Surgical treatments compared with early structured physiotherapy in secondary care for adults with primary frozen shoulder: the UK FROST three-arm RCT

Article history

The research reported in this issue of the journal was funded by the HTA programme as project number 13/26/01. The contractual start date was in October 2014. The draft report began editorial review in October 2019 and was accepted for publication in June 2020. The authors have been wholly responsible for all data collection, analysis and interpretation, and for writing up their work. This report has been published following a shortened production process and, therefore, did not undergo the usual number of proof stages and opportunities for correction. The HTA editors and publisher have tried to ensure the accuracy of the authors’ report and would like to thank the reviewers for their constructive comments on the draft document. However, they do not accept liability for damages or losses arising from material published in this report.

Copyright statement

© Queen’s Printer and Controller of HMSO 2020. This work was produced by Brealey et al. under the terms of a commissioning contract issued by the Secretary of State for Health and Social Care. This issue may be freely reproduced for the purposes of private research and study and extracts (or indeed, the full report) may be included in professional journals provided that suitable acknowledgement is made and the reproduction is not associated with any form of advertising. Applications for commercial reproduction should be addressed to: NIHR Journals Library, National Institute for Health Research, Evaluation, Trials and Studies Coordinating Centre, Alpha House, University of Southampton Science Park, Southampton SO16 7NS, UK.

2020 Queen’s Printer and Controller of HMSO

Frozen shoulder

Frozen shoulder (also known as adhesive capsulitis) occurs when the capsule, or soft tissue envelope, around the ball-and-socket shoulder joint becomes inflamed and then scarred and contracted. This makes the shoulder very painful, tight and stiff. It starts with pain, which increases in intensity as stiffness develops. 1 The exact cause of this condition is unknown. Reported associations include diabetes mellitus, cardiovascular disease, trauma, stroke, neurosurgery and thyroid disease. 1 In the absence of a known association, the condition is labelled by clinicians as ‘idiopathic’ or ‘primary’ frozen shoulder. The pathology of the capsule involves chronic inflammation, and proliferative fibrosis has been reported. 2 Myofibroblasts contribute to matrix deposition and fibrosis, with the underlying pathology considered similar to Dupuytren’s contracture. 2,3 The macroscopic appearance of these changes can be seen in the shoulder during arthroscopic visualisation of the rotator interval capsule. People with this condition may struggle with basic daily activities, suffer serious anxiety and have sleep disturbance due to shoulder pain. There is a tendency for spontaneous resolution, but recovery may be slow or incomplete. Even after an average of ≥ 4 years from onset, around 40% of patients can have mild to severe symptoms. 4 Figure 1 illustrates the pathology of frozen shoulder.

FIGURE 1.

Diagram showing site of pathology of frozen shoulder. This figure has been re-used from www.local-physio.co.uk/articles/shoulder-pain/frozen-shoulder/ with permission from the copyright holders.

This figure has been re-used from www.local-physio.co.uk/articles/shoulder-pain/frozen-shoulder/ with permission from the copyright holders.

Three clinical phases have historically been recognised for this condition,5 where the duration of each phase is indicative but varies considerably between patients:

1. painful phase, which may last 3–9 months

2. adhesive phase, with stiffness lasting for 4–6 months

3. phase of resolution or ‘thawing’, lasting for 5–24 months.

These phases have considerable overlap, and therefore the current favoured terminology is ‘pain predominant’ and ‘stiffness predominant’ phases. 6

The cumulative incidence of frozen shoulder is estimated at 2.4 per 1000 population per year, based on a Dutch general practice sample. 7 Frozen shoulder most commonly affects individuals in their sixth decade of life, and a large primary care-based study in the UK found that frozen shoulder affects 8.2% of men and 10.1% of women of working age. 8 By contrast, an incidence of 1% was reported by a UK shoulder surgeon in his specialist hospital practice. 9 This discrepancy in estimated incidence can be explained by the use of different populations as the denominators in the different studies. 6 Although not clearly established, when associated with diabetes mellitus, frozen shoulder is considered to be more resistant to treatment. 10

Diagnosis of frozen shoulder

A diagnosis of frozen shoulder is based on clinical criteria that include history of insidious onset deep-seated pain in the shoulder and upper arm with increasing stiffness, as well as clinical findings of limited active and passive external rotation in the absence of crepitus. 11 Radiographs are reported as not routinely required,6 but are usually performed in secondary care to exclude pathology like glenohumeral arthritis or posterior glenohumeral dislocation that could manifest with similar clinical signs. There is no reference standard for comparison, which explains the lack of diagnostic test accuracy data. 11 The key examination findings were originally described by Codman as restriction of elevation and external rotation. 12 As visual estimation of external rotation has fair to good reliability,13 restrictions (typically with pain) in both passive and active external rotation have been used as diagnostic criteria in clinical studies. 11,14–17 It can be difficult, however, to correctly diagnose the problem, as highlighted in a qualitative study of patients’ perceptions and priorities when living with primary frozen shoulder. 18 This accords with other studies, which have found that general practitioners (GPs) in the UK and the USA lack confidence in making shoulder diagnoses. 19,20

Treatments for frozen shoulder

The aims of treating a patient with frozen shoulder are to provide advice, education and reassurance; achieve pain relief; improve shoulder mobility; reduce the duration of symptoms; and facilitate return to normal activities. 21 Generally, less invasive treatments are provided in a primary care setting in the UK to those in the earlier phases of the disease, particularly for controlling pain. These may include oral analgesia, physiotherapy, acupuncture, and glucocorticoid (steroid) injection. 21 The treatments utilised in secondary care, when stiffness has become more established, were confirmed in a UK survey of health professionals conducted in 2009 as physiotherapy, manipulation under anaesthesia (MUA) and arthroscopic capsular release (ACR). 22

Physiotherapy treatment includes combinations of advice, exercises, therapist-applied mobilisation techniques, and thermo- and electrotherapies. The modalities of treatment recommended for use are described in the UK national physiotherapy guidelines for frozen shoulder,6 which are based on a systematic review. These are provided either in isolation or as a supplement to other interventions, such as intra-articular corticosteroid injection or surgical interventions (MUA or ACR). Intra-articular corticosteroid injection helps reduce inflammation of the joint capsule and reduce pain, which may facilitate the performance of exercises and hence enhance the effects of physiotherapy. Intra-articular corticosteroid injection has been shown to provide short-term benefits, with better improvements in pain, function and range of movement (up to 6–7 weeks for all three improvements) than placebo13 and probably than isolated manual therapy and exercise. 6

Manipulation under anaesthesia is a procedure the surgeon undertakes when the patient is under general anaesthesia. The affected shoulder joint is manipulated in a controlled fashion to stretch and tear the tight shoulder capsule. The joint is often injected with corticosteroid as part of this procedure. MUA is thought to facilitate recovery by releasing the tightness in the capsule, with the injection helping control capsular inflammation and pain. This is followed by physiotherapy for mobilisation of the arm and shoulder to restore mobility and function.

Arthroscopic capsular release is a ‘keyhole’ surgical procedure performed under general anaesthesia. The keyholes are used to view the joint and divide (release) the contracted capsule using typically arthroscopic radiofrequency ablation. This is thought to allow more accurate and controlled release of the tight capsule. The procedure is completed by performing MUA to complete and confirm full release of the contracted capsule. ACR is also followed by physiotherapy of the arm and shoulder to restore mobility and function.

Rationale for the UK FROzen Shoulder Trial

It is unknown whether a combination of physiotherapy and steroid injection or either of the surgical interventions (MUA or ACR) followed by physiotherapy is more effective. 13 Similarly, there is uncertainty about the benefits of MUA compared with other treatment options,23,24 and only limited evidence on ACR is available from randomised controlled trials (RCTs). 13,25

Systematic reviews have identified large gaps in the evidence base and uncertainty about the effectiveness of treatments for frozen shoulder and, therefore, a need for high-quality primary research. 26 In a systematic review13 commissioned by the National Institute for Health Research Health Technology Assessment (HTA) programme, 28 RCTs, one quasi-experimental study and two case series were included. The review found insufficient studies with a similar intervention and comparator to quantify effectiveness. Most studies were rated as having a high risk of bias, did not report adequate methods of randomisation, allocation concealment and outcome assessment, and seemed to be inadequately powered. Few studies reported collecting data on harms.

In view of the paucity of high-quality evidence to guide current practice, considerable uncertainties remain about the management of frozen shoulder. With the intention of facilitating quicker recovery, more invasive surgical interventions (MUA and ACR) are being used in spite of the lack of good evidence. 13 There is a clear need for a well-designed, high-quality RCT to determine the clinical effectiveness and cost-effectiveness of commonly used interventions for the treatment of frozen shoulder.

The findings of a national survey22 of health-care professionals in the UK, conducted in 2009, were used to determine the most commonly used interventions that need to be tested in a RCT in a secondary care setting. Physiotherapy, MUA and ACR were those that health-care professionals recommended be compared in a RCT. Only 6% of respondents at the time suggested using hydrodilatation as a comparator in a trial, which did not make this a feasible intervention to test in a RCT. This survey informed our decision to compare early structured physiotherapy (ESP) combined with intra-articular steroid injection with the two most frequently used, invasive and costly surgical interventions, namely MUA and ACR. 22 It is important to emphasise that, although physiotherapy is a common treatment in NHS practice, the ESP intervention was a specifically designed and standardised physiotherapy pathway to test the optimal delivery of physiotherapy in the NHS. As evidence about patients’ experiences of frozen shoulder is also limited,18 participants were interviewed about their experience and the acceptability of treatment, as were health professionals (physiotherapists and surgeons).

Aim and objectives

The strategic aim of UK FROST (UK FROzen Shoulder Trial), underpinned by the key treatment uncertainties, was to provide evidence of the clinical effectiveness and cost-effectiveness of three common interventions currently provided in the UK NHS for the treatment of frozen shoulder in a hospital setting. The following objectives were defined to achieve this overarching aim:

1. The primary objective was to determine the effectiveness of ESP compared with MUA compared with ACR for patients referred to secondary care for the treatment of frozen shoulder. This was achieved using a parallel-group RCT, with the Oxford Shoulder Score (OSS) (a patient-reported outcome measure) as the primary outcome at 3, 6 and 12 months. The primary time point was 12 months after randomisation.

2. To compare the cost-effectiveness of the three interventions, to identify the most efficient provision of future care, and to describe the resource impact that the various interventions for frozen shoulder would have on the NHS.

3. To qualitatively explore the acceptability of different interventions for frozen shoulder to patients and health-care professionals and to provide important patient-centred insight to further guide clinical decision-making.

4. To update the HTA programme-funded systematic review examining the management of frozen shoulder by assessing current RCT evidence for the effectiveness of interventions used in secondary care. This would allow the trial findings to be considered in the context of the existing evidence for the interventions under evaluation.

5. To widely disseminate the findings of this study to all stakeholders through networks of health-care professionals, patients, health service managers and commissioning groups. This would be in addition to publishing the results of the study in key journals and publishing the National Institute for Health Research HTA report.

Trial design

This was a pragmatic, multicentre, stratified (diabetes present or not) superiority trial comparing three parallel groups (MUA vs. ACR vs. ESP, with unequal allocation of 2 : 2 : 1) in adult patients referred to secondary care in England, Wales and Scotland for the treatment of primary frozen shoulder, and for whom surgery was being considered.

Participants

Patients with primary frozen shoulder were identified through clinical examination and plain radiographs. 28 To minimise diagnostic uncertainty, clinical examination included the key diagnostic assessment of restriction of passive external rotation in the affected shoulder,29 for which there is evidence of good inter-rater agreement on whether or not restriction is present30 and a high threshold (50% restriction of movement) for inclusion. Plain radiographs (anteroposterior and axillary projections) were obtained routinely for all patients to see whether or not these were normal and could allow the exclusion of glenohumeral arthritis and other pathology that could lead to similar clinical presentation (e.g. locked posterior dislocation).

Setting

The trial recruited from the orthopaedic departments of 35 NHS hospitals in the UK across a range of urban and rural areas. This comprised 28 hospitals in England, six in Scotland and one in Wales. Two additional hospitals in England screened for patients but did not recruit into the trial. Recruitment started in April 2015 and the final follow-up was in December 2018. All 37 participating hospitals are listed in Appendix 1.

Interventions

The components and standardisation of the surgical trial interventions were informed by a survey of 53 surgeons, who were the principal investigators of two multicentre shoulder surgical RCTs. 31,32 The standalone physiotherapy and post-procedural physiotherapy programmes were developed using evidence from a systematic review,13 UK guidelines,6 previous surveys of UK physiotherapists33,34 and consensus of expert shoulder physiotherapists in secondary care derived from a Delphi survey specific to UK FROST. 18 Ethics approval for the last of these was obtained from the School of Health and Social Care Research Governance and Ethics Committee of Teesside University on 23 May 2014 (Research Ethics Committee reference 069/14). The physiotherapy programmes developed are available online. 35 It is important to emphasise that, although physiotherapy is a common treatment in NHS practice, the ESP intervention was a specifically designed, standardised, new physiotherapy pathway to test the optimal delivery of physiotherapy in the NHS based on the best available evidence and expert consensus.

Participants assigned to either of the two surgical procedures were placed on the surgical waiting list and underwent routine preoperative screening. In keeping with NHS waiting time targets, both surgical procedures were expected to be performed within 18 weeks of randomisation. Participants would undergo these procedures under general anaesthetic and were expected to be admitted as day cases.

Physiotherapy was delivered by qualified physiotherapists (i.e. not students or assistants), and participating surgeons were familiar with the surgical procedure(s). There was no minimum number of surgical procedures that the surgeon had to have performed, and no grades of surgeon were excluded. Which surgeon operated on participants and whether or not the individual surgeon needed to be supervised by a consultant was at the discretion of the participating site, and followed normal care pathways and practices. The experience of physiotherapists and surgeons delivering the trial treatments was quantified and recorded in terms of their salary bands and the number of frozen shoulder patients they treated in a typical month.

Outcomes

Sample size

The primary trial outcome was the OSS, and this was assessed for three treatment comparisons: ESP with MUA, ESP with ACR, and MUA with ACR.

Data suggest that a 5-point improvement can be found on the OSS (standard effect size of 0.42) between surgically and non-surgically treated patients,59 with a stable standard deviation (SD) of 12 points across different populations. This larger effect size was required to justify the higher costs and potential risks associated with surgery when comparing ESP with MUA, and ESP with ACR. 42 A smaller difference of 4 points on the OSS (standard effect size of 0.33) was expected to distinguish between MUA and ACR.

To observe the above effect sizes with 90% power and 5% two-sided significance, adjusting for a moderate estimate (r = 0.4) of the correlation between OSS over 12 months and allowing for 20% attrition, a total sample size of 500 patients was required (MUA, n = 200; ACR, n = 200; ESP, n = 100). The sample size calculation was not adjusted for multiple comparisons owing to the a priori specified sequence of treatment comparisons and the analysis of the primary outcome in a single analysis model. 60

There were no planned interim analyses for the trial or stopping guidelines. An internal pilot from which data contributed to the final analyses was performed to confirm the feasibility of the trial, and this is explained in the following section.

Internal pilot study

There were two phases of the internal pilot study.

Recruitment

Initial estimates for recruitment were based on Hospital Episode Statistics for NHS hospitals in England in 2009/10 and 2010/11. These excluded post-trauma or secondary referrals from other specialties, giving a stable rate of 210 per million patients treated for frozen shoulder. Assuming that 50% of frozen shoulder patients presenting in secondary care met the inclusion criteria and 40% of these consented, this left around 40 patients per million to be recruited into the trial. It was estimated that, to recruit 500 trial participants from trusts each serving catchment areas of around half a million people, 25 hospitals would be required to recruit for a minimum of 1 year. This assumed that there would be no delays in set-up or problems at any subsequent time point, that all surgeons at the sites would be willing to participate and that all potential participants would be screened for eligibility. Following the pilot phase, the number of sites required was increased to ensure that recruitment was achieved to target.

Patients who had been referred for a frozen shoulder to an outpatient hospital clinic were identified by the research nurse or assessing clinician. In the clinic, a designated individual within the shoulder team (e.g. surgeon or physiotherapist) completed the study eligibility form (see Report Supplementary Material 8) to confirm whether or not the patient was eligible and, when applicable, approached the patient about the study. The research nurse then provided the patient with an information sheet (see Report Supplementary Material 9) and answered any questions. The patient was able to consent at that time or they could take up to 1 week to decide. When a patient consented (see Report Supplementary Material 10), he or she was asked to complete the baseline form (see Report Supplementary Material 11). The research nurse completed the consent status form (see Report Supplementary Material 12) to confirm the patient’s status.

When a patient did not consent, the research nurse recorded the reason and the treatment plan in another section of the consent status form. The patient was also offered an optional patient preference form (see Report Supplementary Material 13) to complete if they wanted to provide more information about why they chose not to take part.

Training in recruitment was provided to hospital staff as part of the site initiation visit, and a trial site manual was prepared that included guidance on consenting patients into the trial and how to answer questions that might arise during consent-taking. In addition, a poster was provided to publicise the trial to hospital staff and patients. During the trial, training and reminders were implemented using regular e-mail bulletins and face-to-face meetings with principal investigators and research nurses, with trial co-ordinators providing support and guidance to staff as required.

