Trial methods

Summary of trial design

ARTISAN was a multicentre, randomised, pragmatic trial. People presenting at 41 UK hospitals in the NHS with a TASD for non-surgical management were randomised 1 : 1 to receive advice or advice and a programme of physiotherapy (see Appendix 1).

Ethics, registration and oversight

The National Research Ethics Committee approved this study on 26 July 2018 (18/WA/0236), with each trial site granting individual NHS Trust approval prior to recruitment at each site. The ARTISAN protocol was accepted for publication on 13 October 2020 and first published on 19 November 2020.11

The trial was conducted in accordance with the principles of the Declaration of Helsinki and the MRC good clinical practice guidelines as well as all applicable UK legislation and University of Warwick standard operating procedures (SOPs). Trial oversight was provided by a Trial Management Group (TMG) and had independent oversight from a Data Monitoring Committee (DMC) and Trial Steering Committee (TSC).

Settings and locations

There were 41 trauma research teams at UK NHS Trust sites who screened adults with a first-time TASD confirmed radiologically, being managed non-operatively.

Participants

Participant screening

Adults with a primary (first-time) TASD were screened against the eligibility criteria to take part in the trial. Broad eligibility criteria ensured that the results of the study could readily be generalised to the wider patient population.

Inclusion criteria

• Provision of written informed consent.
• Aged 18 years or over.
• They have a primary (as reported by the potential participant) traumatic acute shoulder dislocation, confirmed radiologically.

Exclusion criteria

• Bilateral shoulder dislocation at time of injury.
• Having first-line surgical treatment (indications include a displaced greater tuberosity fracture, for example).
• Cannot receive first session of physiotherapy within 6 weeks of injury.
• In the opinion of the assessing clinician there is a significant neurovascular complication associated with TASD (e.g. brachial plexus injury).
• Unable to adhere to trial procedures or complete questionnaires (e.g. a history of permanent cognitive impairment).
• Previous randomisation in the present trial.

If a trial participant were to sustain a contralateral TASD during the trial period, the second TASD would not be included in the study because the outcomes of this intervention would not be independent from the first intervention.

All potential participants meeting the entry criteria were checked for eligibility and entered on the monthly screening log. Potential participants who were willing to be approached by a suitably trained member of the research team were provided with verbal and written information about the study. They were then asked if they wished to take part in the study. All new non-operatively managed potential participants had a maximum of 6 weeks from date of injury to make a final decision and be randomised. If a patient was eligible and consenting, a member of the local research team completed the informed consent process, enrolment, baseline and pre-injury data collection.

Participants were placed on the waiting list for physiotherapy, with a typical wait of up to 2 weeks. The eligibility was reconfirmed by the treating physiotherapist at the first appointment, and potential participants were excluded at this stage if there had been a change in status. This allowed someone who had consented to the study to be deemed not eligible at the point of randomisation and to be excluded, using the predefined exclusion criteria.

Informed consent

Written informed consent was obtained by a suitably trained member of the research team at each site as per the delegation log, after allowing sufficient time for the potential participant to consider their decision and ask questions about the trial. If participants were identified through a virtual fracture clinic setting or had left the face-to-face clinical setting before being approached, verbal consent was gained in the first instance by telephone communication, and the participant was then posted a paper informed consent document to be completed and handed to the research team prior to randomisation.

The principal investigator (PI) or co-PIs (an orthopaedic consultant and/or physiotherapy lead) retained overall responsibility for informed consent at their site and ensured that any person delegated responsibility to participate in the informed consent process was duly authorised, trained, qualified and competent.

As there is a delay of a number of weeks between consent and randomisation (due to the waiting list for physiotherapy), people who had entered the study had the option to withdraw before treatment started if for any reason they changed their mind.

The participants remained free to withdraw at any time without giving reasons and without prejudice to any further treatment and were provided with a contact point where they could obtain further information about the trial if required.

Randomisation

Pre-randomisation eligibility checks were carried out to ensure that potential participants met the eligibility criteria. Written informed consent for entry into the trial and baseline assessment were obtained prior to randomisation. Participants were randomised once they had been registered as eligible for randomisation on the web-based system and completed their physiotherapy advice session. Allocation concealment was maintained by an independent randomisation team who were responsible for generation of the sequence and had no role in the allocation of participants.

