The ProFHER (PROximal Fracture of the Humerus: Evaluation by Randomisation) trial - a pragmatic multicentre randomised controlled trial evaluating the clinical effectiveness and cost-effectiveness of surgical compared with non-surgical treatment for proximal fracture of the humerus in adults

Article history

The research reported in this issue of the journal was funded by the HTA programme as project number 06/404/53. The contractual start date was in March 2008. The draft report began editorial review in December 2013 and was accepted for publication in June 2014. 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

Amar Rangan has obtained grants and personal fees from De Puy Ltd and grants from JRI Ltd; both are outside the submitted work. In addition, he has a UK and European patent application pending. None of these influenced the trial or the report.

Copyright statement

© Queen’s Printer and Controller of HMSO 2015. This work was produced by Handoll 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.

Proximal humeral fractures

Fractures of the proximal humerus are those occurring at the top end of the humerus or upper arm bone. An estimated 706,000 proximal humeral fractures occurred worldwide in 2000. 1 In the same year, an incidence of 63 proximal humeral fractures per 100,000 adults was reported for the population (534,715 adults) served by the Edinburgh Royal Infirmary,2 and an incidence of 79 humeral fractures, of which 61 were likely to be proximal humeral fractures based on fracture distribution statistics,2 per 100,000 adults aged > 20 years was reported in England and Wales. 3 Based on 2001 UK census data, this amounted to an estimated 23,788 proximal humeral fractures in England and Wales (population 38,997,087 aged > 20 years). 4 These fractures accounted for 5–6% of all adult fractures in studies conducted in Scotland2 and Denmark. 5 Their incidence increases rapidly with age, with the majority occurring in people aged > 65 years. 2,5 Women are affected between two and three times as often as men. 2,5,6 Most of these fractures result from low-energy trauma, predominantly a fall from standing height. 5,7 The primary injury mechanism, the gender ratio and the age distribution point to underlying osteoporosis. 7,8 Similar to other primarily osteoporotic fractures, the incidence and age-specific incidence of these fractures are increasing. 8

The immediate and long-term consequences of proximal humeral fractures are substantial. Especially in the elderly, who will often have other comorbidities, these fractures can result in lengthy inpatient rehabilitation and, for some, a loss of independence such as a permanent move to a nursing home. 9 Hospital Episode Statistics (HES) data for England10 over a 9-year period (2003–12) consistently show a mean inpatient stay of > 1 week and that just over half of these episodes relate to patients aged ≥ 75 years (see Appendix 1). The outcome following treatment of these fractures is frequently unsatisfactory. Poor shoulder function and pain are common long-term outcomes, with a substantial proportion of patients, even those with less severe injuries, continuing to report some or severe disability at 2 years. 11 These injuries are also associated with an increase in mortality. 12

Although dwarfed by the costs of hip (proximal femur) fracture, for which the bulk of specific economic cost data on osteoporotic fractures are available, the costs of proximal humeral fractures are still considerable and rising. One study reporting data collected from 1989 to 1991 on residents of Olmsted County, MN, USA, reported a 0.21 cost ratio of treating a humeral fracture compared with a hip fracture in the first year. 13 Again in the USA, a retrospective study of fracture-related direct medical costs in 2000 found that the cost per fracture per year for a humeral fracture was 21% that of a hip fracture (US$5567 vs. US$26,856 at 2003 Medicare fee schedule payment levels). 14 Given the 3.2% increase in the surgical management of proximal humeral fractures (most hip fractures are managed surgically) between 1999–2000 and 2004–5, it is likely that the relative cost of these fractures has increased in the USA. 15 A report providing direct health-care costs of injuries to the shoulder, arm and wrist in the Netherlands, adjusted to 2007 prices, reported that upper arm fractures (annual average of 9038 proximal humeral and humeral shaft fractures) were the most expensive injuries per case (€4440) and that their overall annual cost was approximately €40M out of an estimated €290M cost for upper extremity injuries. 16 It was suggested that the increase in the cost of fracture care in ‘elderly women’ from a previous report of costs in the Netherlands was partly because of a higher number of patients receiving operations for these fractures. 16

Fracture morphology, classification and epidemiology

Over 99% of proximal humeral fractures are closed fractures in which the overlying skin remains intact. 17 The most commonly used classification of these fractures is that of Neer. 18 Neer based his 16-category classification primarily on the relative positions of the four main segments of the proximal humerus: the humeral head, the greater tuberosity, the lesser tuberosity and the humeral shaft (see Figure 1). Although these may be delineated by fracture lines, a segment is only considered a ‘part’ if there is displacement of > 1 cm or 45° angulation. A ‘minimally displaced’ or ‘undisplaced’ fracture occurs when the displacement criteria are not met for any of the four segments. For ease of description and to avoid the misinterpretation that these are all ‘undisplaced’ or ‘minimally displaced’ fractures, in common with other researchers,19 we generally refer to such fractures as one-part fractures in this report. The other categories – two-part, three-part and four-part fractures – involve the relative displacement of two, three or all four segments respectively. Additional Neer categories involve fractures associated with an anterior or posterior humeral head dislocation and fractures involving the articular surface of the humeral head. Appendix 2 shows the Neer classification system, with numbering of the categories by Sidor et al. 19

FIGURE 1.

Proximal humerus anatomy and the four segments. (1) Shaft, (2) greater tuberosity, (3) lesser tuberosity and (4) head.

The dotted lines in Figure 1 represent lines of epiphyseal scar where the four segments that ossify separately fuse at skeletal maturity. Codman made a key observation that fractures tend to generally occur along lines of these epiphyseal scars in the proximal humerus. 20

A large prospective epidemiological study involving 1027 fractures found that around half (51%) of proximal humeral fractures are displaced according to Neer’s criteria. 7 Of these, the largest groups involved two-part surgical neck fractures (28% of the whole population) and three-part greater tuberosity and surgical neck fractures (9%). Three-part lesser tuberosity and surgical neck fractures were rare (0.3%). Four-part fractures without fracture dislocation comprised 2% of the total. Overall, approximately 40% of proximal humeral fractures involve a displaced surgical neck. Figure 2 shows an example of a two-part fracture of the surgical neck and subsequent internal fixation.

FIGURE 2.

Two-part fracture of the (a) surgical neck with (b) subsequent internal fixation with plate and screws.

Limitations of the Neer classification include the arbitrary thresholds for displacement; the absence of key potentially prognostic fracture patterns, particularly those relating to the relative positions of the shaft and the humeral head (varus and valgus displacement); and the findings of only fair to moderate interobserver and intraobserver agreement for classification. 21,22

Treatment: care pathways

People with these injuries typically present with a very painful and swollen shoulder to an accident and emergency (A&E) department soon after their injury. As well as having severely limited movement of the shoulder, there is often deformity and bruising. On radiographic (generally plain radiography) confirmation of their fracture, patients are either sent home with their arm in a sling and given analgesics or admitted to hospital for social or other health-related reasons. (In rare instances there are serious complications, such as an open fracture or major vascular injury, necessitating emergency surgery.) Subsequently, patients are seen by an orthopaedic specialist in the next fracture clinic at the hospital. A second set of radiographs may be taken to further characterise the fracture. These are likely to be at least two views from the standard shoulder trauma series: anteroposterior in the scapular plane, scapular Y-lateral and axillary or modified axillary. 23,24

The majority of these fractures, including all undisplaced fractures, are treated non-surgically. Closed reduction of displaced fractures to reposition the fractured parts is generally not carried out. Non-surgical treatment generally involves ‘immobilisation’ and support of the injured arm in a sling for around 3 weeks25 to facilitate initial fracture healing before allowing shoulder movement. Surgical treatment, which is often undertaken for more complex displaced fractures, usually involves either closed or open reduction of the fracture and internal fixation using various devices or replacing the humeral head with a hemiarthroplasty. In some cases, total joint replacement (including reverse polarity shoulder arthroplasty in which the ball part is uppermost, fixed to the socket, and the cup is fixed to the upper end of the humerus) may be performed. Common methods of internal fixation are plating, in which a metal plate is placed alongside the repositioned bone and secured using screws (see Figure 2b); intramedullary nailing, in which a nail is inserted into and along the medullary canal and usually ‘locked’ into place, such as with screws; and percutaneous fixation using pins, screws and/or wires after closed reduction. Bone grafts or substitutes may be used to fill bony voids. Typically, patients are placed on the next available list of a specialist surgeon and operated on within 3 weeks of their injury. Post-operative treatment generally involves a period of immobilisation. Whether treated surgically or non-surgically, rehabilitation, comprising a mixture of advice, instruction, physiotherapy and home exercises, is required, with the aim of restoring shoulder function and achieving functional independence. 25

Evidence underpinning practice

Historically, there has been a paucity of evidence from randomised controlled trials (RCTs) to inform the management of these fractures. In 2007, at the time of developing the PROximal Fracture of the Humerus: Evaluation by Randomisation (ProFHER) trial protocol, the Cochrane review (search date September 2006) on this topic included just 12 small trials (578 participants). 26 Of these, three heterogeneous and methodologically flawed trials (102 participants in total) compared surgery with non-surgical treatment. 27–29 The review concluded that there was ‘insufficient evidence from RCTs to determine which interventions are the most appropriate for the management of different types of proximal humeral fractures’. 26 More specifically, it concluded that ‘It is unclear whether operative intervention, even for specific fracture types, will produce consistently better long term outcomes’. 26 The review noted that difficulties in patient recruitment seemed a key problem in this area, particularly in trials involving surgery, as shown by the abandonment of three trials and the slow rates of recruitment and lower than planned recruitment rates in other trials.

Subsequent updates of the Cochrane review have noted a continuing inadequacy of the evidence. The update published in 2012 (search date January 2012) included six trials (270 participants) comparing surgery with non-surgical treatment but noted also six ongoing trials that aim to recruit 1052 patients in all. 30 All three of the more recent trials31–33 recorded validated patient-reported outcome measures of upper limb function. Such measures, which include the Disabilities of the Arm, Shoulder and Hand questionnaire34 and the Oxford Shoulder Score (OSS),35 were unavailable previously.

As well as a lack of evidence to inform on all aspects of management of these fractures (including optimal conservative or surgical treatment), there is evidence of substantial variation in practice. This includes the length of immobilisation and the timing and provision of physiotherapy for less severe fractures25 and the use of surgery. 15,36,37 Additionally, technology is changing all the time and there are various pressures for early implementation of new fixation devices15,37 and surgical techniques, for example. 38

Rationale for a trial comparing surgical with non-surgical treatment

Surgery is rarely essential for these fractures and there is a clinical consensus that it is not needed for the majority of non-displaced or minimally displaced fractures. As shown by the large variation in practice, there is uncertainty about the role of surgery for the majority of displaced fractures, most of which involve the surgical neck. The commonly perceived advantages of surgery are restoration of anatomy, the potential avoidance of symptomatic nonunion and malunion, and improved functional outcome. In particular, this last presumption is unproven. 30 Moreover, surgery involves additional trauma and risks in terms of anaesthesia and complications (such as wound infection) and sometimes surgical errors. 39 There is a need also to justify the greater initial costs associated with surgery, including the potentially increased inpatient stay. Later surgery for failed non-surgical treatment or major revision surgery for surgical complications is usually technically more difficult and less successful. There is some limited evidence indicating that initial surgery is associated with a higher risk of subsequent surgery. 15,30

Aim of the ProFHER trial

This key treatment uncertainty formed the rationale for the ProFHER trial. This was a pragmatic multicentre RCT whose primary objective was to evaluate the clinical effectiveness and cost-effectiveness of surgical compared with non-surgical treatment of the majority of displaced fractures of the proximal humerus involving the surgical neck in adults.

