Clinical effectiveness and cost-effectiveness of open and arthroscopic rotator cuff repair [the UK Rotator Cuff Surgery (UKUFF) randomised trial]

Article history

The research reported in this issue of the journal was funded by the HTA programme as project number 05/47/02. The contractual start date was in July 2007. The draft report began editorial review in July 2014 and was accepted for publication in May 2015. The authors have been wholly responsible for all data collection, analysis and interpretation, and for writing up their work. 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.

Declared competing interests of authors

none

Copyright statement

© Queen’s Printer and Controller of HMSO 2015. This work was produced by Carr et al. under the terms of a commissioning contract issued by the Secretary of State for Health. 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.

Rotator cuff tear

The prevalence of shoulder complaints in the UK is estimated to be 14%, with 1–2% of adults consulting their general practitioner (GP) annually regarding new-onset shoulder pain. 1 Rotator cuff pathology, including tendonitis, calcific tendonitis and rotator cuff tears, reportedly accounts for up to 70% of shoulder pain problems. 2 Painful shoulders pose a substantial socioeconomic burden. Disability of the shoulder can impair the ability to work or perform household tasks and can result in time off work. 3,4 Shoulder problems account for 2.4% of all GP consultations in the UK and 4.5 million visits to physicians annually in the USA. 5,6 More than 300,000 surgical repairs for rotator cuff pathologies are performed annually in the USA, where the annual financial burden of shoulder pain management has been estimated to be US$3B. 7

Rotator cuff pathology is associated with progressive change in the shape of the acromion, with ‘spurs’ forming at its anteroinferior margin. Some reports suggest that these spurs narrow the subacromial space, thereby making physical contact more likely in certain positions of the arm. This is most notable in abduction and elevation of the arm and is sometimes referred to as ‘painful arc’ or impingement because pain is maximal in the mid-range of movement. This process is argued to result in inflammation of the rotator cuff tendons (particularly the supraspinatus tendon) and the overlying subacromial bursa. A conflicting theory suggests that such mechanisms are not causative and that intrinsic age-related degeneration of the tendon is the main determinant of inflammation and symptoms. 8,9

Rotator cuff tear refers to structural failure in one or more of the four muscles and tendons that form the rotator cuff. Any tear that does not extend all the way through the tendon is termed a partial-thickness tear. Asymptomatic full-thickness tears of the rotator cuff are very common in the general population. It is estimated that the overall prevalence of tears is 34% and that risk increases significantly with age. 10 Partial tears are more prevalent than full-thickness tears. 11

Conservative management

Conservative treatment may include rest, exercise, topical non-steroidal anti-inflammatory drugs (NSAIDs), oral corticosteroids, oral paracetamol, opioid analgesics, physiotherapy, activity modification, acupuncture, platelet-rich plasma injections, extracorporeal shockwave therapy, suprascapular nerve block, laser treatment, autologous blood injections, intra-articular NSAID injections, subacromial corticosteroid injections, electrical stimulation, ice and ultrasound.

A search of MEDLINE, EMBASE and The Cochrane Library up to August 2009 for treatment of shoulder pain was undertaken. 12 Harm alerts from relevant organisations, such as the US Food and Drug Administration (FDA) and the UK Medicines and Healthcare products Regulatory Agency (MHRA), were included. The review found 71 systematic reviews, randomised controlled trials (RCTs) or observational studies that met the inclusion criteria. Grading of Recommendations Assessment, Development and Evaluation (GRADE) of the quality of evidence for interventions was performed. 2 It is not known whether topical NSAIDs, oral corticosteroids, oral paracetamol or opioid analgesics improve shoulder pain, although oral NSAIDs may be effective in the short term in people with acute tendonitis/subacromial bursitis. If pain control fails, the diagnosis should be reviewed and other interventions considered. Physiotherapy may improve pain and function in people with mixed shoulder disorders compared with placebo. Platelet-rich plasma injections may improve the speed of recovery in terms of pain and function in people having open subacromial decompression for rotator cuff impingement, but further evidence is needed. Acupuncture may not improve pain or function in people with rotator cuff impingement compared with placebo or ultrasound. Extracorporeal shockwave therapy may improve pain in calcific tendonitis. There is some evidence that suprascapular nerve block, laser treatment, arthroscopic subacromial decompression and rotator cuff repair may be effective in some people with shoulder pain. There is no evidence to support the use of autologous blood injections, intra-articular NSAID injections, subacromial corticosteroid injections, electrical stimulation, ice or ultrasound. Concern exists regarding the potential longer-term damaging consequences of corticosteroid injection. 13

Role of imaging

Imaging is most useful in directing treatment in secondary care if conservative care has failed. A large proportion of the general population will demonstrate abnormalities on imaging of the rotator cuff. 14 Imaging findings need to be interpreted in the context of symptoms, disability and response to treatment. A high proportion of patients with rotator cuff pain will respond to conservative treatment. 15 The only reliable non-interventional method of determining if a rotator cuff tear has healed is use of postoperative imaging, either magnetic resonance imaging (MRI) or ultrasonography.

Surgical management

The most frequent indications for surgery are persistent and severe pain combined with functional restrictions that are resistant to conservative measures. Symptoms of pain and weakness typically disrupt daily activities and night pain affects sleep. Symptoms of a minimum of 3 months’ duration that are sufficiently severe to disrupt daily activities and rest or sleep and failure of standard conservative care (analgesics, rest and physiotherapy and cortisone injection) are usually required before surgery is considered. Surgical repair may be advised in cases of full-thickness rotator cuff tear with persistent pain and weakness after conservative treatment. A rotator cuff repair operation aims to reattach the torn tendons to the humeral bone. In general, two approaches are available for surgical repair. Open surgery involves the rotator cuff being repaired under direct vision through an incision in the skin. Arthroscopic surgery is keyhole surgery and involves the repair being performed through arthroscopic portals into the shoulder. A subacromial decompression (SAD) or acromioplasty to create space around the repaired tendon is usually performed in association with the tendon repair. Reports of the outcome of such surgery are conflicting and evidence for effectiveness is unclear. 16–18 An assessment of the treatment cost of impingement suggests that the addition of surgery, in comparison with exercise treatment alone, is not cost-effective. 19