The treatment group were allocated by computer using a minimisation algorithm with a random element and stratification by participant age (≤ 39 years old or ≥ 40 years old), hand dominance and treating centre. The physiotherapist, following delivery of the control intervention, randomised all trial participants. Physiotherapists were only able to obtain the randomisation code after verifying that the initial control advice session had been complete. Because participants were only made aware of their group allocation after attendance at the control treatment, the impact of resentful demoralisation was minimised to avoid participants failing to engage with the control intervention.

Minimisation was a better option than stratification with variable block sizes, due to the relatively small number of participants expected in some strata. The randomisation service was available 24 hours a day, 7 days a week to facilitate the inclusion of all eligible participants. A confirmation e-mail was automatically generated to the research site containing the randomisation details.

In an open trial of this nature, it was not possible to control for the effects of demoralisation bias on engagement with the interventions. Failure to engage with an intervention is part of the reality of clinical practice regardless of the trial being conducted, meaning that the effect size observed is likely to reflect real-world effectiveness. We were cognisant of the risk of differential loss to follow-up between the two groups. We monitored this closely.

Post-randomisation withdrawals

Unless a participant explicitly withdrew consent, data were collected until trial end. For those withdrawing consent for follow-up procedures, trial data obtained up until the point of withdrawal were included. Participants could withdraw from follow-up but continue to provide routine NHS data for the purposes of the trial.

Participants who withdrew were not replaced in the trial, and a corresponding withdrawal case report form (CRF) was completed. Participants could be withdrawn from treatment and, if necessary, the trial at the discretion of the investigator and/or TSC due to safety concerns.

Participants could also be withdrawn post randomisation by the TMG if participants were found during routine site quality assurance checks not to have had ‘radiological confirmation’ of the primary traumatic dislocation on checking source data. In these cases, participants were withdrawn from the randomised total and replaced, but were still followed up and analysed as part of a preplanned sensitivity analysis.

Trial interventions

Details of the intervention development were first published in December 2021.12 The written trial materials can be found in Appendices 2 and 3. The website reflected the same material content but in a series of short animations.

Control

All participants had an initial period where the injured arm was supported in a sling, and then received an appointment for a physiotherapy advice session within 6 weeks of their injury. At this first encounter, consenting participants were provided with a web link to phase 1 of the advice materials and provided with a paper-based booklet version of the same content.12 [Reproduced with permission from Liew et al.12 This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) license, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: https://creativecommons.org/licenses/by/4.0/. The text below includes minor additions and formatting changes to the original text.]

1. Phase 1 covered the below information:
   1. What has happened to me?
   2. What can go wrong?
   3. How do I stop this happening again?
   4. How long do I have to wear my sling?
   5. Should I move my arm?
   6. How do I control my pain?
   7. When can I return to usual activities?
   8. What if something goes wrong?

The expectation was that all participants would receive an appointment for physiotherapy within 2 weeks of injury (i.e. the time at which the immobilisation would be expected to be removed). However, to reflect that some clinicians/sites may recommend immobilisation to be worn for greater or lesser time, and to offer achievable time frames for all physiotherapy services, an upper time limit of 6 weeks from date of injury was accepted; 6 weeks was chosen as the upper period limit to reflect the time point at which a soft-tissue injury is no longer considered acute.

All participants received a single session of advice to aid self-management. This lasted up to 1 hour and was administered by an ARTISAN-trained physiotherapist. Following routine assessment, the physiotherapist delivered a core set of phase 2 intervention components that included education and discussion reiterating points a–h above and on the following:

1. points of contact if complications occur or expected recovery times are not achieved
2. a core set of progressive phase 2 range of movement exercises and what they aim to achieve
3. enhancing self-management behaviours through the addition of goal-setting, exercise planning and diaries.

The outlined participant information (a–i) aimed to improve understanding of the condition and its management, to counter any participant misconceptions. Points j and k aimed to agree with the participant an exercise (or other) goal (e.g. repetition, duration, frequency) and to prompt them to think of possible factors (obstacles and facilitators) influencing the behaviour (e.g. controlling the pain) and come up with strategies to overcome them. They were also prompted to make detailed planning of performance of the behaviour or behaviours (e.g. exercise or pain management) to include at least one of context, frequency, duration or intensity; for example, they were encouraged to complete one set of exercises every day after work and as soon as they returned home.