This objective was supported and supplemented by the following secondary objectives:

• To describe the study population in terms of key characteristics and in particular to categorise, using optimised methods including training, the trial fractures according to the Neer classification through an independent assessment of baseline radiographs of all randomised patients in the ProFHER trial.

• To ensure that both groups of patients receive comparable and good standards of care. This included remedying identified ‘gaps’ such as through the development of a leaflet to advise patients on self-care during sling immobilisation and a physiotherapy protocol in consultation with specialist shoulder physiotherapists and promoting the need to encourage home exercise.

• To explore the effect of age (< 65 years vs. ≥ 65 years), type of fracture (involving none vs. one or both tuberosities) and patients’ treatment preference on the primary outcome (OSS).

• To explore surgeon treatment preferences, including the relationships between surgeons’ decisions to exclude patients because of a lack of equipoise, and patient age (< 65 years vs. ≥ 65 years) and fracture type (involving none or one or both tuberosities).

Summary of trial design

The ProFHER trial was a pragmatic UK-based multicentre RCT, with a concomitant economic evaluation, of surgical compared with non-surgical treatment of acute displaced proximal humeral fractures involving the surgical neck in adults. 40 The choice of surgical intervention was the decision of the surgeons, who had to use techniques with which they were experienced and fully familiar. Good standards of non-surgical care, including care-provider competence, and a comparable provision of rehabilitation for both groups of patients were expected. The latter was facilitated through the provision to patients of an information leaflet on personal care during initial sling immobilisation and the development of a written physiotherapy protocol. We aimed to recruit and randomise 250 patients with these fractures and follow them up for 2 years. Our primary outcome was the OSS,35,41 collected at 6, 12 and 24 months via a postal questionnaire. There was no blinding of outcome assessment. However, coding, including the independent categorisation of baseline fractures, was carried out blind to allocation. All analyses were based on intention to treat (ITT). The final version of the study protocol (version 7), shown in Appendix 3, was published in 2009. 40 Appendix 4 summarises the changes to the protocol.

Randomisation: sequence generation and allocation concealment

Patients were randomised on an equal basis to surgery or non-surgical treatment by the remote randomisation service (telephone or web based) provided by the York Trials Unit (YTU), University of York. Throughout, randomisation was stratified by tuberosity involvement (yes or no) with allocation using random block sizes (these were four, eight and 12). The original intention, after a specified time, to use minimisation based on the minimisation factors of tuberosity involvement (yes or no) and study centre was not implemented (see Appendix 4). These measures were taken to avoid the risk of prediction of allocation. As fewer patients than expected were being recruited at centres, the initial concern about an imbalance leading to an excess in the allocation to surgery at a centre, with the associated cost and logistics of performing this surgery, no longer applied. Allocation concealment was assured by the treatment being assigned only after obtaining patient identifiers and key baseline data.

Participant eligibility

Only adults with an acute displaced unilateral fracture of the proximal humerus involving the surgical neck were eligible for inclusion. Allowance was made for surgical neck displacement that did not meet the exact displacement criteria of the Neer classification (1 cm and/or 45° angulation of displaced parts) when this reflected an individual surgeon’s equipoise (e.g. whether or not the surgical neck fracture was displaced enough to be considered for surgical treatment). Fractures involving dislocation of the humeral head out of the glenohumeral joint were explicitly excluded. Although fractures involving the articular surface (head-splitting fractures) were not explicitly excluded, it was anticipated that these rare fractures would not be considered eligible.

The following inclusion and exclusion criteria, which were applied throughout patient recruitment, were summarised on the study eligibility form (see Appendix 5). (The order of the exclusion criteria was adjusted for presentational reasons.)

Sample population

All patients meeting the main inclusion criteria (age ≥ 16 years, presenting within 3 weeks of injury, with a displaced fracture of the proximal humerus involving the surgical neck) were considered potentially eligible. Ineligible patients were defined as those who were excluded for one or more reasons given in the list of exclusion criteria or for another reason, such as a surgeon’s preference for one of the treatment options, reflecting lack of surgical equipoise, stated by the surgeon. Some eligible patients were not included because of a lack of patient consent (non-consenting patients), reflecting a strong preference for a specific treatment option or refusal of randomisation.

Study setting

Trial recruitment was undertaken in the orthopaedic departments of 33 of the 35 participating acute NHS hospitals; 33 participating hospitals were based in England and one each was based in Scotland and Wales. Patient care pathways including outpatient and community-based rehabilitation were as per routine practice. Patient recruitment, which was from fracture clinics or orthopaedic wards, lasted from 17 September 2008 until 13 April 2011, when the last patient was recruited. Follow-up was for 2 years.

Interventions

Each centre in the trial had to agree to forgo the introduction of radically novel and experimental interventions for these fractures during the recruitment period. The centres were also responsible for identifying surgeons with sufficient expertise for participation in the trial. We emphasised to the participating centres that central to obtaining reliable evidence was that good standard care, both surgical and non-surgical, was provided throughout the trial. It was expected that, as part of good standard care, surgery for these types of fractures would usually be carried out by specialist surgeons, who were experienced in operating on these fractures. When possible, decisions about the method of surgery, when allocated, and non-surgical treatment were left to the clinical judgement of the participating clinicians. Participating surgeons were advised that they must, however, use surgical interventions and procedures with which they were familiar (and thus experienced with). This was to avoid learning curve problems. 42,43 Similarly, other clinicians, including physiotherapists, were advised that they should use procedures with which they were familiar. Additionally, prescribed deviations from the planned surgery and the physiotherapy protocol for rehabilitation provided for the trial had to be documented and reasons given.

Outcome measures

Classification of study fractures

Copies of all baseline radiographs for all randomised patients were collected. These were processed to facilitate an independent and blinded classification of the study fractures based on the Neer classification system at the end of study recruitment by two orthopaedic specialist shoulder surgeons experienced in assessing and operating on proximal humeral fractures. This was preceded by a training session on the Neer classification, which has been shown to improve inter-rater agreement. 22

Sample size

The primary outcome for the trial is differences in patients’ subjective assessments of pain and ADL as measured by the OSS. To justify both the increased cost of surgery and the exposure to the hazards of surgery, there need to be greater improvements in patients’ subjective assessments of pain and ADL in the surgery group than in the conservative treatment group. There continues to be a lack of data to inform on the minimum clinically important difference for any validated patient-reported outcome measure for these fractures. Observational data collected by the chief investigator and his colleagues found that those patients who had surgery without subsequent complications [1- to 5-year data from 50 patients treated with a PHILOS^® plate (Synthes Ltd, Welwyn Garden City, UK) between 2001 and 2005] had a 5-point differential improvement in OSS compared with patients treated conservatively (3- to 12-month data from 103 patients treated between April 2000 and July 2003). 50 This 5-point difference was judged to be of clinical importance. Given a standard deviation (SD) of 12 this equated to an effect size of 0.42. It was calculated that to detect a minimum effect size of 0.4 with 80% power and a two-sided 5% significance level 100 patients in each treatment group would be required. After allowing for a potential loss to follow-up of 20%, we proposed to recruit and randomise 250 patients (125 in the surgery group and 125 in the control group). (This did not take account of any potential cluster effect, such as from similarities between surgeons within centres, as we did not expect there to be many patients treated by individual surgeons. 51)

Estimating the numbers of patients and centres needed to recruit

There were insufficient good-quality prospectively collected data to inform our estimates of the number of trial centres and duration of recruitment needed to achieve our target of 250 trial participants. Although data from the epidemiological study conducted in Edinburgh7 were used, particularly in terms of the relative proportions of Neer categories in our intended fracture population, using these as a basis for estimating the sample populations in individual hospitals was hampered by the unavailability of accurate data on the catchment populations of individual hospitals. There is also the unknown but expected variation in risk factors in these populations. Another approach of directly asking putative principal investigators when developing the study proposal yielded rough estimates without reliable data on how many proximal humeral fractures were treated at a centre and what proportion of these might be treated surgically. In December 2006 we sent each of the six putative principal investigators a form requesting information that would inform on the recruitment potential of the site. This included estimates of the numbers of people presenting with proximal humeral fractures at individual centres each year, how many of these had surgery and how many would be eligible for inclusion. We received two completed forms, one provided by the chief investigator. This lack of information on centre activity relating to proximal humeral fractures was mirrored throughout the trial. Furthermore, we used HES data to obtain a general impression of overall activity at a hospital (taken as a proxy for catchment population) and specific proximal humeral fracture activity, while recognising that patients with less serious (minimally displaced) fractures may not have been admitted. Appendix 6 gives details of the calculations used to estimate the size of our population sample using the HES data.

Governance

Data handling

The patient questionnaires and hospital forms (i.e. CRFs) were designed using TeleForm software (version 10; Cardiff Software, Cambridge, UK). The CRFs were marked up in TeleForm with variable names and appropriate scoring. When the CRFs were completed and returned to the YTU they were prepared for scanning by a data entry clerk using the TeleForm software. When a form would not scan, the data were manually entered. When a form was scanned, the data were then verified depending on what TeleForm had identified as requiring correction. The verified data were then downloaded into the study database and were available for second checking. This involved each hard copy of the CRF being compared against the entry stored in the study database and correcting the electronic data as necessary. All data that had been scanned, downloaded and second checked were then ready for validating. The data validation was undertaken by the data manager who applied predetermined rules to check that the data recorded for the variables on the CRF, such as whether or not the dates when the patient injured his or her shoulder and was assessed for eligibility fell within the 3-week time frame for being eligible for the trial. Data that had been validated were then available to the trial statistician and health economist.

Data checks, cleaning and processing

The MREC did not allow any return to patients for missing data. Data checks, including those conducted by the data manager, alerted us to recurrent problems with completion of hospital forms during trial recruitment.

Data ‘cleaning’, before the staged closure of the data set according to pre-established time points, was carried out in a systematic manner with consultation on and discussion of any amendments.

Categorisation and interpretation of text comments that were recorded on the various data collection forms was generally performed by two independent and blinded members of the TMG, who then discussed disagreements to reach a consensus. The criteria for assessment were documented, often beforehand. When categorisation was performed by one rater, a second rater was asked to check the results and criteria.

Clinical effectiveness analysis

All analyses pertaining to clinical effectiveness were conducted to a prespecified analysis plan, endorsed by the DMEC. Signatures of approval of the final version (version 19; August 2013) were obtained from the chief investigator (AR), trial manager (SB) and ‘co-applicant’ (HH) before release of results by the statistician (AK). ITT analysis, including all randomised patients in the groups to which they were randomised, was conducted throughout. All main analyses were performed using Stata 12 (StataCorp LP, College Station, TX, USA) using two-sided significance tests at the 5% significance level.

