beliefs about physical activity using five items scored on a 0 (strongly disagree) to 6 (strongly agree) scale. The total score range for FABQ-PA is 0–24, with higher scores representing greater levels of fear-avoidance behaviour.

The PSEQ40 is a well-established 10-item measure of pain self-efficacy (i.e. a belief in one’s ability to carry out activities despite pain). PSEQ-2 is a two-item, short measure of pain self-efficacy. 40 The two items reflect the confidence in ability to work and lead a normal life despite the pain, on a 0 (not at all confident) to 6 (completely confident) scoring scale. The total PSEQ-2 score is summed from the two items, giving a range from 0 to 12, with higher values representative of higher confidence levels despite the pain.

The ISI41 is a brief self-report measure of a patient’s perception of their insomnia, targeting the subjective symptoms and consequences of insomnia, as well as the degree of concerns or distress caused by those difficulties. The ISI has seven items rated on a 0–4 scale and the total score is a summation of these items, with a value ranging from 0 to 28. Higher scores are suggestive of more severe insomnia. A detailed interpretation of the ISI total score is as follows:

• A score of 0–7 indicates no clinically significant insomnia.

• A score of 8–14 indicates subthreshold insomnia.

• A score of 15–21 indicates clinical insomnia of moderate severity.

• A score of 22–28 indicates severe clinical insomnia.

The Global Impression of Treatment (GIT)42 is a simple method of measuring change in health status, with respect to shoulder problems, by charting self-assessed clinical progress on an 11-point scale that ranges from –5 (very much worse) to 5 (completely recovered). Psychometric properties include a minimum detectable change of 0.45 points and a minimally clinically important difference (MCID) of 2 points.

Return to desired activities is a self-reported outcome that aims to measure physical function during social life, recreational activities and work. RDA is an adapted version of the Disabilities of the Arm, Shoulder and Hand (QuickDASH), using three questions with a five-point Likert scale answer option, with lower scores indicating better function.

Other secondary outcomes (including medication usage, work disability, health-care use and out-of-pocket expenses) are analysed separately as part of a health economics analysis (see Chapter 5).

Sample size

The target sample size for the trial was 704 randomised participants (176 in each treatment arm). This sample size was based on 90% power and 1% two-sided statistical significance to detect a minimally clinically important between-group difference of 8 points on the SPADI total scale,10 assuming a baseline standard deviation (SD) of 24.3 (chosen as representative of the patient population50). This difference is the equivalent of a standardised effect size of 0.33, which required a sample size of 550 participants [Power Analysis and Sample Size 13 NCSS Statistical Software Kaysville, UT, USA; URL: www.ncss.com (accessed 20 May 2021)]. Allowing for a potential loss to follow-up of 20% at 12 months inflated the sample size to 688. We further inflated the sample size to take into account the potential for a small clustering by physiotherapist effect in the progressive-exercise intervention group. We used an interclass correlation (ICC) of 0.001, based on our experience with individually tailored physiotherapy interventions51 and the expectation that each physiotherapist would treat approximately 20 participants in the progressive-exercise intervention group. This lead to an inflation of:

and increased the sample size to the total of 704 participants.

This sample size was based on the assumption that there was no interaction effect and was powered for the two main effect comparisons: (1) progressive exercise compared with best-practice advice and (2) corticosteroid injection compared with no injection. However, this number of participants also provided 80% power and 5% two-sided significance to detect an interaction standardised effect size of 0.35, if an interaction effect did exist. The interaction effect was tested before the main effect comparisons were undertaken. It should be noted that a non-significant interaction effect did not preclude a smaller interaction that this study was not powered to detect. We chose 90% power and 1% two-sided significance to provide more convincing evidence of any treatment effects discovered. No further adjustments to the sample size were made because of multiple testing. The Data Monitoring and Ethics Committee (DMEC) reviewed the sample size assumptions after 338 participants were recruited and no changes were made to the final sample size.

Interaction

An interaction between the two main effect comparisons was not expected, but the trial was powered to identify a moderate standardised interaction effect of 0.35. The presence of an interaction between the two interventions was formally investigated before testing their effects on the primary outcome. An initial regression model was fitted for the primary outcome to predict the outcome of interest and included the two effects of interest [i.e. (1) individually tailored progressive-exercise programme compared with best-practice advice and (2) subacromial corticosteroid injection compared with no injection] and their interaction.

In the presence of a non-statistically significant treatment interaction (i.e. p ≥ 0.05), a factorial analysis was planned to determine the success of the trial. The effects for the individually tailored progressive-exercise programme and corticosteroid injection were determined separately from this model as mean differences (MDs) with associated 99% confidence intervals (CIs), as appropriate, adjusted for the relevant covariates. The model used did not include an intervention interaction term.

If a statistically significant treatment interaction (i.e. p < 0.05) had been detected then the effect of the individually tailored progressive-exercise programme and corticosteroid injection would have been evaluated from comparisons within the intervention groups, referred to as ‘inside-the-table’ comparisons (i.e. group A vs. group B to test the effect of the individually tailored progressive exercise programme and group B vs. group D to test for the effect of the corticosteroid injection) (see Table 2). The main effects and their interaction terms would have been included in the analysis model, their regression coefficient with corresponding 95% CI for the interaction terms presented and the reduced statistical power of this model noted.

Primary outcome analysis

The difference in SPADI between the two intervention groups was estimated overall and at each data collection time point using a repeated measures linear mixed-effects regression model. 53 The model was adjusted for the fixed effects of age, sex and baseline SPADI, and random intercepts by centre and observations within participants. Robust standard errors (SEs) for treatment effects from all time points were reported. Clustering by physiotherapist in the progressive-exercise group was accounted for using cluster-robust SEs as part of the mixed-effects model. The final trial results were based on the adjusted model. A non-parametric statistical test (e.g. Mann–Whitney test for comparison of means) with no adjustment and medians and interquartile ranges (IQRs) were reported where approximate normality for the model residual terms was not established. Statistical significance was set at the 1% level and corresponding 99% CIs were reported for the primary outcome. Flooring effects in the SPADI outcome over the 12 months were explored and reported for each treatment group.

Secondary outcome analyses

The intervention effects on secondary outcomes were analysed following the analysis method described for the primary outcome on the basis that the outcomes are clinically similar. If there was no statistically significant evidence of an interaction effect for the primary outcome, then no interaction effect would be assumed for the secondary outcomes. Likewise, if a statistically significant interaction effect was identified for the primary outcome, then the secondary outcomes would be analysed, assuming an interaction effect was present. Continuous secondary outcomes analyses were conducted following similar methods to the outline for the primary outcome analysis, using linear regression for continuous outcomes and logistic/multinomial logistic regression (i.e. logit, mlogit or ologit, as appropriate) for binary and ordinal outcomes. Statistical significance for secondary outcomes was set at 5% and 95% CIs were reported.

Missing data

Missing data were reported and summarised by treatment group. Item-level imputation for the primary outcome SPADI was carried out for items where no more than two out of five items in the pain subscale were missing and no more than three out of eight items in the function subscale were missing, given that no more than 10% of cases had missing data. 54,55 Missing continuous primary and secondary outcomes were handled as part of the likelihood-based estimation of the repeated measures mixed-effects model, assuming the data were missing at random. 56 This method took account of missing observations owing to missed visits or a participant leaving the study prematurely. The distribution of missing data was explored to assess the assumption of data being missing at random. Full details are provided in the SAP. 37

Sensitivity analysis

Sensitivity analyses were used to assess the robustness of the trial results in the light of the assumptions made about the underlying missing data mechanism. Most analyses assume data to be missing at random or missing completely at random. The sensitivity analysis, therefore, assumed missing not at random, such that missing outcomes were assumed to be worse or better than the observed outcomes.

Prespecified subgroup analysis

Subgroup effects in the following prespecified subgroups were analysed for the primary outcome, utilising subgroup-by-treatment interactions:

• Age: ≤ 64 years vs. ≥ 65 years. (Rationale: increasing age has been shown to be associated with poorer outcome. 57,58)

• Sex: male vs. female. (Rationale: prevalence is higher in males than in females. 1)

• Smoking status: never smoked vs. former smoker or current smoker. (Rationale: smoking has been shown to be associated with a negative effect on tendon healing. 59)

• Higher SPADI score at baseline. (Rationale: higher pain and functional disability at baseline may be associated with poorer outcome. We defined a higher SPADI score as ≥ 50 at baseline when the SPADI is converted to the 0 to 100 scale. 19,57)

• Higher pain self-efficacy (PSEQ-2) score at baseline. (Rationale: higher belief in one’s ability to carry out activities despite pain may be associated with better outcome. We define a higher PSEQ-2 score as ≥ 8 at baseline when the PSEQ-2 is converted to the 0–12 scale.)

Supplementary/additional analyses

A complier-average causal effect (CACE) analysis was used to investigate the role of compliance in the treatment effect, given the CACE assumptions described in the SAP. Compliance with intervention was defined in the SAP37 as follows. For the progressive exercise intervention, participants were considered compliant with treatment if they had been signed off for completing treatment or if they receive all six physiotherapy sessions. For corticosteroid injection intervention, participants were considered compliant if they received at least one injection. If no evidence of a statistically significant interaction between the two treatments was identified, then this analysis was planned to be conducted ‘at the margins’ and using a similar mixed-effects model. If a statistically significant treatment interaction for the primary outcome was identified, the CACE analysis would be conducted ‘inside the table’.