The physiotherapist provided details of web-based materials, which included all the core components above in written and video format, and included a dedicated area for participants to set goals and keep diaries. The physiotherapist also discussed with the participant that the website resources also contained progression to a core set of progressive phase 3 strengthening exercises and what they aimed to achieve, and later-stage information in phase 4 on how to return to sports. Participants were offered paper-based alternatives. Offering different formats (e.g. written and digital resources) enhanced adherence, as it adapts to a variety of individual needs. Table 1 summarises the advice materials received by all participants.

Following completion of the advice appointment, the participant was randomised, allocating them to this advice session alone or to this advice session plus the offer of additional physiotherapy.

Participants randomised to advice only were provided with a contact point to self-refer back to the clinical team if recovery did not occur. Participants who self-referred back to the clinical team were considered to be per protocol.

Table 2 summarises the key components of the ARTISAN intervention according to the Template for Intervention Description and Replication (TIDieR) criteria, as described in the published intervention development paper.12

Intervention

The additional course of physiotherapy consisted of the offer of at least one additional physiotherapy session after the pre-randomisation session. Each additional session lasted for up to 30 minutes, over a maximum duration of 4 months from date of randomisation; no upper limit on the number of sessions was set. It was tailored, supervised and taught incorporating common methods to increase adherence by a physiotherapist trained by the trial team or local lead trial therapist. The course of physiotherapy involved teaching and supervising the ‘core set’ of progressive exercises offered to the control arm and published on the web-based resources, in addition to being able to tailor through offering additional exercise components from a trial manual menu which provided a range of exercises at differing levels which the physiotherapist could then choose and set specified frequency, loads and number of repetitions at their discretion.

Monitoring intervention delivery and compliance

Following site set-up, the trial team implemented mechanisms to ensure treatment fidelity. This was based on a standardised approach of evaluating fidelity.13

1. Direct observations: with additional permissions, a member of the trial team observed trial-related procedures and the delivery of the two intervention arms (permission was sought from the trial participants to observe treatment sessions). An adherence evaluation form consisting of items that reflect the occurrence or non-occurrence of an event formed the basis of the assessment.
2. Audio recordings: with additional permissions, and in addition to the adherence form, the interactions between the therapist and trial participant were recorded during the above observation (additional permission was sought from trial participants to record treatment sessions). This was used to assess success or failure of the therapist to introduce the aims/rationale of each component and consolidate participant learning at the end of each component. Assessments were given as ‘yes/demonstrated’, ‘no/not demonstrated’ or ‘unsure’.
3. Therapist self-report: the adherence evaluation form was also self-reported by the site therapist. CRFs were collected on intervention delivery including number of treatment sessions attended, materials provided and exercise components prescribed. This was completed for every trial participant.

Points (1) and (2) were evaluated twice annually for the duration of recruitment and intervention delivery. Any issues identified were discussed on a case-by-case basis by the TMG, who were responsible for recommending appropriate action. If issues with individual sites were not resolved following the recommendations, they were escalated to the TSC.

Changes to the intervention

Recruitment and trial delivery took place prior to and during the COVID pandemic. Prior to the pandemic the control and intervention were delivered face to face. During the pandemic some NHS Trusts switched to virtual delivery of physiotherapy services. No protocol amendment was required to accommodate this change in service delivery.

Outcome measures

Primary

Oxford Shoulder Instability Score

The OSIS is a self-completed outcome measure containing 12 questions (0–4 points each), with possible scores from 0 (worst function) to 48 (best function).14,15 These questions relate to activities of daily living particularly relevant to patients exhibiting shoulder instability. The OSIS has been specifically designed to assess outcome of therapy (both surgical and non-surgical) by measuring activities of daily living and pain of patients exhibiting shoulder instability. The development of the score demonstrated reproducibility and internal consistency and was shown to correlate well with existing related clinical and generic patient-reported outcome measures (PROMs).

Secondary

QuickDASH

The QuickDASH is a self-completed shortened version of the Disabilities of the Arm, Shoulder and Hand (DASH) questionnaire. Instead of 30 items, the QuickDASH uses 11 items to measure physical function and symptoms in people with any or multiple musculoskeletal disorders of the upper limb. The questionnaire was designed to help describe the disability experienced by people with upper-limb disorders and also to monitor changes in symptoms and function over time.16

EQ-5D-5L

The EQ-5D-5L is a well-validated, generic health-related quality-of-life measure consisting of five dimensions each with five levels of response. Each combination of answers can be converted into a health utility score. It has good test–retest reliability, is simple for participants to use, and gives a single preference-based index value for health status that can be used for broader cost-effectiveness comparative purposes.17

Complications

Serious adverse events (SAEs) were reported through the following mechanisms: (1) participant reported during routine collection of follow-up data; (2) local research teams reported any additional investigations or treatment of participants; (3) local physiotherapists delivering the trial interventions reported any events occurring during treatment sessions; and (4) medical records of non-responding participants were retrieved by local research teams at site.