Economic analysis

Economic analysis was likewise conducted to a prespecified analysis plan shared with the DMEC. Key differences from the protocol are described in Appendix 4. The analysis was carried out from the perspective of the UK NHS and Personal Social Services. 55 Data on health utilities, obtained from the EQ-5D data collected from patient questionnaires, were converted into quality-adjusted life-years (QALYs) for each patient using the area under the curve (AUC) method. Costs were expressed in UK pounds sterling at a 2012 price base.

Differences in mean costs and QALYs at 2 years were used to derive an estimate of the cost-effectiveness of surgery and non-surgical treatment. Multiple imputation56,57 instead of the method of Lin et al. 58 was used to derive the data set for the base-case analysis. ITT analysis was conducted throughout. An additional analysis was conducted for complete cases (in which patients with any missing data are excluded). All analysis and modelling were undertaken in Stata 12. The full methods used are detailed in Chapter 7.

Standardisation of trial procedures and care programmes

Throughout the conduct of the trial we emphasised the importance of good practice from clinical and research perspectives; standardised protocols and care pathways; and comparable and sufficient expertise of care providers in the two groups. As far as possible we adopted procedures that reflected and were compatible with usual practice and that did not delay treatment, add unnecessarily to the clinical workload or place a burden on participants. Specific areas of focus are described in the following sections.

Patient recruitment

This section describes the typical processes involved in the identification of potential participants and trial recruitment. Trial recruitment packs, which included all relevant forms and materials for potentially eligible patients (thus those meeting the trial inclusion criteria), were provided by the YTU. Each potentially eligible patient had a unique four-digit identification number that was prerecorded on the individual forms in the individual recruitment packs. Freepost envelopes were provided. To assist centres further in keeping track of form returns, a participant log listing the various forms and materials for return to the YTU was provided for the designated person to record the dates when individual forms were returned to the YTU.

Obtaining informed consent and randomisation

The designated person provided the patient with the patient information leaflet (see Appendix 9) and answered their questions before asking for their consent to participate in the trial. Although the patient could decide immediately, it was made clear that he or she could defer the decision for up to 24 hours to think about it further and/or discuss it with family or friends. The result of the decision was recorded by the designated person on the consent status form (see Appendix 15). When the patient did not consent, the designated person requested that the orthopaedic surgeon answer the questions on the consent status form on their advised treatment for the patient, the patient’s preferred treatment if it had been expressed and the agreed treatment. This form and the study eligibility form were then returned by the designated person to the YTU.

When the patient consented to take part, he or she signed the consent form (see Appendix 16) and completed the baseline information form (see Appendix 17) before randomisation. Of note is that the baseline information form specifically gave advice about the desirability of stopping smoking for those who admitted they smoked. The designated person then contacted the YTU either by a free telephone service during weekday working hours or via the internet at any time to access a password-protected randomisation service. When randomising the patient, the designated person provided key identification information (e.g. name and date of birth) and basic descriptive details about the patient (gender, fracture tuberosity involvement, location of recruitment: ward or fracture clinic) and answered a series of questions in relation to study eligibility. On receipt of these details and confirmation of study eligibility and patient consent, the random allocation (surgery or not surgery) was given. This was recorded on the baseline information form in addition to informing the patient about the treatment that he or she was to receive.

On randomising an eligible patient, the designated person returned the study eligibility form, consent status form, a copy of the consent form, the baseline information form and copies of the baseline radiographs to the YTU in the Freepost envelope provided. A copy of the consent form was also given to the patient, the original being kept in the medical notes. A letter was sent from the YTU to each trial participant to thank them for agreeing to take part in the trial. Additionally, each patient’s GP was notified of the patient’s enrolment into the ProFHER trial and provided with a copy of the patient information sheet.

A flow chart showing the patient recruitment and hospital data collection processes is shown in Figure 3.

FIGURE 3.

Flow chart of patient recruitment and hospital data collection. a, The series of radiographs used to confirm the fracture should be, for example, anteroposterior, axillary (or modified axillary) and scapular Y lateral views; b, after assessing eligibility, patients may need at least 24 hours to decide whether or not to consent to take part; c, the surgical form was also to be completed for any patient randomised to non-surgical treatment who then underwent surgery; and d, the inpatient episode form for a patient not randomised to surgery was not required if the patient was not admitted to hospital.

Data collection

Questionnaire on use of the ProFHER trial sling immobilisation leaflet and immobilisation

At the end of trial recruitment, a two-page questionnaire on sling use was sent to the designated person at each participating centre (see Appendix 30). The questions on the first page were aimed at gathering evidence on the implementation of a key standard of care by obtaining feedback on the use of the sling leaflet provided for the ProFHER trial or alternative leaflets on care during immobilisation provided to patients eligible for participation in the trial. The second page asked about immobilisation (type and duration) typically provided to non-surgical patients and to surgical patients after surgery who would have been eligible for the ProFHER trial. This was to obtain additional insights on this aspect of care, in particular on post-surgical immobilisation, for which we had not collected patient data.

Data checks and ‘cleaning’: specific actions

Data checks alerted us to recurrent problems with completion of two hospital forms during trial recruitment. Early on, missing data for the consent status form in terms of which treatment surgeons would advise for non-consenting patients was remedied by adding in a box to allow for uncertainty (see Appendix 15). The surgical form (see Appendix 21) proved troublesome in terms of data entry for numbers of personnel involved, which were important data for the economic analysis. A follow-up visit to centres by the trial manager and advice in a monthly newsletter were strategies used to address this.

During data ‘cleaning’, two sources of problems were identified and corrective measures taken. A scrutiny of ‘other reasons to exclude the patient’ in the study eligibility forms revealed a very few cases in which lack of patient consent was recorded rather than non-eligibility. There were also some sequence and duplication problems in the recording of dates of physiotherapy sessions in the physiotherapy treatment logs (see Appendix 23).

Processing of data on complications and new therapy including surgery

A file of blinded data for each trial patient from the relevant fields in three forms [inpatient episode form (see Appendix 22), 1-year follow-up form (see Appendix 25) and 2-year follow-up form (see Appendix 26)] was compiled by the trial statistician (AK) to facilitate the independent screening and coding of complications and new therapy by two raters (HH and AR). Any disagreement on the processing, coding and categorisation of these items was resolved by discussion.

In the first stage, the complications and associated treatments were grouped into seven categories: complications related to the proximal humeral fracture and/or its treatment, gastrointestinal events, cardiac and peripheral vascular events, malignancy, falls and other fractures, respiratory events and ‘other’. Processing of the first of these categories is described in the following section.

Adverse events classification

A protocol was developed to inform the reporting and classification of adverse events. This established the following:

• Appraisal of adverse events would be based only on the adverse event reporting forms and thus would be independent of information on complications from hospital forms such as the 1-year follow-up form.

• Events reported after 2 years’ follow-up would not be included.

• We would present separate results, divided by treatment group, for SAEs and non-SAEs. In these two categories, we would report the number of participants who had an adverse event and, as some participants had multiple events, also the number of adverse events.

• We would report the classification of the adverse event based on the judgement of the principal investigator at the centre reporting the adverse event of whether or not the event was (a) related and (b) expected.

• Two reviewers (HH and AR) would independently group, without knowledge of treatment allocation, the adverse events based on the categories used on the hospital forms. Facilitated by the trial manager (SB), consensus would be reached through discussion when there was disagreement on the categories and on the brief descriptions of the events added to the data file by the reviewers. Box 1 presents the agreed categories.

Baseline radiographs: processing and quality assessment

Initial processing of the radiographic images, all of which were in digital format, involved checks on blinding (removal of patient and hospital details), labelling with the correct patient identifier and ensuring that the images could be accessed and transferred to a YTU computer and CD. The designated person and/or the contact in the radiography department was contacted to resolve difficulties at individual centres. For blinding purposes, before sending out for quality checks (see following paragraph) and eventual classification, the sets of radiographs were renumbered using a three-digit code (e.g. 039), with letters used to label each of the radiographs within a set (e.g. 039a, 039b).

At various times during trial recruitment, and subsequently to complete the process, there was a review of the quality of the copies of the radiographic images for each trial participant provided by trial centres. This aimed to facilitate the Neer classification of the study fractures by two independent assessors by ensuring that such images were available and of sufficient quality for this purpose. We undertook a two-phased assessment for each centre whereby the quality of the radiographs for the first five participants at each centre was independently assessed by two orthopaedic surgeons, one of whom was the chief investigator (AR) and other of whom was a principal investigator (JC); the assessment of the radiographs of subsequent participants from each centre was carried out by AR only. When there were concerns about the quality of the radiographs being provided, this was to be raised by AR with the principal investigator of the participating centre to see whether or not better copies could be obtained. The criteria that were used to judge the quality of the radiographs were:

• Are there at least two projections in planes perpendicular to each other? (yes/no)

• Are the proximal humerus and glenohumeral joint seen on each projection? (yes/no)

• Is it possible on the two views to clearly identify the shaft, greater tuberosity, lesser tuberosity, head of the humerus and the glenohumeral joint? (yes/no).

The review also included monitoring for a clear breach of the main inclusion criterion; thus, fractures should involve the surgical neck. The trial manual alerted centres to this intention to ‘monitor the eligibility of patients entering the trial’, stating also that any findings from the interim review would ‘not be referred back to individuals but applied generally’. However, there was the potential to report to the TSC for consideration of discontinuation of recruitment at a site where there were ongoing problems with poor-quality images and serious concerns about patient eligibility.

On discussion with the TSC of the interim findings on radiograph quality, the independent orthopaedic surgeon member of the TSC (Professor Andrew Carr) volunteered to act as a third rater for resolving disagreements in the final stage of the quality audit. This staged audit identified 46 radiograph sets in which at least one out of the three quality criteria had not been met, as judged either by AR alone (n = 28) or, after arbitration by AC, by two independent raters (n = 18). This meant that 18% of the baseline radiograph sets were likely to present future difficulties for the independent Neer classification of fractures.

Preparations for the independent assessment and classification of radiographs based on the Neer classification

Key to the Neer classification of the radiographs is the facility to assess whether or not there is ≥ 1 cm displacement of parts, with clear delineation of the edge of the bones. (The assessment of angular displacement is fortunately not size or scale independent.) Hence, to measure this displacement, a life-sized image or a scale is required. In retrospect it was discovered that, for some Picture Archiving and Communications Systems (PACS), the scale as well as patient details were removed when the images were anonymised. On investigating this, we discovered that it could be remedied by saving the radiograph image as a picture and then removing the patient details manually. However, this may not be an acceptable procedure at some hospitals and is still subject to scaling errors. We learnt, through discussions with a radiographer (Brian Cox) at James Cook University Hospital, Middlesbrough, that a calibration object such as a scaling ball, placed in an appropriate location, is required at image acquisition to enable the accurate measurement of scale. We identified two albeit suboptimal approaches of limited applicability to assist the assessment of linear displacement in our study sample:

• For a limited number of radiographs for which standard left/right markers are used, the stems of the ‘L’ and ‘R’ are 9 mm. It was therefore possible to measure the stem of the ‘L’ or ‘R’ marker on the image to adjust for the scale when measuring for displacement.