Ethics committee approval

The GRASP trial protocol and all related documentation (e.g. informed consent forms, participant information leaflets, patient questionnaires and any proposed advertising material) was approved by the Berkshire B Research Ethics Committee (reference 16/SC/0508) and Integrated Research Application System (ID 199243). The trial was also approved by the UK competent authority the Medicines and Healthcare products Regulatory Agency, as it was classified as a clinical trial of an investigational medicinal product (EudraCT number 2016-002991-28). The trial was conducted in accordance with the principles of the Declaration of Helsinki61 and the Medical Research Council’s GCP guidelines. 62

Trial Management Group

A Trial Management Group, consisting of the core trial team, chief investigator and co-applicants, was responsible for the day-to-day running of the trial and met monthly to report on progress and ensure that milestones were met. A trial manager oversaw all aspects of the day-to-day trial management.

Trial Steering Committee

A TSC was responsible for monitoring the trial’s progress and providing independent advice. The TSC comprised an independent clinician, two specialist physiotherapists, a statistician, a health economist and two patient representatives.

Data Monitoring and Ethics Committee

A DMEC was responsible for monitoring the trial’s progress and providing independent advice. It advised the chairperson of the TSC if, at any time, in its view, the trial should be stopped for ethics reasons, including concerns about participant safety. The DMEC comprised an independent clinician, health service researchers, a specialist physiotherapist and a statistician.

Intervention development

In Chapter 1, the lack of evidence for particular exercises, treatment intensities or durations were outlined. With no clear evidence or clinical consensus, advice from clinicians, patient and public representatives and other experts was crucial to the development of the GRASP trial interventions. Twenty-six clinicians, researchers, patients and public representatives attended a GRASP intervention development meeting (June 2016). Delegates discussed and evaluated a comprehensive list of 22 exercise types commonly used in clinical practice and/or reported in trials of shoulder pain treatments (see Appendix 3, Table 26). The delegates categorised the exercises into essential exercises, optional exercises or exercises considered not important for managing rotator cuff disorders.

Those selected as essential exercises generally strengthen the posterior rotator cuff muscles (i.e. supraspinatus, infraspinatus and teres minor) and load the shoulder into an elevated position, consistent with current trial evidence. 17,64 The GRASP trial, therefore, included these exercise types as an essential component of both the progressive-exercise and best-practice advice exercise interventions. Delegates agreed that both exercise interventions should be as practical and simple as possible. Some exercises were considered important for patients with specific presentations or in particular circumstances. These were included as optional exercises in the progressive-exercise intervention. They were not incorporated into the best-practice advice intervention, which prioritised a simple, progressive set of exercises that were likely to benefit most patients and could be easily understood and performed at home without supervision.

Delegates agreed that stretching was not a priority target in rotator cuff disorders and that range of motion could be incorporated into other active exercises. A posterior capsule/soft tissue stretch for the shoulder was included as an option in the progressive-exercise intervention as we recognised that these stretches feature in many exercise programmes evaluated in other clinical trials. 14 Isolated exercises to correct posture towards a theoretical ideal were not included, as there is limited evidence for this approach. 65 Experts disagree on whether exercises can or should provoke symptoms or should be symptom-free. There is limited evidence for these alternatives, although evidence is building for the acceptability of mild-to-moderate pain symptoms during exercise. 66,67 The participant information booklet, which was included as part of the GRASP trial intervention materials, and treating physiotherapists advised participants that some pain during the exercises is acceptable, provided that they found it manageable and symptoms resolved to an acceptable level within a few hours.

After the meeting, we developed detailed intervention manuals for the treating physiotherapists that described the key components of the progressive-exercise and best-practice advice interventions. We also developed patient-facing materials, which were reviewed by patient representatives.

Internal pilot

As part of the GRASP trial, we ran an internal pilot from February to June 2017 at three sites, recruiting 42 participants. One of the aims of the internal pilot was to test and refine the recruitment process and explore treatment acceptability. The physiotherapists delivering the GRASP exercise interventions were also asked to provide feedback on the delivery of the intervention. The exercise intervention manuals and training materials were subsequently refined to clarify identified misconceptions. The interventions did not require significant modifications.

GRASP trial exercise interventions

Progressive-exercise intervention

Participants received up to six sessions with a physiotherapist over 16 weeks. The initial session was set to last up to 60 minutes for assessment and treatment initiation, followed by up to five 20- to 30-minute follow-up sessions. The physiotherapist and participant decided how frequently to schedule the review appointments within the 16-week time frame. Six sessions over 16 weeks was set to allow the exercise intensity to be progressed and because it was a volume of physiotherapy within the range of feasible delivery in the NHS. It was also deemed to be a sufficient time for a physiological response in the neuromuscular system. 46 The participant information booklet was provided in a file so that instructions on the progressive exercises selected could be inserted, including detailed guidance and photos for each exercise. The physiotherapist and participant jointly chose one to three exercises to address the problems identified during the assessment (Figure 1). We anticipated participants presenting with different problems and pain irritability and, therefore, the exercises were categorised into three difficulty levels (level 1, 2 and 3 exercises), with progressions within each level and different aims and guidelines to match indications for use.

FIGURE 1.

Process map of the GRASP trial progressive exercise intervention.

Level 1 exercises: simple shoulder movements

Level 1 exercises aimed to reduce fear, encourage normal movement, improve shoulder mobility and build confidence in exercising independently at home (see Appendix 3, Figure 17). The exercises were considered appropriate for participants with irritable and/or severe shoulder pain and/or fear avoidance. The frequency and the number of sets and repetitions for each exercise were agreed between the physiotherapist and participant. Depending on symptom severity, it was recommended that some participants could reasonably move straight to level 2 exercises.

Level 2 exercises: progressive structured resistance training

Level 2 exercises included the essential exercises that target strengthening of the posterior rotator cuff muscles and other optional exercises (see Appendix 3, Figure 18). The essential exercises focused on movements commonly affected by a rotator cuff disorder (i.e. resisted external rotation, flexion and abduction of the shoulder). 64 Resistance exercises for other shoulder movements were optional. If participants were prescribed exercises at level 2, at least one had to be from the essential exercise category.

The American College of Sports Medicine guidance for progressive resistance training recommends two to three sessions of resistance training per week,20 whereas many studies of resistance training in patients with musculoskeletal disorders use daily exercise programmes. 21,72 We attempted to strike a balance, ensuring that the resistance training was effective, but also regular enough to address other aims, such as building confidence in moving the arm and re-learning motor skills. We asked GRASP trial participants to do their exercises 5 days per week, with 2 non-consecutive recovery days.

Physiotherapists were taught to regulate the exercise intensity using the modified Borg scale of perceived exertion, an 11-point version of the Rating of Perceived Exertion (RPE) scale,73 validated for quantifying the intensity of resistance exercise. 74 Participants started at a moderate load (3 or 4 on the Borg scale) to enhance motivation and adherence and reduce the likelihood of symptom flare-up. Figure 2 contains detailed guidance on the scale and how level 2 exercises were initiated, progressed and regressed. Where appropriate, resistance bands or hand weights were incorporated into the exercise, with the level of resistance recommended being guided by the RPE scale feedback from the participant.

FIGURE 2.

Guidance for setting level 2 resistance exercises in the GRASP trial. (a) Modified Borg scale; (b) setting the initial exercise; and (c) progression and regression decision guidance.

Level 3 exercise: patient-specific functional restoration

It was recommended that participants progress to level 3 when their initial problems (e.g. weakness) began to resolve. The physiotherapist altered the exercise programme to be more task specific (e.g. returning to sport) by devising a new exercise in consultation with the participant. Level 3 exercises aimed to modify the essential resistance-training exercises towards the specific movements required to achieve the participant’s individual functional goals. The exercise could be high or low load, using many or few repetitions, depending on the participant’s needs. It was recognised that not all GRASP trial participants would reach or need this stage of the programme. We anticipated that level 3 exercise instructions would be given face to face and reinforced with written guidance to aid recall.

Optional stretching exercise

Posterior capsule and/or soft tissue stretches of the shoulder (see Appendix 3, Figure 19) were included as optional exercises at any stage. We recommend selective use, generally for younger adults engaged in throwing or other overhead athletic or physical activities75 who have posterior capsule tightness. We anticipated these exercises to be suitable for participants with low irritability and if they did not provoke symptoms. Although discomfort and stretching sensations were considered acceptable during stretches, we did not anticipate provocation of anterior shoulder pain or reproduction of the participant’s specific symptoms. We advised holding stretches for 20–30 seconds, but the physiotherapist and participant decided the number of repetitions and frequency.

Exercise progression

During the physiotherapist training, emphasis was placed on the need to progress the exercise interventions as much as possible. To determine that this had in fact occurred, treatment logs completed for each session by the physiotherapists [detailing prescribed exercises, number of sets/repetitions (i.e. volume) and exercise intensity73] were analysed to allow an estimation of treatment progression for the progressive-exercise intervention (see Chapter 4). Exercises were categorised and ranked in order of difficulty level (1 = least difficult, 8 = most difficult) according to the following scale:

1. passive movements

2. isometric

3. active assisted

4. active (unassisted)

5. loaded exercises (isotonic) in neutral (0 °)

6. loaded exercises (isotonic) at 90 °(supported)

7. as for 6, but unsupported

8. as for 7, but through range.