Serious adverse events not related to the intervention or TASD event were recorded on the SAE form but were not formally analysed or reported. SAEs that were predefined complications directly related to the trial interventions or directly caused by the primary TASD event were recorded as complications and not reported.

Resource use questionnaires

The primary health economic analysis concentrated on direct intervention and healthcare/PSS costs, while wider impact (societal) costs were included within the sensitivity analyses. Participants completed resource use questionnaires at baseline and all follow-up points, to collect resource use data associated with the interventions under examination.

Qualitative interviews

One of the secondary objectives for ARTISAN was to qualitatively explore the participant experiences of receiving the trial treatments, and facilitators and obstacles to adhering to them.

At a point soon after the return of the 12-month follow-up questionnaire, a purposive sample informed by treatment allocation, gender, age and outcome of up to 50 participants was invited for a one-off face-to-face interview [by telephone or via Microsoft Teams^® (Microsoft Corporation, Redmond, WA, USA)]. The aim of the interviews was to explore the participant experience of receiving the trial treatments, and facilitators and obstacles to adhering to them.

Follow-up

Core outcomes were completed over the telephone or by e-mail if postal copies were not returned. Text messages were also sent to participants; text messages were only sent to participants who gave prior consent.

Multiple contact details were recorded, such as addresses and telephone numbers, mobile telephone numbers and e-mail addresses, and contact details of next of kin, to prevent loss to follow-up. This information was held separately from the trial data, on a password-protected database, to uphold anonymisation, in line with current regulations. If the participant was lost to follow-up at a certain time point, reasonable efforts (e.g. phone calls, mobile text messaging, post) were used to acquire outcome data at each time point.

Changes to trial outcomes

Outcome collection took place prior to and during the COVID pandemic. Prior to the pandemic, postal data collection was the primary method of data collection. During the pandemic, due to the mandate to work from home and not leave the house, the primary method of data collection became telephone. Once the national restrictions were lifted, the team reverted back to postal mechanisms as the primary collection method. A protocol amendment was made to clarify the different methods of acceptable data collection.

Due to the national instruction to close non-COVID research to recruitment for a period of 3 months in 2020, there was delay to achieving recruitment milestones. A request was submitted to the funder and it was agreed that the date for the end of the trial would be extended to 30 November 2022, the impact being that all patients randomised after 1 August 2021 were only consented for 6-month follow-up and not 12 months. This allowed the primary research question to be answered, while minimising cost to the funder.

Adverse events and serious adverse events

Adverse events

An adverse event (AE) was defined as any untoward medical occurrence in a participant which did not necessarily have a causal relationship with this treatment/intervention. Foreseeable AEs related to the management of TASD occurring as a result of the trial intervention(s) were not recorded as part of the trial because advice and physiotherapy are part of normal clinical practice, with a good safety profile. Examples of such AEs include pain and reduced shoulder movement.

Serious adverse events

An AE was considered a SAE if it was an untoward medical occurrence that fulfilled one or more of the following criteria:

• resulted in death
• was immediately life-threatening
• required hospitalisation or prolongation of existing hospitalisation
• resulted in persistent or significant disability or incapacity
• was a congenital abnormality or birth defect
• was an important medical condition.

In the context of this protocol, ‘hospitalisation’ referred to any hospital event including day surgery and single A&E attendances. SAEs that may be expected as part of the interventions were predefined and recorded on the participant’s CRF for routine return to the ARTISAN central office and reported to the relevant oversight committees, but not recorded on a SAE form; they were instead defined as a complication. AEs/SAEs that were expected as part of the TASD and defined as a complication were: damage to nerves or blood vessels, fractures, re-dislocation, torn ligaments or muscles, persistent exacerbation of shoulder pain, restriction of range of movement, adhesive capsulitis (frozen shoulder) and persistent instability.

All participants experiencing non-predefined SAEs related to the intervention or TASD injury were entered onto the appropriate reporting form and reported to Warwick Clinical Trials Unit (WCTU) using a dedicated ARTISAN and quality assurance resource account within 24 hours of the investigator becoming aware of them.