Comparative studies of subacromial decompression and non-operative treatment options such as physiotherapy have not shown any significant difference in outcome between the two treatment modalities. 20–23 A growing number of studies have tried to assess the effectiveness of subacromial decompression against a control. Three studies of patients undergoing rotator cuff repair, including or excluding subacromial decompression in their operative treatment, did not demonstrate any difference in outcome between the groups. 24–26 A RCT of subacromial decompression plus subacromial bursectomy compared with bursectomy alone reported no significant difference in clinical outcome between the two groups. This finding suggests that removing acromial spurs might not be necessary. 27

The management of partial tears is particularly controversial and patients with such tears have commonly been treated conservatively. Favourable results have been reported following debridement of partial tears in association with subacromial decompression. 28 Higher rates of re-rupture are associated with repairs of larger tears, increased patient age and increased fatty degeneration of the cuff muscles. 29–32 Partial tears are most commonly managed without repair, but some authors advocate repair to prevent progression to full-thickness tears. The evidence supporting this approach is weak. 11 There is also uncertainty regarding the relative value of conservative care, repair surgery and debridement surgery for large and massive tears. 33–36 High failure rates of 13–68% have been reported for surgical repair of rotator cuff tears, irrespective of the surgical technique employed. 37–39 Some studies have suggested that re-rupture rates are associated with poorer outcomes. 40 Surgical decision-making in the management of rotator cuff tears was reviewed by Dunn et al. 41 They surveyed surgeons in the USA and found considerable variation in decision-making. This included the type of surgery, the surgical techniques employed and the type and duration of conservative treatment, including cortisone injections, physiotherapy, rest, analgesia and home exercises. Rates of medical visits for rotator cuff pathology in the USA were reviewed between 1996 and 2006. The volume of rotator cuff repairs had increased by 141% and the unadjusted number of arthroscopic repairs increased by 600% compared with a 34% increase in open repairs. 42 The volume of arthroscopic subacromial decompressions has also increased significantly over time. Recent figures from the USA report a 240% increase (from 30.0 to 101.9 per 100,000 people per year) in use of the procedure in New York state between 1996 and 2006. 43 This compares to a 78.3% increase in ambulatory orthopaedic surgery overall. Similar increases have recently been reported in the UK. 44 The introduction of less invasive arthroscopic techniques accounts for some of the overall increased rate of surgery, but does not explain regional variation. Patient and disease characteristics have not changed over time and there is a growing concern that this procedure is being overused. Observational studies of subacromial decompression surgery show positive results in terms of pain reduction and functional outcome, with high patient satisfaction rates. However, equally good outcomes have been noted in two studies following patients who had arthroscopic rotator cuff debridement or open rotator cuff repair in the absence of a subacromial decompression.

Rationale for the study design

The objective of the original commissioned call was to conduct a pragmatic multicentre randomised clinical trial to obtain good-quality evidence of the effectiveness and cost-effectiveness of conservative care compared with arthroscopic surgical repair compared with open surgical repair for the treatment of degenerative rotator cuff tears. Because of high crossover from the conservative arm to surgery the study was reconfigured to a comparison of the two surgical techniques only. There is conflicting evidence regarding the effectiveness of open and arthroscopic repair. 12,45–47 Proponents of arthroscopic rotator cuff surgery suggest that the procedure may have advantages over standard open techniques by causing less trauma to the deltoid muscle and overlying soft tissue. Arguably, this causes less postoperative patient discomfort together with earlier return of movement. However, the success of the repair depends partly on the ability of the surgeon to achieve a secure attachment of tendon to bone. This may be more easily and reliably achieved by open/mini-open surgery. Other potential disadvantages of the arthroscopic approach include increased technical difficulty and longer time in theatre. There is a need to compare the outcomes of the two surgical techniques.

Literature update since the call

A review updating the literature published since the original commissioned call was undertaken to inform this report and set the results in context. Only reports of RCTs were included. Quasi-RCTs, which use methods of allocating participants to a treatment that are not strictly random, for example date of birth, hospital record number or alternation, were excluded.

Search methods for identification of studies

We searched Ovid MEDLINE from 2006 to March 2014 for possible reports of RCTs. The search strategy used was based on one developed by Coghlan et al. 16 for a Cochrane review of surgery for rotator cuff disease. This search strategy was modified to account for changes to the medical subject heading (MeSH) terms since the original search was conducted in 2006, the addition of free-text terms and the replacement of the original RCT filter used with the Cochrane sensitivity- and precision-maximising version RCT filter [2008 version; see http://handbook.cochrane.org/chapter_6/box_6_4_b_cochrane_hsss_2008_sensprec_pubmed.htm (accessed 23 August 2015)] (see Appendix 4 for search strategy). We also searched the World Health Organization (WHO) International Clinical Trials Registry Platform51 to identify reports of any ongoing RCTs. One author screened the titles and abstracts of all retrieved records. Full articles were then obtained for any potentially eligible studies and assessed using the predefined eligibly criteria described earlier.

Results

Reconfigured study design

A high rate of crossover (77%) of the 214 patients in the rest-then-exercise programme to surgery was observed and so the trial was adapted and reconfigured on the instruction of the funder in 2009 after consultation with the Trial Steering Committee (TSC) and the Data Monitoring Committee (DMC). Crossover did not occur at a consistent or predictable time point. The reconfigured design was a two-way parallel-group RCT of open compared with arthroscopic rotator cuff repair (UKUFF reconfigured REC reference number 10/H0402/24, April 2010). At the time of reconfiguration there were 131 participants in stratum A (n = 43, arthroscopic surgery; n = 44, open surgery; and n = 44, rest then exercise), 181 in stratum B (n = 91, arthroscopic surgery; and n = 90, rest then exercise) and 162 in stratum C (n = 82, open surgery; and n = 80, rest then exercise). The 87 patients already randomised between arthroscopic and open surgery (stratum A) were carried through to the subsequent reconfigured trial. After the reconfiguration it was calculated that a further 180 patients should be recruited and followed up for 2 years as per the original protocol, leading to a total of 267 patients treated with surgery.