Progression of the progressive-exercise intervention was defined as an increase in exercise difficulty, or volume and/or load across attended sessions. Maintained indicates no changes in exercise parameters. Regressed is defined as a decrease in exercise level, or volume and/or load.

Centre characteristics

Twenty English NHS trusts participated in the GRASP trial (see Appendix 4, Table 28). The recruitment period (20 February 2017 to 2 May 2019) at each trust varied from 9 to 26 months, with resulting variation in the number of participants recruited at each site.

Participating clinicians

A total of 223 clinicians across the 20 trial sites were involved in providing the study treatment interventions. Fifty-three physiotherapists and three doctors delivered subacromial corticosteroid injections to 329 (97%) and 10 (3%) participants, respectively. The best-practice advice and progressive-exercise interventions were delivered by 83 and 104 physiotherapists to 324 and 339 participants, respectively (see Appendix 4, Table 29). The physiotherapists and doctors worked in hospital physiotherapy, orthopaedic outpatient departments or at satellite clinics (e.g. general practices). Nine physiotherapists at two sites were working for a private company contracted to provide services to NHS patients [Buckinghamshire Musculoskeletal Integrated Care Service (High Wycombe, UK) and Medway Community Healthcare (Rochester, UK)]. The number of physiotherapists involved at each site varied from 3 to 43 (see Appendix 4, Table 28). Several physiotherapists provided both the corticosteroid injection and either the best-practice advice or progressive-exercise intervention. Two physiotherapists swapped treatment groups during the trial because of staffing issues at sites and, therefore, delivered both exercise interventions.

The level of experience of each physiotherapist (as determined by Agenda for Change pay banding)72 is described in Appendix 4, Table 29. Sixty-four per cent (34/53) of physiotherapists delivering injections were band 8. Exercise interventions were predominantly delivered by band 6 and 7 physiotherapists (73% and 80% for best-practice advice and progressive-exercise interventions, respectively).

Screening

Screening and recruitment took place between February 2017 and April 2019. A total of 2287 patients were screened and assessed to see whether or not they met the GRASP trial eligibility criteria (Table 5). A total of 1284 (56%) patients were eligible and 708 agreed to participate. Of those participants who did not meet the eligibility criteria (1003/2287), the most common reasons were owing to another shoulder disorder (e.g. inflammatory arthritis, frozen shoulder, glenohumeral joint instability) or red flags consistent with BESS criteria (n = 434, 43%), or had received corticosteroid injection or physiotherapy for shoulder pain in last 6 months (n = 167, 17%). Of those participants who were eligible but declined to participate (n = 576), the most common reasons given were not interested in taking part in research (n = 175, 30%) or had a treatment preference for not receiving injection (n = 158, 27%). For full details see Appendix 4, Table 30. The mean age of those who declined to participate was 55 years (median 56 years, IQR 46–65 years) and 53% (n = 306) were female.

A total of 708 participants were recruited between March 2017 and May 2019 (see Appendix 4, Figure 24). This exceeded the target of 704 participant; four participants were randomised once the target was reached as they were already in the process of joining the study.

Baseline characteristics of participants

The baseline characteristics of participants recruited into the trial are shown in Table 6 (for more detail see Appendix 4, Tables 31–33). Intervention groups were generally well matched in terms of baseline demographic data and primary and secondary outcome measurements. Stratification factors used in the randomisation are shown in Appendix 4, Table 34. The majority of participants were white British (89.6%), with similar numbers of men and women (50.7% vs. 49.3%). The overall mean age was 55.5 (SD 13.1) years. The majority of participants reported being married (64.8%) or living with a partner (13.1%). More than two-thirds of participants were classified as either overweight (38.7%) or obese (30.6%). Approximately half (58.9%) reported being in employment, with one-quarter (25.8%) being retired. The majority of participants reported being right-handed (87%) and the numbers were similar in terms of whether the right or left shoulder was affected (49.3% vs. 46.9%). The average duration of rotator cuff symptoms was 4 (median 4, IQR 3–6) months.

The overall mean baseline SPADI score was 54.1 (SD 18.5) (on a scale of 0–100, with higher values being worse), suggesting a moderate level of disability, with higher levels of shoulder pain (mean SPADI pain subscale 63.9, SD 17.1) compared with impairment of shoulder function (mean SPADI function subscale 44.3, SD 22.1). Overall baseline scores for secondary outcome measures were mean 15.1 (SD 5.5) for FABQ-PA, mean 9.7 (SD 2.3) for PSEQ-2, mean 10.5 (SD 6.3) for ISI and mean 7.9 (SD 2.6) for RDA. Mean baseline values were in the moderate to low range, suggesting that a participant’s rotator cuff disorder may not be having considerable impact on sleep and daily activities despite higher levels of pain.

Withdrawals

Overall, 26 (3.7%) participants withdrew from the trial: 15 withdrew from the intervention delivery only and 11 withdrew from both the intervention and follow-up questionnaire completion. Numbers were very similar across intervention groups. One participant withdrew completely from the trial and withdrew permission to use their data. Details of the reasons for withdrawal by intervention group are shown in Table 7. Two participants died and both deaths were unrelated to the trial intervention(s).

Compliance with intervention

Of the 708 participants randomised, the majority received treatment as allocated, with 94% (339/360) of participants randomised to receive subacromial corticosteroid injection prior to either the best-practice advice or progressive-exercise intervention receiving the injection as allocated (Table 8). For the exercise interventions, there was no difference in attendance rates between those receiving the exercise interventions only and those who received exercise in conjunction with a corticosteroid injection. Overall, 92% (324/352) of participants of those allocated to receive the best-practice advice intervention attended, and 95% (339/356) of participants allocated to receive progressive exercise either partially or fully completed the exercise intervention.

The proportion of participants not attending any treatment sessions was roughly equal between intervention groups. For the progressive-exercise intervention, just over half of participants received four out of a maximum of six sessions, with almost 80% receiving at least three sessions. Note that, participants could be discharged prior to attending six sessions if their symptoms had improved sufficiently.

A small number of participants received extra treatment over and above the interventions defined by the GRASP trial protocol. Two sites reported that two participants received a second subacromial corticosteroid injection in addition to that received as part of their treatment allocation. With regard to the exercise interventions, eight participants allocated to receive best-practice advice received a total of 10 additional sessions. For progressive exercise, five participants received a total of eight additional sessions over and above the six-session maximum. Three of the 13 participants who received additional sessions required further reassurance regarding their condition and/or the exercise programme, four required further session(s) before treatment was concluded satisfactorily, three reported a re-occurrence or suffered re-injury and three reported no improvement in symptoms. There was no sizeable difference in the proportion of participants receiving additional sessions (nor the reasons for further contact) between the four intervention groups.

Timing of delivery of interventions

Appendix 4, Table 36, presents median time from randomisation to first appointments for each intervention group. Subacromial corticosteroid injections prior to receiving the best-practice advice or progressive-exercise interventions were largely delivered within 10 days of randomisation in accordance with the GRASP trial protocol. The first session of the exercise interventions was also delivered within the required time frames for all trial groups, with the majority of participants being seen between 14 and 28 days after randomisation. Participants randomised to injection received their first exercise session 6 days later, on average, than those who were not allocated to receive an injection, possibly because of logistical issues with booking appointments. The progressive-exercise intervention was generally completed within the recommended 16-week time frame, with the median time corresponding to between 13 and 15 weeks for both intervention groups. The initial delay experienced by those who received an injection prior to their exercise sessions is reflected in the longer time to the last session, although the ‘actual’ exercise treatment period (i.e. first to last exercise session) was almost identical for both progressive-exercise arms.

Content and adherence to interventions

Injecting clinicians and physiotherapists completed a treatment log for each session, detailing the type of treatment provided.

Exercise progression

Appendix 4, Table 38, describes the proportion of participants who were progressed, maintained or regressed, as defined previously in Chapter 3. Two-thirds (227/339) of participants who received the progressive-exercise intervention had their exercise progressed in accordance with the instructions provided in training. A small minority of participants had their exercise programme regressed (39/339, 12%) overall, whereas the remaining participants maintained (73/339; 22%) exercising at the same level as the initial session over the course of their treatment.

Participant-reported adherence to exercise

As part of the participant follow-up questionnaires, participants were asked whether or not they were currently performing the exercises the physiotherapist asked them to complete at home as part of the GRASP trial. A total of 77.8% of participants reported ‘yes’ at 8 weeks, but this dropped to 49.3% and 32.1% at 6 and 12 months, respectively. Participants were also asked how often they were performing these exercises. Appendix 4, Table 39, shows the difference in proportions across intervention groups. At 8 weeks, participants who received progressive exercise were more likely to report performing their exercises 5 days per week than those receiving best-practice advice (60.4% vs. 43.2%) and there was little difference at 6 and 12 months. At 8 weeks and 6 months, participants who received injection were more likely to report performing their exercises 5 days per week than those who received no injection (8 weeks, 57.5% vs. 46.0%; 6 months, 21.4% vs. 12.1%). There was little difference at 12 months.