Reporting serious adverse events

Serious adverse events and the associated management of them that may be expected as part of the interventions, and which were predefined, were recorded on the participant’s CRF only for routine return to the ARTISAN central office and reported to the relevant oversight committees. SAEs were entered onto a SAE form and once received, causality and expectedness were confirmed by either the PI or Chief Investigator.

Serious adverse events that were deemed to be unexpected and possibly, probably or definitely related to the trial interventions were notified to the Research Ethics Committee (REC) within 15 days. All such events were reported to the TMG at their next meeting. All SAEs that occurred between the date of randomisation and the end of 12-month follow-up for the participant were reported. For each SAE the following information was collected:

• details of event that occurred from participant CRF or direct from site.

If the event was identified on the participant CRF, this resulted in the site being contacted to collect:

• full details in medical terms and case description
• event duration (start and end dates, if applicable)
• action taken
• outcome
• seriousness criteria
• causality (i.e. relatedness to intervention), in the opinion of the investigator
• whether the event would be considered expected or unexpected.

If the event was identified by site, this resulted in the site contacting WCTU with the above details via a SAE form.

Any change of condition or other follow-up information was communicated to the sponsor as soon as it was available. Events were followed up until the event had resolved or a final outcome was reached. The research team liaised with the investigator to compile all the necessary information. The trial co-ordinating centre was responsible for reporting any related and unexpected SAEs to the sponsor and REC within required timelines.

Blinding

Following randomisation it was not possible to blind participants or treating clinicians to treatment allocation. However, both the treating clinician and the participant were blind to treatment allocation during the initial advice session. All staff involved in follow-up data collection were blind to treatment allocation; any unblinding was reported to the TMG. The central research team members were blinded until after data analysis was complete, with the exception of the trial statisticians, who had access to treatment assignment for the purposes of data monitoring and safety, and data-entry personnel who entered data from questionnaires, including some details of treatments received.

Statistical methods

All data have been analysed and reported in accordance with the Consolidated Standards of Reporting Trials (CONSORT) guidelines.18,19 A detailed statistical analysis plan and a data-sharing plan were agreed with the DMC prior to any formal analyses being conducted. These are publicly available. All statistical analyses were carried out using R version 4.0.3 (The R Foundation for Statistical Computing, Vienna, Austria).

Power and sample size

The OSIS is the only PROM recommended by UK BESS/BOA guidelines and was used by the most recent Cochrane review,1,2 so was chosen as the primary outcome for the study. The standard deviation (SD) of the OSIS 6 months after injury is around 10 points;20,21 however, the literature has predominantly included a younger population. Given that we planned to recruit a wider range of ages, the SD for this study might be expected to be larger. We estimated a required sample size with two-sided significance set at 5% for various scenarios of difference, power and SD (Table 3). The bolded figure of 191 participants per treatment arm represented the most likely scenario, based on a SD of 12, and our current knowledge when beginning the study, for 90% power to detect the selected, worthwhile, difference. This corresponds to a small standardised mean difference of 0.3. This represents a conservative evaluation of the sample size required based on the above literature.

Allowing a SD of 12 and for 20% loss during follow-up, this gives a figure of 478 participants in total. Therefore, 239 participants randomised to each group would provide 90% power to detect a difference of 4 points (corresponding to a small standardised mean difference of 0.3) in OSIS at 6 months at the 5% level.15,20

To address the possibility that therapist effects might adversely affect our statistical power, we planned an interim analysis to estimate the intracluster correlation coefficients (ICCs) for the pooled study data when we had 3-month data available on around 200 participants. We chose to use the earlier follow-up point because if there were therapist effect present, we would expect it would be maximal soon after the end of the treatment phase and to attenuate over longer-term follow-up, allowing us to conduct the analysis earlier. As the analysis used pooled data, adjustments to control the type I error were not needed.22

We modelled the ICC using the same mixed-effects model; we used the fixed effects planned for the primary analysis, with the term for treatment allocation removed and the physiotherapist conducting the initial physiotherapy session and subsequent randomisation added as a random effect. The 95% confidence interval (CI) of the calculated ICC was then estimated using bootstrap methods.23

This adaptive design had the additional advantage that we had actual data on the SD of our primary outcome at 3 months, which allowed us to further refine our sample size estimate.

Summary of baseline data and flow of participants

Baseline data were summarised to check comparability between treatment arms, and screening data checked to highlight any characteristic differences between those individuals in the study, those ineligible and those eligible but withholding consent. Standard statistical summaries were constructed for the primary outcome measure (i.e. OSIS) and all secondary outcome measures.