• Software such as Microsoft Office Picture Manager (Microsoft Corporation, Redmond, WA, USA) could be used to allow a JPEG file to be opened and viewed at the original size that it was saved at. (This still may not be a true result if the image was manipulated in some way before it was saved.)

In addition to being able to present the best-quality images available with the right scale to accurately assess displacement, we also considered what training was necessary for the two raters who would be reviewing the radiographs. We therefore revisited a systematic review22 and various individual reports of studies investigating inter-rater and intra-rater reliability for classifying proximal humeral fractures using the Neer system, noting in particular the approaches taken for assessing displacement and training. 19,63–65 We also approached Stig Brorson (Herlev University Hospital, Herlev, Denmark) for details of the training programme used in his RCT, which showed that interobserver agreement is improved with training. 65 Brorson reported that the authors of the RCT had chosen to include only material from Neer’s original articles (22 and 29 March 2012, personal communication). 18,66 Eighteen illustrations from the first article and two from the second article were reproduced and incorporated into two 45-minute teaching sessions. We decided against adopting this approach as we considered that using actual ProFHER trial images was more relevant for facilitating the training of the two raters. This decision was consistent with our aim to have continuity in the images available for the decision-making process – from those seen by the recruiting surgeon to those seen by the independent assessor.

When considering which of the baseline radiographic images should be available for the two raters to review for the Neer classification we decided that it would be in excess to have more than four images per set of radiographs. This is because we expected that some images were poorer-quality duplicates of standard views or uninformative radiographs that would not contribute to the Neer classification. On provision of the baseline radiographs and a table listing all of the radiograph image sets, AR selected the best-quality projection (based on prespecified criteria) in each perpendicular plane. The duplicates in these planes that were of poorer quality were removed, with the reasons for exclusion being recorded. The decisions that AR made were checked by two others (SB and HH), who sought clarification when necessary.

We developed a protocol, training presentation and data collection tool (proforma) to aid in the description and assessment of the quality of each set of radiographs and the Neer classification of the study fractures. Throughout, the purpose of quality assessment was to establish if the sets of radiographs were adequate to permit their classification and not if they were adequate for use in clinical practice. As well as collecting data on the displacement of structures according to the Neer classification, the proforma facilitated the recording of ‘involvement’ (any indication of a fracture), a fully displaced surgical neck fracture (bone parts/fragments do not overlap), whether or not the head segment was in varus or valgus and whether or not the fracture was impacted.

The training presentation and data collection process were piloted with the help of two senior orthopaedic registrars at the James Cook University Hospital, Middlesbrough, on 6 August 2012. For this pilot, AR selected 10 sets of patient radiographs based on the following criteria: with and without adequate markers (i.e. with markers that were added to the cassette at the time of image acquisition and with markers that were added using PACS); a selection of two, three and four views (which for the last should include a couple with similar anteroposterior views); and a choice of two- compared with three- and four-part fractures so that some would have tuberosity involvement. After AR had delivered the presentation, which described the objectives of the Neer classification study and provided training in the interpretation and classification of radiographs, he and the two registrars independently classified 10 sets of radiographs according to the Neer classification using the proforma. A ruler and goniometer were made available to the raters, who were advised to use these. The trial manager (SB) acted as a general facilitator, including timing the various processes and documenting the proceedings, including the inter-rater differences after each of the two batches of five assessments. This enabled the three raters to discuss their differences and decide on the Neer classification for each fracture. The insights from the pilot led to some adjustment of the proforma (see Appendix 33 for the final version), clarification of the advice on interpretation, development of a briefing document on the Neer classification of radiographs and a form to record discrepancies between the two raters and their resolution and development of a realistic timetable for the main assessment of the baseline radiographs and classification of the study fractures according to Neer.

The main Neer classification of the radiographs was undertaken by two independent consultant orthopaedic surgeons (from the UK) who were experienced in treating proximal humeral fractures and who had comparable experience to that of surgeons in the ProFHER trial. This differed from our original intention to use an independent panel of musculoskeletal radiologists or orthopaedic surgeons with specific expertise in the Neer classification. However, it was in keeping with the pragmatic design of the trial and, from our experience of the audit described earlier and the literature, it was clear that involving more raters does not improve agreement. When agreeing to perform this role, both surgeons confirmed that they had the written support of their clinical director to commit to the requirements for participation. This comprised attendance at the training day (held on 17 January 2013), protected clinic time (five sessions: 20 hours) to undertake assessment of the radiographs and up to a day to achieve consensus when required. They also confirmed that they would ‘complete the assessment of radiographs for the whole study population according to the agreed procedures and undertake to deliver a definitive categorisation of the fractures, if necessary via an adequately-documented consensus process, within the timeline for this study’ (letter from AR to the two orthopaedic surgeons setting out the terms, 6 November 2012).

The same format and number of sets of radiographs were used for the main study training day as in the pilot. Thus, after the presentation by AR with examples, the two independent surgeons looked at the first five sets of radiographs and reviewed discrepancies; this was followed by assessment of the second five sets of radiographs, with a review of discrepancies. Subsequently, the surgeons returned copies of their completed data collection forms to SB, who collated the results and returned a table indicating where there were differences in the Neer classification for individual fractures. The two surgeons then met up to resolve these differences and document the decisions behind each of the final verdicts.

Patient recruitment

The first patient was recruited on 17 September 2008, which was during the short 3-month pilot phase, and the 250th patient was recruited on 13 April 2011. Thus, patients were recruited over a 31-month period; the trial stopped when the recruitment target was met. Figure 4 shows the monthly recruitment and accrual set in the context of the revised projections submitted with our successful application for an extension to recruitment from the original 18 months (from the official start time of the main trial recruitment of 1 October 2008) to 36 months.

FIGURE 4.

Patient recruitment and accrual in the ProFHER trial. Cum., cumulative.

Ultimately, 32 out of 35 UK-based centres that had been set up to recruit after approvals had been received recruited a patient. One of the three centres that did not recruit had started screening but identified only ineligible patients. A list of the participating centres, and the primary care trusts for which R&D approval for follow-up was obtained, is given in Appendix 34. A full list of health-care collaborators at the participating centres is presented in Appendix 35.

The 35 participating centres recruited a median of five patients [interquartile range (IQR) 2–9 patients]. The lead centre, James Cook University Hospital, Middlesbrough, recruited 43 patients and achieved the original centre recruitment target of one patient per month. Figure 5 shows the distribution of number of patients recruited by centre, ordered by when each site was set up to recruit.

FIGURE 5.

Patient recruitment by centre listed in order of when the centre was set up to recruit.

Population sample

Based on the return of correctly completed study eligibility forms and consent status forms, 1250 patients with a suitable fracture were identified, of whom 563 (45%) were considered eligible; of these 563 patients, 250 (44%) consented to take part in the study. Thus, of 1250 patients identified with a suitable fracture, 250, or 20%, were enrolled into the study. This figure compared favourably with our original expectation to recruit only 11% of patients with fracture types suitable for inclusion in the trial.

Patient consent

The percentage of eligible patients who consented to take part in the trial varied substantially in the 32 centres that recruited a patient, from 20% to 100% (Figure 6). Patient treatment preference data were collected but otherwise, patients were purposefully not asked for their reasons for not consenting. It is likely that there are various reasons for this variation in the rate of consent. It should be noted that all of the designated persons involved in consenting were provided with advice and training on the process involved but still had variable skills and experience in recruiting patients into a trial.

FIGURE 6.

Non-consenting and consenting patients by centre.

Participant flow

The flow of patients from screening to the 2-year follow-up as well as the data used for the primary analysis are summarised in a Consolidated Standards of Reporting Trials (CONSORT) flow diagram in Figure 7. Of the 250 patients randomised in equal numbers to surgery or not surgery, 215 (86%) were available for analysis at 2 years’ follow-up (85% in the surgical arm, 87% in the non-surgical arm). The trial was designed to allow for a 20% dropout rate.

FIGURE 7.

Participant flow through the trial. Max., maximum; min., minimum.

Baseline characteristics: all screened patients

The characteristics of the screened, ineligible, eligible consenting and eligible non-consenting patients at baseline are presented in Table 6. Eligible patients tended to be younger, presenting slightly later following injury, and were more likely to have fractures that involved tuberosities than ineligible patients. There were no marked differences between consenting and non-consenting patients.

Baseline characteristics of randomised patients

Consenting patients were stratified by tuberosity involvement (yes or no) at randomisation. Their baseline demographics and fracture details by treatment group are given in Tables 7–9. The treatment groups appeared to be balanced for these characteristics, including Neer category, with the exception of smoking status. Patients in the non-surgical group were more likely to be smokers (smoking status was included in a sensitivity analysis for the primary outcome; see Chapter 5). The same observations applied to the baseline characteristics of the 215 patients with OSS data at 24 months (see Tables 7 and 8).

Table 9 shows the results of the independent classification of the fractures according to Neer (see Chapter 3). The categories in bold (categories 3, 8, 9 and 12) are those that fully meet the ProFHER inclusion criteria, accounting for 221 of the 250 participants (88%). Consistent with the pragmatic study design, there were other fractures involving the surgical neck that did not meet the Neer criteria but which were sufficiently displaced for inclusion of these patients into the trial because of their potential suitability for surgery. The two surgeons undertaking the classification confirmed that all 250 fractures met the main study criterion, that is, that the fracture involved the surgical neck. Only category 10, involving humeral head dislocation, is a protocol violation. It was noteworthy that both fracture dislocations were categorised at the consensus meeting; before this, the two independent surgeons had assigned each of these fractures to category 8. This indicates some uncertainty and difficulty in the classification of these two fractures.

For the purposes of the sensitivity analysis, secondary to a subgroup analysis of tuberosity involvement, categories 1 and 3–5 were combined to form the one- or two-part fractures group, with the other four categories forming the three- or four-part fractures group (see Chapter 5). More detailed results and analysis of the baseline radiographs and the Neer classification results are presented in Chapter 6.

Interventions: surgery

Of the 125 patients allocated to surgery, 109 received surgery and 16 (13%) were treated non-surgically for the following reasons: difference in opinion of surgeon (n = 2), patient changed mind (n = 8) and patient unfit for surgery (n = 6).

Surgery took place on average 4.8 days (median 4 days; range 0–21 days) from randomisation and an average of 10.4 days (median 9 days; range 1–33 days) from the date of injury.

Interventions: non-surgical care

Brief details of the prescribed non-surgical treatment for these patients, exemplified by ‘sling immobilisation’, were collected using the treatment confirmation form. These details are summarised in Table 13. Of the 125 patients allocated to non-surgery, 78 (62.4%) were prescribed sling immobilisation, 35 (28.0%) were prescribed a collar and cuff; four (3.2%) were prescribed a polysling or equivalent and three (2.4%) were prescribed a hanging cast followed by a collar and cuff in one case and sling immobilisation in a second case. Data were missing for five patients. There was reference made to early mobilisation in six cases, to exercises in three cases and to physiotherapy in 18 cases. The prescribed duration of immobilisation was stated in four cases: 2 weeks, 2–3 weeks, 3 weeks and 6 weeks.