Randomisation

The randomisation sequence was based on a computer-generated randomisation algorithm provided by a remote randomisation service (telephone or online access) at York Trials Unit, University of York. The unit of randomisation was the individual patient, allocated to the trial interventions MUA, ACR and ESP in the ratio of 2 : 2 : 1, stratified by the presence of diabetes,62 using random blocks sizes of 10 and 15. The research nurse used the remote randomisation service to register eligible and consenting patients before computer generation of the allocation. This ensured treatment concealment and immediate unbiased allocation.

The research nurse then informed the treating clinician and the patient of the treatment allocation.

Blinding

Given the nature of the trial treatments, comparing surgical and non-surgical treatment options, the blinding of participants and clinicians to treatment allocation was not possible or desirable in this pragmatic trial. Therefore, patients and clinicians were informed of the treatment allocation after randomisation. The statistician was blind to treatment allocation until after data were hard locked and no further changes could be made.

Statistical methods

Analyses were conducted for the three treatment comparisons of interest, ACR with ESP, MUA with ESP, and ACR with MUA, according to the principle of intention to treat (ITT). All analyses were conducted in Stata^® version 15 (StataCorp LP, College Station, TX, USA) using two-sided statistical significance at the 0.05 level. The statistical analyses plan was completed prior to completion of data collection on 12 February 2019.

Data management

A central database at York Trials Unit was used to manage data collection, including the sending and return of participant questionnaires (see Report Supplementary Material 16) and hospital CRFs. This included automated e-mail reminders to participating sites to help ensure the timely return of hospital CRFs. Participant questionnaires and hospital CRFs were designed using TeleForm software (version 10; Cardiff Software, Cambridge, UK) and marked up with variable names and appropriate scoring. To maximise data quality, when hospital CRFs were returned to York Trials Unit the key variables required for the statistical analysis and checking adherence in the delivery of the treatments were reviewed for completion and accuracy by a research data administrator, who resolved any queries with the research nurse at the site. The hospital site was reimbursed for the completion of all CRFs up to a maximum value of £124.00. This was agreed by the trust and trial sponsor using a clinical trial agreement during the site set-up. No checks of the quality of data in the postal questionnaires were made on return of the questionnaires to York Trials Unit, although a trial co-ordinator checked whether the participant had given extreme responses to the last EQ-5D-5L question and/or given a free-text response that indicated they could be at harm. When either of these occurred, the principal investigator, research nurse and chief investigator were notified by e-mail. After this initial check, all postal questionnaires and hospital CRFs passed through a process of scanning in the TeleForm software, second checking and validating against predetermined rules.

Active and systematic follow-up of all randomised participants by post included pre-notification reminders, 2- and 4-week letter reminders and the option to complete an abridged questionnaire (a minimum of the OSS and EQ-5D) over the telephone after 6 weeks. At 12 months, the primary end point, an unconditional incentive of £5 was included. If the patient agreed at the time of consent, text messages were sent on the day the participant was sent the postal questionnaire64 and newsletters were circulated to trial participants. 65 Trial participants could withdraw entirely from the study at any time for any reason, but any data collected up to that point were included in the analysis. The participant could agree to being withdrawn from only postal questionnaire collection or from only hospital CRF collection.

Essential trial documentation were kept with the trial master file and investigator site files, allowing the conduct of the trial and quality of the data produced to be evaluated. The documentation will be retained for a minimum of 5 years after the conclusion of the trial. The postal questionnaires and hospital CRFs will be stored for a minimum of 5 years after the conclusion of the trial as paper records, and for a minimum of 20 years in electronic format.

Adverse event management

All AEs and SAEs were recorded by the site principal investigator or delegated clinician and returned to the trial office on a CRF (see Report Supplementary Materials 17 and 18). In accordance with good clinical practice, SAEs were reported within 24 hours and AEs were reported within 5 days of the investigator becoming aware.

Once this information was received, the chief investigator determined causality and expectedness. The Research Ethics Committee was notified of SAEs that were unexpected and related to the trial within 15 days for a non-life-threatening event and within 7 days for a life-threatening event. For non-serious AEs, the central office was notified within 5 days of the event being known. All AEs and SAEs were reported to the DMEC, TSC and TMG. Expected AEs for this shoulder condition included infection; bleeding; delayed wound healing; conversion of a planned day-case procedure to an overnight stay for control of pain; post-procedural worsening of shoulder pain; injury to adjacent structures such as nerve, tendon, bone or joint; recurrent stiffness requiring further treatment; transient hyperglycaemia, steroid flare or joint sepsis following corticosteroid injection; and injuries related to the heating or cooling of tissues. The chief investigator reviewed follow-up reports 1 month later (see Report Supplementary Material 19) to ensure that adequate action had been taken and progress had been made.

Ethics approval and monitoring

Patient and public involvement

Two patients who had previously received treatment for frozen shoulder at the lead site (James Cook University Hospital) and the independent patient representative member of the TSC were invited to comment on the patient information leaflet, the patient-facing data collection forms and the consent process for trial participation. The need to develop a leaflet to provide general information about frozen shoulder was identified following a qualitative study of patients with frozen shoulder using semistructured interviews. 18 The two patient representatives were invited to attend the TMG during the early stages of the study and it was later agreed to seek their opinion outside the meetings when necessary. Recruitment was steady and the target was met on time. The retention of participants also went well and the target was exceeded. Therefore, there was little further contact with the two patient representatives during the trial, although they did advise on the newsletters to be sent to trial participants.

Following the initial analyses of the study results, we sought the advice of the two patient representatives and a wider group of seven patients with frozen shoulder at the lead site. Study results, associated risks for individual trial treatments and their health economic impact were discussed. Members shared their thoughts on their preferred choice of treatment based on the study results and agreed to support the trial team with dissemination to various platforms. This included contributing to the Plain English summary of this report, journal publications and web-based outputs, such as updating the entry about management of frozen shoulder on Wikipedia and helping to develop content for other appropriate web pages. These patients will also meet with local (shoulder research users group) and national shoulder patient groups (British Elbow and Soulder Society patient liaison group) to ensure that the current evidence base for treatment options is available and disseminated appropriately to patients and the wider public.

Summary findings of the internal pilot

The objective of phase 1 of the internal pilot (months 4–9) was to have a minimum of four sites recruiting during the 6 months that had recruited 24 patients (i.e. evidence that they could recruit the expected one participant per month). To ensure that adequate progress had been made with setting up sites to recruit, 12 sites were to be set up (i.e. 50% of the total sites). At the end of month 9, we had recruited 20 patients (i.e. 83% of the target). This was in spite of not starting recruitment until month 7 and having only three of the four pilot sites set up. There were, however, two sites at which we were waiting on approval, and 16 out of 26 sites with which we had held a preliminary meeting.

We also reviewed other aspects of the study. In summary, of the 34 patients who had a clinically confirmed frozen shoulder, four met the exclusion criteria, 10 did not consent and 20 consented. Early data showed that treatment preference was the reason for all non-consent, rather than patients being too busy or not wanting to be involved in research. The time taken to consent ranged from 15 minutes to 1 hour. Participating sites had confirmed that they had received sufficient training and supporting documentation. Except at one of the four pilot sites, all surgeons were supportive of the study. At this one site, one surgeon was taking part, another surgeon felt that he did not see a sufficient number of patients to take part, and a third surgeon lacked equipoise to consent patients. All three surgeons, however, had agreed to deliver the surgical interventions to which patients were allocated. No patient non-compliance with treatment had yet been reported.

Although we had not met our patient recruitment or site set-up targets, both oversight committees were satisfied with the overall progress made during phase 1.

The primary objective of phase 2 was to review the feasibility of the ESP intervention and whether non-compliance in the ESP arm did not exceed 20–30%. At the end of this phase (month 27), of the 65 trial participants who had been allocated to ESP, 37 had ended their treatment and could be assessed for non-compliance. The remaining 28 participants either had started their treatment or were waiting to start treatment. Of the 37 participants who had ended their treatment, 29 (78%) met our criteria for completing the intervention as had been agreed with the trial team and independent committees: the participant had attended eight sessions or more (n = 19); or had attended fewer than eight sessions but both the participant and the physiotherapist were satisfied with their progress (n = 9); or had attended fewer than eight sessions and declined to attend more because they were satisfied with their progress, their ability to manage independently, or both (n = 1). Therefore, non-compliance with the ESP intervention applied to 22% of participants, which was within the threshold of 20–30%. The oversight committees agreed that this was an acceptable degree of non-compliance and that the trial should continue with all three treatment arms.

Another aspect of trial feasibility that was reviewed at month 27 was non-consent into the trial. This was because there was concern that patients would often have already received physiotherapy in primary care and that this could affect their decision to take part in the trial, given that the ESP intervention was one of the treatment options. It was found that 55% (n = 72) of the 131 reasons for patients not taking part was because they either ‘want surgery’ or ‘do not want physiotherapy’. This compared with 18% (n = 24) of patients who ‘want physiotherapy’ or ‘do not want surgery’. Other reasons for non-consent were infrequent. The main treatment that non-consenting patients went on to have was keyhole surgery (45%, n = 67). Of patients randomised into the trial, the majority had no treatment preference (53%, n = 159), over one-third preferred surgery (39%, n = 116) and the rest preferred physiotherapy (8%, n = 25). Despite the preferences for surgery, this did not have an impact on the feasibility of the trial, with 36 sites set up, compared with the target of 25 sites, and 325 participants recruited against the target of 250 participants. We also reviewed the timing of the delivery of interventions, which confirmed that only one site had regularly failed to deliver surgery on time because of local pressures. The local trial team and principal investigator were very engaged and responsive to the trial team’s concerns and prioritised the trial participants for the surgical procedures.

In short, UK FROST was being delivered on time and to target, with an acceptable degree of non-compliance with the ESP intervention. The oversight committees were satisfied with the progress of all aspects of the trial and for it to continue as planned.

Summary findings of the shoulder capsule tissue and blood samples study

The primary aim of this nested study was to determine the key molecular processes and changes seen in the shoulder capsular tissue of patients with frozen shoulder in order to better understand these processes; and to determine the relationship between tissue changes, serum biomarkers and clinical symptoms and signs at presentation. This was done by determining the molecular and cellular abnormalities in shoulder capsular tissue obtained during surgery, by determining serum protein and cytokine signatures, and by correlating any tissue and serum abnormalities detected with the clinical presentation.

Following research ethics approval from the Oxford Musculoskeletal Biobank (09/H0606/11) and National Research Ethics Service Committee Newcastle and North Tyneside (14/NE/1176), appropriate informed consent was sought from UK FROST participants randomised to receive the ACR intervention. For a small sample of 16 patients who consented to the study, the shoulder capsular tissue and a venous blood sample were collected. The findings from the analysis of the capsular tissue samples were then compared with data available in the Oxford Tissue Biobank of findings from both healthy and diseased rotator cuff tendon tissues.

Inflammation signatures differed between tissue from frozen shoulder and that from tendon tears. Compared with tendon tear tissue, frozen shoulder capsular tissue showed reduced expression of nuclear factor-κB response genes, including TNFA, IL6 and IL8, and increased expression of IL10, CD14, CD163 and C1QA messenger RNA. The fibroblast activation markers podoplanin (PDPN), CD106 (VCAM1) and CD248 and the fibroblast activation protein were highly expressed in adhesive capsulitis and torn tendons, compared with healthy tendons. However, fibroblast activation marker CD90 was significantly reduced in adhesive capsulitis compared with healthy and diseased tendon tissue. Proresolving receptors mediating resolution of inflammation, including ALX/FPR2, CMKLR1 and GPR32, were highly expressed in frozen shoulder capsular tissue. 37

This study in patients of a similar age has provided some insight into why inflammation ultimately resolves in frozen shoulder but persists in tendon tears. This study suggests that the phenotypes of fibroblast subsets populating diseased shoulder tissues differ between conditions with self-limiting and and those with persistent inflammation. CD90 therefore represents an important pathogenic marker and possible molecular checkpoint regulating persistent stroma-mediated inflammation in common soft tissue diseases of the shoulder. Proresolving proteins were highly expressed in frozen shoulder tissue compared with established shoulder tendon tears. These findings have provided a novel insight into the disease mechanism of frozen shoulder, which points towards a resolving inflammatory environment. Further studies to better understand the biological mechanisms governing successful resolution of inflammation should inform new therapeutic strategies to accelerate disease resolution in frozen shoulder.

Recruitment into UK FROST

A total of 37 sites screened patients for the UK FROST trial, of which 35 randomised at least one patient. Appendix 3 presents the number of patients screened and randomised at each site, as well as the number of participants who withdrew before the end of the study.

Flow of participants

The flow of participants from screening to randomisation, treatment, follow-up and analysis is illustrated in the CONSORT flow diagram (Figure 2). Of 914 screened patients, 503 were randomised into the UK FROST trial. The reasons for exclusion were not meeting eligibility criteria (n = 95), non-consent (n = 295) and other reasons (n = 21). The most frequent reason for exclusion was having frozen shoulder symptoms secondary to trauma that required hospital care. Most patients who provided information about why they were not willing to join the trial said that this was because they had already had physiotherapy and wanted to have surgery (Table 1).

FIGURE 2.

The CONSORT flow diagram. IQR, interquartile range.

Treatment allocations were 2 : 2 : 1 to MUA with steroid injection (n = 201), ACR with MUA (n = 203), and ESP with steroid injection (n = 99). Follow-up rates at 3, 6 and 12 months post randomisation were between 85% and 89%, above the target of 80% assumed in the sample size and with no evidence of differential dropout in any of the treatment arms. The primary analysis at 12-month follow-up included all participants with OSS outcome data at one or more follow-ups, and therefore 94% of participants could be included in the analysis.

Baseline characteristics

Eligible patients who did and eligible patients who did not consent to participate in the trial were comparable in their baseline characteristics (Table 2). The demographic and clinical characteristics of participants at baseline are presented in Table 3, comparing the profile of the total patients randomised (n = 503) with that of participants included in the primary analysis (n = 473). No systematic differences between the two populations were evident. The characteristics of patients in the three randomised arms were broadly similar, with the exception of a greater number of participants in the MUA arm being currently in paid work and some group imbalance in having had a similar shoulder problem on the opposite side to the reference shoulder.

Intervention delivery

The criteria for having completed each of the three trial interventions were agreed and documented in the statistical analysis plan. For MUA and ACR, this was the completed surgical procedure, regardless of the completion of any PPP. In the ESP arm, completion of the intervention was defined as receipt of a minimum of eight physiotherapy sessions, unless the patient was discharged as satisfied with their progress earlier.

Table 4 shows that 82% of patients completed MUA, 80% of patients completed ACR and 81% of patients completed ESP. Overall, 16 participants (3%) crossed over to a different trial treatment, and 17 (3%) received an alternative non-trial treatment. Participants who did not start or complete any trial or non-trial treatment were classed as ‘no treatment recorded’ (n = 64, 13%).

The profile of surgeons and physiotherapists treating patients who completed their randomised intervention is presented in Appendix 4. Based on the available data, operating surgeons were predominantly consultants who had experience of routinely performing the trial operations up to once per month. Physiotherapists delivering ESP or PPP were most frequently band 6, treating between two and three frozen shoulders per month.

Waiting times to the start of each randomised intervention varied considerably. Table 5 shows that ESP patients received their first physiotherapy session or steroid injection within a median of 14 days, whereas patients waited a median of 56.5 days for MUA and a median of 71.5 days for ACR. Seventy patients in the ESP arm received a steroid injection on average within 12.8 days of randomisation. Nearly half (46%, n = 32) of these injections were administered on the day of randomisation.

Following completion of their randomised treatment, a number of patients received further treatment, as detailed in Table 6. Most commonly, this was ACR for patients randomised to MUA and further physiotherapy for patients randomised to ESP. Patients in the ACR arm received fewest further treatments.

As part of the surgical treatments, optimal release was reported as achieved in 92% of MUA procedures and in 98% of ACR procedures (Table 7). Steroid injection was delivered for all completed MUAs and 28% of ACRs. Steroid injection was also given to 80% of patients randomised to ESP (Table 8).

The number of delivered physiotherapy sessions is presented in Table 9. Participants who completed the ESP intervention attended a median of 9 sessions, whereas PPP following surgical procedures attended slightly fewer sessions (median of 7 for MUA and 8 for ACR). Individual therapeutic elements delivered as part of ESP and PPP sessions are summarised in Appendix 5.

Primary outcome

Descriptives

The OSS was the trial primary outcome and was collected using questionnaires at baseline and at 3, 6 and 12 months post randomisation. When OSS data from either the 6-month post-randomisation or the 6-month post-treatment questionnaire were available, and the two questionnaires had been sent to patients within 28 days, the available responses were used to complete any missing OSS outcomes. A summary of descriptive statistics of OSS scores is presented in Table 9 and Figure 3 (see Appendix 6 for the scores split by pain and function subdomains). By 12-month follow-up, many participants (24%) had regained function up to the top OSS score of 48, and a ceiling effect of OSS scores for all three arms could be observed. This restricted variability of scores at the top end meant that the ability of the trial to detect clinically meaningful differences at the primary end point was reduced.

FIGURE 3.

Unadjusted mean OSS and 95% CIs by treatment arm.