Complier-average causal effect analysis

The primary outcome was analysed in the ITT population. An additional CACE analysis was used to investigate the role of compliance with intervention on the trial effects for the primary outcome. Compliance with intervention was defined in the SAP37 as follows. For the progressive-exercise intervention, participants were considered compliant with treatment if they had been signed off for completing treatment or if they received all six physiotherapy sessions. For the injection intervention, participants were considered compliant if they received at least one injection. Compliance with treatment by intervention group is summarised in Table 8. As no evidence of a statistical interaction effect between interventions was identified, the CACE analysis was conducted at the margins for the two main effect comparisons.

Progressive exercise compared with best-practice advice

The proportion of compliers in the progressive-exercise intervention group was 78%. Compliance with the progressive-exercise intervention did not have a significant effect on the primary outcome. The CACE estimate analysis for progressive exercise over 12 months was (mean adjusted difference) –0.27 (95% CI –2.69 to 2.16; p = 0.830) (see Table 13). The model included 1869 participant data points and was adjusted for sex, age, baseline SPADI and centre. The variable for primary physiotherapist was omitted from this analysis because of collinearity.

Injection compared with no injection

The proportion of compliers in the injection treatment group was 94%. Compliance with injection did not have a significant effect on the primary outcome. The CACE estimate analysis for injection over 12 months was mean adjusted difference –1.50 (95% CI –3.61 to 0.61; p = 0.164) (see Table 14). The model included 1869 participant data points and was adjusted for sex, age, baseline SPADI and centre. The variable for primary physiotherapist was, again, omitted because of collinearity.

Progressive exercise compared with best-practice advice

Table 15 presents the adjusted analysis results for the comparison of progressive exercise with best-practice advice for each of the secondary outcomes at 8 weeks and at 6 and 12 months (for unadjusted results see Appendix 4, Table 42).

SPADI: shoulder pain subscale

There was an improvement in shoulder pain in both intervention groups over time (Figure 6). However, there was no statistically significant difference between the progressive-exercise intervention and best-practice advice in shoulder pain when analysed over 12 months (adjusted mean SPADI pain subscale difference between groups –1.61, 95% CI –4.94 to 1.72) or when analysed at 8 weeks or at 6 and 12 months (adjusted MD at 12 months –4.01, 95% CI –8.11 to 0.09).

FIGURE 6.

Marginal adjusted mean SPADI pain values from the repeated measures mixed-effects model and associated 95% CIs for the two treatment groups from baseline to 12 months: best-practice advice vs. progressive exercise.

SPADI: shoulder function subscale

There was an improvement in shoulder function in both intervention groups over time (Figure 7). However, there was no statistically significant difference between the progressive-exercise intervention and best-practice advice in shoulder function when analysed over 12 months (adjusted mean SPADI function subscale difference between groups –0.15, 95% CI –2.92 to 2.61) or when analysed at 8 weeks or at 6 and 12 months (adjusted MD at 12 months –2.61, 95% CI –6.03 to 0.81).

FIGURE 7.

Marginal adjusted mean SPADI function values from the repeated measures mixed-effects model and associated 95% CIs for the two treatment groups from baseline to 12 months: best-practice advice vs. progressive exercise.

Fear Avoidance Belief Questionnaire – Physical Activity

There was an improvement in fear avoidance behaviour in both intervention groups over time. However, there was no statistically significant difference between the progressive-exercise intervention and the best-practice advice intervention in fear avoidance behaviour when analysed over 12 months (adjusted mean FABQ-PA score difference between groups –0.35, 95% CI –1.16 to 0.46) or when analysed at 8 weeks, 6 months or 12 months (adjusted MD at 12 months –0.97, 95% CI –1.99 to 0.05).

Pain Self-Efficacy Questionnaire

There was no statistically significant difference between the progressive-exercise intervention and best-practice advice intervention in patient-reported pain self-efficacy when analysed over 12 months (adjusted mean PSEQ-2 score difference between groups 0.02, 95% CI –0.23 to 0.27) or when analysed at 8 weeks, 6 months or 12 months (adjusted MD at 12 months 0.12, 95% CI –0.22 to 0.45). A ceiling effect was noted for this outcome, with most participants reporting higher confidence levels despite experiencing pain.

Insomnia Severity Index

There was an improvement in patient-reported perception of insomnia in both intervention groups over time. However, there was no statistically significant difference between the progressive-exercise intervention and best-practice advice in patient-reported perception of insomnia when analysed over 12 months (adjusted mean ISI difference between groups 0.04, 95% CI –0.69 to 0.77) or when analysed at 8 weeks, 6 months or 12 months (adjusted MD at 12 months –0.52, 95% CI –1.40 to 0.36).

Return to desired activities

There was an improvement in patient-reported RDA in both intervention groups over time. However, there was no statistically significant difference between the progressive-exercise intervention and best-practice advice intervention in patient-reported RDA when analysed over 12 months (adjusted mean RDA difference between groups –0.06, 95% CI –0.36 to 0.23) or when analysed at 8 weeks, 6 months or 12 months (adjusted MD at 12 months –0.14, 95% CI –0.51 to 0.24).

Global Impression of Treatment

Progressive exercise resulted in an improvement in patient-reported GIT success over 12 months (adjusted mean GIT difference between groups 0.38, 95% CI 0.10 to 0.66) and at 6 and 12 months (adjusted MD at 12 months 0.51, 95% CI 0.14 to 0.87), compared with best-practice advice. However, this result was not seen at 8 weeks and the difference was not clinically significant (MCID on the GIT scale is 2 points on an 11-point scale).

Injection compared with no injection

Table 16 presents the adjusted results for the comparison of injection with no injection for each of the secondary outcomes at 8 weeks, 6 months and 12 months (for unadjusted results, see Appendix 4, Table 42).

SPADI: shoulder pain subscale

There was an improvement in shoulder pain in both groups over time (Figure 8). Injection resulted in an improvement in shoulder pain when analysed at 8 weeks, compared with no injection (adjusted mean SPADI pain subscale difference between groups –7.38, 95% CI –11.10 to –3.67). This difference was not clinically significant (MCID on the SPADI scale is 8 points on a scale of 0–100). There was no statistically significant difference between injection and no injection when analysed over 12 months (adjusted MD over 12 months –1.55, 95% CI –4.46 to 1.37) or when analysed at 6 and 12 months (adjusted MD at 12 months 2.05, 95% CI –1.72 to 5.81).

FIGURE 8.

Marginal adjusted mean SPADI pain values from the repeated measures mixed-effects model and associated 95% CIs for the two treatment groups from baseline to 12 months: injection vs. no injection.

SPADI: shoulder function subscale

There was an improvement in shoulder function in both groups over time (Figure 9). Injection resulted in an improvement in shoulder function when analysed at 8 weeks compared with no injection (adjusted mean SPADI function subscale difference between groups –3.84, 95% CI –6.95 to –0.73). This difference was not clinically significant. There was no statistically significant difference between injection and no injection when analysed over 12 months (adjusted MD over 12 months –0.59, 95% CI –3.02 to 1.83) or when analysed at 6 and 12 months (adjusted MD at 12 months 1.97, 95% CI –1.18 to 5.11).

FIGURE 9.

Marginal adjusted mean SPADI function values from the repeated measures mixed-effects model and associated 95% CIs for the two treatment groups from baseline to 12 months: injection vs. no injection.

Fear Avoidance Belief Questionnaire – Physical Activity

There was an improvement in fear avoidance behaviour in both groups over time. However, there was no statistically significant difference between injection and no injection in fear avoidance behaviour when analysed over 12 months (adjusted mean FABQ-PA score difference between groups over 12 months 0.02, 95% CI –0.74 to 0.77) or when analysed at 8 weeks, 6 months or 12 months (adjusted MD at 12 months 0.47, 95% CI –0.51 to 1.44).

Pain Self-Efficacy Questionnaire

There was no statistically significant difference between injection and no injection in patient-reported pain self-efficacy when analysed over 12 months (adjusted mean PSEQ-2 score difference between groups over 12 months –0.06, 95% CI –0.31 to 0.19) or when analysed at 8 weeks, 6 months or 12 months (adjusted MD at 12 months –0.30, 95% CI –0.63 to 0.03). A ceiling effect was noted for this outcome, with most participants reporting higher confidence levels despite experiencing pain.

Insomnia Severity Index

There was an improvement in patient-reported perception of insomnia in both groups over time. Injection resulted in an improvement in patient-reported perception of insomnia when analysed at 8 weeks compared with no injection (adjusted mean ISI difference between groups –1.50, 95% CI –2.32 to –0.68). There was no statistically significant difference between injection and no injection when analysed over 12 months (adjusted MD over 12 months –0.41, 95% CI –1.08 to 0.26) or when analysed at 6 and 12 months (adjusted MD at 12 months 0.32, 95% CI –0.51 to 1.15).