During the period between 2007 and 2010, surgical opinion had changed and an increased number of surgeons were in equipoise between open and arthroscopic surgery. The UKUFF trial was reconfigured as a pragmatic multicentre study involving 20 surgeons from 16 UK centres; 15 of these surgeons had originally recruited to stratum A.

Patient and public involvement and engagement

From the outset patients were involved and engaged in the design of the trial. A patient representative (D Farrar-Hockley) was a member of the group designing the trial. He subsequently became a member of the TSC.

Interventions

Study population

Eligible patients were those for whom care had been provided by a participating surgeon and who were deemed suitable for rotator cuff repair surgery, with the surgeon uncertain which surgical procedure was better. In addition, patients had to be aged ≥ 50 years, have symptoms from a degenerative full-thickness rotator cuff tear and be able to give informed consent.

Study registration/consent to randomise

Recruitment of patients occurred through a two-step process. A patient’s eligibility was assessed by the local consultant orthopaedic surgeon, who introduced the trial to the patient using a prompt sheet and a patient assessment form. If the patient was interested in participating, the surgeon then provided the patient with a copy of the patient information sheet (see Appendix 6), which summarised what the study involved and answered any questions the patient might have.

If the patient was willing to enter the trial then an initial consent form was signed, which allowed the patient’s details to be forwarded to the central study office in Oxford. The office then issued to the participant, by post, an invitation letter, the comprehensive patient information sheet (see Appendix 6), a consent form and a baseline questionnaire (see Appendix 7) with a prepaid return envelope. Patients were encouraged to contact the office or their surgeon if they had any further questions or concerns. Patients who had not returned their questionnaire and consent form within a week were telephoned by a member of the study team in Oxford. This contact allowed the patient to ask questions about the study and permitted the team to assess whether the patient was still willing to participate. When the full consent form and baseline questionnaire had been returned to the Oxford office the patient then officially entered the trial and was randomised to one of the surgical options. A copy of the signed consent form was returned to the patient.

Randomisation was by computer allocation using the service provided by the Centre for Healthcare Randomised Trials (CHaRT) at the Health Services Research Unit, University of Aberdeen. Allocation was minimised using surgeon, age and size of tear. After randomisation the participant was considered irrevocably part of the trial for the purpose of the research, irrespective of what occurred subsequently.

Outcomes

The primary outcome measure was the Oxford Shoulder Score (OSS)62 completed at 24 months after randomisation. The OSS is a 12-item shoulder-specific PROM that was developed, with patients, for the assessment of shoulder pain and function in the context of shoulder surgery, particularly in trials. Items refer to the past 4 weeks and each offers five ordinal response options. Originally, these were scored from 1 to 5 (5 = most severe) and then summed to produce a summary score ranging from 12 to 60. Subsequently, the recommended method of scoring was changed. 63 Under the new system, each item on the OSS is scored from 0 to 4, with 4 representing the best outcome (i.e. the opposite direction from the original method of scoring). When the 12 items are summed, this produces an overall score ranging from 0 to 48, with 48 being the best outcome. The OSS has been demonstrated to be reliable, valid, responsive and very acceptable to patients.

The primary measure of cost-effectiveness was the incremental cost per quality-adjusted life-year (QALY).

The secondary outcome measures were used to further assess functional outcome and patient health-related quality of life. These assessed a range of symptoms often experienced with rotator cuff tears, for example pain, weakness and loss of function. These included:

• Shoulder Pain and Disability Index (SPADI)64 at 8, 12 and 24 months after randomisation. The SPADI is a self-administered questionnaire, developed by a panel of rheumatologists and a physiotherapist, to measure shoulder pain and disability in an outpatient setting. 64 It contains 13 items that assess two domains: shoulder pain (five items) and disability (eight items), all with reference to the last week. The original version scored each item on a visual analogue scale (VAS). A second version, used in this trial, replaced the VAS with a 0–10 numerical rating scale. 65 Item responses within each subscale are summed and transformed to a score out of 100. A mean is taken of the two subscales to give a total score out of 100, with a higher score indicating greater impairment or disability. The SPADI is reliable, valid and responsive. 64

• Mental Health Inventory 5 (MHI-5)66 at 8, 12 and 24 months after randomisation. The MHI-5 is the mental health subscale of the Short Form questionnaire-36 items (SF-36) generic health status measure. 67 It contains five items that address anxiety, depression, loss of behavioural or emotional control and psychological well-being, all with reference to the past 4 weeks. Each item offers responses on a 6-point scale (ranging from ‘all of the time’ to ‘none of the time’). The total score is calculated by reversing the answers to two items (the third and fifth), summing the scores, and transforming the raw scores to a scale ranging from 0 to 100. A higher score indicates better mental health. The MHI-5 has been demonstrated to be good at detecting major depression, affective disorders generally and anxiety disorders. 66

• European Quality of Life-5 Dimensions three levels (EQ-5D-3L)65,68 at 8, 12 and 24 months after randomisation. The EQ-5D-3L is a standardised generic instrument for use as a self-completed measure of health outcome. It provides a simple descriptive profile and a single index value for health status that can be used in the clinical and economic evaluation of health care, as well as population health surveys, and consists of five items on mobility, self-care, pain, usual activities and psychological status, with three possible answers for each item (1 = no problem, 2 = moderate problem, 3 = severe problem). The response for each item/domain is converted to a quality-of-life estimate using an algorithm (see Chapter 6 for further details) to produce an index score for each patient. Negative scores represent health states worse than death, 0 represents the state of worst health and 1.00 represents full health.

• The OSS62 completed at 8 and 12 months after randomisation.