Secondary analyses

Analysis incorporating different waiting times

In addition to questionnaires completed at post-randomisation follow-ups, participants were asked to complete the OSS just before and 6 months following receipt of treatment in order to account for the differential waiting times for each trial treatment. Descriptive results for these outcomes at these two points are given in Table 11 and are presented together with OSS scores at baseline and at 12 months’ post-randomisation follow-up in Figure 4. The OSS scores appeared to stay stable between baseline and the start of any of the treatments, which was later for the surgical arms (95% CI of mean difference ESP vs. MUA –2.0 to 2.7, ESP vs. ACR –1.5 to 3.2, MUA vs. ACR –1.4 to 2.5). Six months following treatment, scores had improved to a greater extent in the surgical arms than in the ESP arm (95% CI of mean difference ESP vs. MUA –5.5 to –2.2, ESP vs. ACR –6.1 to –0.6, MUA vs. ACR –2.6 to 1.6) and were similar to final follow-up scores by 8 months.

TABLE 11 - Unadjusted OSS by treatment arm: pre treatment and 6 months post treatment

Time point	Treatment arm			Total
	MUA	ACR	ESP
Pre treatment
n	159	157	77	393
Mean (SD)	21.5 (8.79)	21.0 (8.92)	21.8 (8.02)	21.4 (8.68)
Median	21	21	22	21
Minimum, maximum	3, 46	1, 42	6, 42	1, 46
Post treatment (6 months)
n	157	152	81	390
Mean (SD)	39.0 (9.03)	39.4 (9.68)	36.1 (10.67)	38.5 (9.70)
Median	42	43	39	42
Minimum, maximum	6, 48	2, 48	6, 48	2, 48

FIGURE 4.

Unadjusted mean OSS and 95% CIs by treatment arm (using scores at baseline, follow-up before treatment and 6 months after treatment; and at 3, 6 and 12 months post randomisation).

A linear mixed random intercept model incorporated time as a continuous variable, included data from all available time points for each patient (up to five measurements) and adjusted for the same covariates as the primary analysis model. OSS estimates at 3, 6 and 12 months post-randomisation follow-up were derived from the model and are presented in Table 12. Compared with the primary analysis model, group differences tended to be of smaller magnitude, with the exception of the difference between ACR and ESP at 12 months (3.26 points in favour of ACR, 95% CI 1.18 to 5.35 points). The 95% CI interval still included the minimal clinically important difference for this comparison of 5 OSS points.

TABLE 12 - Estimated mean OSS differences by treatment arm (estimates from model incorporating follow-ups before and after treatment in addition to post-randomisation outcomesa)

Linear mixed random intercept model adjusted for age, sex, diabetes, OSS at baseline (fixed effects) and site (random effect).

Time point	Treatment arm		Difference, mean (95% CI)	p-value
	Mean (95% CI)	Mean (95% CI)
	MUA	ESP
3 months	28.2 (27.1 to 29.3)	29.4 (27.8 to 30.9)	–1.18 (–3.10 to 0.73)	0.23
6 months	32.5 (31.5 to 33.5)	32.7 (31.2 to 34.1)	–0.15 (–1.90 to 1.60)	0.87
12 months	41.1 (40.0 to 42.3)	39.2 (37.5 to 40.9)	1.92 (–0.16 to 4.00)	0.07
	ACR	ESP
3 months	26.0 (24.9 to 27.2)	29.4 (27.8 to 30.9)	–3.33 (–5.25 to –1.40)	< 0.01
6 months	31.5 (30.5 to 32.5)	32.7 (31.2 to 34.1)	–1.13 (–2.88 to 0.62)	0.21
12 months	42.5 (41.3 to 43.7)	39.2 (37.5 to 40.9)	3.26 (1.18 to 5.35)	< 0.01
	ACR	MUA
3 months	26.0 (24.9 to 27.2)	28.2 (27.1 to 29.3)	–2.14 (–3.71 to –0.57)	0.01
6 months	31.5 (30.5 to 32.5)	32.5 (31.5 to 33.5)	–0.98 (–2.40 to 0.44)	0.18
12 months	42.5 (41.3 to 43.7)	41.1 (40.0 to 42.3)	1.35 (–0.33 to 3.02)	0.12

Analysis incorporating treatment compliance

Baseline characteristics for participants who did and participants who did not complete their randomised treatment according to the trial definitions are presented in Table 13. The profile of non-completers in each treatment arm tended to be different.

TABLE 13 - Comparison of baseline characteristics by treatment compliance

Characteristic	MUA		ACR		ESP
	Completed treatment (N = 164)	Did not complete treatment (N = 37)	Completed treatment (N = 162)	Did not complete treatment (N = 41)	Completed treatment (N = 80)	Did not complete treatment (N = 19)
Sex, n (%)
Male	54 (33)	18 (49)	63 (39)	14 (34)	29 (36)	6 (32)
Female	110 (67)	19 (51)	99 (61)	27 (66)	51 (64)	13 (68)
Age (years)
n	164	37	162	41	80	19
Mean (SD)	54.0 (7.4)	56.8 (8.8)	54.3 (7.5)	52.3 (8.3)	55.6 (7.7)	49.8 (6.3)
Median (minimum, maximum)	54 (30, 57)	56 (36, 73)	54 (33, 76)	52 (34, 71)	55 (39, 77)	50 (39, 69)
Diabetic, n (%)
No	115 (70)	26 (70)	118 (73)	25 (61)	55 (69)	14 (74)
Type 1	12 (7)	–	8 (5)	4 (10)	4 (5)	1 (5)
Type 2	37 (23)	11 (30)	36 (22)	12 (29)	21 (26)	4 (21)
Employment status summary, n (%)
In paid work	113 (69)	16 (43)	95 (59)	23 (56)	44 (55)	9 (47)
Not in paid work	48 (29)	21 (57)	65 (40)	17 (41)	36 (45)	10 (53)
Missing	3 (2)	–	2 (1)	1 (2)	–	–
Duration of symptoms (months)
n	160	36	161	40	79	19
Mean (SD)	10.6 (8.5)	10.3 (8.9)	11.5 (10.5)	10.56 (7.5)	10.3 (6.1)	13.0 (15.8)
Median (minimum, maximum)	8 (2, 60)	7.5 (2, 48)	9 (2, 96)	9 (0, 36)	9 (3, 36)	8 (2, 72)
Previous physiotherapy for affected shoulder, n (%)
Yes	100 (61)	25 (68)	99 (61)	25 (61)	49 (6)	10 (53)
No	64 (39)	12 (32)	63 (39)	14 (34)	31 (39)	8 (42)
Missing	–	–	–	2 (5)	–	1 (5)
OSS, points (0–48)
n	163	37	161	41	80	19
Mean (SD)	20.4 (8.9)	20.8 (8.9)	19.0 (7.6)	19.9 (8.4)	20.7 (7.8)	18.3 (8.4)
Median (minimum, maximum)	20 (2, 48)	20 (3, 36)	19 (1, 37)	19 (4, 35)	20 (2, 42)	18 (4, 34)

Owing to the three active treatments under investigation and multiple alternative treatment pathways for each patient, the scope for conducting CACE analysis was limited, as assumptions of the analysis did not hold. Only one treatment comparison was conducted at the primary end point at 12 months, that of compliance with ESP as defined in the fidelity section of this report (Table 8).

Instrumental variable regression was implemented predicting OSS at the primary end point at 12 months in order to quantify the effect of compliance with ESP. From the model, outcomes for ESP compliers remained lower than for patients in the other treatment arms (–1.84 OSS points, 95% CI –4.41 to 0.74 OSS points; p = 0.157); however, the difference was not statistically significant. Based on Appendix 8, patients tended to have better outcomes if they completed their randomised treatment.

Missing data

Possible predictors of missing OSS data at 3-, 6- or 12-month follow-up are presented in Table 14. Only age (younger participants being more likely to have missing data) and OSS outcomes prior to the time of missing data (participants with poorer outcomes being more likely to have missing data) were significant predictors of missingness. As these are already covariates in the primary analysis model, no model adjustments were undertaken.

TABLE 14 - Comparison of patient characteristics by missingness of OSS over time

Statistically significant at the 5% level.

	Not missing	Missing	p-value
3-month follow-up	N = 447	N = 56
Age (years), mean (SD)	54.6 (7.6)	51.5 (8.4)	0.01a
Male, n (%)	161 (36%)	23 (41%)	0.46
Diabetic, n (%)	128 (29%)	22 (39%)	0.10
In employment, n (%)	270 (60%)	30 (54%)	0.45
Duration of symptoms (months), mean (SD)	11.1 (9.4)	9.6 (7.1)	0.26
Previous physiotherapy, n (%)	281 (63%)	27 (48%)	0.07
Baseline OSS (points), mean (SD)	20.1 (8.2)	18.5 (8.6)	0.19
6-month follow-up	N = 430	N = 73
Age (years), mean (SD)	54.7 (7.6)	51.6 (8.1)	< 0.01a
Male, n (%)	155 (36%)	29 (40%)	0.55
Diabetic, n (%)	122 (28%)	28 (38%)	0.09
In employment, n (%)	257 (60%)	43 (59%)	0.90
Duration of symptoms (months), mean (SD)	11.0 (9.1)	10.2 (9.9)	0.50
Previous physiotherapy, n (%)	266 (62%)	42 (58%)	0.65
Baseline OSS (points), mean (SD)	20.2 (8.2)	18.2 (8.4)	0.06
3-month OSS (points), mean (SD)	30.4 (10.9)	26.2 (11.9)	0.05a
12-month follow-up	N = 446	N = 57
Age (years), mean (SD)	54.5 (7.5)	52.1 (9.0)	0.03a
Male, n (%)	164 (37%)	20 (35%)	0.80
Diabetic, n (%)	136 (30%)	14 (25%)	0.36
In employment, n (%)	266 (60%)	34 (60%)	0.82
Duration of symptoms (months), mean (SD)	11.0 (9.4)	10.4 (7.8)	0.68
Previous physiotherapy, n (%)	278 (62%)	30 (53%)	0.26
Baseline OSS (points), mean (SD)	20.1 (8.2)	18.3 (8.7)	0.11
3-month OSS (points), mean (SD)	30.4 (10.8)	25.9 (13.4)	0.05
6-month OSS (points), mean (SD)	37.4 (10.1)	36.5 (10.7)	0.72

Based on the low drop-out rate at the primary end point at 12 months (11%), and the fact that nearly all patients could be included in the primary analysis (94%), further adjustments for missing data, such as multiple imputation, were not implemented.

Other secondary analyses

Further secondary analyses excluded responses received beyond 6 weeks of each intended follow-up and adjusted for the observed baseline imbalance in employment status (see Appendix 9). The results were similar to those observed in the primary analysis.

Subgroup analyses

The possibility of differential treatment effects were explored for subgroups based on diabetes status, receipt of previous physiotherapy and baseline treatment preference, and, in addition, length of frozen shoulder symptoms at baseline following advice from the trial oversight committee. Interaction terms between treatment allocation and subgroups were added to the primary analysis model and p-values for interactions for each treatment comparison were derived (Table 15). None of the interaction terms was statistically significant, although the study was not powered to detect such interactions and the number of participants in some of the subgroups in each treatment arm was very small.

TABLE 15 - Subgroup analyses summary

Allocated to non-preferred vs. allocated to preferred treatment.

No treatment preference vs. allocated to preferred treatment.

Subgroup	Treatment arm (n)			Treatment comparison	Contrast	95% CI	p-value of allocation interaction with subgroup
	MUA	ACR	ESP
Diabetes
Diabetic (n = 150)	60	60	30	MUA vs. ESP	–0.34	–4.58 to 3.90	0.88
				ACR vs. ESP	0.09	–4.16 to 4.34	0.97
Not diabetic (n = 353)	141	143	69	ACR vs. MUA	0.43	–2.98 to 3.85	0.80
Previous physiotherapy
Had previous physiotherapy (n = 308)	125	124	59	MUA vs. ESP	–2.08	–6.04 to 1.89	0.30
				ACR vs. ESP	–0.87	–4.86 to 3.12	0.67
Did not have previous physiotherapy (n = 192)	76	77	39	ACR vs. MUA	1.21	–2.02 to 4.44	0.46
Patient treatment preference
Allocated to preferred treatment (n = 131)	56	64	11	MUA vs. ESP	2.10a	–4.32 to 8.52	0.81
					1.28b	–4.66 to 7.22
Allocated to non-preferred treatment (n = 105)	40	27	38	ACR vs. ESP	3.11a	–3.50 to 9.73	0.65
					2.18b	–3.73 to 8.09
Had no treatment preference (n = 263)	103	111	49	ACR vs. MUA	1.01a	–3.84 to 5.87	0.87
					0.90b	–2.70 to 4.50
Duration of symptoms at baseline
< 9 months (n = 249)	103	95	51	MUA vs. ESP	–2.41	–6.29 to 1.46	0.22
				ACR vs. ESP	–2.00	–5.85 to 1.85	0.31
≥ 9 months (n = 246)	93	106	47	ACR vs. MUA	0.41	–2.73 to 3.56	0.80

Possible trends are illustrated in Figures 5–8 and descriptive tables are in Appendix 10. Diabetic patients tended to have poorer outcomes than non-diabetic patients at all time points, and especially at the 3-month follow-up for patients in the ACR arm. Patients who had previous physiotherapy tended to have worse outcomes if they were randomised to ESP, especially at the 3- and 6-month follow-ups, whereas patients who had indicated a prior preference for physiotherapy tended to have better outcomes if they were randomised to ESP and worse outcomes if they were randomised to either surgical treatment. Participants who reported frozen shoulder symptoms for ≥ 9 months before entering the trial tended to have worse outcomes at 3 months if they were randomised to ACR and better outcomes at 3 months if they were randomised to ESP.

FIGURE 5.

Unadjusted mean OSS function items and 95% CIs by treatment arm and diabetes: (a) diabetic; and (b) not diabetic.

FIGURE 6.

Unadjusted mean OSS function items and 95% CIs by treatment arm and previous physiotherapy: (a) had previous physiotherapy; and (b) did not have previous physiotherapy.

FIGURE 7.

Unadjusted mean OSS function items and 95% CIs by treatment arm and baseline preference: (a) preference for physiotherapy; (b) preference for surgery; and (c) no preference.

FIGURE 8.

Unadjusted mean OSS function items and 95% CIs by treatment arm and length of time with symptoms: (a) symptoms for < 9 months; and (b) symptoms for ≥ 9 months.

Secondary outcomes

Among the secondary outcomes, QuickDASH and shoulder pain followed a similar pattern to the OSS, in that ACR patients were observed to have significantly poorer outcomes at 3 months (note that many patients had only recently had or were still waiting for their surgery at this point) but better outcomes at 12 months post randomisation than those allocated to MUA or ESP (Tables 16 and 17). Unadjusted means are presented and illustrated in the tables and figures below (see Appendix 11 and Figures 9 and 10).

QuickDASH

FIGURE 9.

Unadjusted mean QuickDASH scores and 95% CIs by treatment arm.

TABLE 16 - Estimated mean QuickDASH differences by treatment arma

Linear mixed covariance pattern model adjusted for age, sex, diabetes, QuickDASH at baseline (fixed effects) and site (random effect).

Time point	Mean (95% CI)	Mean (95% CI)	Mean difference (95% CI)	p-value
	MUA	ESP
3 months	38.8 (35.7 to 42.0)	37.1 (32.7 to 41.4)	1.77 (–3.41 to 6.96)	0.50
6 months	25.7 (22.6 to 28.7)	29.2 (24.9 to 33.5)	–3.55 (–8.68 to 1.58)	0.18
12 months	22.9 (19.8 to 26.0)	23.4 (19.0 to 27.8)	–0.50 (–5.70 to 4.70)	0.85
	ACR	ESP
3 months	44.4 (41.3 to 47.5)	37.1 (32.7 to 41.4)	7.33 (2.16 to 12.49)	< 0.01
6 months	27.4 (24.4 to 30.4)	29.2 (24.9 to 33.5)	–1.82 (–6.94 to 3.31)	0.49
12 months	18.2 (15.1 to 21.3)	23.4 (19.0 to 27.8)	–5.20 (–10.42 to 0.02)	0.05
	ACR	MUA
3 months	44.4 (41.3 to 47.5)	38.8 (35.7 to 42.0)	5.55 (1.32 to 9.78)	0.01
6 months	27.4 (24.4 to 30.4)	25.7 (22.6 to 28.7)	1.73 (–2.39 to 5.86)	0.41
12 months	18.2 (15.1 to 21.3)	22.9 (19.8 to 26.0)	–4.71 (–8.91 to –0.50)	0.03

Shoulder pain

FIGURE 10.

Unadjusted mean shoulder Numeric Rating Scale for Pain and 95% CIs by treatment arm.

TABLE 17 - Estimated shoulder Numeric Rating Scale for Pain differences by treatment arma

Linear mixed covariance pattern model adjusted for age, sex, diabetes, Numeric Rating Scale for Pain at baseline (fixed effects) and site (random effect).