Return to desired activities

There was an improvement in patient-reported RDA in both groups over time. Injection resulted in an improvement in patient-reported RDA when analysed at 8 weeks compared with no injection (adjusted mean RDA difference between groups –0.53, 95% CI –0.89 to –0.17). There was no statistically significant difference between injection and no injection when analysed over 12 months (adjusted MD over 12 months –0.12, 95% CI –0.40 to 0.16) or when analysed at 6 and 12 months (adjusted MD at 12 months 0.21, 95% CI –0.15 to 0.57).

Global Impression of Treatment

Injection resulted in an improvement in patient-reported GIT success when analysed at 8 weeks compared with no injection (adjusted mean GIT difference between groups 0.69, 95% CI 0.35 to 1.03). There was no statistically significant difference between injection and no injection when analysed over 12 months (adjusted MD over 12 months 0.20, 95% CI –0.06 to 0.46) or when analysed at 6 and 12 months (adjusted MD at 12 months –0.14, 95% CI –0.49 to 0.20).

Participant-reported shoulder condition

As part of the follow-up questionnaires, a total of 23 (3%) participants reported ‘yes’ to the question ‘have you been told you will need to have surgery because of your shoulder problem’. Two, six and 15 participants reported ‘yes’ at 8 weeks, 6 months and 12 months, respectively, and the numbers were similar across intervention groups. Three participants reported that they had been admitted to hospital as an NHS inpatient because of their shoulder. The type of surgery recorded was surgery to repair rotator cuff tear (n = 1; injection plus progressive-exercise intervention), frozen shoulder surgery diagnosed after randomisation (n = 1; best-practice advice intervention), subacromial decompression and cuff repair (n = 1; injection plus best-practice advice intervention).

Participant-reported injection outside the trial

As part of the follow-up questionnaires, 16 (4%) participants in the injection group and four (1%) participants in the no injection group reported ‘yes’ to the question ‘have you had a steroid injection as a result of pain in your shoulder?’ at 8 weeks. Twelve (3%) participants in the injection group and 18 (5%) participants in the no injection group reported ‘yes’ to the question at 6 months, and 20 (6%) participants in the injection group and six (2%) participants in the no injection group reported ‘yes’ at 12 months. This excluded the injection(s) participants may have received as part of the GRASP trial.

Progressive exercise compared with best-practice advice

There were no statistically significant subgroup differences between the progressive-exercise intervention and best-practice advice intervention in shoulder pain and function when analysed over 12 months (Figure 10) or when analysed at 8 weeks, 6 months or 12 months (see Appendix 4, Figure 25).

FIGURE 10.

Subgroup-adjusted SPADI analysis for progressive exercise over 12 months.

Injection compared with no injection

There were no statistically significant subgroup differences for age, sex, smoking status and pain self-efficacy between injection and no injection interventions in shoulder pain and function when analysed over 12 months (Figure 11) or when analysed at 8 weeks, 6 months or 12 months (see Appendix 4, Figure 28). At 8 weeks, the effect of injection was stronger in participants with a higher baseline SPADI score (adjusted MD –9.67, 99% CI –15.37 to –3.97) than in those who received injection but had a lower baseline SPADI score (adjusted MD –0.36, 99% CI –6.87 to 6.16). This difference was clinically significant (see Appendix 4, Figure 28). There was no statistically significant subgroup difference when analysed over 12 months or when analysed at 6 and 12 months.

FIGURE 11.

Subgroup-adjusted SPADI analysis for injection over 12 months.

Aim

The aim of the GRASP trial’s economic evaluation was to address the following question:

What is the cost-effectiveness of individually tailored progressive exercise compared with best-practice advice, with or without corticosteroid injection, in people with a new episode of a rotator cuff disorder?

The within-trial economic analysis was performed using individual patient-level data from the GRASP trial. The analysis uses data from the GRASP trial only and did not combine this trial with any external evidence because, to the best of our knowledge, no study has evaluated exactly the same progressive-exercise intervention evaluated in this study. The analytical approach took the form of a cost–utility analysis. Based on trial evidence, net monetary benefit (NMB) statistics were calculated as QALYs multiplied by willingness-to-pay threshold minus cost to enable comparison between treatment groups.

The economic analysis compared the costs and consequences of each intervention group over the 12-month period following randomisation, with no extrapolation beyond the study period of 12 months, as prespecified in the health economics analysis plan, because there was no statistical difference in clinical outcome between treatment groups at 12 months. 85

Measurement of resource use and costs

Resource use data for the economic evaluation were collected during the trial period from questionnaires sent to participants (at 8 weeks, 6 months and 12 months after randomisation to the GRASP trial) and from treatment logs completed by the treating physiotherapists at sites. The questionnaire captured both NHS- and PSS-perspective resource use and costs borne by the participant and their family attributable to a rotator cuff disorder. This included the frequency of use of inpatient care, outpatient care and community-based health care (both private and NHS) that was not part of the GRASP trial. It also recorded direct medical costs that were not part of the trial (e.g. medications and steroid injections) and direct non-medical costs (e.g. help with housework/childcare and travel), the latter being excluded from the base-case economic evaluation. Free-text responses (applicable to all the ‘other’ options) were reclassified to the appropriate cost category, were removed if deemed unrelated/irrelevant to the trial by clinical experts (e.g. shoulder specialists) or were analysed collectively as ‘other’ in the descriptive analysis.

Some of the assumptions made when cleaning, analysing or costing the data included (1) if a patient answered ‘no’ to a prompt question about resource utilisation, then we assumed that the frequency of service use for this category of resources was equal to zero, (2) injections performed outside the GRASP trial were assumed to have been given by a GP within the duration of a typical GP visit and (3) for the self-reported questions on prescribed and over-the-counter medication, when participants failed to specify the duration or reported ‘as needed’, ‘when in pain’ or ‘occasionally’ and similar, we assumed an intake duration of 3 weeks based on clinical expert opinion.

Subacromial corticosteroid injection

Participants randomised to receive corticosteroid injection were given either methylprednisolone or triamcinolone acetonide (median dose of 40 mg) with a local anaesthetic. Injections were mainly given by extended-scope physiotherapists (bands 6–8a); in three cases injections were given by doctors (two orthopaedic consultants and one specialist registrar in orthopaedics). The consultation time varied from 20 minutes to 45 minutes based on the information provided by the physiotherapists. We based the cost of injections on the median injection administration time of 30 minutes, as a median is more robust against outliers and the weighted average cost per hour for each physiotherapist/clinician delivering injections. 89 Most participants also received local anaesthetic: either 1% lidocaine (up to 5 ml) or 0.5% bupivacaine hydrochloride (up to 10 ml) (see Appendix 5, Table 43 for unit costs of corticosteroid injection and anaesthetic). The total cost of administering the injection to each participant was calculated by adding the weighted mean administration cost per participant to the mean cost of corticosteroid injections and mean cost of anaesthetic per participant.

Best-practice advice

Participants randomised to best-practice advice received one 45- to 60-minute session with a physiotherapist (band 5–8a), when they were provided with a set of eight 2-week exercise diaries (i.e. a normal A4 printed sheet totalling 16 pages), a three-page document printed on non-carbon copy paper that served as an action planner, an information booklet, one piece of resistance band (with an average of 1 yard per participant) and either a DVD or details of how to access online exercise videos (Table 17). The cost of physiotherapists’ time was calculated by multiplying the median therapist cost per hour by the median estimate (in minutes) of the exercise session.

Progressive exercise

Participants randomised to progressive exercise received a median of four sessions with a physiotherapist (band 5–8a). The first session lasted between 45 and 60 minutes and the rest generally lasted between 20 and 30 minutes. Participants were given a set of eight 2-week exercise diaries (i.e. a normal A4 printed sheet totalling 16 pages), a three-page document printed on non-carbon copy paper that served as an action planner, an information booklet and at least one piece of resistance band if they were prescribed an exercise that required this (an average of 1 yard per participant) (see Table 17). Again, the cost of physiotherapists’ time was calculated by multiplying the median therapist cost per hour by the median estimate (in minutes) of the exercise session.

In both the best-practice advice and progressive-exercise interventions, we estimated costs based on median estimates, as they are more robust against outliers. Storage boxes given to study centres were excluded from the costings, as they were deemed to be protocol-driven resources.

Base-case analysis

The base-case analysis aimed to reflect the NHS cost of face-to-face training as it was delivered in the GRASP trial. It included the time cost for the physiotherapists attending the face-to-face training and the time cost for the physiotherapist delivering the training to the sites (Table 19). The cost of refresher training and the trainers’ travel costs to each site, venue hire and NHS parking charges were excluded from the analysis. In the base-case analysis, we calculated the total cost of training each physiotherapist and divided it by the mean number of patients treated per physiotherapist (among physiotherapists receiving this training) to estimate the cost of training as it was delivered in the trial. The total cost of training each physiotherapist was divided by the mean number of participants treated by each physiotherapist in the trial. We estimated the mean number of participants per physiotherapist separately for best-practice advice and progressive exercise, as physiotherapists giving the best-practice advice intervention could treat more people per week than those delivering progressive exercise. The mean numbers of participants treated per physiotherapist in the best-practice advice and the progressive-exercise groups were 3.9 and 3.6, respectively.