Primary outcome analysis

The main analysis investigated differences in the primary outcome measure (OSIS), 6 months after randomisation, between the two treatment groups. Unadjusted and adjusted regression analyses were used to estimate the between-group difference. The adjusted analyses accounted for the stratification variables (intervention, age group, dominant arm injured) and the baseline score. More specifically, adjusted mixed-effects modelling was used where the recruiting centre was included as a random effect to allow for possible heterogeneity in participant outcomes due to the recruiting centre. This adjusted model was predefined as the primary efficacy analysis of the study.

Secondary outcome analyses

Descriptive statistics of each PROM data set (i.e. QuickDASH and EQ-5D-5L) at each time point were constructed, and between-group analyses were performed following the method set out for the primary analysis. The patterns of recovery were also explored.

The secondary analyses carried out chi-squared tests to compare the number of dislocations and other complications between allocation groups. It was specified in the statistical analysis plans that Kaplan–Meier curves would be constructed for important complications; however, as there were only a few complications, this was omitted. The secondary outcomes were also modelled at each time point. Temporal effects on the intervention effects were also investigated using a multilevel model of all follow-up data. A mixed-effects regression model was fitted with an interaction term (treatment allocation group and time point), controlling for age group and dominant arm injured. This was analysed and reported for each of the primary and secondary outcomes.

Subgroup and exploratory analyses

Two pre-specified subgroup analyses were undertaken to measure whether there was any difference in intervention effects for hand dominance (injured shoulder in dominant arm vs. injured shoulder in non-dominant arm) and age group (younger participants vs. older participants). The analyses followed the methods described for the primary analyses, with additional interaction terms incorporated into the mixed-effects regression model.

The age group distribution was also analysed to define the best cut-off point between younger and older participants. We did this by fitting a Gaussian mixture model with a fixed support size of two, using an expectations maximisation algorithm.24 We predefined the cut point to be when the probability of membership on either distribution was 0.5. If this cut point was found to be more than 10 years different to our initial age cut point (40 years), we would use the new boundary as a sensitivity analysis.

Furthermore, we suspected the events of the COVID epidemic may have influenced the study. Therefore, we conducted exploratory analyses to investigate the effects this had on the trial follow-up rates as well as participants’ anxiety levels pre and post the date when the UK Government enforced lockdown (20 March 2020).

Criteria for the premature termination of the trial

The incidence of complications, AEs and SAEs in each group was also analysed for the interim analyses and presented to the DMC. If there were concerns regarding the patient-reported incidences of complications, the DMC could decide that further investigations needed to be made. The trial team contacted study recruiting centres to obtain confirmed complications, which were reported in the primary analyses.

Participant population

The primary analysis and secondary analyses were conducted on an intention-to-treat basis on the randomised population. That is, they included any participant randomised into the study, regardless of whether they received study intervention and regardless of protocol deviations, unless specified above.

Procedures to account for missing data

Every effort was made to ensure compliance and return of questionnaires. The impact of COVID created new challenges to obtain participant data; however, steps were taken to minimise the loss of participants’ follow-up questionnaires. Also, due to the population demographics, there were a number of participants lost to follow-up. The missingness and crossovers were carefully monitored and reported. As a result, multiple imputation was used to account for the missing data. The multiple imputation by chained equations (MICE) method using predictive-mean-matching technique was used to impute missing outcomes, with variables chosen in conjunction with the health economic analysis (see below). The baseline scores of the primary outcome (post-injury OSIS scores), and the randomisation strata (dominant arm injured, age group), were used as predictor variables for the missing values. Twenty-five imputed data sets were generated and pooled using Rubin’s rules and then a fixed-effects linear regression model was fitted, adjusting for treatment allocation, age group, dominant arm injured and baseline OSIS score, with site fitted as a fixed effect.

Health economic evaluation

Overview

The health economic objective was to assess the comparative cost-effectiveness of ARTISAN compared to ARTISAN Plus in the management of patients presenting with TASD. Resources associated with each trial arm were collected alongside information on quality of life at 6 weeks, 3 months, 6 months and 12 months. Quality-of-life information was collected using the EQ-5D-5L at baseline, 6 weeks, 3 months, 6 months and 12 months. The primary analysis adopted a NHS and PSS perspective in line with the recommendations from the National Institute of Health and Care Excellence (NICE).25 A societal perspective was used for the sensitivity analysis, and this included private costs incurred by patients because of the interventions and loss of earnings due to work absences. Outcomes were analysed using cost–utility analysis and expressed in terms of incremental costs per QALY gained. The time horizon covered the period from randomisation to 12 months post randomisation. Cost and outcomes were not discounted due to the 12-month time horizon.