Time point	Mean (95% CI)	Mean (95% CI)	Mean difference (95% CI)	p-value
	MUA	ESP
3 months	4.1 (3.8 to 4.5)	3.7 (3.2 to 4.2)	0.43 (–0.17 to 1.03)	0.16
6 months	2.8 (2.4 to 3.1)	3.0 (2.5 to 3.5)	–0.19 (–0.78 to 0.40)	0.53
12 months	2.4 (2.1 to 2.8)	2.5 (2.0 to 3.0)	–0.08 (–0.66 to 0.50)	0.78
	ACR	ESP
3 months	4.7 (4.3 to 5.1)	3.7 (3.2 to 4.2)	1.02 (0.42 to 1.61)	< 0.01
6 months	2.8 (2.5 to 3.2)	3.0 (2.5 to 3.5)	–0.14 (–0.74 to 0.45)	0.63
12 months	1.7 (1.4 to 2.0)	2.5 (2.0 to 3.0)	–0.81 (–1.39 to –0.23)	< 0.01
	ACR	MUA
3 months	4.7 (4.3 to 5.1)	4.1 (3.8 to 4.5)	0.59 (0.10 to 1.07)	0.02
6 months	2.8 (2.5 to 3.2)	2.8 (2.4 to 3.1)	0.05 (–0.43 to 0.52)	0.85
12 months	1.7 (1.4 to 2.0)	2.4 (2.1 to 2.8)	–0.73 (–1.20 to –0.25)	< 0.01

Extent of recovery

There was no evidence of statistically significant differences between treatment arms in the reduction in frozen shoulder symptoms as measured by the extent of recovery (‘To what extent would your frozen shoulder symptoms in the past 24 hours prompt you to ask for further treatment?’; response on visual analogue scale of 0–100). Descriptives are presented in Appendix 11 (see Table 53) and illustrated in Figure 11, and the results of the analysis are in Table 18.

FIGURE 11.

Unadjusted mean extent of recovery visual analogue scale and 95% CIs by treatment arm. Scale is 0–100, where 100 is equivalent to maximum belief that symptoms require further treatment.

TABLE 18 - Estimated mean extent of recoverya visual analogue scale differences by treatment armb

Scale is 0–100, where 100 is equivalent to the maximum belief that symptoms require further treatment.

Linear mixed covariance pattern model adjusted for age, sex, diabetes, symptom severity at baseline (fixed effects) and site (random effect).

Time point	Mean (95% CI)	Mean (95% CI)	Mean difference (95% CI)	p-value
	MUA	ESP
3 months	51.4 (45.8 to 56.9)	53.9 (46.3 to 61.5)	–2.55 (–11.68 to 6.58)	0.58
6 months	31.9 (26.5 to 37.2)	38.6 (30.9 to 46.3)	–6.71 (–15.83 to 2.42)	0.15
12 months	27.3 (22.4 to 32.3)	26.9 (20.0 to 33.8)	0.46 (–7.79 to 8.70)	0.91
	ACR	ESP
3 months	54.0 (48.5 to 59.5)	53.9 (46.3 to 61.5)	0.11 (–9.02 to 9.23)	0.98
6 months	34.7 (29.3 to 40.0)	38.6 (30.9 to 46.3)	–3.93 (–13.06 to 5.21)	0.40
12 months	21.2 (16.3 to 26.2)	26.9 (20.0 to 33.8)	–5.65 (–13.91 to 2.61)	0.18
	ACR	MUA
3 months	54.0 (48.5 to 59.5)	51.4 (45.8 to 56.9)	2.66 (–4.84 to 10.15)	0.49
6 months	34.7 (29.3 to 40.0)	31.9 (26.5 to 37.2)	2.78 (–4.50 to 10.06)	0.45
12 months	21.2 (16.3 to 26.2)	27.3 (22.4 to 32.3)	–6.11 (–12.86 to 0.64)	0.08

Other analyses

Systematic review: integrating the new evidence

A systematic review was undertaken to assess the effectiveness of MUA, ACR, hydrodilatation and physiotherapy with steroid injection in the management of patients with a primary frozen shoulder in order to place the findings of UK FROST in the context of existing evidence for these treatments. Nine relevant studies were identified, including UK FROST, which provided by far the most, and the most robust, evidence.

Owing to considerable heterogeneity of the interventions and study populations, only two studies could be pooled as part of a meta-analysis, comparing long-term shoulder functioning for patients receiving either ACR or physiotherapy in UK FROST and one other trial. The pooled effect favoured ACR (standard effect size 0.32, 95% CI 0.08 to 0.56), which was largely determined by the UK FROST results, as the second trial was much smaller (n = 44). The pooled effect was of smaller magnitude than the clinical threshold of 5 OSS points, equivalent to a standard effect size of approximately 0.42.

Full details are presented in Report Supplementary Material 20.

Objective

The objective of this economic evaluation is to assist decision-making to identify the most efficient provision of future care for the management of frozen shoulder in secondary care in the NHS.

Overview

A prospective economic evaluation was conducted alongside the UK FROST trial, the aim being to estimate the cost-effectiveness of the three most commonly used interventions for the management of frozen shoulder in secondary care. The three interventions in the study were ESP with an intra-articular steroid injection compared with MUA with a steroid injection or ACR followed by manipulation. Both surgical interventions were followed with a programme of PPP.

Costs and health benefits were compared for the three groups over 12 months, and hence discounting was not required. All costs were expressed in Great British pounds at a 2017–18 price base. Health benefits were expressed in terms of QALYs based on patients’ health-related quality of life (HRQoL) assessed using the EQ-5D-5L. 50,66 Differences in mean costs and mean QALYs at 1 year were used to derive an estimate of the cost-effectiveness of surgery and of non-surgical treatment.

The base-case analysis was conducted on an ITT basis as with the statistical analyses in Chapter 3. The perspective of the UK NHS and Personal Social Services was adopted for the analysis, and hence costs incurred by families and informal carers were excluded from the base case. A secondary analysis was undertaken from a broader perspective. The National Institute for Health and Care Excellence (NICE) guidelines were applied to all methods used in this economic analysis. 67

Owing to the impact of missing data, the base-case analysis was conducted as an imputed analysis;68 the choice of method for handling missing data (multiple imputation) was grounded in the assumed missing data mechanism (MAR), which in turn was supported by the UK FROST data set. The impact of alternative assumptions about the missing data mechanism was carefully assessed in sensitivity analyses.

Methods

Cost-effectiveness analyses

The main cost-effectiveness analyses were conducted following multiple imputations of all cost and outcomes data. The mean difference in costs and QALYs for the base-case analysis was estimated using regression methods for data on costs and QALYs, adjusting for baseline characteristics.

A bivariate regression model – seemingly unrelated regression (SUR) – of costs and QALYs was used to calculate incremental estimates, using conventional decision rules and estimating incremental cost-effectiveness ratios (ICERs) when appropriate. 82 SUR allows outcomes to be estimated jointly and so brings efficiency gains over ordinary least squares. 83 In the bivariate model, incremental costs and QALYs are estimated simultaneously from two separate ordinary least squares regressions, assuming correlation between the error terms in both regressions. 84 The SUR model used the same set of covariates as the mixed-effect regression model used for the clinical effectiveness analysis [age, sex, baseline EQ-5D score, baseline OSS score and diabetes (yes/no)].

The cost-effectiveness results were expressed in terms of ICERs. These were estimated as the difference in mean costs divided by the difference in mean QALYs between the trial comparators. The ICER is estimated to inform decision-makers about the optimal use of NHS resources. According to standard cost-effectiveness decision rules, four different eventualities are plausible when comparing incremental costs and QALYs. If the new intervention provides better outcomes (positive incremental QALYs) at lower costs (negative incremental costs), it is considered a dominant intervention and, hence, cost-effective. If the new intervention achieves poorer outcomes (negative incremental QALYs) at higher costs (positive incremental costs), it is considered a dominated option and, hence, not cost-effective. Thus, the ICER is considered only if either intervention does not dominate, that is both incremental costs and incremental QALYs are positive (or negative). In these last two situations, to determine whether or not the incremental health gain is worth the incremental cost, the ICER needs to be compared against a threshold value. For positive incremental costs and QALYs (the most frequent situation in HTA), an intervention will be considered cost-effective only if the ICER is lower than the threshold. According to NICE, the willingness-to-pay threshold for an additional QALY ranges from £20,000 to £30,000. 67 This threshold has been used by NICE for more than a decade; however, it has recently been suggested that the threshold should be decreased to £13,000 per QALY gained. 85,86 According to the current established decision rules, if the result of this cost–utility analysis, namely the estimated cost per QALY, is below the £30,000 threshold, the intervention would be considered cost-effective in terms of QALYs gained.

To compute the probability that each intervention is cost-effective at a given cost-effective threshold, the SUR was conducted in a bootstrapping approach on five imputed data sets to generate 10,000 replicates of incremental costs and benefits. These replicates were represented graphically as cost-effectiveness acceptability curves (CEACs). The probability that each intervention is cost-effective is reported at the cost-effectiveness threshold of £20,000–30,000 per QALY applied by NICE and at a threshold of £13,000 per QALY as suggested by recent research.

Results

Health-related quality of life and quality-adjusted life-years

A similar decrease in complete EQ-5D-5L questionnaires was seen throughout the trial follow-up across the three arms. Responses for the complete follow-up (i.e. baseline and 3, 6 and 12 months) were available for 369 (73%) participants: 156 (78%) in the MUA arm, 149 (73%) in the ACR arm and 64 (65%) in the ESP arm. The extent of incomplete EQ-5D-5L responses as a result of missing data strengthened the justification for using the imputed data sets as the base case (see Appendix 16).

The proportions of participants who reported the EQ-5D-5L levels (1–5) by dimension, group and time point are in Appendix 16. Comparing self-care levels between baseline and 12 months, we found that more people reported no problems across all the treatments, but the percentage was slightly higher in the ACR arm. When looking at usual activities, all groups had a similar increase over 12 months in the number of participants who classed themselves as level 1. A smaller percentage of participants were level 1 for pain/discomfort after 12 months in the ESP arm than in the MUA and ACR arms. When looking at anxiety and depression, we found that there was, again, a comparable increase in the percentage of participants in level 1 between baseline and 12 months across all treatments.

The overall distribution of the EQ-5D-5L scores (utilities) for the different follow-up assessments is illustrated in Figure 12. Patients allocated to MUA started from a higher utility value than patients allocated to ACR and ESP.

FIGURE 12.

The EQ-5D-5L scores distribution at the different time points over the 12 months.

Table 27 summarises the mean EQ-5D-5L scores reported at each follow-up point for all available patients. Adjusted analysis shows that patients allocated to ACR and MUA had similar utility values at 12-month follow-up [ACR (mean 0.739) vs. MUA (mean 0.734)]. Similarly, patients allocated to the surgical arms had better utility values than those allocated to the ESP arm (mean 0.693). The QALY estimates at 1-year follow-up (adjusted for baseline utility) show that patients allocated to MUA accrued more QALYs than those in the other two arms: MUA (0.6765) >ESP (0.6492) >ACR (0.6475).

TABLE 27 - Health-related quality of life: EQ-5D-5L summary scores (available cases) at each time point adjusted for baseline utility

QALY estimates (complete cases) adjusted for baseline utility.

Follow-up point	Treatment arm, mean (SE)
	MUA	ACR	ESP
Baseline	0.456 (0.263)	0.428 (0.234)	0.402 (0.294)
3 months	0.632 (0.017)	0.567 (0.017)	0.606 (0.024)
6 months	0.729 (0.016)	0.677 (0.016)	0.680 (0.024)
12 months	0.734 (0.018)	0.739 (0.184)	0.693 (0.027)
QALYs (adjusted utility) (95% CI)	0.6765 (0.651 to 0.702)	0.6475 (0.621 to 0.674)	0.6492 (0.609 to 0.690)

As for total costs, the HRQoL and QALYs estimates shown in this section are of limited value, as these estimates correspond exclusively to patients with complete EQ-5D-5L data (i.e. at baseline and at 3, 6 and 12 months). The mean differences in QALYs between the groups and corresponding 95% CIs, adjusted for all relevant covariates and taking into consideration the correlation between costs and QALYs, are shown in Cost-effectiveness analysis.

Missing data

The UK FROST study collected data using the EQ-5D-5L at 3, 6 and 12 months. Health-care resource use was elicited from patients by postal questionnaire at 3, 6 and 12 months, and from health-care professionals by hospital forms 52 weeks after randomisation. A description of the economic variables in UK FROST can be found in Appendix 17.

Overall, the proportion of participants with complete economic data remained similar between treatment arms (see Appendix 17): 46.46% in the ESP arm, 58.21% in the MUA arm and 57.14% in the ACR arm. In all arms, more individuals are observed in month 12 than in month 6. Therefore, the missing data do not follow a monotonic pattern; in other words, there are participants with intermittent missing data (e.g. lost to follow-up at 6 months but remained subsequently). Hence, inverse probability weighting would be inappropriate under such a pattern. Similarly, complete-case analysis would be, as a minimum, inefficient because it would discard observed data from individuals with some missing outcomes.

Figures showing the pattern of missing data are in Appendix 17. As discussed above, missing data are shown to be non-monotonic, as individuals with missing data at one follow-up point may provide data subsequently.

Logistic regressions of indicators of missing cost and QALY data on treatment allocation and a selection of baseline variables showed that lower EQ-5D-5L scores at baseline are associated with missing cost and QALY data (see Appendix 17). Baseline age was also found to be a significant predictor of missing data on HRQoL. This suggests that the data are unlikely to be missing completely at random. The other baseline covariates (sex and diabetes) were associated with missingness but were not statistically significant at 5%. However, diabetes was significant predictor of costs and QALYs at 6 months and 1 year, which would support both covariate-dependent missing completely at random and MAR assumptions.

We also explored whether or not missingness is associated with previously observed outcomes by regressing indicators of missing costs or QALYs at each year on their previously observed values (e.g. regressing missing costs and QALYs at 1 year on costs and QALYs in previous months). Most regressions produced statistically insignificant results (p > 0.05), with two exceptions: missing QALYs at 1 year were significantly associated with QALYs at 3 months, and missing costs at 1 year were significantly associated with QALYs at 3 months and QALYs at 6 months. Although these regressions are likely to be affected by multicollinearity, they provide an indication that data are unlikely to be covariate-dependent missing completely at random.

Therefore, data were assumed to be MAR, and multiple imputation by chained equations was selected to handle missing data for this economic analysis. In the analysis, missingness is assumed to depend on baseline covariates (sex, diabetes, age, EQ-5D-5L at baseline and OSS score at baseline) and observed costs and QALYs but to be independent of unobservable costs at QALYs at 1 year. As it is impossible to know whether data are MNAR or MAR from the observed data, a mixed model is presented as per sensitivity analysis. Complete-case analysis, which is not valid when data are MAR, is presented for comparison only.

The multiple imputation model was validated by comparing the distribution of the observed UK FROST data with the imputed data (see Appendix 17).

Cost-effectiveness analysis

Base-case analysis

A bivariate regression, in the form of a seemingly related regression, conducted in the imputed data set was used to estimate the incremental costs and incremental health outcomes (i.e. QALYs) associated with the interventions (Table 28). Patients allocated to MUA showed a (non-significant) QALY gain compared with those allocated to ESP (mean difference 0.0396, 95% CI –0.0008 to 0.0800). Similarly, patients allocated to ACR showed a (non-significant) QALY gain compared with those allocated to ESP (mean difference 0.0103, 95% CI –0.0304 to 0.0510). Overall, those allocated to ACR had worse (non-significant) QALYs than those allocated to MUA at the 12-month follow-up (mean difference –0.0293, 95% CI –0.0616 to 0.0030).

TABLE 28 - Adjusted mean differences in QALYs and costs between interventions (base case)

SUR, seemingly unrelated regression.

Compared with ESP, as it is the alternative with lower costs and worse health outcomes.

Treatment arm	Adjusted difference in means with SURa (95% CI)
Difference in costs (£)
MUA vs. ESP	276.507 (65.67 to 487.35)
ACR vs. ESP	1733.78 (1529.48 to 1938.06)
ACR vs. MUA	1457.26 (1282.73 to 1631.79)
Difference in QALYs
MUA vs. ESP	0.0396 (–0.0008 to 0.0800)
ACR vs. ESP	0.0103 (–0.0304 to 0.0510)
ACR vs. MUA	–0.0293 (–0.0616 to 0.0030)
	ICER (£ per QALY)	Probability that intervention is cost-effective at £13,000/QALY	Probability that intervention is cost-effective at £20,000/QALY	Probability that intervention is cost-effective at £30,000/QALY
MUA	6984	0.7942	0.8632	0.8978
ACR	> 100,000	0.0000	0.0002	0.002
ESP	–	0.2058	0.1366	0.1002

The results of the fully incremental cost-effectiveness estimates and probability that each intervention is cost-effective at a threshold of £20,000 per QALY are also shown in Table 28. Compared with physiotherapy, MUA cost a mean of £276 more per patient (95% CI £65.67 to £487.35) and allowed patients to experience improved health outcomes at the end of the trial (on average 0.0396 more QALYs per participant than ESP, 95% CI –0.0008 to 0.0800). The resulting ICER for MUA was £6984 per additional QALY. ACR is significantly more costly than ESP [on average £1733.78 more expensive per participant (95% CI £1529.48 to £1938.06)], and, despite the QALY gain accrued by ACR participants (on average 0.0396 more QALYs per participant than ESP, 95% CI –0.0008 to 0.0800), this was not sufficient to prove that ACR is a cost-effective use of NHS resources when compared with ESP (i.e. ICER above recommended NICE threshold). Similarly, ACR is dominated by MUA, with higher mean costs and fewer QALYs.