Information on sling immobilisation after surgery was not collected for individual patients. To gain further insights about this and the use of the ProFHER sling immobilisation information leaflet (see Appendix 12), we sent a survey to hospitals asking about current practice.

Findings from the questionnaire survey on sling immobilisation

All 35 participating sites completed the post-recruitment survey (see Appendix 30) on the provision of sling care information leaflets and sling immobilisation to patients eligible for participation in the ProFHER trial.

Physiotherapy

Details of the physiotherapy provided to each treatment group, collected from the physiotherapy logs (see Appendix 23), are provided in Table 17. This demonstrates equivalence between the treatment groups in access to and implementation of physiotherapy, including advice given for home exercises and performance of home exercises by patients.

Return of data collection forms

Other returns

Missing data

The multilevel model analysis used assumed any missing data to be missing at random (MAR). The extent of and reasons for missing OSS data at each time point are detailed in Table 22. Complete response is defined here as sufficient data (maximum of two missing OSS items) to allow the calculation of the OSS total score.

Overall, 209 patients (84%) had complete OSS responses at all follow-up time points (surgery: 103 patients; not surgery: 106 patients), whereas 41 patients (16%) did not respond or only partially responded at one or more time points (surgery: 22 patients; not surgery: 19 patients). The latter group are defined as non-responders in the following tables and analysis.

Demographic and fracture characteristics at baseline for responders and non-responders are presented in Tables 23 and 24 respectively. Each baseline variable was entered into a logistic regression model predicting non-response, none of which reached statistical significance (p < 0.10). Consequently, no additional variables were included in the primary analysis.

Table 25 provides the descriptive OSS total scores for complete responders and those non-responders for whom data were available at some of the time points. Complete responders had better OSS outcomes than non-responders at all time points. OSS scores were entered into a logistic regression model predicting non-response and the observed differences were statistically significant at 12 months’ follow-up (p = 0.025) but not at 6 or 24 months’ follow-up. Although this is a clear indication that data were not MAR, this is not anticipated to be a substantial problem given the relatively small number of non-responders.

Descriptive Oxford Shoulder Score statistics

Descriptive OSS statistics at 6, 12 and 24 months are presented in Table 26 and illustrated in Figure 8. Mean scores improved over time for both treatment groups. At 6 months, the surgery group showed better scores than the non-surgery group (a mean difference of 3.0 points), although the 95% confidence interval (CIs) overlapped. This difference was smaller than the sought effect size of 5 points. Mean scores at 12 and 24 months were very similar between the groups.

FIGURE 8.

Mean OSS scores (with 95% confidence intervals) over time by treatment group.

As suggested by the discrepancy between the means and the medians, the distribution of OSS scores was found to be left skewed (Figure 9), that is, a greater number of participants reported higher (better) shoulder function.

FIGURE 9.

Distribution of OSS scores over time. (a) 6 months; (b) 12 months; and (c) 24 months.

Primary analysis

The primary outcome was analysed using a multilevel model. Patients were treated as a random effect and time (6, 12 and 24 months) was nested within patients. The effect of treatment group was assessed while adjusting for time, group × time interaction, baseline EQ-5D index, gender, age group and tuberosity involvement at baseline. As no baseline characteristics were found to predict non-response, no further covariates were added.

Different covariance structures were applied to this model. An unstructured pattern that models all variances and covariances separately was judged to be the most intuitive. The more restrictive patterns of independent and exchangeable structures were tried and resulted in a worse model fit. Therefore, the unstructured covariance pattern was selected.

Diagnostics of model fit revealed non-normal standard residuals that were not uniform against predicted model values. This was likely to be a result of the left-skewed distribution of OSS scores. Therefore, all analyses were carried out on squared OSS values, which reduced skew and resulted in acceptable fit diagnostics. All group means and differences reported here were back transformed to the original OSS scores.

Adjusted OSS means and group differences for the model as specified above are presented in Table 27. The primary analysis (and subgroup analyses) included data from 231 patients (114 surgery, 117 not surgery) with a valid OSS score for at least one follow-up time point and complete baseline covariates. The analysis showed no significant differences between treatment groups at any time point, although the difference of 2.25 score points in favour of the surgery group at 6 months approached significance (p = 0.058). There was no overall effect of treatment group (difference of 0.75 score points in favour of the surgery group; p = 0.479).

Subgroup analyses

Age

The first subgroup analysis was undertaken to test the hypothesis that younger patients (aged < 65 years) would benefit more from surgical treatment than older patients. The relationship between age and treatment group in terms of OSS is illustrated in Table 28 and Figure 10. Older patients generally had lower OSS scores (including poorer shoulder functioning), and scores in the surgery group were higher than those in the non-surgery group at all time points for this age group, although the CIs overlapped in each case. There was a trend for younger patients in the non-surgery group to improve more rapidly than those in the surgery group and older patients.

TABLE 28 - Descriptive OSS statistics over time by treatment group and age group

Max., maximum; min., minimum.

Follow-up	Surgery (n = 125)	Not surgery (n = 125)	Total (n = 250)
Age: < 65 years (n = 108)
6 months
n	46	53	99
Mean (SD)	35.5 (10.87)	33.9 (10.11)	34.7 (10.45)
Median (min., max.)	39 (11, 48)	35 (3, 48)	38 (3, 48)
12 months
n	48	53	101
Mean (SD)	37.0 (11.6)	37.8 (10.53)	37.4 (11.01)
Median (min., max.)	42 (5, 48)	42 (4, 48)	42 (4, 48)
24 months
n	45	50	95
Mean (SD)	38.7 (9.95)	41.0 (8.36)	39.9 (9.17)
Median (min., max.)	43 (16, 48)	43 (14, 48)	43 (14, 48)
Age: ≥ 65 years (n = 142)
6 months
n	67	66	133
Mean (SD)	36.4 (9.40)	32.4 (11.72)	34.4 (10.76)
Median (min., max.)	39 (14, 48)	34.5 (3, 48)	37 (3, 48)
12 months
n	63	61	124
Mean (SD)	36.8 (10.20)	35.2 (11.09)	36.0 (10.63)
Median (min., max.)	40 (1, 48)	39 (7, 48)	40 (1, 48)
24 months
n	61	59	120
Mean (SD)	37.9 (9.94)	36.2 (12.41)	37.1 (11.21)
Median (min., max.)	41 (16, 48)	39 (1, 48)	40.5 (1, 48)

FIGURE 10.

Mean OSS scores (with 95% CIs) over time by treatment group and age group. (a) < 65 years and (b) ≥ 65 years.

The same multilevel model as for the primary analysis was applied but an additional treatment group × age group interaction was included. Adjusted OSS means and group differences for this model are presented in Table 29. Similar to the primary analysis, there was no significant overall effect of treatment group over the 2-year follow-up period and no significant effect of treatment group at individual time points. A 2-log likelihood comparison with the primary analysis model was not significant [χ²(1) = 1.24, p = 0.265]; therefore, inclusion of the interaction did not significantly improve the model.

TABLE 29 - Difference in mean OSS scores over time by treatment group: subgroup analysis 1 – age (< 65 years/≥ 65 years)a

Multilevel model adjusted for treatment group, time (6, 12 and 24 months), group × time interaction, baseline EQ-5D index, gender, age (< 65 years/≥ 65 years), group × age interaction and tuberosity involvement at baseline (yes/no).

Follow-up	Surgery (n = 125; n = 114 in analysis), mean (95% CI)	Not surgery (n = 125; n = 117 in analysis), mean (95% CI)	Difference (95% CI)	p-value
Overall	38.98 (37.21 to 40.68)	38.40 (36.66 to 40.07)	0.58 (–1.52 to 2.68)	0.590
6 months	37.74 (35.83 to 39.56)	35.69 (33.72 to 37.55)	2.06 (–0.28 to 4.40)	0.085
12 months	39.14 (37.29 to 40.91)	38.89 (37.08 to 40.62)	0.26 (–1.96 to 2.47)	0.822
24 months	40.03 (38.15 to 41.82)	40.48 (38.68 to 42.21)	–0.46 (–2.71 to 1.79)	0.690

Fracture type

The second subgroup analysis was undertaken to test the hypothesis that patients with more complex fractures would benefit more from surgical treatment than patients with less complex fractures. Baseline radiographs were initially assessed by surgeons at the time of assessing patient eligibility in terms of tuberosity involvement (primary subgroup analysis) and were later categorised by the two independent surgeons using the Neer classification (sensitivity subgroup analysis).

Tuberosity involvement at baseline

Surgeons indicated at baseline whether a fracture involved the greater tuberosity, the lesser tuberosity or both in addition to the surgical neck. This information was summarised as tuberosity involvement (yes or no). The relationship between tuberosity involvement and treatment group in terms of OSS scores is illustrated in Table 30 and Figure 11. Patients with fractures that involved one or both tuberosities generally had lower OSS scores (poorer shoulder functioning). Again, the scores in the surgery group were noticeably higher than those in the non-surgery group at 6 months, although the CIs overlapped. The differences between the two groups were greater for patients with fractures that involved neither tuberosity than for those with fractures with tuberosity involvement.

TABLE 30 - Descriptive OSS statistics over time by treatment group and tuberosity involvement

Max., maximum; min., minimum.

Follow-up	Surgery (n = 125)	Not surgery (n = 125)	Total (n = 250)
Neither tuberosity involved (n = 57)
6 months
n	24	29	53
Mean (SD)	37.6 (9.81)	33.6 (11.89)	35.4 (11.09)
Median (min., max.)	41.5 (18, 48)	36 (3, 48)	38 (3, 48)
12 months
n	22	28	50
Mean (SD)	38.7 (10.58)	36.6 (11.92)	37.5 (11.29)
Median (min., max.)	44 (14, 48)	43 (7, 48)	43 (7, 48)
24 months
n	21	26	47
Mean (SD)	39.0 (9.86)	40.0 (12.34)	39.6 (11.19)
Median (min., max.)	43 (16, 48)	45.5 (1, 48)	44 (1, 48)
Greater and/or lesser tuberosity involved (n = 193)
6 months
n	89	90	179
Mean (SD)	35.7 (10.05)	32.9 (10.78)	34.3 (10.48)
Median (min., max.)	39 (11, 48)	33.5 (3, 48)	37 (3, 48)
12 months
n	89	86	175
Mean (SD)	36.4 (10.84)	36.4 (10.57)	36.4 (10.68)
Median (min., max.)	40 (1, 48)	40 (4, 48)	40 (1, 48)
24 months
n	85	83	168
Mean (SD)	38.1 (9.97)	37.9 (10.53)	38.0 (10.23)
Median (min., max.)	41 (16, 48)	42 (3, 48)	41.5 (3, 48)

FIGURE 11.

Mean OSS scores (with 95% CIs) over time by treatment group and tuberosity involvement. (a) Neither tuberosity involved and (b) greater and/or lesser tuberosity involved.

The same multilevel model as for the primary analysis was applied but an additional treatment group by tuberosity involvement interaction was included. Adjusted OSS means and group differences for this model are presented in Table 31. Similar to the primary analysis, there was no significant overall effect of treatment group over the 2-year follow-up period and no significant effect of treatment group at individual time points. A 2-log likelihood comparison with the primary analysis model was not significant [χ²(1) < 0.01, p = 0.954]; therefore, inclusion of the interaction did not significantly improve the model.