Resource use

Data on health and social care services were recorded when used during the study time horizon. Societal costs included private medical costs and productivity losses due to injury. All costs were expressed in GBP using 2020/2021 prices. If necessary, costs were inflated or deflated to current prices using the NHS Cost Inflation Index.26 Resources used were collected at each follow-up point over the time horizon using questionnaires. The questionnaires captured details of the following resource use categories: medication, outpatient and emergency attendances, encounters with primary or community health and social services, inpatient and day case admission, walking and adaptive aids related to injury, and number of days off work due to injury.

It was anticipated that hospital physiotherapy visits would be key to the analysis of cost as these formed the core of the ARTISAN Plus intervention. Hospital physiotherapy visits were recorded by two methods in both groups: site records and patient recall. Neither method is perfect, since site records cannot differentiate trial-related and unrelated physiotherapy visits and patient recall might be vulnerable to under-reporting. Participant recall was used as the primary method of analysis, given its consistent use across all resource items. Site-reported hospital physiotherapy use was substituted in a sensitivity analysis.

Intervention costs

The difference between the trial arms was the offer of at least one course of physiotherapy (ARTISAN Plus) following a single advice session (ARTISAN). The recall period of the 6-week resource use questionnaire covered the period during which at least a single course of physiotherapy was offered to participants. Physiotherapy contacts reported in CRFs could not be distinguished from those that formed part of the intervention; hence intervention costs were not applied, to avoid double counting.

However, a sensitivity analysis was conducted by replacing all patient-reported physiotherapy contacts with site-reported physiotherapy visits over the follow-up period.

Valuation of resource use

Resources used were valued in accordance with methods recommended by the NICE Guide to Methods of Technology Appraisal.25 Unit costs were derived for each resource use item from national databases. The key databases used to derive unit costs for resource use items include: Department of Health and Social Care Reference Costs, Personal Social Services Research Unit’s unit cost compendium, 2021 NHS Prescription Cost Analysis database for England, 2021 volumes of the British National Formulary, and the NHS Supply Chain Catalogue 2019.27–30 Data from the Office for National Statistics were used to estimate loss of earnings due to time off work.31

Summary statistics were generated for resource use variables by treatment allocation and assessment point. Mean differences in costs between treatment groups for patients with complete data were compared using the two-sample t-test. Statistical significance was assessed at a 5% significance level.

Measuring outcomes

The primary health outcome for economic evaluation was the QALY, in accordance with NICE guidelines.25 The health-related quality of life of trial participants was assessed at baseline (pre and post injury), 6 weeks, 10 weeks, 16 weeks, 24 weeks and 12 months post randomisation using the EQ-5D-5L instrument.17 Responses to the EQ-5D-5L instrument were converted to utilities by mapping responses from the EQ-5D-5L to the EQ-5D-3L valuation set using a mapping function as recommended by NICE.32,33 QALYs were generated for each participant using the area under the curve assuming linear interpolation across each temporal measurement point. QALYs accrued over the follow-up period were summarised across each time point and reported by trial group. Between-group differences in QALYs for patients with complete data were compared using a two-sample t-test.

Missing data

Missing data are common in RCTs. Participants are likely to be lost to follow-up for various reasons. Due to trial-related systematic differences in costs and outcomes, participants with missing data may systematically differ from those with fully observed data. Hence, missing data need to be handled in a principled way, underpinned by the missing data mechanism. Missing costs and health utility data were imputed under the missing at random (MAR) assumption, at each time point, using fully conditional MICE implemented through the MICE package in Stata^® 17 (StataCorp LP, College Station, TX, USA).34 The appropriateness of using the MAR assumption was evaluated by investigating the missing data patterns and comparing attributes with and without missing cost and health-related quality-of-life data at each follow-up time point. Predictors of missingness were identified using a stepwise logistic regression model at each follow-up time point, adjusting for baseline covariates. The multiple imputation model used baseline covariates (age, gender and dominant arm; i.e. whether the injured shoulder was the dominant arm).