Worst-case scenario

This was the same as the base-case analysis, but also included the cost of refresher training, travel, venue hire and NHS parking charges as part of the training delivery cost. Refresher session cost was calculated by multiplying the 2.5-hour training time by the hourly cost of a grade 7 physiotherapist. Travel cost was calculated as the mileage from Oxford to the site and back by car multiplied by a flat cost per mile (£0.45 for the first 10,000 business miles). 90 Venue hire was calculated based on an average cost of hiring an NHS room of a maximum capacity of 30 people. 91 We also assumed an average of £2 per hour NHS parking charge for the duration of the training. Accommodation charges for overnight stay were not included.

Best-case scenario

The best-case scenario assumes that if the trial interventions were to be implemented in routine clinical practice, the training would be delivered on a free-to-access online platform. The cost of developing the online training materials were based on the team’s experience of developing training materials for the NIHR HTA-funded BeST (Back Skills Training) trial for the treatment of low back pain in primary care. 45 This includes 6 weeks of a full-time grade 8 physiotherapist’s time to develop the training materials for the progressive-exercise intervention and 1 month to develop the best-practice advice intervention. It also includes the cost of a grade 7 physiotherapist researcher to maintain it for a 10-year period (spending 30 hours/year). There are approximately 10,000 users per year using the online learning platform [URL: www.futurelearn.com/courses/back-skills-training-programme (accessed 24 May 2021)]. We, therefore, assumed that this researcher would support all 10,000 learners. We included the cost of the physiotherapists treating participants based on having a one-off training session of 3.5 hours for best-practice advice and of 4.5 hours for progressive exercise. We assumed that the training would last 10 years without a refresher session and that each physiotherapist would treat patients for 10 years after being trained. We assumed that each physiotherapist would treat approximately 100 patients per year, based on an assumption of 2.74 new patients per week over 1 year (9.6 patients per week and 3.5 patients per session on average).

The best- and worst-case scenarios have been presented as part of the sensitivity analysis.

We averaged and applied the cost of training for each intervention across all participants randomised, regardless of how many sessions they attended. Other intervention costs were estimated at the individual patient level, based on the recorded number of sessions that they attended. The estimated costs for different numbers of sessions attended of progressive exercise, best-practice advice, progressive exercise preceded by a corticosteroid injection and best-practice advice preceded by a corticosteroid injection are shown in Table 19. Following our base-case assumptions, the cost of best-practice advice is £93.45 and best-practice advice plus injection is £133.55. For a progressive-exercise patient attending all six sessions, the cost is £206.13, compared with £246.23 for a patient attending all recommended sessions in the progressive-exercise and injection treatment arm.

Valuation of resource use

Unit costs of direct medical care that is not part of the trial, such as inpatient care, outpatient care and NHS community care, were sourced from the latest available NHS Reference Costs92 (see Appendix 5, Table 44). The unit costs of medications related to rotator cuff disorders have been sourced using the latest available British National Formulary (BNF)93 (see Appendix 5, Table 43). Costs of medications for individual participants were estimated based on their reported doses and frequencies, when these were available, or based on an assumed daily dose using BNF recommendations. When a dose range was reported as ‘as required’ or when the quantities were not recorded, expert opinion was sought to make reasonable assumptions. The cost of NHS health-care resource use per participant was computed by multiplying the frequency of health resource utilisation reported by the participant by the unit cost of each resource item. In the case of non-NHS costs, participants self-reported the cost of additional expenses. All costs were expressed in 2018/19 Great British pounds. No discounting was applied, as the time horizon of the analysis did not exceed 12 months.

Calculation of utilities and quality-adjusted life-years

Participants’ questionnaires contained the EQ-5D-5L questionnaire for self-completion at baseline and at 8 weeks, 6 months and 12 months post randomisation. The EQ-5D-5L instrument94 facilitates the generation of a utility score from a person’s health-related quality of life while reducing the ceiling effect and being more sensitive than its three-level predecessor. 95 A utility score refers to the preference of the general population for any particular set of health outcomes. As per the NICE position statement, the responses to the EQ-5D-5L were converted into multiattribute utility scores using the approved ‘crosswalk’ to the three-level instrument and applying the mapping function developed by van Hout et al. ,96 and the converted responses were valued using the established time trade-off utility algorithm for the UK. 97 QALYs were calculated as the area under the utility curve of utility scores from baseline, 8-week, 6-month and 12-month data using the trapezoidal rule. 98 Deceased patients were assigned a utility of zero from the date of death. We assumed that utility remained constant between the last utility measurement and the date of death.

Missing data

Because within-trial health economic evaluations draw on many sources of information on patient characteristics, treatments, outcomes and resource use over the whole trial, incomplete data are a particular issue that require careful attention. Consequently, the base-case analysis imputed missing data using fully conditional multiple imputation under chained equations, using the Stata command ‘mi impute chained’. Within multiple imputation under chained equations, regression models were used to impute unobserved costs and utilities at each time point using the baseline covariates [i.e. age, sex and EuroQol-5 Dimensions (EQ-5D)] as predictor variables. The imputation model included a dummy variable for allocation to injection, a dummy variable for allocation to progressive exercise and an interaction term equal to the product of these two variables, following recommended practice for factorial trials. 88 Different components of costs and EQ-5D utility scores at each time point contributed as both predictors and imputed variables.

Multiple imputation was used to generate 25 data sets (or ‘draws’) using predictive mean matching, which provides plausible values when costs and utilities are non-normally distributed. In line with recommended practice,99 the imputation model was validated by comparing the distributions of the imputed data with the observed data. The imputation was run following the rule of thumb that the number of imputations (M) should be similar to the percentage of incomplete cases (in this case M = 25). 99

Cost-effectiveness analysis

The analysis was conducted based on the ITT principle and incremental cost-effectiveness ratios (ICERs) were calculated as the difference in mean costs divided by the difference in mean QALYs between a pair of interventions. The NICE100 cost-effectiveness threshold of £20,000 per additional QALY was used to identify which of the following treatments represent best value for money (i.e. has the highest NMB): (1) best-practice advice, (2) best-practice advice plus corticosteroid injection, (3) progressive exercise or (4) progressive exercise plus corticosteroid injection.

In addition to calculating and reporting ICERs and NMBs, the results are presented graphically in cost-effectiveness planes, and cost-effectiveness acceptability curves (CEACs) are used to show the probability that each of the four treatment groups has the highest NMB. Measures of uncertainty (SEs and CIs) are also reported around the mean costs and QALYs (95% CIs are presented around ICERs if they are defined). SEs and CEACs were generated using non-parametric bootstrapping with 1000 replicates, as described in the next section. This accommodates sampling (or stochastic) uncertainty and varying levels of willingness to pay for an additional QALY. We made no adjustment for clustering of participants by physiotherapist when analysing costs, QALYs or cost-effectiveness, as the randomisation was carried out on an individual basis, stratified by centre, rather than using cluster randomisation, and the subacromial corticosteroid injections and physiotherapy sessions were delivered in accordance with a standard protocol.

As the GRASP trial is a factorial trial, it was important to consider the interactions [i.e. to examine whether or not the differences in costs, QALYs or NMB between best-practice advice and progressive exercise were affected by the use of corticosteroid injection (or vice versa)]. In the GRASP trial clinical analysis, the primary outcome was analysed at the margins, assuming no interactions and that interaction terms were included in the model only if interactions were significant at the 0.05 level. From the economics point of view, this approach was not appropriate, as health economics findings are interpreted in terms of the absolute magnitude of ICERs, rather than focusing on hypothesis testing, and several mechanisms have been suggested that may introduce large but non-significant interactions for economic end point, but not clinical end point. 88 It is more important to avoid the bias that may result from ignoring interactions rather than maximising statistical power. We, therefore, compared specific treatment combinations (e.g. best-practice advice, injection only, progressive exercise only and injection plus progressive exercise) incrementally and identified the combination that represents best value for money, rather than making separate decisions on injection and exercise, as the former provides more relevant information for decision-makers if there is any interaction.

For the GRASP trial, the base-case economic analysis was prespecified as regression analysis with an interaction term, although regression analysis without interaction terms has been used as a sensitivity analysis to assess whether or not the assumptions about interactions change the conclusions of the analysis. Benefits of using regression analysis in the context of factorial design trial are that it allows for variation in sample size between groups, adjusts for the effect of the other intervention, facilitates adjustment for baseline utility and can predict the mean outcomes for each cell in the factorial design.

Regression analysis with interaction term (base-case analysis)

Linear regression analyses that predicted both costs and QALYs were calculated for each bootstrap sample on each imputed data set. Randomisation to corticosteroid and randomisation to exercise were included as dummy variables and the base-case analysis also included an interaction between these two variables. The ordinary least squares regression that predicted QALYs also controlled for baseline utility to avoid the bias that would otherwise arise from any imbalance in baseline utility between groups. 101 A total of 1000 bootstrap samples were drawn for each imputed data set. Mean costs, QALYs and the NMB in each of the four treatment groups were estimated based on the regression coefficients for each bootstrap of each imputed data set.