TABLE 31 - Difference in mean OSS scores over time by treatment group: subgroup analysis 2 – tuberosity involvement (yes/no)a

Multilevel model adjusted for treatment group, time (6, 12 and 24 months), group × time interaction, baseline EQ-5D index, gender, age (< 65 years/≥ 65 years), tuberosity involvement at baseline (yes/no) and group × tuberosity involvement interaction.

Follow-up	Surgery (n = 125; n = 114 in analysis), mean (95% CI)	Not surgery (n = 125; n = 117 in analysis), mean (95% CI)	Difference (95% CI)	p-value
Overall	39.05 (37.10 to 40.90)	38.33 (36.48 to 40.10)	0.71 (–1.79 to 3.22)	0.577
6 months	37.81 (35.73 to 39.79)	35.61 (33.53 to 37.57)	2.20 (–0.54 to 4.94)	0.115
12 months	39.20 (37.18 to 41.12)	38.82 (36.91 to 40.64)	0.38 (–2.22 to 2.98)	0.772
24 months	40.09 (38.05 to 42.02)	40.42 (38.52 to 42.23)	–0.33 (–2.94 to 2.28)	0.804

Independent Neer classification (sensitivity analysis)

The Neer classification categories were grouped into one- or two-part fractures and three- or four-part fractures. The relationship between Neer grouping and treatment group in terms of OSS scores is illustrated in Table 32 and Figure 12. Patients with Neer three- or four-part fractures had lower OSS scores (poorer shoulder functioning), and scores in the surgery group were higher than those in the non-surgery group at all time points, although CIs overlapped. Patients with Neer one- or two-part fractures seemed to improve shoulder functioning at a slower rate over time in the surgery group than in the non-surgery group.

TABLE 32 - Descriptive OSS statistics over time by treatment group and Neer classification

Max., maximum; min., minimum.

Follow-up	Surgery (n = 125)	Not surgery (n = 125)	Total (n = 250)
Neer one- and two-part fractures (n = 146)
6 months
n	69	67	136
Mean (SD)	37.4 (9.69)	34.6 (10.59)	36.0 (10.20)
Median (min., max.)	41 (12, 48)	37 (3, 48)	39 (3, 48)
12 months
n	66	62	128
Mean (SD)	38.2 (10.51)	38.4 (10.62)	38.3 (10.52)
Median (min., max.)	42.5 (5, 48)	43 (4, 48)	43 (4, 48)
24 months
n	63	58	121
Mean (SD)	39.0 (10.40)	40.8 (10.34)	39.9 (10.37)
Median (min., max.)	43 (16, 48)	45 (1, 48)	44 (1, 48)
Neer three- and four-part fractures (n = 104)
6 months
n	44	52	96
Mean (SD)	34.0 (10.21)	31.1 (11.30)	32.4 (10.86)
Median (min., max.)	36.5 (11, 48)	32 (7, 48)	33 (7, 48)
12 months
n	45	52	97
Mean (SD)	35.0 (11.03)	34.1 (10.80)	34.5 (10.86)
Median (min., max.)	37 (1, 48)	36 (10, 48)	37 (1, 48)
24 months
n	43	51	94
Mean (SD)	37.2 (9.16)	35.6 (11.08)	36.3 (10.23)
Median (min., max.)	39 (16, 48)	39 (3, 48)	39 (3, 48)

FIGURE 12.

Mean OSS scores (with 95% CIs) over time by treatment group and Neer grouping. (a) Neer one- and two-part fractures and (b) Neer three- and four-part fractures.

A similar multilevel model to the primary analysis was applied but the Neer grouping term replaced the tuberosity involvement term and the model additionally included a treatment group × Neer grouping interaction. Adjusted OSS means and group differences for this model are presented in Table 33. Similar to the primary analysis, there was no significant overall effect of treatment group over the 2-year follow-up period and no significant effect of treatment group at individual time points. A 2-log likelihood comparison with an adjusted primary analysis model (including Neer grouping instead of tuberosity involvement) was not significant [χ²(1) = 0.05, p = 0.823]; therefore, inclusion of the interaction did not significantly improve the model.

TABLE 33 - Difference in mean OSS scores over time by treatment group: subgroup analysis 2 – Neer grouping (one- and two-part fractures/three- and four-part fractures)a

Multilevel model adjusted for treatment group, time (6, 12 and 24 months), treatment group × time interaction, baseline EQ-5D index, gender, age (< 65 years/≥ 65 years), Neer grouping (one- and two-part fractures/three- and four-part fractures) and treatment group × Neer grouping interaction.

Follow-up	Surgery (n = 125; n = 114 in analysis), mean (95% CI)	Not surgery (n = 125; n = 117 in analysis), mean (95% CI)	Difference (95% CI)	p-value
Overall	38.20 (36.50 to 39.82)	37.65 (36.00 to 39.23)	0.55 (–1.58 to 2.68)	0.613
6 months	36.93 (35.09 to 38.69)	34.86 (32.98 to 36.64)	2.08 (–0.29 to 4.45)	0.086
12 months	38.36 (36.58 to 40.06)	38.15 (36.42 to 39.80)	0.21 (–2.03 to 2.46)	0.853
24 months	39.26 (37.46 to 40.99)	39.77 (38.05 to 41.43)	–0.51 (–2.79 to 1.77)	0.661

Sensitivity analysis: smoking

A sensitivity analysis was undertaken for the primary outcome model to explore whether or not outcomes were robust when smoking status was taken into account. Smoking is associated with a number of complications and additionally was found to be imbalanced at baseline (surgery group: 19% smokers; not surgery group: 32% smokers).

Descriptive OSS statistics by treatment group and smoking status at 6, 12 and 24 months are provided in Table 34 and illustrated in Figure 13.

FIGURE 13.

Mean OSS scores (with 95% CIs) over time by treatment group and smoking status. (a) Smoker and (b) non-smoker.

The OSS scores improved for both smokers and non-smokers over the trial period. Because of the relatively smaller number of participants who were smokers, the variability of scores around the mean was greater for this group. Although scores at 6 and 12 months’ follow-up were comparable between smokers and non-smokers, at 24 months the mean OSS score was around 2 score points higher for non-smokers. Again, surgical group scores were noticeably higher than those for the non-surgery group at 6 months, although CIs substantially overlapped.

The same multilevel model as for the primary analysis was applied but an additional smoking status was included as a covariate. Adjusted OSS means and group differences for this model are presented in Table 35. Similar to the primary analysis, there was no significant overall effect of treatment group over the 2-year follow-up period and no significant effect of treatment group at individual time points. Therefore, the results of the primary analysis are robust.

Patients’ baseline treatment preferences

Treatment preference was recorded for all eligible patients. Surgeons recorded treatment preference for non-consenting patients, and consenting patients recorded whether or not they had any treatment preference at baseline, before randomisation.

Patient preferences at baseline are shown in Table 36. The majority of non-consenting patients (72%) expressed a preference for non-surgical treatment, whereas this applied to only 24% of consenters. Of the consenters, just under half (46%) had no treatment preference and the remainder were split slightly in favour of surgical treatment (29% vs. 24% for non-surgical treatment). Patients in the surgery group showed a greater preference for surgical treatment (33% vs. 25% in the non-surgery group), whereas patients in the non-surgery group were more likely to have no preference (50% vs. 42% in the surgery group).

Descriptive OSS statistics by treatment group and patient preference at 6, 12 and 24 months are detailed in Table 37 and illustrated in Figure 14. Although OSS scores improved for all groups over time, patients who expressed a preference for surgery generally reported the lowest OSS scores followed by patients who had no preference; patients who expressed a preference for no surgery had the highest OSS scores. The difference between treatment groups in favour of surgery at 6 months’ follow-up was most pronounced for patients who had no preference for either treatment.

FIGURE 14.

Mean OSS scores (with 95% CIs) over time by treatment group and patient preference. (a) Preference surgery; (b) preference not surgery; and (c) no preference.

Patients who preferred surgery tended to be older (by 3 years on average) and presented earlier (by 2 days) with more complex fractures (7% more factures involving one or both tuberosities) than patients who preferred not to have surgery; this gives an indication why these patients had poorer OSS outcomes.

The same multilevel model as for the primary analysis was applied but additionally patient preference at baseline and its interaction with allocated treatment group were included as covariates. Adjusted OSS means and group differences are presented in Table 38. Again, there was no significant overall effect of treatment group over the 2-year follow-up period or at individual time points.

Of all the analyses adding further terms to the primary model (age, fracture type, smoking status, patient preference), this model reduced the magnitude of the treatment effect the most (from an overall group difference of 0.75 score points to 0.50 score points), reflecting the additional variability in OSS scores explained by patient preference, as seen in Figure 14. However, the interaction between patient preference and allocated treatment was not statistically significant (F = 0.29, p = 0.751), even against a relaxed significance level of 0.1.

Exploration of centre variability

In our analysis plan we revised our intention to explore the potential for clustering, from individual surgeons to the level of the centre, in essence quantifying the performance of the surgical team(s) rather than that of an individual surgeon within the centre. This was because of the low numbers of patients treated by an individual surgeon. Among the 33 centres that screened patients, around 60% (n = 148) of the 250 participants were recruited by eight sites (12–43 patients each, 18.5 patients on average) and around 40% (n = 102) were recruited by the remaining 25 centres (0–9 patients each, 4.1 patients on average).

Mean OSS scores and 95% CIs for randomised patients in each centre at each follow-up time point are illustrated in Figure 15. These ranged from 22 to 48 score points, but no centre appeared to be substantially outlying. When CIs were either exceptionally wide or did not include the overall mean, this was predominantly because of the small number of patients in those centres. Extreme CI limits are not displayed in full here. When no CIs are displayed, the sample size was equal to 1, except in one case in which there were two patients with identical scores.

FIGURE 15.

Mean OSS scores (with 95% CIs) over time by recruitment site. Max., maximum; min., minimum.

The distribution of scores is further examined using box plots of OSS scores (Figure 16). These show a wide spread of scores within each centre, again affected by small sample sizes in some centres, and highlight a small number of outliers.

FIGURE 16.

Distribution of OSS scores (box plots) by recruitment site.

Summary of the Oxford Shoulder Score multilevel modelling results

Table 39 summarises the OSS estimates for each treatment group as well as group differences following multilevel analysis. These are generated using a mixed-effect model with a random slope of time within patients. The outcome is the squared OSS (estimates are back transformed in the table). Fixed effects for each model are given in the footnotes.

12-item Short Form health survey

The 12 SF-12 questions were summarised into eight domain scores, each based on one or two questionnaire items. If at least one question had been completed for a domain, any missing items were replaced with the available score for that domain. Weighted estimates of the eight SF-12 domains were then used to calculate the PCS score and MCS score, each ranging from 0 to 100 score points, with higher scores representing a better health status. All domain scores needed to be available for the calculation of the PCS and MCS scores. Scores were standardised against the general US population (mean 50, SD 10). 69 Using standard effect sizes, differences of 2, 5 and 8 score points would correspond to small, medium and large differences respectively.