• Participants’ view of the overall state of their shoulder compared with an earlier time point (‘transition item’) at 8, 12 and 24 months after randomisation. There were five possible responses to this item: ‘much better’, ‘slightly better’, ‘no change’, ‘slightly worse’ or ‘much worse’.

• Participants’ rating of how pleased they were with their shoulder symptoms at 12 and 24 months after randomisation. There were four possible responses to this item: ‘very pleased’, ‘fairly pleased’, ‘not very pleased’ or ‘very disappointed’.

• Surgical complications (intra- and postoperative) at 2 and 8 weeks post surgery and at 12 and 24 months after randomisation.

• 12-month postoperative imaging.

Data collection

Outcome assessment was conducted using questionnaires that participants self-completed and, as such, interviewer bias and clinical rater bias were avoided. This form of outcome measurement has consistently performed well compared to clinician-based assessments and general health status measures. All participants, including those who had withdrawn from their allocated intervention but who still wished to be involved in the study, were followed up, with analysis based on the intention-to-treat (ITT) principle.

Participants received questionnaires at the following time points:

• baseline (see Appendix 7) – questionnaire completed before randomisation

• 2 and 8 weeks post treatment (see Appendix 8) – questionnaire completed by telephone

• 8, 12 and 24 months post randomisation (see Appendix 9) – questionnaire completed by post.

The baseline and 12 and 24 months’ post-randomisation questionnaires also incorporated a section that measured cost-effectiveness. This included questions relating to primary care consultations, other consultations, out-of-pocket costs and the work impact of the intervention received.

The study team based at the Health Services Research Unit, University of Aberdeen, contacted participants whose questionnaires had not been returned. In the first instance this was through a reminder letter by post or e-mail, depending on participant preference. If a questionnaire had still not been returned within the specified time frame, the study team telephoned the participant and addressed any administrative issues that may have arisen, such as change of address or loss of questionnaire. If any clinical issues were identified, the study team in Oxford contacted participants, if appropriate, and addressed these issues. The time period allocated to the follow-up checks depended on which outcome assessment was involved.

Magnetic resonance imaging and ultrasound scans

Postoperative imaging was performed on patients who had undergone a repair. It was not performed if a repair was either impossible to perform or when no tear was found. Both the MRI and the ultrasound scans were undertaken locally to the participant and were arranged by the study office in Oxford, at a time agreed by the trust and the participant. The scans were collected centrally. The MRI scans were reported by an independent consultant radiologist who was blinded to the type of surgery that was performed. Because of the operator-dependent nature of the ultrasound scans, an independent report on these was not valid. The report obtained from the site was used to determine the tear status. Any re-tears were not reported to the participating surgeons, so that no deviation occurred from their normal practice.

Statistical analysis of outcomes

Statistical analyses were based on all people randomised, irrespective of subsequent compliance with the randomised intervention. The principal comparison was all those allocated arthroscopic surgery compared with all those allocated open surgery. When an effect size is shown for the intervention, a negative sign on the effect size indicates that an open procedure is favoured over an arthroscopic procedure.

Reflecting the possible clustering in the data, the outcomes were compared using repeated-measures mixed models with centre as a random effect and with adjustment for minimisation variables (size of tear and age) and participant baseline values (when available) as fixed effects. Statistical significance was at the 5% level, with corresponding confidence intervals (CIs) derived. All participants remained in their allocated group for analysis (ITT).

Preplanned subgroup analyses on the primary outcome included exploration of tear size (small/medium vs. large/massive) and age (≤ 65 years vs. > 65 years); these analyses were conducted by including a subgroup by treatment interaction term in the primary outcome model described above. Conservative levels of statistical significance (p < 0.01) were sought, reflecting the exploratory nature of these subgroup analyses.

Health economics methods

A cost-effectiveness analysis was performed. A simple patient cost-related questionnaire was sent out at baseline and at 12 and 24 months post randomisation to obtain information on primary care consultations, other consultations, out-of pocket costs, the work impact of the intervention received and return to work (when relevant). Although longer intervals between questionnaires may result in recall errors (in particular under-reporting of health-care use70) more frequent data collection can result in a higher proportion of missing responses, which introduces uncertainty. It has been argued that there is no optimal interval for self-reported data collection70 and as such the timing of questionnaire distribution was chosen to coincide with those for the clinical study outcomes. Unit costs came from national sources and participating hospitals. The patient questionnaire was also used to administer the European Quality of Life-5 Dimensions (EQ-5D), which was obtained at baseline. The main health economic outcome was within-trial and extrapolated QALYs, estimated using the EQ-5D. 71

Incremental cost-effectiveness was calculated as the net cost per QALY gained for arthroscopic surgery compared with open surgery. Power calculations (see following section) were based on clinical effectiveness rather than cost-effectiveness outcomes, which were estimated rather than used in hypothesis testing. Cost-effectiveness ratios and acceptability curves were calculated.

An important component of this trial was the assessment of cost. Therefore, obtaining an accurate record of procedures at each of the proposed centres was essential. To evaluate the costs of each type of surgery, information was collected from the operating theatres. Resources used, equipment costs and standard procedures for rotator cuff repairs were examined. Per-case information was also analysed. A checklist of equipment, consumables, implants, time and staff utilised during each case was completed by theatre staff. Information from theatres was collected by the Oxford office and used in a cost comparison of the arthroscopic and open surgery approaches.

Sample size

In the original UKUFF trial, with three randomised strata, the sample size was constructed to detect a difference in the OSS 24-month postoperative score of 0.38 of a standard deviation (SD) for the comparison between arthroscopic surgery and open surgery at 80% power. We did not propose any amendment to that clinically important difference in the reconfigured study. This defined difference was based on our experience of developing the OSS score and using it in a variety of settings; a 3-point score difference (0.33 of a SD) was deemed a clinically important difference. In the original UKUFF trial, the detectable difference of 0.38 was constructed by combining evidence from a direct randomised comparison with indirect (non-randomised) comparison data from the other strata. Incorporating indirect effects is a suboptimal approach to measuring effectiveness because unmeasured confounders can bias the outcomes. The proposed change in this proposal was to achieve the detectable difference of 0.38 of a SD by direct randomised comparison data only at 85% power. As described earlier, such an approach was feasible by 2010 because of the increased number of surgeons in equipoise between the arthroscopic approach and the open approach.