The corresponding CEACs showing the probability that each treatment is cost-effective across a range of thresholds are shown in Figure 13. The probability that MUA surgery is cost-effective is 0.88 at a threshold of £20,000 per QALY gained. The CEAC indicates that, regardless of the value of the cost-effectiveness threshold, the probability that ACR is cost-effective does not exceed 0.002.

FIGURE 13.

Base-case CEACs.

Conclusion

This economic analysis has provided robust evidence on whether or not surgical management is cost-effective for the treatment of frozen shoulder. Over the trial period, the base-case analysis with the ITT approach showed that MUA was the intervention most likely to be cost-effective. The resulting ICER for MUA was £6984 per additional QALY when compared with ESP; over common threshold values of a QALY, the probability that MUA was cost-effective was high (> 85% from an NHS perspective). The finding indicates that ACR is dominated by MUA (higher mean costs and fewer QALYs), and ACR showed very low probability of being cost-effective (< 5% from an NHS perspective).

The positive cost-effectiveness profile of MUA is robust to plausible departures from MAR. Similarly, these results were robust to a number of sensitivity analyses, showing that MUA was the intervention most likely to be cost-effective at a threshold of £20,000 per QALY, with probabilities ranging across scenarios from 48% (complete-case analysis) to 99%. The only exemption was when we used the societal perspective to estimate the costs; ACR appeared to be dominated by MUA across all scenarios.

Discussion

The economic evaluation alongside the UK FROST trial was conducted following NICE methodological standards. We implemented a comprehensive strategy to handle missing data in accordance with methodological guidelines, and we used a number of analytical tools to address uncertainty, including sampling and methodological uncertainty. The results of the analyses suggest that MUA is a cost-effective option for the treatment of frozen shoulder in terms of QALYs gained calculated using the EQ-5D-5L. Compared with ESP, MUA cost a mean of £276 more per patient (95% CI £65.67 to £487.35) and allowed patients to experience improved health outcomes at the end of the trial [on average 0.0396 more QALYs per participant than ESP (95% CI –0.0008 to 0.0800)]. The ICER for the ITT approach in the imputed data set was £6984 per additional QALY. The probability that MUA is cost-effective is > 85%, whereas the probability that ESP is cost-effective does not exceed 20%. ACR is significantly more costly than ESP [on average £1733.78 more expensive per participant (95% CI £1529.48 to £1938.06)], and, despite the QALY gain accrued by ACR participants [on average 0.0396 more QALYs per participant than ESP (95% CI –0.0008 to 0.0800)], this was not sufficient to prove that ACR was a cost-effectiveness use of NHS resources when compared with ESP (i.e. the ICER was above the recommended NICE threshold). Despite the ACR arm having fewer additional interventions, ACR was dominated by MUA, with higher mean costs and fewer QALYs. Therefore, the worse outcomes observed in the ACR arm than in the other two arms, along with ACR’s higher costs, make this treatment difficult to justify. The results of the base-case analysis remained robust to several sensitivity analyses that assessed the impact of areas of uncertainty around a number of study components.

There are two potential limitations to consider when interpreting these results. The first relates to the issue of missing data. Although the use of hospital forms reduced the number of incomplete data, the presence of missing data was unavoidable. We followed a comprehensive analysis to explore whether or not the MAR assumption is plausible given the actual missing data mechanism of the UK FROST data set. Our analysis showed that MAR is a plausible assumption fitting the UK FROST data set; therefore, multiple imputation was selected to handle missing data for the base-case analysis. Furthermore, the results were robust to alternative assumptions about the pattern of missing data, showing that the positive cost-effectiveness profile of MUA is robust to plausible departures from the MAR assumption. It is, therefore, highly unlikely that such assumptions regarding missing data will change the conclusions of our analysis. The second limitation relates to the duration of the study, which at 12 months might still be considered too short in terms of potential functioning. The clinical results showed that nearly 50% of the patients were only 3 points away from being in perfect health in the OSS, which, in turn, should influence positively the associated quality of life. Consequently, the clinical trends observed during the trial would also suggest that it is unlikely that any important difference in QALYs would emerge beyond the trial follow-up. It is notable that fewer QALYs were observed in the ACR group than in the other two treatment arms at 3 months, which is consistent with the results of the OSS. Conversely, the ACR group had more QALYs and higher OSS scores at 12 months. Moreover, although MUA had marginally higher estimates of OSS scores than ESP and ACR for the average treatment effect over 12 months, this also applied to those in the MUA arm accruing more QALYs over the duration of the study. The extra additional cost for MUA to maximise QALYs would be considered good value for money to the UK NHS at the NICE threshold of willingness to pay. Regarding costs, we are confident that important costs, including costs of complications, have been captured during the trial, especially for MUA, as the most significant risk of this is fracture and it is very unlikely that this happened beyond follow-up. It should be noted that sensitivity analysis to explore whether the results were sensitive to under-reporting of complications did not change the positive cost-effectiveness results in favour of MUA.

Evidence presented in this analysis relates to interventions conducted in the UK. However, given the pragmatic design of the UK FROST trial, the results are generalisable to other health-care systems when patients are referred to secondary care with frozen shoulder, where the decision is whether to offer surgery relatively early or to continue to control symptoms with physiotherapy.

Introduction

Qualitative research is often conducted before, during or after a clinical trial to explore the personal perspectives of trial participants and/or health professionals on, commonly, the trial feasibility, participation, the data collection process and the effects of trial interventions. 87,88 A comprehensive understanding of the subjective experiences to complement the quantitative evaluation of clinical and process outcomes in a trial will better inform patient-centred care and evidence-based practice. 88 The use of qualitative research methods in standalone studies or as part of clinical trials is gaining momentum in research on a wide range of musculoskeletal conditions. 89–91 However, there are very few published qualitative studies of people with frozen shoulder. 18,92 Currently, no qualitative exploration of trial participants’ and health professionals’ experiences is available within frozen shoulder trials. Therefore, we conducted a qualitative study embedded within UK FROST27 to provide trial participants’ and health professionals’ insights to guide clinical decision-making.

The objectives of the qualitative study were to explore (1) trial participants’ experience and acceptability of the treatments and taking part in the trial and (2) health professionals’ (surgeons and physiotherapists) experience of the treatments they delivered in the trial.

Methods

The UK FROST trial participants, surgeons and physiotherapists who had agreed to be contacted by the study team were invited to participate in the interviews. The trial participants were invited approximately 12 months after randomisation at the time of the primary end point of the trial. 8 This allowed time for post-surgical recovery and for trial participants to reflect on their experience of the intervention received. Men and women with and without diabetes were included. Surgeons who delivered both surgical interventions and physiotherapists who delivered physiotherapy in all three arms of the trial were invited.

The study information sheet and consent forms were sent to the trial participants, surgeons and physiotherapists by post or e-mail. Non-respondents were sent reminders the second and fourth weeks after the invitation was sent. Once signed consent for participation and audio-recording of interviews had been received, a convenient date and time were arranged for a face-to-face or telephone interview. A physiotherapy researcher (CS) trained in qualitative research methods and not involved in the delivery of UK FROST treatments conducted the interviews. Interviews with the trial participants were semistructured, using open questions about their experience of living with frozen shoulder and the treatments in the trial. An interview schedule (see Report Supplementary Material 21) was used that was developed following a literature review and discussions with the research team, people with frozen shoulder, a physiotherapist and a surgeon with expertise in this area. The interview schedule for surgeons and physiotherapists (see Report Supplementary Material 22) covered the routine clinical management of frozen shoulder, their experience of treating participants in the trial, their personal treatment preferences, and the barriers to and enablers of positive treatment outcomes for frozen shoulder.

We planned to interview to the point of theoretical saturation93 until no further useful categories emerged. We proposed to interview up to 45 trial participants and 15 health-care professionals.

The data were analysed in two ways:

1. The interviews were analysed using constant comparative methods. 94,95 This involves comparing similarities and differences and developing themes with a shared essence. The data were coded and categorised into themes by CS. Another qualitative researcher (FT) reviewed these themes, and the two researchers discussed and reached an agreement. CS used NVivo 11 qualitative data software (QSR International, Warrington, UK) to organise the analysis.

2. The World Health Organization’s International Classification of Functioning, Disability and Health (ICF)96,97 is a biopsychosocial framework used to conceptualise functioning and disability as a dynamic interaction between the following components: (1) body functions (denoted as ‘b’) and structures (denoted as ‘s’), (2) activities and participation (denoted as ‘d’), and (3) contextual factors (environmental factors denoted as ‘e’). Each component is arranged in hierarchal domains and has up to four levels of categories coded with the alphanumeric system. The first letter of the coding refers to the component followed by the first-level category (ICF chapter number designated for each component). For example, in d5, ‘d’ denotes the activities and participation component and ‘5’ denotes the chapter on ‘self-care’. A second-level category, d510, depicts a self-care problem, ‘washing oneself’. Third- and fourth-level ICF categories are also available for some components. For example, b2801 denotes ‘pain in body part’ and b28013 denotes ‘pain in the back’. A specific set of ICF categories that relate to common functional problems for different health conditions is available.

We aimed to map the problems reported by the UK FROST trial participants with a reference of second-level ICF categories (19 in body functions and structures, 34 in activities and participation, and 8 in environmental factors) identified in a previous study on chronic shoulder conditions, including frozen shoulder. 98

Results

Sixty interviews (trial participants, n = 44; surgeons, n = 8; physiotherapists, n = 8) were completed between August 2016 and January 2018. All interviews with the trial participants were conducted over the telephone. The majority (75%) of the interviews with surgeons and physiotherapists were telephone based and a few were face to face (surgeons, n = 2; physiotherapists, n = 2).

Interviews with trial participants

The flow diagram of interviews with the trial participants is presented in Figure 14. The characteristics of the trial participants are presented in Table 31. This includes participants who were allocated to ESP, MUA or ACR. All participants who were interviewed received their allocated treatment, except one participant allocated to MUA and another allocated to ACR, both of whom received ESP.

FIGURE 14.

Flow diagram of the trial participant interviews.

TABLE 31 - Characteristics of the trial participants interviewed

IQR, interquartile range.

Treatment arm	Sex (n)		Age in years, median (IQR)	Diabetes status (n)
	Male	Female		Diabetic	Non-diabetic
ESP	5	9	58 (51–63.5)	3	11
MUA	8	7	55 (53–57.5)	5	10
ACR	7	8	59 (53–69)	5	10

The five themes of the trial participants’ experiences are described in the following sections. There was nothing to indicate that the UK FROST themes found for men and women, and those found for people with and people without diabetes, were different.

Living with frozen shoulder

Trial participants described how frozen shoulder had a major impact on all areas of their life. They perceived it as a combination of painful, restrictive and disabling. Many were not able to identify what had caused their shoulder problem. Some reported a previous shoulder injury and others attributed the problem to tasks such as gardening and lifting heavy weights:

How would I describe it? Well it is one hell of a pain. You are restricted of movement. You cannot move it like say, very far out from your body. You try and lift something and even when you are grabbing something, you can feel the pain in your shoulder. Like say if you were trying to lift up a cup of tea, you could not. I mean it is very awkward.

61 years, female (ESP)

It was a dramatic impact on my life. I felt I could hardly use my left arm at all, so it was restricting everything that I did.

53 years, female (ACR)

Trial participants had mixed experiences of the onset of symptoms. Some had a sudden sharp pain or a constant dull ache that gradually progressed to reduced movements and function. During the course of frozen shoulder, they experienced pain and movement impairments, sleep disturbances, limitations in day-to-day activities, and restrictions on participating in leisure, work and social activities. Tables 32–34 present the narratives of the trial participants mapped to 19 ICF categories (five in body functions and structures, 12 in activities and participation, and two in environmental factors) from a previous study. 15

TABLE 32 - Participant-reported problems mapped to ICF categories: body functions and structures

Trial participants’ quotations	ICF category
It was like a stabbing pain; it was very severe73 years, male (ACR)	b280 sensation of pain
I couldn’t move my shoulder at all. So it was stuck to my side59 years, female (ESP)	b710 mobility of joint functions
Definitely lost my strength in my arm66 years, male (MUA)	b730 muscle power functions
I didn’t sleep at all48 years, female (ESP)	b134 sleep functions
I was generally getting very depressed. I would have happily, towards the end, I would happily have them amputate the arm50 years, male (MUA)	b152 emotional functions

TABLE 33 - Participant-reported problems mapped to ICF categories: activities and participation

Trial participants’ quotations	ICF category
Carrying was probably a bit of a problem because I couldn’t move that arm so well64 years, female (ESP)	d430 lifting and carrying objects
Reaching things from the tall shelves in the kitchen, reaching stuff out of the top of the wardrobe . . . And it’s things you take for granted really57 years, male (MUA)	d445 hand and arm use
I couldn’t drive my car even. I couldn’t change gear55 years, male (ESP)	d475 driving
I couldn’t lift my hand up, my arm up, you know, to wash my hair or anything in the bath. And eventually I couldn’t get into the bath properly76 years, male (ESP)	d510 washing oneself
It was the pain and the stiffness, particularly the stiffness, and the inability to address and attend to myself at the toilet. It was becoming a personal hygiene issue for me53 years, male (ACR)	d530 toileting and d520 caring for body parts
I couldn’t get myself dressed. I couldn’t get my tops on above my head. I had to wear slack things, so I could get my clothes on58 years, female (ACR)	d540 dressing
I couldn’t open anything, I couldn’t use a tin opener, so I couldn’t cook neither, I couldn’t do anything55 years, female (ACR)	d630 preparing meals
Doing housework was nigh impossible54 years, female (MUA)	d640 doing housework
I’m a coachbuilder by trade . . . I couldn’t work overhead, I couldn’t lift my arms up and I couldn’t stretch my arms out, I just couldn’t do it, so I wasn’t, I didn’t go to work, I was off for about 7 months55 years, male (MUA)	d850 employment
Well I decorate cakes. I do novelty cakes, all kinds of decorating. That literally stopped64 years, female (ACR)	d920 recreation and leisure
I like gardening and I was doing the decorating on my house, all sorts of jobs like that, sport, tennis, anything like that it sort of ruined everything really66 years, male (MUA)	d920 recreation and leisure
I couldn’t sleep on my left side any more and if I turned over in the night and tried to sleep on that side it instantly woke me up57 years, male (MUA)	d410 changing basic body position

TABLE 34 - Participant-reported problems mapped to ICF categories: environmental factors

Trial participants’ quotations	ICF category
. . . my wife had to help me put my socks on, things like that, get in and out of the shower, she had to do all the gardening, shopping, things like that. It was just really, really sore72 years, male (ESP)	e310 and e320 immediate family and friends; facilitators
I had a lot of support from my work, but the household work, I couldn’t manage because my husband had to do that for me53 years, female (MUA)	e310 and e320 immediate family and friends; facilitators

Participants described putting off making a GP appointment until their symptoms had worsened, and some described delays in receiving NHS care. Participants thought, in retrospect, that having a quicker NHS care pathway in terms of diagnosis and further specialist referrals was important. Participants also emphasised seeking early medical help and referrals from their GP:

I would say go straight to your doctor and get them to refer you. That would be the first thing because it doesn’t just seem to go away of its own accord which I possibly thought initially and that’s why I delayed in going to the doctor’s in the first place.

64 years, female (ESP)

Just speed it up. It was, I think I went to see the GP at the beginning of November, and it wasn’t really till the following January before I got any kind of treatment other than pain relief, by which time I’d lost all movement.

50 years, male (MUA)

Some participants reported that they had had a range of treatments before the trial, such as painkillers, physiotherapy, acupuncture and steroid injections, whereas others had had no treatment at all. Participants felt that painkillers and injections used before the trial had not helped them. Similarly, participants found that physiotherapy treatments had not helped to increase their range of movement, partly because of the difficulty that they had exercising as a result of the pain:

I had a couple of sessions of physio before I was referred to [consultant’s name]. And I will be honest with you; the physio basically said there was not much they could do for me at the time.

59 years, male (MUA)

Participants were concerned that they were stuck with the disability from frozen shoulder and were eager to get it sorted. This was their main motivation for taking part in UK FROST:

Just the fact that my life would seem to be on hold because I couldn’t function properly, you know that was my main concern, I didn’t want to be left like this permanently, I wanted something done about it, I didn’t want to be continually taking painkillers, which I seem to be living on just to ease the pain, and I didn’t want to be doing that. I thought, ‘I need to get something done’, that was my main concern.

59 years, female (ACR)

Improvements in outcomes and participant satisfaction following the trial

Trial participants considered pain relief and the return of shoulder movement and function to be important treatment outcomes:

Going back to normal . . . When you had nil pain and full flexibility and movement within your shoulder. No sleepless nights and that.

62 years, female (ESP)

Trial participants said that they had experienced significant pain relief after their treatment. Participants in the ESP arm said that the steroid injections had reduced their pain and allowed them to start physiotherapy:

When I went to the surgeon I was injected into my shoulder and the pain down my arm that more or less went straight away. After that I went on that course for frozen shoulder, for therapy, and I went for 12 weeks running once a week. It seemed to go, and it’s been fine since.

76 years, male (ESP)

So at the beginning I said the pain was 10 and now after all my physios, I’d say it was, I’d say it was about 2 now.