We combined uncertainty around missing data with sampling uncertainty using the MI Boot pooled sample approach of Schomaker and Heumann,102 which has been shown to yield valid inference and to be equivalent to nesting bootstraps within imputations and combining results using Rubin’s rule. Pooling bootstraps is simpler to implement than Rubin’s rule and facilitates presentation of CEACs. In addition, a simulation study has shown it to give good coverage with ≥ 20 imputations. To implement this in our data set, we averaged across the 25 imputed data sets for the original (non-bootstrapped) sample to get point estimates and estimated 95% CIs as the 2.5th and 97.5th percentiles across all 25,000 bootstraps, adapting the code used previously. 103 CEACs were estimated across all 25,000 bootstraps. The regression models used to predict cost and QALYs are presented in Appendix 5, Model 1: regression analysis model with interaction term (base-case analysis).

Regression analysis without interaction term (sensitivity analysis)

Regression techniques with an interaction term provide an alternative to at-the-margins analysis, which also assumes no interaction. 88 The description of the analysis and the model are presented in Appendix 5, Model 2: regression analysis without interaction term (sensitivity analysis).

Completion rate

Among the 708 participants randomised in the trial, 174 were randomised to best-practice advice, 178 were randomised to best-practice advice plus injection, 174 were randomised to progressive exercise and 182 were randomised to progressive exercise plus injection. The completion rates of all health resource items by treatment intervention for each time point are displayed in Table 20. In addition, Table 21 shows the response rate of EQ-5D-5L by follow-up points and treatment group.

Base-case results

Table 25 presents the costs and QALYs associated with the four interventions under investigation. This analysis evaluated the impact of exercise treatment and injection and interactions between these two interventions inside the table, while imputing missing values and adjusting for age, sex and baseline utility.

When all randomised patients were included in the analysis and missing values were imputed using multiple imputation, patients receiving best-practice advice accrued an average of 0.737 (95% CI 0.710 to 0.763) QALYs and an NHS cost of £195 over the 12-month period (see Table 25). In the base-case analysis, adding injection to best-practice advice gained 0.021 QALYs (p = 0.184) and increased the cost by £10 per participant (p = 0.747) compared with best-practice advice alone. Progressive exercise alone was £52 (p = 0.247) more expensive per participant than best-practice advice, while gaining 0.019 QALYs (p = 0.220). However, there was a non-significant interaction for cost (p = 0.397), which meant that when injection was added to progressive exercise it generated an additional cost of £60 per participant (i.e. £50 more than the difference between best-practice advice and injection only). There was also a non-significant qualitative interaction for QALYs (p = 0.106), whereby adding injection to best-practice advice increased QALYs, but adding injection to progressive exercise reduced QALYs; the groups receiving injection alone or progressive exercise alone accrued more QALYs than the group receiving best-practice advice, whereas the group that received both of these treatments had lower QALYs than either of the groups receiving only one treatment. Best-practice advice plus injection cost £475.59 more per QALY gained than best-practice advice alone. Progressive exercise alone and progressive exercise plus injection were both strongly dominated by best-practice advice plus injection, being both more costly and accruing fewer QALYs.

The 2013 NICE cost-effectiveness threshold of £20,000 per additional QALY100 was used to identify which treatments represent best value for money (i.e. has highest NMB). The interactions for cost and QALYs combine to give a qualitative, but non-significant interaction for NMB (p = 0.100). Best-practice advice plus injection had a 54.93% probability of being the most cost-effective treatment at a ceiling ratio of £20,000 per QALY (Figure 12). Progressive exercise alone had the second highest NMB and had a 35.64% probability of being cost-effective at a ceiling ratio of £20,000 per QALY.

FIGURE 12.

The CEAC for the comparison between treatment groups (base-case analysis). The frontier indicated which treatment is economically preferred at different threshold values for cost-effectiveness.

Aim and overview of trial findings

Shoulder pain in the UK is very common, with the most common attribution being the rotator cuff, which accounts for around 70% of new episodes of shoulder pain presenting in primary care. 2 The majority of shoulder pain is managed in primary care or at primary care interface services by physiotherapists and GPs. Current treatments aim to improve pain and function with standard care that includes rest, advice, analgesia, physiotherapist-prescribed exercise and corticosteroid injection; however, there are no NICE clinical guidelines for this area. 4 Prior to the GRASP trial, limited evidence existed regarding the most clinically effective and cost-effective form of physiotherapist-prescribed exercise and delivery mechanism associated with the best outcomes for people with a rotator cuff disorder. There was also uncertainty around the long-term benefits and harms associated with corticosteroid injection. In the GRASP trial, we aimed to assess the clinical effectiveness and cost-effectiveness of whether or not (1) an individually tailored progressive, home exercise programme prescribed and supervised by a physiotherapist provided greater improvement in shoulder pain and function over 12 months (measured using the SPADI score) compared with a best-practice advice session with a physiotherapist and provision of high-quality self-management materials, and (2) subacromial corticosteroid injection provided greater improvement in shoulder pain and function over 12 months compared with no injection.

For the primary outcome, there was no evidence of a difference in the SPADI scores over 12 months between participants randomised to receive the progressive-exercise intervention and those who received best-practice advice. Likewise, there was no evidence when analysed at the 8-week and 6-and 12-month time points. In both intervention groups, shoulder pain and function did improve over time, although SPADI scores at 12 months showed that the condition did not resolve completely, as most participants still reported some symptoms. There was also no difference between groups for secondary outcome measures, with the exception of progressive exercise, which resulted in an improvement in patient-reported GIT over 12 months and at the 6- and 12-month time points. There were no significant subgroup differences in shoulder pain and function across the different time points when assessed for age, sex, baseline smoking status, SPADI and pain self-efficacy.

Over 12 months, there was no evidence of a difference in SPADI scores between participants randomised to receive corticosteroid injection and those receiving no injection, nor when analysed at the 6- and 12-month time points. There was a small difference in SPADI scores at 8 weeks, in favour of injection, but this was just below the threshold for a clinically important difference in the SPADI. There were no differences between groups for secondary outcome measures, with the exception of corticosteroid injection at 8 weeks, which resulted in a small improvement in shoulder pain, shoulder function, sleep disturbance, RDA and GIT. Prespecified subgroup analysis showed that the effect of corticosteroid injection was stronger at 8 weeks in people with a higher baseline SPADI score (i.e. SPADI ≥ 50) than in those who received injection but had a lower baseline SPADI score. No differences were observed for other prespecified subgroup analyses.

At 8 weeks and 6 months, participants who received injections were more likely to report performing their exercises 5 days per week, in accordance with the advice from the treating physiotherapist, than participants who did not receive injections. This suggests that, although the effect of corticosteroid injection is short lived, it may facilitate engagement with prescribed home exercises. There were no SAEs recorded as part of the trial as a result of either corticosteroid injection or physiotherapy.

A cost–utility analysis investigated the impact of progressive exercise, injection and the interaction between progressive-exercise treatment and injection, accounting for missing values and adjusting for baseline imbalance. There were no statistically significant differences between the treatment groups in terms of costs or QALYs over the 12-month period. More specifically, the addition of injection to best-practice advice produced a negligible and non-statistically significant QALY gain, with a small extra cost. Progressive exercise was more expensive per participant than best-practice advice, but, again, with a negligible and non-statistically significant QALY gain. Progressive exercise alone and progressive exercise plus injection were both strongly dominated by best-practice advice plus injection, being more costly and accruing fewer QALYs. Among the treatments under investigation, the one expected to provide the best value for money was the best-practice advice session with a physiotherapist and injection, although there is substantial uncertainty around this conclusion. Although the benefits of corticosteroid injection were limited both in size and to the early phase of recovery, this combination of interventions dominated in terms of their cost-effectiveness.

Internal validity and methodology

The GRASP trial was a pragmatic multicentre superiority 2 × 2 factorial randomised controlled trial. Given the factorial design, we first formally tested for an interaction effect. In the absence of any significant interaction, we were able to assess the effects of our two main comparisons: (1) progressive-exercise programme compared with best-practice advice session and (2) subacromial corticosteroid injection compared with no injection. In accordance with the sample size estimation (see Chapter 4), data from 704 participants were required to detect a standardised effect size of 0.33 (equivalent to 8 points on the SPADI total score) with 90% power and 1% two-sided statistical significance, allowing for 20% loss to follow-up at 12 months and potential for a small clustering effect by physiotherapist. 36 The DMEC reviewed the sample size assumptions after 338 participants had been recruited and did not recommend any changes to the final sample size. We recruited a total of 708 participants and had a lower than estimated loss to follow-up rate of 13% at 12 months, and so the trial was adequately powered to detect a statistically and clinically important difference between interventions. Measurements for the primary and secondary outcomes were collected by postal questionnaires at 8 weeks, 6 months and 12 months. When postal questionnaires were returned to the GRASP trial office, data were checked and if missing data were identified, either for part or a whole outcome measure, then the participant was contacted and information collected by telephone. As a result, we had limited missing outcome data from the completed questionnaires.

Randomisation was generated using the centralised computer randomisation service provided by OCTRU. Randomisation was stratified by site, age and sex using variable block size, ensuring that participants were balanced across interventions groups and minimising the chances of research staff anticipating treatment allocation prior to randomisation. It was not possible to blind participants and treating physiotherapists because of the nature of the interventions being tested. Both the primary and secondary outcomes were patient reported and collected using postal questionnaires. There was potential for bias as participants were aware of treatment allocation, but this reflects the difficultly of achieving and maintaining blinding in pragmatic rehabilitation trials of this nature. 72 In the small number of cases where outcome data were collected by telephone (e.g. participants who did not respond to postal reminders), the researcher was blinded to the treatment allocated and participants were asked not to disclose which intervention they had received.