Physical component summary score

Descriptive SF-12 PCS score statistics at 6, 12 and 24 months are shown in Table 40 and illustrated in Figure 17. The distribution of scores is presented in Figure 18.

TABLE 40 - Descriptive SF-12 PCS score statistics over time by treatment group

Max., maximum; min., minimum.

Follow-up	Surgery (n = 125)	Not surgery (n = 125)	Total (N = 250)
6 months
n (%)	106 (84.8)	110 (88.0)	216 (86.4)
Mean (SD)	45.3 (10.01)	42.7 (11.25)	43.9 (10.71)
Median (min., max.)	46.7 (15.4, 64.1)	43.6 (11.9, 62.4)	45.0 (11.9, 64.1)
12 months
n (%)	108 (86.4)	110 (88.0)	218 (87.2)
Mean (SD)	45.2 (10.98)	43.7 (10.98)	44.5 (10.93)
Median (min., max.)	45.2 (14.9, 63.7)	44.1 (17.6, 61.3)	44.7 (14.9, 63.7)
24 months
n (%)	105 (84.0)	105 (84.0)	210 (84.0)
Mean (SD)	45.2 (11.30)	44.1 (11.58)	44.6 (11.43)
Median (min., max.)	46.1 (14.9, 65.7)	45.2 (14.3, 60.7)	46.0 (14.3, 65.7)

FIGURE 17.

Mean SF-12 PCS scores (with 95% CIs) over time by treatment group.

FIGURE 18.

Distribution of SF-12 PCS scores over time. (a) 6 months; (b) 12 months; and (c) 24 months.

Overall, physical functioning remained largely unchanged over the trial period. The distribution of scores was approximately normal at all time points. Mean PCS scores were slightly higher for the surgery group at all time points, with group differences being largest at 6 months. There was substantial overlap in the CIs between groups.

A similar multilevel model to that in the OSS analysis was applied, using SF-12 PCS score as the outcome (no transformation was found to be necessary) and including the same predictors. Adjusted SF-12 PCS score means and group differences for this model are presented in Table 41. The results mirrored those of the OSS analysis, with slightly better functioning in the surgery group by an average of between 1.3 and 2.6 score points over the 2-year follow-up period. None of the differences between the two groups, overall or at individual time points, was statistically significant. Similar to the OSS outcome, the largest group difference was observed at 6 months’ follow-up.

TABLE 41 - Difference in mean SF-12 PCS scores over time by treatment groupa

Multilevel model adjusted for treatment group, time (6, 12 and 24 months), group × time interaction, baseline EQ-5D index, gender, age group (< 65 years/≥ 65 years) and tuberosity involvement at baseline (yes/no).

Follow-up	Surgery (n = 125; n = 111 in analysis), mean (95% CI)	Not surgery (n = 125; n = 115 in analysis), mean (95% CI)	Difference (95% CI)	p-value
Overall	45.64 (43.44 to 47.84)	43.87 (41.75 to 45.99)	1.77 (–0.84 to 4.39)	0.184
6 months	45.73 (43.44 to 48.02)	43.18 (40.97 to 45.39)	2.55 (–0.21 to 5.32)	0.070
12 months	45.51 (43.22 to 47.80)	44.22 (42.01 to 46.43)	1.29 (–1.48 to 4.06)	0.361
24 months	45.68 (43.28 to 48.08)	44.20 (41.87 to 46.54)	1.48 (–1.48 to 4.43)	0.327

Mental component summary score

Descriptive SF-12 MCS score statistics at 6, 12 and 24 months are shown in Table 42 and illustrated in Figure 19. The distribution of scores is presented in Figure 20.

TABLE 42 - Descriptive SF-12 MCS score statistics over time by treatment group

Max., maximum; min., minimum.

Follow-up	Surgery (n = 125)	Not surgery (n = 125)	Total (N = 250)
6 months
n (%)	106 (84.8)	110 (88.0)	216 (86.4)
Mean (SD)	49.2 (10.84)	49.8 (11.46)	49.5 (11.14)
Median (min., max.)	52.2 (20.3, 63.7)	52.1 (17.6, 68.4)	52.2 (17.6, 68.4)
12 months
n (%)	108 (86.4)	110 (88.0)	218 (87.2)
Mean (SD)	48.8 (10.51)	50.8 (10.67)	49.8 (10.61)
Median (min., max.)	51.2 (17.8, 66.2)	54.0 (15.8, 67.4)	52.6 (15.8, 67.4)
24 months
n (%)	105 (84.0)	105 (84.0)	210 (84.0)
Mean (SD)	50.1 (11.64)	51.5 (9.96)	50.8 (10.83)
Median (min., max.)	55.2 (8.8, 64.4)	54.2 (21.2, 66.1)	54.3 (8.8, 66.1)

FIGURE 19.

Mean SF-12 MCS scores (with 95% CIs) over time by treatment group.

FIGURE 20.

Distribution of SF-12 MCS scores over time. (a) 6 months; (b) 12 months; and (c) 24 months.

Overall, mental functioning improved slightly over the trial period. The distribution of scores was slightly left skewed at 6 and 12 months, which became more pronounced at 24 months. Unlike physical functioning, as assessed by the OSS and PCS score, the mean MCS score was slightly higher for the non-surgery group at all time points, with the group difference being greatest at 12 months. There was substantial overlap of the 95% CIs of the two groups.

A similar multilevel model to the OSS analysis was applied, using SF-12 MCS score as the outcome (no transformation was found to be necessary) and including the same predictors. Adjusted SF-12 MCS score means and group differences for this model are presented in Table 43. There were better scores in the non-surgery arm by an average of between 0.5 and 2.0 score points over the 2-year follow-up period. This difference was not statistically significant over the 2-year follow-up period or at individual time points.

TABLE 43 - Difference in mean SF-12 MCS scores over time by treatment groupa

Multilevel model adjusted for treatment group, time (6, 12 and 24 months), group × time interaction, baseline EQ-5D index, gender, age group (< 65 years/≥ 65 years) and tuberosity involvement at baseline (yes/no).

Follow-up	Surgery (n = 125; n = 111 in analysis), mean (95% CI)	Not surgery (n = 125; n = 115 in analysis), mean (95% CI)	Difference (95% CI)	p-value
Overall	48.66 (46.55 to 50.77)	49.96 (47.92 to 52.00)	–1.28 (–3.80 to 1.23)	0.317
6 months	48.43 (46.07 to 50.80)	48.95 (46.66 to 51.24)	–0.52 (–3.44 to 2.41)	0.729
12 months	48.24 (45.96 to 50.53)	50.20 (47.98 to 52.41)	–1.95 (–4.76 to 0.85)	0.172
24 months	49.30 (46.97 to 51.64)	50.69 (48.40 to 52.97)	–1.38 (–4.27 to 1.51)	0.349

Surgical and shoulder fracture-related complications

Surgical and other shoulder fracture-related complications were recorded at the end of patients’ orthopaedic inpatient episodes when applicable and at 1- and 2 years’ follow-up. Clinic staff could select from a list of predefined complication categories or indicate ‘other’. For each, descriptions could be added as free text and resulting treatments could be detailed in a further section on each form.

As described in Chapter 3, to ensure the validity and interpretability of the entered complications, two expert raters assessed each reported complication in the context of the information provided elsewhere on the same form or previous/subsequent forms as well as the other trial data available for that patient and agreed on any changes. Using clinical judgement, some records were recategorised and entries under ‘other’ were grouped when possible. Duplicate and irrelevant information was removed and additional complications were appended to the data if these were clearly reported elsewhere, for example when included in the description of further treatment (see Appendix 36).

The final data set consisted of 59 surgical and shoulder fracture-related complications reported for 53 patients (Table 44).

Apart from haematoma formation at the surgical site, all predefined complications were reported at least once. In the surgery group, a total of 30 patients (24%) experienced at least one complication over the 2-year follow-up period. These included six patients with two complications, which were concurrent for three of these patients. The majority of complications were reported at the 1-year follow-up. The most common complications in the surgery group were metalwork problems (n = 10) and post-traumatic stiffness (n = 6). It is noteworthy that three of the four reported malunions occurred for crossover patients who were allocated to surgery but who did not receive the intervention. In the non-surgery group, a total of 23 patients (18%) experienced one complication each over the trial period. All but four of these were reported at the 1-year follow-up. The most common complications in this group were malunion (n = 5), nonunion (n = 5) and post-traumatic stiffness (n = 5). Although marginally more patients in the surgery group experienced these complications, a chi-squared test revealed that there was no statistically significant difference between the treatment groups [χ²(1) = 1.17, p = 0.279].

To ascertain whether or not certain attributes may predispose patients to experience a surgical or shoulder-related complication, selected baseline characteristics for patients who suffered at least one complication (n = 53) and patients for whom no complications were reported (n = 197) were compared (Table 45). No marked between-group differences were evident. [Although patients with complications appeared to be marginally younger (49% were aged < 65 years compared with 42% of patients without complications), the mean ages of the treatment groups were similar.]

Treatments for surgical and shoulder fracture-related complications

Clinical staff were asked to detail any secondary surgery or increased/new therapy for these complications at the end of the orthopaedic inpatient episodes and again at the 1- and 2-year follow-ups.

After processing there were 33 valid entries reported for 33 patients, all of whom had received the treatment that they were randomised to. Each treatment corresponded to a surgical or shoulder fracture-related complication as detailed in the previous section. A breakdown of reported treatments by time and treatment group is presented in Table 46.

Secondary surgery to the shoulder was reported for 22 (9%) of the 250 participants, with 11 in each treatment group. Increased/new shoulder-related therapy was reported for 11 (4%) of the 250 participants, of whom seven were in the surgery group and four were in the non-surgery group. Although numbers were too small for formal testing, there appeared to be no marked group differences for secondary surgery or increased/new therapies. There was no statistically significant difference when combining secondary surgery and increased/new therapies to compare the total number of treatments between groups [χ²(1) = 0.31, p = 0.575].

Using all available patient information, the two expert raters assessed whether a reported surgical or shoulder fracture-related complication was addressed by further surgery or increased/new therapy and whether or not each complication was ultimately resolved (Table 47). Proportionally more complications were treated surgically in the non-surgery group (48% vs. 33% in the surgery group). The opposite was true for increased or new therapy (21% in the surgery group vs. 13% in the non-surgery group). Information regarding treatment outcome was ascertainable for only half (50%) of the recorded observations, and only a small number (9%) were categorised as being resolved or resolving. There were no marked between-group differences in outcome.

The treatments and outcomes associated with the listed complications are shown for surgery in Table 48 and for non-surgery in Table 49. Linking this information in this way shows, for example, that the three malunions for cross-over patients in the surgery group were ‘not treated’. One patient with malunion died before further treatment could be discussed. Four of the five nonunions in non-surgical group had secondary surgery; the fifth patient died before referral for orthopaedic assessment.

Medical complications during the inpatient stay

Clinical staff were asked to select from a list of medical complications or specify any other medical complications that occurred during the patients’ orthopaedic inpatient stay when applicable. End of orthopaedic inpatient episode forms were received for 111 patients in the surgery group and 23 patients in the non-surgery group.