56 years, female (MUA)

I mean the pain in the beginning was just horrendous, it was really, really sore, really painful but after I’d had the physiotherapy, it was . . . I’ve got no pain at all now.

55 years, female (ACR)

Trial participants in all treatment arms reported increased shoulder movement:

Virtually full movement. My shoulder is fine as far as movement is concerned.

68 years, male (ESP)

Basically had all my full movement back.

58 years, male (MUA)

I got my life back again. I can walk my dogs. I can hold the dog leads. I can do my shopping. I can carry things again.

58 years, female (ACR)

Trial participants described how the physiotherapy sessions (ESP and PPP) had helped to improve their shoulder movement:

I could tell initially straightaway that my movement was starting to come, within a few days I could tell a difference of doing the exercises and as the weeks went on, it was just got better and better and by the time the 12 weeks was up, I virtually had full movements with no pain or anything, it was brilliant!

64 years, female (ESP)

After a few days I was doing my exercises and I was quite surprised already how much movement I had back and then it was regular physio appointments up at the hospital just to keep moving things around and that went really well . . . the physiotherapy was actually really, really beneficial.

45 years, female (MUA)

I felt that the physiotherapy I received was marvellous and improved the range of movements or showed me how to keep that range of movements much quicker than they did on the right-hand side, so I felt that everything went along fine, and I’ve got no complaints at all, none.

53 years, female (ACR)

Participants in the ACR group felt that their recovery in terms of pain and movement had been quicker than they had expected. Some had experienced improvements as early as after 1 or 2 weeks of physiotherapy after surgery:

It is almost like you have had a quick fix to fix your shoulder then you move on and I think personally for me because the surgery went very well and almost after a couple of weeks I was back to normal.

44 years, male (ACR)

Participants in all treatment arms said that their ability to do routine activities had improved:

I can lift my arm above my head now, you know? I can carry stuff, and I can lift it above my waist, and I can actually go swimming, you know? I can swim now.

54 years, male (ESP)

My little everyday things have come back; I have come back, yes.

52 years, female (MUA)

I can do everything – there’s nothing that I can’t do; I can wash my back, I can put my bra on, fasten it at the back, I can fasten my skirt at the side and the back now, there’s nothing I can’t do before I had the frozen shoulder everything I could do then I can now do again.

59 years, female (ACR)

In spite of achieving pain relief and improved function, participants experienced mild and occasional pain and restrictions during certain end-range activities:

I do get occasional pains in my arm, but it’s very mild and yeah, I’m aware that I still don’t have full movement in my shoulder, but it is much better than it was.

45 years, female (ESP)

There’s still a wee bit of pain there, but it’s nothing. You see, I’m not concerned about it.

55 years, male (MUA)

I still get twinges now and again but it’s nothing, and that’s only when I try to put my arm right around my back.

73 years, male (ACR)

Trial participants were satisfied with the UK FROST treatments they had received:

I’m absolutely delighted with the treatment that I was given. I feel as though it did everything that I wanted it to do and expected it to do.

64 years, female (ESP)

I would say I’m like, 100% happy with the treatment, and the study was, like, 100%, it’s good, I didn’t mind it.

56 years, female (MUA)

Very satisfied. I have no complaints at all.

61 years, male (ACR)

However, two participants were not satisfied with ESP. One had been treated by a private physiotherapist before the trial and did not improve after physiotherapy during the trial. The other was not pleased that the exercise sessions had been supervised by an unfamiliar physiotherapist:

Waste of time . . . I wouldn’t be recommending it to a friend . . . Well because by the time I got to see the NHS physio, the private physio had already, if you like, done the hard work and they just had to, if you like, pick up the pieces and keep it OK, and they didn’t.

50 years, female (ESP)

May I say 50%? That is mainly because when I saw the physio one to one, I was 100% happy and then when I went to the gym and the physiotherapist sort of left you to your own devices, they didn’t really know who I was, why I was there, and they certainly didn’t. It was basically just like going to any old gym and being supervised by someone who didn’t know you from Adam. It was very, very disappointing.

59 years, female (ESP)

Trial participants’ adherence to home exercises

Participants found that the exercises were difficult to begin with but eased off in subsequent sessions:

It [exercise] was difficult and painful. But I could tell week on week the pain was reducing, and my movement was increasing. So it was obviously working quite well.

57 years, male (MUA)

However, they were aware of the benefits of exercise and persevered with doing their home exercises regularly:

I did persevere, and I was doing what I was told, which was obviously you’ve got to have the pain to get back to normal.

48 years, female (ESP)

Most participants said that they had not continued their home exercises after the trial because they felt that they had regained their normal shoulder function. They shifted from doing the structured home exercise regime to daily functional activities to keep their shoulder mobile. A few participants did some shoulder stretches occasionally:

I’m working with my shoulder all the time so I’m not doing the exercises that the hospital gave me, because I’m working my, I’m swimming, I’m doing . . . I go on long walks, I take the dog out and what have you, so I’m using my arm.

54 years, male (ESP)

Trial participants’ treatment preferences before and after the trial

Participants had various treatment preferences before the trial. Some of the participants allocated to MUA and ACR felt that physiotherapy would be ineffective because it had not worked for them previously or they felt that physiotherapy would be difficult with their painful shoulder. MUA was perceived as less invasive and ACR was perceived as an effective treatment. A few preferred physiotherapy so that they could avoid the risks of surgery. Some did not have a particular preference at all. Three participants with diabetes felt that it would take longer to recover after ACR because of their diabetes:

I’m not really too sure why I wouldn’t choose physiotherapy. I just think surgery seems a more final option. Physiotherapy, it might work, it may help, it may not. But to me, if I was given surgery, the surgery would work. I had more faith in the surgery working than the physiotherapy itself.

53 years, male (ACR)

Well the first option would have been more intense physiotherapy which I didn’t find would have been successful because I’d already had physiotherapy. The second option was to have been invasive surgery which means cutting open, an actual surgical procedure which I wasn’t that keen on to be honest with you.

55 years, female (MUA)

Well I didn’t want to go to surgery or anything like that, so I just had the needle in my shoulder. I don’t think I’d have wanted to go surgery at my age.

76 years, male (ESP)

Just I didn’t think, as I said, I didn’t know if any was better than the other, I just went along with what was there.

55 years, male (MUA)

Despite preferences at the outset, at the end of the trial there was a sense from participants that they would choose the same treatment that they had been allocated, particularly those who had received ACR:

Oh no I would have the same treatment. As I say I was only there for 10 weeks and I mean, on my 11th week I was still hell of a lot better.

61 years, female (ESP)

Probably the same, because I had to get it moving to start with, it just felt as though it was never going to move. So again, the manipulation, it kind of kick-started it and got it moving, because I honestly thought it was never, ever going to move.

53 years, female (MUA)

I think if it happened to me again, I would be looking to be referred for keyhole surgery again. I think it was an excellent course of treatment and if it had to happen again, that’s the treatment I would want.

53 years, male (ACR)

A few participants in the ESP and MUA arms wanted to choose ACR for a permanent and quicker solution for their shoulder problem:

If I had a recurrence of the frozen shoulder in the same joint, I would obviously look for alternative treatment for the simple reason because obviously that treatment, although it alleviated the symptoms, hasn’t completely got rid of the symptoms then because if it recurs. So you would look for a permanent solution . . . I would think if it came back again, I would prefer to have the keyhole surgery, yes.

68 years, male (ESP)

Trial participants’ experience of participating in the trial

Trial participants had altruistic and personal reasons for participating in the trial. They desired to help other people and contribute to research, and some expected to have their shoulder problem treated quickly:

Because I’m all in favour, if you can do something to help other people not go through the misery that you’ve been through, and gone through, then I would do it.

62 years, female (ESP)

That I would be seen to sooner than if I didn’t do it. And I would have the opportunity to have any treatments much sooner than being on the waiting list. It’s not a very nice reason for you to hear, but that’s what I did it for.

53 years, female (ACR)

Trial participants found the trial questionnaires relevant and simple to complete. A few felt that the questions were lengthy and repetitive. Some had difficulties answering the questions about comparing their health status with that of the previous month:

It was quite easy, they were simple questions, and I just sort of flew through the questionnaire with no problem.

64 years, female (ESP)

Very repetitive [laughter]. Very repetitive. It went on and on and on. Just when you thought you’d finished there would be another page added on. They are a bit of a pain really.

57 years, male (MUA)

Trial participants said that the physiotherapists who delivered the physiotherapy sessions (ESP and PPP) were supportive and helpful:

My physiotherapist was very nice and he didn’t push me to do anything that wasn’t in my ability and yeah, we just took it at a nice, steady pace.

41 years, female (ESP)

The physiotherapist really knew what she was doing and straight away assessed exactly where I was at and what I needed to be doing, and that worked really well.

50 years, male (MUA)

They also liked seeing the same physiotherapist so that good connections and rapport could be developed throughout the programme:

I think seeing the same person is always helpful because otherwise it must be time saving as well because you haven’t got to read up on the case every time and you get a good rapport like that.

53 years, male (MUA)

Participants in the MUA and ACR arms said that the surgical procedures were explained and went well:

The day in the surgery, I was told everything, how long I would be off work for, and I had targeted physiotherapy immediately after, I knew that was part of it. I attended on the day of my operation; everything went very, very smoothly.

58 years, female (MUA)

It was just a case of just sitting around, reading books, talking with other people, some people had been there two to three times and had various operations, everyone was just chatting and making everyone feel at ease with you know their own experiences . . . And then it was a case of get in, swabbed up and on the trolley through to the operating theatre, operation I believe obviously went fantastic, went really well. It was just a case of waking up, recovery.

44 years, male (ACR)

Trial participants felt that their treatment packages were well co-ordinated and did not require major modifications. Two MUA participants with diabetes suggested providing more information on the effects of pain block and steroid injections on blood sugar levels. A few participants in the ESP and ACR arms felt that the exercises were time-consuming:

Well personally I found it quite difficult to make sure I had them that many times in the day. I’m not too sure how somebody working or having a family could actually manage to fit it in because as I say, by the time going towards the middle to the end of the programme . . .

64 years, female (ESP)

Yeah that was quite difficult because it takes up quite a bit of time and it is quite tiring. I’m not the fittest person and I did find it quite tiring to do all the exercises required.

53 years, male (ACR)

Interviews with surgeons and physiotherapists

The flow diagram of the surgeons’ and physiotherapists’ interviews is presented in Figure 15. The characteristics of surgeons and physiotherapists are presented in Table 35.

FIGURE 15.

Flow diagram of surgeons’ and physiotherapists’ interviews.

TABLE 35 - Characteristics of the surgeons and physiotherapists interviewed

IQR, interquartile range.

Health professional	Sex		Median (IQR) years of experience in treating shoulder conditions
	Male	Female
Surgeons	7	1	11 (7–16.25)
Physiotherapists	0	8	13.5 (12–15.75)

The four themes from the interviews with surgeons and physiotherapists are described in the following sections.

A stage-based approach in routine treatment of frozen shoulder

Surgeons and physiotherapists described a stage-based treatment approach (from conservative to surgical interventions depending on the severity and duration of symptoms) in their routine practice. During the early painful phase, surgeons and physiotherapists thought that pain control with steroids would be the priority so that exercises could be initiated. Physiotherapists also provided patients with education as part of their treatment plan:

A humeral steroid injection is beneficial for pain relief. It allows them to do the physiotherapy. So, I do use that a lot.

Surgeon 6

I would always tend to go down the glenohumeral joint injection first to see if that settled things down and to give me a window where I could then push them with regards to their exercises.

Physiotherapist 1

And then it’s education; tell them all about the condition, what we know about the condition, let them know that it is going to get better over time, and it may not resolve fully and so on. So we’ll basically educate them, reassure them.

Physiotherapist 5

Surgeons felt that exercises during the early, painful, phase may aggravate the pain but that exercises after surgery are important for recovery:

Physiotherapy in isolation without an injection I’ve found most patients present having had really pain that’s not manageable because they’re trying to stretch and rehab a painful shoulder from frozen shoulder just increases their pain.

Surgeon 2

We do believe in physiotherapy afterwards. One is to have specialised physiotherapists who can allay their fears and talk about their fitness, talk about their recovery, talk about the time scales and the pain. That’s one compliment of the physiotherapies is some health professionals who can talk to them on a regular basis.

Surgeon 3

During the stiffness-dominant and postoperative phases, physiotherapists focused on improving shoulder movements, strength and function. They prescribed an intensive exercise regimen that included vigorous shoulder stretches, joint manipulation/mobilisation and home or gym-based exercises:

I know that pain is not a problem then I can push them, I can do the more vigorous stretches, I can manipulate their joints or mobilise their joints and I put them on a more gym-based or exercise-based programme . . . I will go and more push them towards reaching their more functional goals and more towards achieving their return to work and that kind of thing task.

Physiotherapist 4

For people with severe or longstanding symptoms and those resistant to conservative treatments, both surgeons and physiotherapists considered surgery as the final treatment option. For most surgeons, ACR was the default procedure as it resulted in faster recovery and low risk of humeral fractures, and they were familiar with the procedure:

So, if a shoulder is extremely stiff, tight and quite painful I would probably avoid doing a manipulation on these patients and my preference would be ACR if the longevity of the stiffness and the symptoms is quite prolonged again my preference would be an ACR.

Surgeon 6

Generally, the people that don’t respond to physio often will end up seeing a surgeon at some point. So the tricky more longer-term frozen shoulder patients are the ones with diabetes that tend to have more problems and end up requiring intervention.

Physiotherapist 2

Physiotherapists felt that the UK FROST physiotherapy programmes and the exercise booklet gave them flexibility to choose exercises that were compatible with their routine practice. A few suggestions were made, for example structuring the treatment components to suit the stages of frozen shoulder, spreading out the 12 weekly sessions over 6 months and having group sessions:

I think the interventions that were on the booklet were what I would use generally. There was always an option there for me to tick off what I would do so I was in agreement with the options that were there and in agreement with the options that they actually, didn’t want you to use.

Physiotherapist 2

I think group sessions would be really useful because patients get a lot from each other, and having experience group sessions with other clients with different pathologies, you know, they find that really reassuring . . .

Physiotherapist 5

Physiotherapists also commented on the feasibility of the UK FROST physiotherapy programmes in the NHS, and there was a sense that it would be difficult to deliver the number of UK FROST physiotherapy sessions in their routine practice:

We don’t normally get the luxury of being able to see patients as often as the FROST trial was letting us. I think it was 12 treatments we could have overall.

Physiotherapist 2

. . . they [trial participants] were seen with the 24 hours post surgery and they had 12 sessions which is a luxury because in our trust, that is never, not going to happen and that never used to happen.

Physiotherapist 4

In general practice, well if somebody is improving and they’re self-managing, then we don’t need to be seeing these patients every week and our service would not allow us to be able to see them every week.

Physiotherapist 7

Treatment expectations and preferences

Surgeons and physiotherapists had mixed treatment expectations. Although surgeons said that they maintained equipoise, they also said that ESP would not be as effective as surgery:

They usually come to me and say, ‘What would you do doctor?’ I just be honest with them that so far, I have done this, and my results are reasonable, but I wouldn’t say that this is the only answer. There are other answers that are equally valid and have to be tested . . . It is an ethical thing to do because all three of the treatments are valid and accepted treatment for frozen shoulder.

Surgeon 3

My expectation is that physio won’t work. My expectation is that the other two are probably equivocal.

Surgeon 7

Some surgeons and physiotherapists expected similar outcomes across treatment arms. A few physiotherapists felt that the surgical arms, especially the ACR arm, would perform better than the ESP arm. Although some considered MUA an outdated intervention with a risk of injury, some felt that it would be comparable with ACR. Some physiotherapists mentioned that post-surgical soreness is common with ACR:

I expected my patients to get better on whichever arm they chose.

Surgeon 3

I’m expecting the MUAs to be surprisingly better than I would expect. I think the arthrolysis do great anyway and the physio is an unfair one, because if we’re seeing them over such a long time the natural history of the frozen shoulder is it will get better. Going against my own profession here; is it the physio that made it better or is it just time.

Physiotherapist 6

I’ve had experience of patients doing well in them in them all and equally patients not doing well in them all. I thought they were; I don’t think they were randomised equally. I thought there were more randomised to surgery. It was more for surgery than physio. So no, I didn’t have any predisposed feelings about it at all.

Physiotherapist 8

Surgeons and physiotherapists would have preferred that hydrodilatation had been one of the UK FROST trial arms. Hydrodilatation was described as an easy-to-use, less invasive and inexpensive procedure and an alternative for reducing NHS waiting lists for surgery:

I’d definitely have a hydrodilatation group because part of your trial is trying to work out if the cheaper operation is better than the more expensive operation and hydrodilatations probably gained quite popularity since we started the trial design and it reflects current practice.

Surgeon 5

It needs hydrodilatation in it. I personally think it gets really good results on a big bulk of patients and it’s a wasted opportunity to have done this study and not have that as one of the arms.

Physiotherapist 6

Factors that influence treatment outcomes

Similar to trial participants, surgeons and physiotherapists perceived that pain relief and improved movements and function were important outcomes. Pain relief was the priority outcome for physiotherapists:

Better in a sentence? That their pain is resolved, and their movements improve, and they return to full function within their daily living, both at work and for recreation and home. That’s what I expected to achieve in any of the arm of the trial.