Attrition was expected to be low (10%), as were the effects of clustering of outcomes by surgeon [intracluster correlation (ICC) < 0.03]. 72,73 Although we did not have a direct estimate from a shoulder trial, other orthopaedic data sets available to our team support this low ICC estimate. Both of these factors required the sample size to be inflated; however, the primary analysis adjusted for baseline OSS score, which conversely allowed the sample size to be decreased by a factor of (1 – correlation squared). 74 Our previous studies showed that the correlation in the OSS score pre surgery to 6 months post surgery in patients similar to the potential trial participants was 0.57. Assuming a conservative correlation of 0.5 implied that the sample size could be reduced by 25% and still maintain the same power. Therefore, a study with a total of 267 participants would still have 85% power to detect a clinically important difference in each comparison, assuming that attrition and clustering accounted for approximately 25% of variation in the data.

Data monitoring

An independent DMC met on four occasions and did not recommend any fundamental changes to the protocol. The decision in 2009 to reconfigure the trial was made by the NIHR HTA programme. The committee did not meet after recruitment was completed.

Recruitment to the study

Participants were recruited in 47 clinical centres, all within the UK (Table 3). Nineteen centres recruited to the randomised arthroscopic surgery and randomised open surgery comparison (referred to as the stratum A comparison). Thirteen of these centres also randomised participants to stratum A prior to reconfiguration of the study in December 2010 (see Chapter 2). Twenty centres recruited to stratum B (arthroscopic surgery vs. rest then exercise) and 18 to stratum C (open surgery vs. rest then exercise). A total of 660 participants were recruited to the study, with 317 in stratum A (n = 136, allocated to arthroscopic surgery; n = 137, allocated to open surgery; and n = 44, allocated to rest then exercise prior to reconfiguration), 181 in stratum B (n = 91, allocated to arthroscopic surgery; and n = 90, allocated to rest then exercise prior to reconfiguration) and 162 in stratum C (n = 82, allocated to open surgery; and n = 80, allocated to rest then exercise prior to reconfiguration). Table 3 shows recruitment by centre. No centre contributed more than 12% of participants to stratum A. Recruitment to the trial began on 9 November 2007 and continued until 28 February 2012, although not all centres enrolled over the total period because of the staggered introduction of centres and early closure for reconfiguration of the study (range 6 months to 39 months from first to last participant randomised in stratum A). Data were closed to follow-up on 31 December 2013.

Study conduct

The derivation of the main study groups and their progress through the stages of follow-up in the trial is shown in Figure 1 This is in the form of a Consolidated Standards of Reporting Trials (CONSORT) flow diagram. In total, 811 patients were considered for trial entry and 38 (5%) of these were found not to meet one or more of the eligibility criteria. Of the 111 patients eligible for the study but not recruited, 84 declined to participate and the remaining 27 could not be randomised while the study underwent reconfiguration (because of the requirement for new research ethics and research and development approvals). Two participants were excluded post randomisation because each had received previous surgery prior to randomisation. Details of the clinical management actually received are provided in Chapter 4. The median [interquartile range (IQR)] time intervals in days between randomisation by the trial office and each subsequent follow-up are shown in Table 4; all were similar between groups, as would be expected. The 2-week and 8-week follow-ups were timed to occur at 2 weeks and 8 weeks after surgery. Table 5 illustrates the success of this strategy in the subgroup of participants who did receive a surgical procedure.

FIGURE 1.

Consolidated Standards of Reporting Trials (CONSORT) diagram. RtE, rest then exercise.

The overall rates of return of follow-up questionnaires at 8, 12 and 24 months were equivalent to > 85% of the study participants allocated to surgery (see Figure 1). There were no substantive differences in response rates between the surgery groups. Seven participants are known to have died by the end of the 2-year follow-up. There was no evidence that these deaths were linked to trial participation.

Description of the groups at trial entry

Participant and sociodemographic factors

Participant and sociodemographic characteristics are shown in Table 7. The average age of the participants was 63 years, 40% were female and 90% were right-handed. The mean period of time that the participants reported having the shoulder problem prior to surgery was approximately 2.5 years; however, the mean was driven by a few extreme values in each group relating to participants who had had shoulder problems for decades. The median (IQR) time that the participants had had the shoulder problem prior to recruitment was 1.2 (0.7–2.5) years. There were no substantive differences within or between strata on any of the sociodemographic factors.

Health status

The health-related quality-of-life measures are shown in Table 8. The mean OSS was approximately 26 across the groups. The EQ-5D, MHI-5 and SPADI measures were broadly similar across and within the strata.

Attitudes to surgery

As expected, there was variation in participants’ attitudes to undergoing surgery in general, but the variation was not different between groups (Table 9).

Summary

There was no evidence of any important imbalances between the groups. With a mean OSS of around 26, the population was representative of those in other shoulder studies (see Chapter 1). Most participants had undergone some form of non-surgical intervention (such as physiotherapy) before the trial and had had symptoms for over a year. In Chapter 4, the results of the randomised trial of arthroscopic compared with open surgery will be reported. A description of the findings from the rest-then-exercise programme will be provided in Chapter 5.

Analysis populations

Throughout the analyses presented in this chapter, the participants in the formal randomised comparison between arthroscopic surgery and open surgery (stratum A) are kept separate from those in the arthroscopic and open groups in strata B and C respectively. All 273 participants who joined the randomised arthroscopic compared with open surgery component in stratum A are referred to as the randomised ITT population; the 148 (n = 63 arthroscopic; n = 85 open) within this group who actually received a repair over the 2-year follow-up period are referred to as the per-protocol population. The non-randomised population refers to the 91 participants allocated to arthroscopic surgery from stratum B and the 82 participants allocated to open surgery in stratum C (included in this chapter for completeness and visual inspection of data). No statistical analysis is performed on the non-randomised population.