After processing (see Appendix 36), there was a total of 10 medical complications for 10 patients (4% of all patients) (Table 50). All complications were reported for patients in the surgery group, although two of these patients were crossovers and did not receive their randomised treatment. Medical complications could be grouped into cardiac or peripheral vascular events (MI, cardiac arrest), respiratory events (two chest infections) and gastrointestinal events (norovirus, jaundice); the remainder were individual complications summarised under ‘other’ (urinary retention, intensive care admission and two blood transfusions).

The small number of complications limits the validity of comparing the baseline characteristics of the 10 patients with medical complications with the baseline characteristics of the 240 patients without medical complications. Nonetheless, consistent with expectations, patients experiencing an early medical complication tended to be older [mean (SD) age 71.8 (10.08) years vs. 65 (11.96) years].

Details of medical complications that occurred after the inpatient stay were collected as treatment for newly diagnosed medical complications at 1 and 2 years’ follow-up (see the following section).

Other admissions and treatments: newly diagnosed medical complications and fractures

Treatments for any serious newly diagnosed medical problems were recorded at the end of the patients’ inpatient episode when applicable and at the 1- and 2-year follow-ups. After processing (see Appendix 36), there were treatments recorded for 51 valid medical problems in 46 patients (Table 51).

Fifteen patients (12%) were treated for a newly diagnosed medical complication in the surgery group (including two crossover patients who did not receive surgical treatment), whereas more than twice as many patients in the non-surgery group (n = 31, 25%) were treated for such medical complications. However, medical complications reported for 10 patients in the surgery group during the inpatient stay (see Table 50) should also be taken into account.

For patients in the surgery group, the majority of treatments were recorded during the second year of follow-up, whereas in the non-surgery group an equal number of treatments were recorded in year 1 and year 2. Five patients, of whom one was in the surgery group and four were in the non-surgery group, were treated for more than one medical complication, once during the first year and once during the second year. Overall, the most common complications were ‘other’ (n = 21), cardiac and peripheral vascular events (n = 13, 10 of which occurred in the non-surgery group) and gastrointestinal events (n = 8).

Admissions to hospital for another fracture as well as any visits to the orthopaedic/fracture clinic post discharge were recorded at the 1- and 2-year follow-ups. After data processing, there were 18 valid entries for 18 patients, which are detailed in Table 51. Eight patients (6%) in the surgery group and 10 patients (8%) in the non-surgery group were admitted to hospital for another fracture. In addition, there was a total of 26 valid visits to orthopaedic/fracture clinics for 25 patients (see Table 51). Fewer patients in the surgery group (n = 10, 8%) than in the non-surgery group (n = 15, 12%) visited an orthopaedic/fracture clinic, and only one patient (non-surgery) had more than one visit for a shoulder fracture in the first year and for another fracture in the second year.

Overall, a total of 28 additional fractures for 25 patients were reported as part of the hospital admissions and clinic visits: 10 in the surgery group and 18 in the non-surgery group. More details are given in Table 52. The most commonly reported fractures were those of the hip (n = 10) and the wrist (n = 8). All three patients for whom multiple fractures were reported were in the non-surgical group.

Mortality

Of the 14 patients who died during the trial period, nine (7%) were in the surgery group and five (4%) were in the non-surgery group. Causes of death by treatment group are presented in Table 53. One death in the surgery group was judged as being trial related (venous thromboembolism). A chi-squared test revealed that there was no statistically significant difference in mortality rate between treatment groups [χ²(1) = 1.21, p = 0.271].

Adverse events

A total of 78 SAEs were reported for 56 patients and 10 non-SAEs were reported for nine patients.

‘Untoward events’ documented in the physiotherapy treatment logs

The screening by two independent raters (LG and HH) of comments and reasons for referral recorded in the physiotherapy treatment logs resulted in the identification of 37 ‘untoward events’ recorded for 32 patients. These were shoulder-related or medical events that impeded physiotherapy or were reasons for non-attendance or for referral. Listed surgical or shoulder-related complications that were not identified by the inpatient and two hospital follow-up forms are summarised in Tables 59 and 60 for the surgical group and non-surgical group respectively. As in the coding for the hospital forms, frozen shoulder is termed post-traumatic stiffness. In screening of the ‘untoward events’, precursors (e.g. severe pain) to a complication recorded on the hospital form (e.g. AVN) and complications subsequent to treatment for a complication were not included.

Given that the physiotherapy logs did not specifically collect data on complications and considering the diffuseness and variability in the commentary, these results should be considered as illustrative of what was recorded. It is clear that a few early complications were not recorded on the hospital forms but, apart from axillary nerve lesion, none of these complications was a named complication on the hospital forms.

Discharge locations after inpatient stay

Only five of the 109 participants who were inpatients in the surgery group and one of the 22 participants who were inpatients in the non-surgery group were discharged to locations other than back home. In the surgery group, one patient was discharged to a nursing home, three were discharged to a rehabilitation unit and one was discharged to a renal ward for dialysis. The patient in the non-surgery group was discharged to another hospital.

Patient treatment preferences at the end of the 2-year follow-up

Consenting patients were asked at baseline in advance of randomisation whether or not they had any treatment preference before agreeing to randomisation. At the end of the 2-year follow-up period, patients were again asked to indicate their treatment preference in the event of a similar shoulder injury given their experience over the past 2 years.

Patient preferences at 2 years’ follow-up are tabulated against baseline preferences in Table 61 and against the allocated treatment in Table 62. More patients expressed a preference for a treatment (either surgery or not surgery) at 24 months (61.9%) than at baseline (53.4%). The slight preference for surgical treatment over non-surgical treatment seen at baseline remained.

Fewer than half of the patients retained their original treatment preference (41.7% for surgery and 46.7% for non-surgery) (see Table 61). Around half of the patients expressed a preference for the treatment that they were allocated to (51.2% in the surgery group and 44.0% in the non-surgery group) (see Table 62). In each case, patients’ second most frequent choice was ‘no preference’ rather than the opposite treatment.

Lack of surgeon equipoise

As reported in Chapter 4, the independent assessment of the ‘other’ reasons given for non-eligibility on the study eligibility form (option j; see Appendix 5) resulted in the identification of 117 patients who were judged to have been excluded because of a lack of equipoise. The two raters also recorded the treatment preference stated for individual patients (surgery, conservative management, unclear) and deduced this when not recorded in the appropriate box on the study eligibility form if other information was available.

Listed below is the categorisation of ‘protocol violations’ for the 159 study eligibility forms on which only option j had been ticked:

• lack of clinical equipoise: 117 (74%) [surgeons’ treatment decision/preference: surgery 45 (38%); not surgery 63 (54%); unclear 9 (8%)]

• existing exclusion criteria: 18 (11%)

• local difficulties (e.g. research nurse not available): 11 (7%)

• new exclusion criteria: 8 (5%)

• other: 5 (3%).

Listed under ‘new exclusion criteria’, which equated to when the two raters considered that the reason for exclusion was justified, were the following:

• severely compromised social circumstances (n = 1)

• repeat injury to previous fracture (n = 1)

• extreme age and frailty (n = 1)

• patient with compromised or exceptional use (e.g. wheelchair user) upper limb function (n = 4)

• head-split fracture (n = 1).

Between one and 23 patients were excluded because of a lack of equipoise in 22 centres; no patients were excluded for this reason in the remaining 11 centres that returned study eligibility forms. Figure 21 shows the numbers of patients excluded because of a lack of equipoise together with the numbers of ineligible patients for each of the 33 centres. Lack of equipoise is most notable in three centres, two with high numbers of inappropriately excluded patients [23 (47% of ineligible patients) and 14 (67% of ineligible patients)] and one in which lack of equipoise was the reason for exclusion of all seven ineligible patients.

FIGURE 21.

Lack of surgeon equipoise and ineligible patients by centre.

Lack of equipoise was recorded for 68 surgeons. In some centres only a small proportion of surgeons (e.g. one of 16 in one centre) listed on study eligibility forms excluded patients because of a lack of equipoise, whereas in other centres this was more prevalent (e.g. 12 of 17 surgeons in one centre). Several surgeons (n = 12) who showed a lack of equipoise screened and excluded only one patient. Of those who excluded more than one patient because of a lack of equipoise, a further 12 excluded half or more of the patients who they screened.

Using the t-test or chi-squared test as applicable, selected baseline characteristics were compared between all eligible patients and patients who were excluded because of a lack of equipoise to identify any differences between the two populations (Table 63). Patients who were excluded because of a lack of equipoise tended to be slightly older (70 years vs. 67 years, p = 0.062) and were more likely to have a fracture not involving either tuberosity (41% vs. 25%, p < 0.001) than the sample of all eligible patients. The proportion of men and women as well as time since injury did not appear to differ meaningfully between the two groups.

Age and tuberosity involvement were tabulated against surgeons’ treatment preference (reported data only) and assessed by chi-squared test for patients excluded because of a lack of equipoise to estimate the direction and strength of any observed preference related to age and fracture type (Table 64). Patients aged ≥ 65 years were more likely to be advised to have non-surgical treatment (p = 0.059) whereas patients with fractures involving tuberosities were just as likely to be advised surgery or non-surgery as patients with fractures involving no tuberosities (p = 0.794).

Surgeon preferences and agreed treatments for non-consenting patients

Availability of radiographic views in the study population

Clinicians completing a study eligibility form (see Appendix 5) recorded whether or not the series of views used to inform the assessment of the fracture included one or more of the following views: the anteroposterior view, the axillary (or modified axillary) view and the scapular Y-lateral view. Our protocol stipulated that a minimum of two radiographic views/projections were required for the assessment of study eligibility. Table 67 shows the breakdown of radiographic views recorded for all screened patients. This demonstrates that the majority (88%) of the study eligibility forms indicated that the minimum requirement was achieved, with the standard trauma series of three perpendicular views59 being available for 18% of screened patients. The anteroposterior view was recorded in 98% of patients, the axillary view in 57% and the scapular Y-lateral view in 50%. Similar findings in terms of views recorded and meeting the minimum requirements of two perpendicular views applied for ineligible (88%), eligible (89%) and randomised patients (89%).

Use of other imaging investigations

Data on the use of other imaging modalities to determine study eligibility were collected only for randomised patients (see Appendix 20). In total, six patients (three in each treatment allocation group), each from a different hospital, received a CT scan. This is consistent with the results of the post-recruitment survey of radiography departments, which found that CT scanning was ‘routinely/frequently used for these fractures’ in only one of the 26 responding hospitals (see Appendix 31).

Independent assessment and classification of baseline radiographs according to the Neer classification

Cull of radiographs in preparation for assessment

A minimum of two radiographs were available for each trial participant, with a maximum of seven being received for two participants. In all, 664 radiographs were available for the 250 study participants. Sixty-two views were excluded based on prespecified criteria, which aimed to retain the best-quality view(s) in each perpendicular plane. Thus, the number of named views in each plane would not have been affected by this cull. The results of the cull are illustrated in Figure 22. For the majority of patients (n = 165), two radiographic views were available post cull, leaving 68 with three views and 17 with four views.

FIGURE 22.

Pre- and post-cull radiographic views per patient.