Physiotherapists were trained to deliver either the best-practice advice intervention or the progressive-exercise intervention to minimise possible contamination between treatment groups. Only two physiotherapists swapped treatment groups during the trial, because of staffing issues at sites, and so ended up delivering both interventions. According to the results of the GIT questionnaire, the interventions were generally well received by the participants. Interventions were also delivered with high levels of fidelity by the physiotherapists, as determined by quality assurance visits conducted at site and review of treatment logs. Attendance rates were 94% for injection, 95% for progressive exercise (either partially or full completed) and 92% for best-practice advice. These attendance rates are above those normally expected for NHS outpatient physiotherapy, with non-attendance rates estimated to be 10.4% according to recent NHS Hospital Episode Statistics data. 41 There was no difference in attendance rates between those receiving the physiotherapist-delivered exercise interventions and those who received physiotherapist-delivered exercise interventions in conjunction with a corticosteroid injection.

Despite some initial concerns from physiotherapists raised during site training regarding the adequacy of a single contact with a physiotherapist to start a self-guided exercise programme, very few participants in the best-practice advice intervention required an additional contact session during the trial. Most participants in the progressive-exercise intervention group attended a median of four sessions (out of a maximum of six sessions) before being discharged. Progression of exercises, defined as an increase in exercise difficulty or volume and/or load, was regarded as key to achieving the overload and subsequent physiological response in the neuromuscular system to improve muscle function. Treatment logs provided evidence that 67% of participants in the progressive-exercise group had exercise progression between their first and last session, 22% maintained the initial exercise dose during treatment and only 12% had to regress their exercise dose over the sessions. The maintenance or regression of the level of exercise over the sessions was consistent with the tailoring and modification of the exercise programme allowed in the intervention protocol in the trial. The intervention was designed to enable therapists to adapt the programme according to the participant’s response to exercising the shoulder.

A small proportion of participants (3.8%) withdrew from the GRASP trial. This was either withdrawal from the intervention only or withdrawal from both the intervention and future follow-up assessment. The most common reason was a change in diagnosis since randomisation, and not dissatisfaction with treatment. Numbers were balanced across intervention groups. Participants were advised that they may seek other forms of treatment/medication outside the GRASP trial. This information was collected as part of the participant follow-up questionnaires. There were no significant differences in medication type or other health-care resources usage across the intervention groups, suggesting that the effects seen in terms of any improvement in shoulder pain and function were due to the trial interventions.

Exercise intervention

Findings from a Cochrane review published in 2016, before the start of the GRASP trial, highlighted the lack of evidence about the long-term clinical effectiveness and cost-effectiveness of physiotherapy for the treatment of rotator cuff disorders, despite its widespread provision. 14 Evidence from several small trials with short-term follow-up also raised uncertainty about which types of exercise and delivery mechanisms were associated with best outcomes. 15,16

We searched MEDLINE, EMBASE and Cumulative Index to Nursing and Allied Health Literature (CINHAL) to identify new evidence relevant to the GRASP trial (date of last search: June 2020). From an initial 2354 records, we identified seven trials67,110–115 published between 2013 and 2020, comparing the effects of supervised exercise with unsupervised exercise, or no intervention, in people with a rotator cuff disorder (excluding those who required surgery). Most of the trials concluded that there was little or no difference between supervised and unsupervised exercise. The populations were comparable to the GRASP trial, with the exception of the trial by Krischak et al. ,110 which evaluated people with atraumatic full-thickness rotator cuff tears. All trials were small, with the exception of the trial by Contreras et al. 113 (n = 271) and the SUPPORT trial115 [which had 64 participants in each group (2 × 2 factorial trial) and 256 participants in total]. Only two67,115 of the seven trials reported on the effect of exercise on shoulder pain and function at 12 months, two reported medium-term follow-up data (24 or 26 weeks111,113) and the remaining three trials reported outcomes at ≤ 6 weeks (see Appendix 6, Table 58). This reinforces the importance of the GRASP trial findings in terms of their definitive nature and length of follow-up.

Corticosteroid injection

At the time of planning the GRASP trial, there was systematic review evidence that, in comparison with placebo, corticosteroid injections had short-term benefit for treating tendinopathy, although there was some uncertainty regarding its use for rotator cuff disorders. There were also concerns about the longer-term safety of corticosteroid injection and its effect on the tendon. 30

We searched MEDLINE, EMBASE, the Allied and Complementary Medicine Database (AMED), CINHAL, the Cochrane Central Register of Controlled Trials (CENTRAL), ClinicalTrials.gov and the World Health Organization International Clinical Trials Registry to identify new evidence relevant to the GRASP trial (date of last search: June 2020). 91 From an initial 794 records, we identified one small trial116 that compared the effects of corticosteroid injection with no injection in people with a rotator cuff tear and found no difference in shoulder pain and function when analysed at 3 or 6 months, measured using the Constant–Murley score (MD at 6 months –1.9, 95% CI –10.89 to 7.09; patients, n = 40).

We identified an additional 10 trials7,90,117–124 that compared the effects of corticosteroid injection with placebo injection (see Appendix 6, Table 59), of which four7,118,121,122 were judged as suitable for inclusion in a meta-analysis. The remaining studies could not be included because of either incomplete or incompatible outcome data. Three trials,118,121,122 with a total of 215 patients, compared subacromial corticosteroid injection with placebo injection for combined shoulder pain and function in the short term (i.e. ≤ 8 weeks), with results favouring corticosteroid over placebo [standardised mean difference (SMD) –0.51, 95% CI –1.02 to 0.00; I² = 69%; n = 3; patients, n = 215; rated as having moderate-quality evidence]. Two of the trials118,122 also reported outcome data for medium-term follow-up (i.e. 3–6 months), in which no difference was apparent (SMD 0.08, 95% CI –0.24 to 0.39; I² = 0%; n = 2; patients, n = 151; rated as having moderate-quality evidence). With regard to shoulder pain only, three trials reported short-term outcomes118,121,122 and two reported medium-term outcomes. 118,122 The results mirrored those seen for combined shoulder pain and function, which favoured corticosteroid injection over placebo in the short term (SMD –0.35, 95% CI –0.65 to –0.05; I² = 17%; n = 3; patients, n = 215; rated as having moderate quality evidence), but not the medium term (SMD 0.05, 95% CI –0.27 to 0.37; I² = 0%; n = 2; patients, n = 151; rated as having moderate-quality evidence). Two trials7,122 reported on shoulder functional improvement in the short and medium term and, again, there was a short-term benefit favouring corticosteroid injection (SMD –0.33, 95% CI –0.67 to 0.00; I² = 0%; n = 2; patients, n = 143; rated as having moderate-quality evidence), but there was no difference seen in medium-term functional outcomes (SMD –0.24, 95% CI –0.59 to 0.10; I² = 0%; n = 2; patients, n = 131; rated as having moderate-quality evidence). No trials provided outcome data beyond 6-month follow-up and none reported any SAEs as a result of injection.

These findings reinforce the importance of the results of the GRASP trial in terms of the short-term benefit of subacromial corticosteroid injection, a benefit which is not maintained at 6- and 12-month follow-up.

University Hospitals of Derby and Burton NHS Foundation Trust

Marcus Bateman (principal investigator).

East Lancashire Hospitals NHS Trust

Alison Hallett (principal investigator) and Helen Thompson.

Gloucestershire Hospitals NHS Foundation Trust

Elaine Willmore (principal investigator).

Birmingham Community Healthcare NHS Trust

Lucy McCann (principal investigator) and Jonathan Price.

Sandwell and West Birmingham Hospitals NHS Trust

Neil Smith (principal investigator).

Buckinghamshire Musculoskeletal Integrated Care Service

Harry Kardamilas (principal investigator).

East Cheshire Hospitals NHS Trust

Matt Hurst (principal investigator).

Bedfordshire Hospitals NHS Foundation Trust

Tim Andrews (principal investigator) and Lori Wells.

Wirral University Teaching Hospital NHS Foundation Trust

Chole De Matas (principal investigator).

Medway Community Healthcare

Arun Jaykumar (principal investigator).

Bristol Community Health

Sean Grove (principal investigator).

Somerset Partnership NHS Foundation Trust

Corinne Birch (principal investigator).

Doncaster and Bassetlaw Teaching Hospitals NHS Foundation Trust

Julie Bury (principal investigator).

Sherwood Forest Hospitals NHS Foundation NHS Foundation Trust

James Blacknall (principal investigator).

Kent Community Health NHS Foundation Trust

Sally Jessep (principal investigator) and Llewelyn Boucher.

Northern Devon Healthcare NHS Trust

Robert Sandbach (principal investigator).

Airedale NHS Foundation Trust

Stacey Lalande (principal investigator).

North West Boroughs Healthcare NHS Foundation Trust

Gill Dickson (principal investigator).

Staffordshire and Stoke on Trent Partnership NHS Foundation Trust

Treena Larkin (principal investigator).

Warrington and Halton Hospitals NHS Foundation Trust

Carole Cummings (principal investigator).