Surgical management

Two-week follow-up

Follow-up measures timed to occur at 2 weeks post surgery are shown in Table 13. Very few participants reported being pain free and approximately two-thirds were taking painkillers. Of those participants who were employed, about 80% were still off sick. There were no clinically important differences between or within any of the randomised or non-randomised groups.

Eight-week follow-up

Follow-up measures timed to occur at 8 weeks post surgery are shown in Table 14. The results were similar to those at the 2-week follow-up with the exception that the percentage reporting no or mild pain improved from 35% to 50% with an apparent concomitant effect of reducing painkiller use from 66% to 55% and increasing the number of participants returning to usual work (no or a little interference) from 28% to 55%. There were no clinically important differences between or within any of the randomised or non-randomised groups.

Outcomes at 8, 12 and 24 months

Outcomes at 8, 12 and 24 months were primarily obtained from questionnaire returns. As described in Chapter 3 and reiterated here in Table 15, the return rates were similar across groups and ranged from 90% at 8 and 12 months to 86% at 24 months. There were no notable differences in baseline characteristics between those who had completed a questionnaire at 24 months and those who had not (Table 16). The only exception to this was a statistically significant difference in housing status, with 84.7% of homeowners in the responder group compared with 71.1% in the non-responder group. Given the possibility of multiple statistical testing, the difference should be interpreted with caution. As described in Chapter 2, these results confirmed that a repeated measures analysis assuming no differential loss to follow-up could be considered.

Health status

Health status measures at 8, 12 and 24 months are shown in Table 17. Full details of the statistical testing of the health status measures can be found in the following sections.

TABLE 17 - Health status at 8, 12 and 24 months

Measure	Randomised, n, mean (SD)		Effect size	95% CI	p-value	Non-randomised, n, mean (SD)
	Arthroscopic	Open				Arthroscopic	Open
OSS
Baseline	129, 26.3 (8.2)	131, 25.0 (8.0)				91, 25.3 (8.9)	82, 25.9 (8.4)
8 months	121, 36.1 (9.2)	127, 37.0 (8.6)	–1.27	–3.21 to 0.67	0.200	85, 37.2 (9.7)	75, 37.4 (9.2)
12 months	122, 38.3 (9.5)	122, 39.6 (8.5)	–1.60	–3.55 to 0.35	0.108	83, 37.4 (10.3)	75, 40.6 (6.8)
24 months (primary outcome)	114, 41.7 (7.9)	115, 41.5 (7.9)	–0.76	–2.75 to 1.22	0.452	78, 41.3 (7.7)	75, 41.3 (7.3)
SPADI
Baseline	128, 60.4 (22.1)	130, 61.8 (21.7)				91, 60.6 (23.1)	82, 60.7 (20.1)
8 months	117, 31.8 (26.6)	124, 31.7 (27.8)	1.02	–4.73 to 6.77	0.728	82, 28.7 (24.9)	73, 27.8 (23.3)
12 months	119, 24.2 (26.3)	120, 23.4 (26.4)	1.83	–3.93 to 7.59	0.533	80, 23.9 (25.8)	74, 18.6 (20.2)
24 months	115, 16.1 (21.7)	118, 17.5 (23.7)	0.50	–5.30 to 6.30	0.866	75, 16.6 (21.8)	71, 17.3 (22.9)
SPADI pain
Baseline	128, 69.2 (19.6)	130, 70.5 (20.4)				91, 70.0 (21.8)	82, 69.6 (19.5)
8 months	118, 36.7 (27.7)	126, 35.4 (28.9)	2.00	–4.26 to 8.26	0.532	84, 32.9 (27.8)	72, 31.7 (24.9)
12 months	119, 28.1 (27.8)	119, 25.9 (27.1)	2.91	–3.39 to 9.21	0.365	80, 26.2 (27.0)	75, 23.1 (23.8)
24 months	114, 18.7 (23.1)	118, 20.1 (26.1)	0.92	–5.42 to 7.26	0.777	75, 18.0 (22.8)	71, 20.3 (24.5)
SPADI disability
Baseline	128, 54.9 (25.1)	130, 56.5 (24.4)				91, 54.7 (25.9)	82, 55.2 (21.9)
8 months	117, 28.7 (26.9)	124, 29.3 (28.1)	0.41	–5.26 to 6.07	0.889	82, 26.3 (24.4)	74, 26.2 (24.9)
12 months	119, 21.8 (26.3)	120, 21.7 (26.9)	1.12	–4.57 to 6.80	0.700	80, 22.4 (26.9)	74, 16.2 (19.5)
24 months	116, 14.8 (22.5)	118, 15.8 (22.8)	0.36	–5.35 to 6.07	0.902	76, 16.3 (22.7)	72, 15.5 (22.5)
MHI-5
Baseline	128, 22.4 (4.9)	130, 22.9 (4.5)				90, 22.4 (5.1)	82, 22.1 (4.7)
8 months	118, 23.8 (4.9)	124, 23.8 (4.4)	0.32	–0.61 to 1.26	0.500	81, 23.8 (4.6)	74, 24.0 (4.1)
12 months	118, 23.5 (5.0)	119, 23.6 (4.6)	0.13	–0.81 to 1.07	0.783	78, 24.4 (4.6)	73, 24.2 (3.8)
24 months	116, 24.4 (4.0)	118, 24.3 (4.5)	0.22	–0.72 to 1.17	0.648	76, 23.8 (4.9)	71, 24.2 (4.2)
EQ-5D
Baseline	129, 0.551 (0.297)	131, 0.518 (0.293)				91, 0.514 (0.326)	81, 0.536 (0.287)
8 months	120, 0.680 (0.300)	124, 0.700 (0.257)	–0.032	–0.092 to 0.028	0.296	79, 0.709 (0.244)	74, 0.711 (0.194)
12 months	119, 0.727 (0.278)	118, 0.711 (0.300)	0.011	–0.050 to 0.071	0.724	80, 0.696 (0.293)	71, 0.761 (0.192)
24 months	116, 0.760 (0.235)	118, 0.778 (0.219)	–0.043	–0.104 to 0.017	0.163	79, 0.760 (0.262)	74, 0.753 (0.258)

Oxford Shoulder Score

Figure 2 graphically displays the OSS over the course of the follow-up period for the randomised and non-randomised groups. All groups followed a similar pattern. The OSS increased markedly from baseline (mean 25.7) to 8 months (mean 36.5) and continued to increase thereafter (at a much slower rate) to 24 months (mean 41.5).