Surgeon 3

So yeah, pain relief first and foremost and then everything else after that; increasing their range of movement, function, return to activities, whether it be sport or work or hobbies and so on. But yeah, definitely pain is the number one.

Physiotherapist 5

I think pain is always the predominant thing with these patients. They just want someone to do something to help with the pain.

Physiotherapist 7

Surgeons and physiotherapists felt that diabetes negatively affects treatment outcomes:

Well my experience with diabetics they have been very bad for a hell of a long time, OK, and that’s again maybe just my experience, I haven’t measured that experience, but I’ve seen plenty of them of male diabetics that say, ‘Well, I’ve been stiff for three years’, or two-and-a-half years, that’s not uncommon.

Surgeon 8

I think my understanding is diabetic patients are slightly more prone to developing a frozen shoulder. It seems to be potentially more complications, slightly more resistant to treatment. And the chances of them maybe developing it again are slightly higher, I think.

Physiotherapist 2

They also felt that participants engaging with treatments and having positive expectations leads to better outcomes:

With any treatment they do they have to engage, and that is something that I emphasise to my patients. So I spent a lot of time in the initial consultations giving them the knowledge. So what makes them better, it’s the patient themselves and their knowledge.

Surgeon 3

I think the expectations and belief is probably the most noticeable factor that affects people’s outcome; if they believe something is the right thing, the best thing for them, they seem to do well.

Physiotherapist 5

Perceptions about trial participants’ experience

Surgeons and physiotherapists felt that the trial participants were happy to be involved in UK FROST. At the same time, participants came to the trial with fixed ideas about the treatment they wanted:

Once they had consented to be part of the study, and they had no problems because they were equally welcome on what treatment they would get . . . all those who once we enrolled them on to the study, were OK with that.

Surgeon 6

So, we saw a lot of frozen shoulders coming in, but a lot of them had fixed ideas of what treatment they wanted. They didn’t want surgery yet, or they didn’t want to take time off work was the other one, but less so. The standout one was they didn’t want an operation, or they wanted to try physiotherapy and injection and then they would opt for surgery. They wanted it to be continuum like that, not a one or the other.

Physiotherapist 8

Surgeons and physiotherapists said that some people declined to take part in the trial because their previous physiotherapy had not worked and so they did not want to be randomised to receive ESP:

Many of them they say, ‘Look, I would love to contribute to the greater good and be involved in clinical trials, but I’ve come to the point that I will not consent for physiotherapy if I was randomised to that’.

Surgeon 8

Like I say, if patients have already had physiotherapy, some of those patients have not wanted to be recruited at risk of repeating what’s already not worked. So that was quite awkward to do.

Physiotherapist 6

Physiotherapists said that the UK FROST interventions were well received. Surgeons and physiotherapists said that participants were surprised with the number of PPP sessions they received during the trial. Surgeons and physiotherapists described how a few participants randomised to ESP felt that they were not improving:

They seem to be quite happy with the intervention [ESP] generally. If they had had an injection they were happier because their pain level was better and they were able to tolerate the exercises a bit better.

Physiotherapist 2

I think most of them were with my experience, the patients were really, really very happy, the ones who went to the manipulation as well as arthroscopy capsular release.

Physiotherapist 4

Discussion

Conclusion

This qualitative study has provided a fuller understanding of the perspectives of UK FROST trial participants and health professionals and has complemented some of the key findings of the main trial. Our findings indicate that although the content of the physiotherapy interventions was acceptable to trial participants and health professionals, they also highlight concerns about delivering this intensity of treatment within the constraints of the NHS. Future trial designs would usefully include qualitative research as part of intervention development to ensure the feasibility of the interventions in the NHS. More primary qualitative studies on people with frozen shoulder are needed to integrate patient perspectives into informing patient-centred care and shared decision-making.

Principal findings of clinical effectiveness

Cost-effectiveness

The base-case economic analysis showed that at 12 months MUA was, on average, £276 more costly per participant (95% CI £65.67 to £487.35) than ESP. MUA was slightly more beneficial in terms of utilities than ESP (on average 0.0396 more QALYs per participant than ESP, 95% CI –0.0008 to 0.0800). The ICER for the ITT approach in the imputed data set between MUA and ESP was £6984 per additional QALY. ACR was more costly than ESP (on average £1733.78 more expensive per participant, 95% CI 1529.48 to 1938.06). Despite the QALY gain accrued by ACR participants (on average 0.0103 more QALYs per participant than ESP, 95% CI –0.0304 to 0.0510), the ICER was over £100,000 per additional QALY. ACR was more expensive than MUA and resulted in slightly fewer QALYs. Therefore, given the limited differences in outcomes observed in the ACR arm compared with the other two treatment options, along with the much higher costs, it is difficult to justify ACR as a first-line treatment option on evidence of cost-effectiveness. MUA was the intervention most likely to be cost-effective at a £20,000 per QALY threshold (MUA 86% > ESP 14% > ACR 0%).

The results of the base-case analysis remained robust to several sensitivity analyses that assessed the impact of areas of uncertainty around a number of study components. This included our analyses being robust to missing data and the assumptions around missing data. However, the cost-effectiveness of MUA compared with ESP was sensitive to the addition of non-shoulder costs and the broader perspective that included private treatment costs and days off work. A key cost driver in these analyses was days off work, at £113.80 per day. During the 12-month follow-up, participants allocated to the ESP arm had a median of no days off work, participants allocated to the MUA arm had a median of 6 days off work, and participants allocated to the ACR arm had a median of 2 weeks off work. This potentially could be related to quicker access to treatment for ESP participants, and may be important in patient decision-making. The analysis was also limited to a 12-month follow-up. However, as the results of the OSS at 12 months show that 50% of the participants were only 5 points away from regaining full function, this suggests it is unlikely that an important difference in QALYs would emerge during longer-term follow-up. Regarding costs, the important costs of treatment, and complications, were expected to have been captured during the 12-month follow-up.

Qualitative study findings

Trial participants described how frozen shoulder had a major impact on all aspects of their life. They were keen to get their shoulder problem resolved, which motivated them to participate in the trial. They thought that seeking early medical help and having a quicker NHS care pathway were important. In general, trial participants were satisfied with the UK FROST interventions and found them acceptable. They reported improvements in pain, shoulder movements and function. Participants who received ACR described recovering more quickly than they had expected. Surgeons and physiotherapists followed a stage-based treatment approach in their routine practice. Both groups of professionals felt that people with diabetes tend to have poorer outcomes. They suggested that hydrodilatation could have been a treatment arm of the trial. Both groups of professionals commented that some people who had received previously ineffective physiotherapy had not wanted to take part in the trial.

Trial validity and minimising bias

Various measures were taken to ensure trial validity and minimise bias, or to explore the potential for bias, of which some are discussed here.

The secure randomisation method helped to ensure comparability in the characteristics of the participants in the three treatment arms. A greater number of participants in the MUA arm were currently in paid work, and there was some arm imbalance in having had a similar shoulder problem on the opposite side from the reference shoulder. A sensitivity analysis of the primary outcome, which included employment status as an additional covariate, found that the results were similar to those observed in the primary analysis. The use of unequal random allocation reflected differential treatment effect expectations. The greater number of participants allocated to the surgery arms than allocated to physiotherapy allowed for a larger effect size to justify the higher costs of and potential risks associated with surgery. 27

To help ensure a good standard of care, surgeons were advised to use techniques with which they were familiar, which also helped to avoid learning curve problems. Most operations were carried out by consultant surgeons for both surgical procedures, and most operating surgeons routinely performed both procedures up to once per month. Both ESP and PPP were delivered by qualified physiotherapists who were predominantly band 6 and treated two or three frozen shoulder patients per month. It is unlikely that not including students or assistants in delivering physiotherapy introduced bias, as this was applied consistently across all treatment arms. The number of physiotherapy sessions across the three trial arms was similar. All participants were provided with standardised, written physiotherapy advice detailing the home exercises they needed to perform.

There were low levels of attrition in the completion of the primary outcome and there was no evidence of differential dropout in any of the treatment arms. There were no systematic differences in baseline characteristics compared between those included in the primary analyses and all randomised participants. The use of a mixed-effect, repeated measures analysis model that included data from any participants with at least one valid follow-up meant that only 6% of participants were not included in the primary analysis. This also increased the statistical power of the analyses compared with the single time point comparison used for the sample size calculation. 27 There was a ceiling effect at 12-month follow-up, in that 24% of participants had regained full function (top OSS score). Although it is encouraging that participants across all three treatment arms were recovering well, it could be argued that this limited the potential to find clinically meaningful differences at the primary end point.

Given the nature of the trial treatments, blinding participants and clinicians to treatment allocation was not possible or desirable in this pragmatic trial. The statistician and health economist were blinded to group allocation until after the data had been hard locked and no further changes could be made. The lack of any subgroup effect of participant baseline preferences on treatment outcome (using the OSS) may in part mitigate concerns about introducing bias from a lack of blinding in the participant self-reported primary outcome.

It could be argued that a potential bias of the primary analyses concerned the different waiting times for treatment delivery, with ESP starting at around 14 days, and MUA and ACR starting at around 57 days and 72 days post randomisation, respectively. This could have benefited those in the ESP arm at the 3-month follow-up when compared with participants who had not yet received a trial intervention or who were recovering from a surgical procedure. To account for different waiting times, participants also completed the OSS on the day of treatment and 6 months later. Reassuringly, the OSS appeared to stay stable between baseline and the start of any of the treatments. Analyses incorporating all data were largely consistent with the primary analysis findings. This analysis is limited, however, as it reflects treatment effects at pragmatic follow-up times accounting for the different outcome trajectories, rather than observing what would have happened had all three trial arms been delivered at similar times.

A further potential threat to study validity is non-compliance because the treatments were not delivered as planned to all participants. This could dilute the treatment effect observed in the ITT primary analysis. Only 16 participants (3%) crossed over to a different trial treatment, and 17 (3%) received an alternative treatment that was not a trial intervention (e.g. steroid injection only). However, around 20% of participants across all three trial arms did not complete their treatment, according to our defined criteria. This was expected, as the natural history is for frozen shoulder to resolve,133–135 particularly for participants awaiting MUA or ACR, who did not receive their allocated treatment because the waiting time was 57 and 72 days, respectively. For ESP patients, although up to 12 sessions of physiotherapy were encouraged, based on existing evidence,38,39 we used strict criteria to define ‘compliers’ as having to complete eight or more sessions, or fewer if the participant and/or physiotherapist were satisfied with their progress. To explore the effect of non-compliance on the OSS at the primary end point of 12 months, an instrumental variable regression was undertaken comparing ESP compliers and those who would have complied in the two surgery groups. ESP outcomes were lower at 12 months, as in the primary analysis, but this was neither statistically significant nor clinically important, with a difference of < 2 points on the OSS. Interestingly, unadjusted OSS at 12 months found that participants who complied with ESP scored, on average, 5 points higher on the OSS than those who did not, which is potentially clinically important. Finally, a steroid injection was delivered for all who received completed MUA and for 80% of patients randomised to ESP, compared with 28% of those who received ACR, who had a steroid injection at the surgeon’s discretion. This could be argued to be a bias against ACR, but it is consistent with our finding from a survey of 53 surgeons, carried out when developing the trial protocol, that only 30% routinely provide a steroid injection alongside ACR. 27 Therefore, this result reflects clinical practice.

Applicability of results

Application of the trial results to clinical practice

The characteristics of the trial participants and duration of their symptoms were as expected. Therefore, at the primary end point of 12 months for the primary outcome, it was encouraging to find that participants in all three treatment arms had improved considerably since they had been enrolled into the trial. Although participants in the ACR arm did a little better, the mean differences between the treatment options were not of the magnitude of the minimally clinically importance difference that we sought. By contrast, at the earlier time point of 3-month follow-up, participants in the MUA arm did a little better than participants in the ACR arm, and a clinically important difference was approached between participants in the ESP and ACR arms in favour of the former. The timing of delivery of these interventions could explain these findings. However, the analyses that attempted to account for variation in waiting times illustrated that the differences between treatment options were smaller than those from the primary analyses, except for a further benefit in favour of ACR compared with ESP, and smaller still than the minimally clinically important difference. The findings for the secondary outcomes showed a similar pattern. Therefore, there is evidence that ESP has potential early benefits compared with ACR. Although it could be argued that these benefits are confounded by waiting times and the surgical procedures being performed in more carefully selected participants whose frozen shoulder had not resolved naturally, pragmatically, this reflects the quicker access to this intervention in clinical practice, with waiting times for surgery likely to be longer than those during the trial. Importantly, participants in the nested qualitative study commented on the need for their frozen shoulder to be resolved and so ESP offers quick access to an effective treatment. Otherwise, the evidence is inconclusive on whether any of the three treatment options is superior in terms of primary and secondary outcomes. For these findings to be replicated in clinical practice, ESP with a steroid injection would need to be delivered as rigorously as it was in UK FROST. Although this might be potentially challenging in routine care in the NHS, the effective delivery of ESP could prevent the ‘opportunity cost’ of using theatre resources for MUA or ACR, and avoid the need for PPP.

These findings also apply to people with diabetes, as the presence or absence of diabetes in participants did not have a statistically significant effect on treatment comparisons.

All three treatments had similar completion rates, at around 80%. There was an optimal release in 92% and 98% of participants who had MUA or ACR, respectively. Overall, among those allocated to ACR, this treatment was more definitive, with further treatment required for 4% of participants. Nearly twice as many participants in the MUA arm (7%) required further treatment and even more in the ESP arm (15%) required further treatment. Serious complications were rare, although the ACR arm was relatively less safe (4%). Only two participants allocated to MUA experienced a serious complication (1%). One of the participants in the ACR arm who was diagnosed with a deep-vein thrombosis actually received non-trial physiotherapy. Therefore, only a marginal difference was seen in the safety profiles of MUA and ESP, with the latter arm having half the participants. The systematic review that was undertaken in an attempt to further underpin UK FROST findings found that most of the comparisons between treatments were limited by the availability of single trials, often in a single centre with small sample sizes, and the considerable presence of bias and the heterogeneity of treatment interventions. None of the included trials helped to produce conclusive findings about the effectiveness of the interventions evaluated in UK FROST. Although hydrodilatation has increased in popularity in clinical practice, and although research in this area has also increased, evidence of its effectiveness was inconclusive in the systematic review. In this context, hydrodilatation will need to be considered carefully as a treatment option for patients with a frozen shoulder.

Finally, ESP was the least expensive intervention, as most participants in this arm did not require a surgical procedure. MUA was the second most expensive treatment option, with participants spending one-third of the time in theatre compared with ACR, for which the latter was by far the most expensive option. MUA, however, resulted in more QALYs over the duration of the study than either ESP or ACR. This meant that MUA had an 86% probability of being a cost-effective intervention at the threshold of £20,000 per QALY threshold if commissioners of services would be willing to pay £6984 per additional QALY.

Conclusion

UK FROST has provided robust clinically relevant evidence that none of the three treatments was clearly superior in improving patient-reported shoulder pain and functioning at 12 months. Our specifically designed ESP pathway can be accessed quickly in the NHS and has lower costs. However, the likelihood of further treatment being required is higher with ESP than with the other two interventions. MUA produced the most QALYs overall. At a modest additional cost, MUA is the most cost-effective option to the NHS, with an ICER of £6984 per additional QALY. Patients who receive ACR are the least likely to need further treatment, but ACR is associated with relatively higher risks and costs. These findings should help inform treatment decisions by patients, providers and commissioners of care.

The conclusions should be interpreted with some caution given the potential confounding effect of waiting times to surgery, which have also lengthened since the trial. This may have meant that participants with a more resistant frozen shoulder were those who received an operation. It also could be challenging to implement the ESP pathway in clinical practice with the same optimal timing of access and to the same standard of delivery as in UK FROST.

Collaborators: principal investigators

Mr Philip Ahrens (Royal Free Hospital NHS Foundation Trust).

Miss Alison Armstrong (University Hospitals of Leicester NHS Trust).

Miss Cheryl Baldwick (Northern Devon Healthcare NHS Trust).

Mr Amit Bidwai (Sherwood Forest Hospitals NHS Foundation Trust).

Mr Andrew Brooksbank (Glasgow Royal Infirmary, NHS Greater Glasgow and Clyde).

Mr Muhammad Butt (The Dudley Group NHS Foundation Trust).

Mr Jamie Candel-Couto (Northumbria Healthcare NHS Foundation Trust).

Mr Charalambos Charalambous (Blackpool Teaching Hospitals NHS Foundation Trust).

Mr Mark Crowther (North Bristol NHS Trust).

Mr Steve Drew (University Hospitals Coventry and Warwickshire NHS Trust).

Mr Sunil Garg (James Paget University Hospitals NHS Foundation Trust).

Mr Richard Hawken (Torbay and South Devon NHS Foundation Trust).

Mr Cormac Kelly (The Robert Jones and Agnes Hunt Orthopaedic Hospital NHS Foundation Trust).

Mr Matthew Kent (Royal Liverpool and Broadgreen University Hospitals NHS Trust).

Mr Kapil Kumar (Woodend Hospital, Aberdeen, NHS Grampian).

Mr Tom Lawrence (University Hospitals Coventry and Warwickshire NHS Trust).