FIGURE 2.

Means and 95% CIs for the OSS across time for the ITT and non-randomised groups.

European Quality of Life-5 Dimensions

Figure 3 displays the EQ-5D score over the follow-up period. The pattern is similar to that seen for the OSS.

FIGURE 3.

Means and 95% CIs for the EQ-5D across time for the ITT and non-randomised groups.

Shoulder Pain and Disability Index

The SPADI overall score, together with the pain and disability subscales (Figures 4–6), followed a similar pattern to that for the OSS (note that a lower SPADI score is a better outcome), with a large improvement from baseline at 8 months followed by a smaller rate of improvement thereafter. All randomised and non-randomised groups followed a similar pattern.

FIGURE 4.

Means and 95% CIs for the SPADI overall score across time for the ITT and non-randomised groups.

FIGURE 5.

Means and 95% CIs for the SPADI disability subscale score across time for the ITT and non-randomised groups.

FIGURE 6.

Means and 95% CIs for the SPADI pain subscale score across time for the ITT and non-randomised groups.

Mental Health Inventory

The MHI-5 scores showed very little change across time from baseline (Figure 7).

FIGURE 7.

Means and 95% CIs for the MHI-5 score across time for the ITT and non-randomised groups.

Per-protocol population

For completeness, data on the per-protocol population are included in Appendix 1. These data are from the 63 participants who were randomised to an arthroscopic procedure and who actually received an arthroscopic repair and the 85 participants who were randomised to an open procedure and who received an open repair. Similarly, the data reported also include the 50 arthroscopic repairs and the 40 open repairs in the non-randomised group.

Recognising that caution must be used when interpreting the per-protocol group, we nevertheless note that the lack of important differences between the arthroscopic and the open ITT groups was also observed in the per-protocol data. The similarity is illustrated visually in Figures 8 and 9, where the OSS and EQ-5D scores, respectively, are contrasted between the ITT population and the per-protocol population.

FIGURE 8.

Means and 95% CIs for the OSS across time for the ITT and per-protocol groups.

FIGURE 9.

Means and 95% CIs for the EQ-5D across time for the ITT and per-protocol groups.

Statistical between-group analysis

Primary outcome

The pre-chosen primary outcome was the OSS score at 24 months’ follow-up, for which the effect sizes and 95% CIs are shown in Table 17. Under ITT analysis there was no evidence of a difference between those randomised to receive an arthroscopic procedure and those randomised to receive an open procedure (difference –0.76, 95% CI –2.75 to 1.22; p = 0.452). The 95% CI was also small enough to exclude the prespecified clinically important difference of 3 points.

To test the sensitivity of the primary outcome to the actual procedure received we also performed a per-protocol analysis of the 24-month OSS using the instrumental variable approach described in Chapter 2. A similar result to that of the ITT analysis was produced, although the CIs were much wider, as expected (difference –0.46, 95% CI –5.30 to 4.39; p = 0.854). We are therefore confident that there was no evidence of important differences between surgical treatments.

Secondary outcomes

The secondary outcomes were the health status measures (EQ-5D, SPADI and MHI-5), the change in symptom measures (transition items regarding shoulder problems and shoulder symptoms) and satisfaction ratings (‘How pleased are you with your shoulder symptoms?’) at 8, 12 and 24 months, the OSS at 8 and 12 months and MRI-diagnosed re-tears at 12 months post surgery. Analyses of these outcomes are shown in Tables 17 and 18.

Health status

There was no evidence of any differences between the randomised groups for any of the health status measures at all follow-up times (see Table 17).

Change in symptoms and satisfaction

There was no evidence of any differences between the randomised groups for either of the symptom transition or satisfaction measures at all follow-up times (see Table 18). For example, at 24 months, when asked how pleased they were with their shoulder symptoms, 84.6% of participants in the arthroscopic group and 82.4% in the open group were either very or fairly pleased (OR 1.37, 95% CI 0.73 to 2.60; p = 0.330).

Magnetic resonance imaging findings

The rate of re-tear was similar across the randomised groups, at 46.4% in the arthroscopic repair group and 38.6% in the open surgery group (difference 9.5%, 95% CI –6.9% to 25.8%; p = 0.256).

Subgroup analysis

Two preplanned subgroup analyses on the primary outcome (OSS at 24 months) were conducted: size of tear at baseline assessment (small or medium vs. large or massive) and age at recruitment (< 65 years vs. ≥ 65 years). Figure 10 displays the means and 95% CIs for the differences in OSS score at 24 months in the subgroups. There was no evidence that any of the subgroups was statistically significantly different at the 1% level (p = 0.843 for tear size and p = 0.024 for age).

FIGURE 10.

Prespecified subgroup analysis of the OSS at 24 months (size of tear; age of patient).

Learning curve

The statistical model for investigating any trend in OSS at 24 months as surgeon experience increased during the trial did not demonstrate any significant learning effect (trend in OSS +0.04 per procedure, 95% CI –0.21 to 0.29; p = 0.744).

Description of the rest-then-exercise group at trial entry

All baseline data collected on these patients were presented in Chapter 3. We replicate a subset of the baseline data here to illustrate that participants randomised to rest then exercise across the three strata displayed broadly similar characteristics and beliefs at baseline (Tables 21 and 22, respectively).

Outcomes in those allocated to rest then exercise

Outcomes of all those allocated to rest then exercise (i.e. by ITT) are presented in Table 23. As with the full trial data presented in previous chapters, these data show that participants demonstrated substantial improvements in quality of life from baseline to final follow-up at 2 years.

Outcomes in those allocated to rest then exercise but who did not complete the 10-week course before progressing to surgery

In total, 77 (36%) of the 214 participants initially randomised to rest then exercise progressed to surgery before completion of the 10-week rest-then-exercise course. Outcomes at each time point in this specific subgroup of participants are presented in Table 24.

Outcomes in those allocated to rest then exercise who completed the full 10-week course but still progressed to have surgery

In total, 88 (41%) of those allocated to rest then exercise completed the full 10-week course but despite this progressed to surgery. Outcomes at each time point in this specific subgroup of participants are presented in Table 25.

Outcomes in those allocated to rest then exercise who completed the full 10-week course and did not proceed to surgery

In total, 36 (17%) of those allocated to rest then exercise completed the full 10-week course and had no surgical intervention. Outcomes at each time point in this specific subgroup of participants are presented in Table 26. A further 14 rest-then-exercise patients withdrew from the study over the course of the study.

Methods: overview

The health economic analyses evaluate the total resource use and costs of rotator cuff repair and quality of life over the follow-up period of the trial. The primary economic outcome is additional cost per QALY gained of arthroscopic repair compared with open repair over a time horizon of 24 months, by ITT analysis. This outcome was assessed using patients allocated to arthroscopic and open repair in stratum A only, although cost and quality-of-life outcomes from stratum B (arthroscopic repair only) and stratum C (open repair only) are also reported. Lifetime extrapolation was not conducted for this analysis.

All analyses were conducted from the perspective of the NHS. Cost components included all initial and subsequent inpatient episodes and outpatient hospital visits, as well as GP, nurse and physiotherapist visits during follow-up.

Resource-use data collection

Unit costs and application to resource use

Quality of life

Quality of life was measured at baseline and 8, 12 and 24 months using the five domains of the EQ-5D-3L questionnaire,68,71 with patients reporting levels of mobility, self-care, usual activities, pain/discomfort and anxiety/depression using a 3-point scale. (For a summary of the proportion of responses at each level of each domain for each time point of follow-up, see Table 38.) The responses for the domains were converted to quality-of-life valuations using the Stata command ‘eq5d’ (Stata/SE version 12; StataCorp LP, College Station, TX, USA), which uses a value set based on the preferences of a large sample of the UK population to produce a EQ-5D index score for each patient at each time point. The possible range for the index scores when using the UK value set runs from –0.594 to 1, with a higher index indicating a better quality of life and an index of zero representing quality of life equivalent to death. An EQ-5D index score of zero was assumed for patients who had died.

The EQ-5D index scores were used to estimate the total QALYs for each patient between the 8-, 12- and 24-month follow-up points, as well as the total QALYs at 24 months. QALYs were calculated from the area under the curve after linear interpolation of the EQ-5D index score between time points. A discount rate was applied to QALYs after 12 months, at a rate of 3.5% per annum. 78

Dealing with missing data

Table 29 shows the number of complete cases (individuals with complete data for all resource-use, cost and QALY outcome variables) at each follow-up point by treatment arm. Missing data were handled using multiple imputation methods. Imputation via chained equations (with the ‘mi impute chained’ command in Stata) was used to impute the missing values in the original data set, based on all other variables included in the imputation model, to produce a specified number of complete data sets. Three imputation models were constructed for (1) resource-use parameters (number of health-care visits and number of nights in hospital), (2) reported out-of-pocket costs (medications, transportation, private health care, lost earnings) and (3) the five EQ-5D-3L domains at each follow-up point. Thirty data sets were produced for this analysis to align with the greatest level of missingness among the imputed variables. The data sets were then combined using the ‘mi estimate’ command in Stata, which applies Rubin’s rules when combining estimates from multiple imputed data sets to account for variation both within and between data sets. Imputation was conducted separately for initial hospitalisation, follow-up resource-use/costs and quality-of-life data. All imputation models included variables for patient characteristics (age, sex, size of tear, duration of shoulder problems at baseline) and indicators for centre and treatment allocation. For quality of life each domain of the EQ-5D-3L was imputed (rather than the index score directly) and baseline EQ-5D-3L domains were also included in the imputation model.

Incremental analysis

Under conventional decision rules, specifically those used by the National Institute for Health and Care Excellence (NICE), the cost-effectiveness of a technology is assessed using the incremental cost-effectiveness ratio (ICER), which is based on the additional cost per QALY gained compared with the current best practice. The guidance for methods of economic evaluation produced by NICE78 states that a technology with an ICER below £20,000 per QALY gained is likely to be deemed cost-effective. Technologies with ICERs above £30,000 are unlikely to be considered cost-effective unless additional justification is provided. In addition, the likelihood of a technology being accepted as cost-effective decreases as the ICER increases from £20,000 to £30,000.

Results

This section presents the results of the economic analysis. Resource use during initial surgery (implants, theatre time) and the follow-up period (appointments with health-care professionals, hospital visits and so on) is presented, along with the associated costs up to 24 months after surgery. Results were adjusted for covariates when specified [EQ-5D score at baseline (for QALY outcomes only), age, tear size, centre]. All results are based on imputed data unless ‘original data’ is specified.

Four analyses are presented for the economic outcomes:

1. ITT, no adjustment for covariates – base case

2. ITT, adjusting for covariates

3. per protocol (only those patients receiving the allocated treatment), no adjustment for covariates

4. per protocol (only those patients receiving the allocated treatment), adjusting for covariates.

Covariates for age, tear size and centre were included when calculating mean cost and effect differences for analyses (b) and (d). Effect differences for analyses (b) and (d) were also adjusted for EQ-5D index score at baseline.