Mr Christopher Little (Oxford University Hospitals NHS Foundation Trust).

Mr Iain Macleod (Hampshire Hospitals NHS Foundation Trust).

Mr Jodi George Malal (Bedfordshire Hospitals NHS Foundation Trust).

Mr Tim Matthews (Cardiff & Vale University Health Board, University Hospital of Wales).

Mr Damian McClelland (University Hospitals of North Midlands NHS Trust).

Mr Neal Millar (West Glasgow Ambulatory Care Hospital, NHS Greater Glasgow and Clyde).

Mr Prabhakar Motkur (United Lincolnshire Hospitals NHS Trust).

Mr Rajesh Nanda (North Tees and Hartlepool Hospitals NHS Foundation Trust).

Mr Chris Peach (Manchester University NHS Foundation Trust).

Mr Tim Peckham (Basildon and Thurrock University Hospitals NHS Foundation Trust).

Mrs Jayanti Rai (East Kent Hospitals University NHS Foundation Trust).

Professor Amar Rangan (South Tees Hospitals NHS Foundation Trust).

Mr Ravi Ray (Basildon and Thurrock University Hospitals NHS Foundation Trust).

Mr Douglas Robinson (Forth Valley Royal Hospital, NHS Forth Valley).

Lt Con Philip Rosell (Frimley Health NHS Foundation Trust).

Mr Adam Ruman (East and North Hertfordshire NHS Trust).

Mr Adnan Saithna (Southport and Ormskirk Hospital NHS Trust).

Mr Colin Senior (Dorset County Hospital NHS Foundation Trust).

Mr Harish Shanker (Royal Alexandra Hospital, NHS Greater Glasgow and Clyde).

Mr Barnaby Sheridan (Taunton and Somerset NHS Foundation Trust).

Mr Kanthan Theivendran (Sandwell and West Birmingham Hospitals NHS Trust).

Mr Simon Thomas (Perth Royal Infirmary, NHS Tayside).

Mr Balachandran Venateswaran (The Mid Yorkshire Hospitals NHS Trust).

Research associates/nurses at collaborating sites

Lynne Allsop, Pam Ashdown, Manjit Attwell, Piet Bakker, Jenna Bardsley, Dawn Barker, Steven Barnfield, Geraldine Belcher, Sue Bell, Helen Black, Lesley Boulton, Rebecca Boulton, Charlotte Boustead, Joanne Bradley-Potts, Sally Breakspear, Tracy Brear, Emma Brennan, Sarah Brown, Sarah Buckley, Ellis Burton, Julie Butler, Melanie Caswell, Diana Cedron, Carla Christie, Louise Clarkson, Rachel Clarkson, Cheryl Cleary, Katherine Coates, Julie Colley, Elise Cooke, Carina Cruz, Win Culley, Carol Dalton, Rachael Daw, Angie Dempster, Christine Dobb, Joanne Donnachie, Alison Dube, Anne Elliot, Barbara Finson, Joanna Forbes, Kelly Goffin, Francesca Gowing, Natalie Grocott, Christian Hacon, Ruth Halliday, Anne Hardwick, Zena Haslam, Melissa Hawkes-Blackburn, Steven Henderson, Christine Herriott, Dennise Hill, Martin Howard, Maggie Hughes, Bianca Hulance, Jane Hunt, Alison Innes, Caron Innes, Juliet James, Joan Joyce, Cherry Kilbride, Tracy Lee, Jane Martin, Richard Mccormick, Kerri Mcgowen, Pauline Mercer, Alanna Milne, Norma Murray, Dominic Nash, Kimberley Netherton, Anne Nicholson, Claire Nott, Helen Nutt, Joshua O’Donnell, Deborah Page-Browne, Pravin Pastil, Ann Quirk, Angelo Ramos, Alex Ramshaw, Joanne Riches, Cheryl Ritchie, Tessa Rowland, Terri-Ann Sewell, Simon Sharpe, Emma Sharpe, Sarah Sheldon, Emma Shinn, Helen Shirley, Amanda Skinner, Carol Snipe, Jan Stephen, Emma Stoddard, Yogita Stokes, Natalie Taylor, Mairead Tennent, Kerry Third, Susan Thomson, Anne Todd, Dawn Trodd, Sylvia Turner, Mark Vertue, Michael Villaruel, Joanna Wales, Bridget Watkins, Thanuja Weerasinghe, Lori Wells, Kim Wheway, Alison Whitcher, Jessica Whiteman, Matthew Williams, Georgina Williamson, Caroline Wrey Brown and Andrea Young.

Surgeons at collaborating sites

A Armstrong, A Baan, A Bidwai, A Brooksbank, A Desai, A Dunkley, A Leonidou, A Modi, A Putti, A Rajput, A Rangan, A Rumian, A Saithna, B Charalambous, B Sheridan, B Venkateswaran, Banclore, C Baldwick, C Jones, C Kelly, C Modi, C Peach, C Senior, D Cairns, D Robinson, F Ashton, H Colaco, H Ingoe, H Shanker, H Singh, Hilley, I Macleod, J Candal-Couto, J Kurian, J Malal, J Relwani, K Kumar, K Mangat, K Theivendran, M Butt, M Crowther, M Kent, M Smith, Morrison, N Bayliss, N Millar, N Modi, P Ahrens, P Holland, P Jenkins, P Moses, P Motkur, P O’Farrell, Qadir, R Aveledo, R Bansal, R Evans, R Hawken, R Jeavons, R Nanda, R Ray, R Sofat, R Taranu, S Barker, S Chan, S Cloke, S Drew, S Garg, S Sawalha, S Srinivasan, S Thomas, S Venkatachalam, T Lawrence, T Matthews, T Peckham, T Yarashi, U Fazzi, Velu and W Tabi.

Physiotherapists at collaborating sites

A Quirk, C Smith, L MacPherson, M Robertson, N Carr, S Barklie, C Powell, G Hill, S Leahy, C McClymont, D Page-Browne, L Steed, L Wells, S Pak, T Andrews, C Blundell, A Daniel, A Smillie, C Argles, C Holmes, C Hubner, D Coyle, D Keighley, E Haynes, E Newby, H Moulton, H O’Brien, J Bardsley, J James, J Phillips, J Winton, K Clay, K Moses, L Bellis, M Dickinson, N Munnelly, P Lewis, R Daw, S Rogers, D Edmonds, K Williams, V Walker, A Doodson, C Clahson, D Buckle, E Palmer, J Neal, N Howarth, N Maher, C Brookes, M Peak, A Deutsch, A Glover, E Phipps, I Carpenter, L Brookes, M Bailey, S Wang, V Hawke, G Whitting, J Rai, K Flexton, R Camp, S Meadows, A Coutts, C Grady, C Meadley, G Watson, K Williamson, L Dawson, M Taylor, B Danks, D Norris, H Alsafar, L Mercer, N Goodfellow, V Forgie, A Kunda, C Gallagher, G Cranfield, H Sell, L Clark, M London, S Garrett, M Bawden, S Flaherty, S Ingram, C Bourne, R Witham, A Butler, A Fairley, A Haddick, A Meddes, A Rafferty, A Warsama, A Wood, B Bull, B Woods, C Birch, C Day, C Mccarthy, C Robinson, C Tarn, D Dernie, D Irvine, D Ryan, G Baker, H Harper, J Watson, J Wilson, N Wild, R Conway, R Cross, R Whittle, S Henderson, S Mtopo, T Docherty, A Mclean, D Holmes, J Darling, L Roberts, R Holman, R Jenks, S Treble, B Davison, C Clark, C Rowell, C Smith, D Robson, J Preston, J Pringle, M Bourke, N Dinsdale, N Watt, P Moss, R Haugh, R Mather, S Chester, T Calvert, T Mckay, T Sanderson, J Moser, C Bennett, J Arnot, L Watson, N Bryson, P Jaggard, J Lee, L Scott, M Tsuro, P Patel, R Littleworth, S Brayshaw, T Richardson, C Andrews, J Lloyd Evans, L Fox, I Gray, Y Edmiston, H Du Toit, J Padkin, L Beaman, N Depala, P Canning, R Jones, R Morgan, S Baynes, S Orbell, D Henderson, M Durell, B Mathew, J Paulraj, K Barak, K Baraks, Y Stokes, A Arnold, A Emmott, A Southgate, D Anthony, D Smith, D Wharton, J Blackmall, J Fletcher, J Grainger, J Moroney, J Rohun, K Budge, K Revitt, L Bardgett, M Korimbocus, M West, R Heyer, S Brown, S Deaton, T Mullally, V Bullock, V Georgopoulos, W Stone, C Gallen, D Mishra, J Curley, M Franklin, M Langley, S Titherley, A Christer, A Donnelly, B Cole, C Usher, J Calvert, J Goodwin, J Hatfield, K Langman, K Thornton, L Goodchild, L Readyhough, L Thomson, L Wildgoose, M Heaton, N Goodenough, R Sheridan, S Boucher, S Cook, S Willingham, C Whiteman, L Hills, M Dagless, C Watts, A Pritchard, C Hegarty, C Khanna, D Selvin, F Deery, H Bush, K Hollway, K Stevenson, L Abbey, L Burton, L Fort, L Nicholls, T Rhodes-Jones, A Din, A Loizou, D Egerton, E Wallis, G Bennett, G Brear, G Cartner, J Farrar, J Taylor, K Madden, R Brooke, R Carter, R Gymer, R Holmes, T Grimson, C Duffield, C Handy, C Mutiziva, F Cunningham, G Betts, G Wright, H Giga, H Tunnicliffe, J Davis, J Leigh, K Aghada, M Hughes, N Cook, N Horsley, P Frostick, P Lewis, P Stanton, R Bove, R Dixey, R Shillcock, S Gaurovski, S Joshi, S Latif, S Mistry, T Swinley, C Gillies, H Hamilton, K Shields, K Taylor and M Kaan.

Trial Steering Committee independent members

Professor Chris Rogers (Chairperson; Professor of Medical Statistics and Clinical Trials, University of Bristol).

Mr Marcus Bateman (Upper Limb Consultant Physiotherapist, Derby Hospitals NHS Foundation Trust).

Mr David Stanley (Consultant Orthopaedic Surgeon, Sheffield Teaching Hospitals NHS Foundation Trust).

Mrs Denise Denton (Independent Member; Hull Ambassador for NHS Hull Clinical Commissioning Group).

Data Monitoring and Ethics Committee independent members

Professor John Norrie [Chairperson; Chairperson of Medical Statistics and Trial Methodology, University of Edinburgh (he is also a member of the National Institute for Health Research HTA and EME Editorial Board)].

Dr Chris Littlewood (Physiotherapist/Senior Lecturer in Clinical Trials, Keele University).

Mr Roger Hackney (Consultant Orthopaedic Surgeon, Leeds Teaching Hospitals NHS Trust).

Contributions of authors

Dr Stephen Brealey (https://orcid.org/0000-0001-9749-7014) (Research Fellow, Trial Manager) was a co-applicant and the trial manager. He contributed to the design and conduct of the trial throughout the study, and contributed to and commented on all drafts of the report.

Dr Matthew Northgraves (https://orcid.org/0000-0001-9260-8643) (Research Fellow, Trial Co-ordinator) contributed to the day-to-day running of the trial and the ongoing development of the study protocol, and contributed to and commented on all drafts of the report.

Dr Lucksy Kottam (https://orcid.org/0000-0002-7976-2416) (Trauma and Orthopaedic Surgery Trial Manager and Trial Co-ordinator) was a co-applicant and trial co-ordinator who contributed to the design and conduct of the trial, and contributed to and commented on all drafts of the report.

Miss Ada Keding (https://orcid.org/0000-0002-1182-887X) (Lecturer in Statistics, Statistician) was a co-applicant and statistician on the trial who contributed to the design and monitoring of the trial throughout the study, conducted the statistical analysis, and contributed to and commented on drafts of the report.

Mrs Belen Corbacho (https://orcid.org/0000-0002-2359-0379) (Research Fellow, Health Economist) led the write-up of the chapter on the economic evaluation and commented on drafts of the report.

Mrs Lorna Goodchild (https://orcid.org/0000-0001-6301-7684) (Clinical Physiotherapist, Co-leader of Physiotherapy Arm) was involved in the design, conduct and delivery of the trial throughout the duration and commented on drafts of the reports.

Dr Cynthia Srikesavan (https://orcid.org/0000-0002-3540-8052) (Postdoctoral Research Assistant in Physiotherapy, Qualitative Researcher) contributed to the data acquisition, analysis and interpretation, and preparation of the HTA chapter on the embedded qualitative study of UK FROST.

Mrs Saleema Rex (https://orcid.org/0000-0002-0318-6125) (Medical Statistician) contributed to the analysis and reporting of the systematic review that combines UK FROST trial results with other relevant RCTs to treat idiopathic frozen shoulder.

Mr Charalambos P Charalambous (https://orcid.org/0000-0003-0680-9970) (Consultant in Trauma and Orthopaedics, Honorary Professor) was a co-applicant who contributed to the design and conduct of the trial, and contributed to and commented on drafts of the report.

Dr Nigel Hanchard (https://orcid.org/0000-0002-6570-4346) (Reader in Orthopaedics, Co-Leader of Physiotherapy Arm) was a co-applicant who contributed to the design and conduct of the trial, and commented on drafts of the report.

Mrs Alison Armstrong (https://orcid.org/0000-0002-6254-0411) (Consultant Orthopaedic Surgeon) was a co-applicant and was the principal investigator at the Leicester site. She contributed to the design and conduct of the trial, and contributed to and commented on drafts of the report.

Mr Andrew Brooksbank (https://orcid.org/0000-0002-9271-4259) (Consultant Orthopaedic Surgeon) was a principal investigator and co-applicant on the original grant. He contributed to study design, trial set up/recruitment and co-ordination of the Scottish centres, and commented on drafts of the report.

Professor Andrew Carr (https://orcid.org/0000-0001-5940-1464) (Nuffield Professor of Orthopaedics) was a co-applicant who contributed to the design of the trial, and contributed to and commented on all drafts of the report.

Mrs Cushla Cooper (https://orcid.org/0000-0002-3790-3428) (Operational and Trials Development Lead) was a trial co-ordinator who contributed to the conduct of the trial and management of participating sites and the development and implementation of the tissue substudy, and contributed to and commented on all drafts of the report.

Professor Joseph Dias (https://orcid.org/0000-0001-5360-4543) (Professor of Orthopaedic and Hand Surgery) was a co-applicant who contributed to the design of the trial, and contributed to and commented on all drafts of the report.

Mrs Iona Donnelly (https://orcid.org/0000-0002-9794-5347) (Trial Co-ordinator) made a substantial contribution to the acquisition of data as well as commenting on drafts of the report.

Professor Catherine Hewitt (https://orcid.org/0000-0002-0415-3536) (Professor of Trials and Statistics/Deputy Director York Trials Unit) was a co-applicant and had overall responsibility for the statistical analyses of the trial data. She contributed to the design and analysis of the study, and commented on all drafts of the report.

Professor Sarah E Lamb (https://orcid.org/0000-0003-4349-7195) (Professor and Director of Oxford Clinical Trials Research Unit) was a co-applicant and co-supervised the qualitative elements and the development and reporting of the intervention. She has contributed to the design and conduct of the study, and contributed to and commented on drafts of the report.

Dr Catriona McDaid (https://orcid.org/0000-0002-3751-7260) (Reader in Trials) was a co-applicant and contributed to the design and conduct of the trial and systematic review, and commented on all drafts of the report.

Professor Gerry Richardson (https://orcid.org/0000-0002-2360-4566) (Professor of Health Economics) was a co-applicant who contributed to the design and conduct of the economic evaluation alongside the trial, and contributed to and commented on all drafts of the report.

Mrs Sara Rodgers (https://orcid.org/0000-0003-1757-541X) (Research Fellow, Trial Co-ordinator) was a trial co-ordinator who contributed to the conduct of the trial and commented on drafts of the final report.

Mrs Emma Sharp (https://orcid.org/0000-0001-7988-9508) (Orthopaedic Research Manager) was a trial co-ordinator who contributed to the study implementation at multiple sites throughout Scotland.

Professor Sally Spencer (https://orcid.org/0000-0001-6330-7200) (Director of Postgraduate Medical Institute, Lead for Patient-reported Outcome Measures) was a co-applicant who contributed to the design of the trial, and contributed to and commented on all drafts of the report.

Professor David Torgerson (https://orcid.org/0000-0002-1667-4275) (Director of York Trials Unit) was a co-applicant and helped write the study protocol and is a trial methodologist responsible for the overall methods of the trial.

Dr Francine Toye (https://orcid.org/0000-0002-8144-6519) (Qualitative Researcher) was a co-applicant who contributed to the design and conduct of the trial of the qualitative research embedded in this trial throughout the duration of the study, and contributed to and commented on all drafts of the report.

Professor Amar Rangan (https://orcid.org/0000-0002-5452-8578) (Professor of Orthopaedic Surgery) was the lead applicant and chief investigator. He contributed to the design and conduct of the trial throughout the duration of the study, and contributed to and commented on all drafts of the report.

Data-sharing statement

All data requests should be submitted to the corresponding author for consideration. Access to anonymised data may be granted following review with the chief investigator and the trial team.

Disclaimers

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