Unobserved costs and QALYs were imputed separately by trial arm at each time point using observed values. The imputation model was assessed for convergence, and the distributions of observed, imputed and completed data for costs and QALYs were compared graphically.35 [Reproduced with permission from Faria et al.35 This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) license, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: https://creativecommons.org/licenses/by/4.0/. The text below includes minor additions and formatting changes to the original text.] It ran 50 times following methodological guidance that the number of imputations be determined by the fraction of missing information rather than the proportion of missing data.36 Bivariate regression using the seemingly unrelated regression model (sureg), within the Stata MI (multiple imputation) framework, was used to estimate the costs and QALYs over the time horizon, controlling for baseline covariates (age and gender) and baseline utility. There were no significant interactions between the interventions and any of the baseline covariates. Joint distributions of costs and outcomes from the original data set were generated through non-parametric bootstrapping of the MI model and incremental costs and QALYs were calculated.34

Presentation of cost-effectiveness results

Cost-effectiveness was presented as an incremental cost-effectiveness ratio (ICER), calculated as the mean difference in costs and QALYs, with ARTISAN as the reference (control) treatment and ARTISAN Plus as the comparator (intervention). Bootstrapped replicates of incremental costs and QALYs were used to populate the ICER plane. Cost-effectiveness acceptability curves showed the probability that ARTISAN Plus was cost-effective compared to ARTISAN at different cost-effectiveness thresholds ranging from £0 to £100,000 per QALY. The net monetary benefit (NMB) of using ARTISAN Plus was calculated at the different cost-effectiveness thresholds. A positive incremental NMB would show that ARTISAN Plus is cost-effective when compared to ARTISAN at the specified cost-effectiveness threshold. The expected value of perfect information was calculated at willingness-to-pay thresholds and represented graphically. The expected value of perfect information reflects the monetary value of removing uncertainty from the cost-effectiveness estimates at different willingness-to-pay thresholds.

Sensitivity and secondary analysis

Sensitivity analyses were conducted to test the robustness of the cost-effectiveness estimates. This included re-estimating the cost-effectiveness estimates under the following scenarios: (1) using complete data; (2) adopting a wider societal perspective that included private costs incurred by the trial participant and productivity losses due to work absences; and (3) using data up to 6-month follow-up due to relatively lower levels of completion rates at 12 months compared to earlier time points.

Due to difficulties in distinguishing physiotherapy visits that were part of the intervention and routine visits that were requested by patients, cost-effectiveness estimates were re-evaluated using site-reported physiotherapy data. The costs of missed appointments were included (assumed to be half the price of an NHS physio visit) and the cost-effectiveness estimates were re-evaluated.

Qualitative data analysis

Interviews were digitally recorded, subject to permission of each participant, and were transcribed verbatim. Data were analysed as follows [Reproduced with permission from Ellard et al.37 This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) license, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: https://creativecommons.org/licenses/by/4.0/. The text below includes minor additions and formatting changes to the original text.]:

• Data familiarisation: reading of complete interview transcripts, listening to original audio recordings and use of field notes.
• Identifying a thematic framework: key issues, concepts and themes identified and an index of codes developed.
• Indexing: whereby the index generated through identification of the thematic framework was applied to all data.
• Charting: a summary of each passage of text was transferred into a chart to allow more overall and abstract consideration of index codes across the data set and by each individual.
• Mapping and interpretation: understanding the meaning of key themes, dimensions and broad overall picture of the data and identifying and understanding the typical associations between themes and dimensions.

The charting process provided an opportunity to code data from numerous vantage points: by demographic factors, such as gender or age; by personality characteristics, such as looking specifically at people who are highly anxious compared to those who are not; or by medical aspects, such as those with a particular condition compared to those without.

The computer package NVivo 7 (QSR International, Warrington, UK) was used to facilitate this process. Researcher bias was minimised through regular cross-checking of data and findings by the members of the team. In addition, transcripts were returned to participants (where necessary), providing them with the opportunity to check the transcripts for accuracy and authenticity and to offer any subsequent reflections.

Data management

Personal data collected during the trial were handled and stored in accordance with applicable UK data protection law. Personal identifying information was brought to WCTU for follow-up purposes. Handling of personal data was documented in the participant information sheet and consent obtained.

Disclosure of confidential information was only considered if there was an issue which may jeopardise the safety of the participant or another person, according to Warwick SOPs and the UK regulatory framework.

Data collection and management

The CRFs were developed by the trial manager in consultation with the chief investigator, statistician, health economist and other relevant members of the trial team to the required trial data. A suitably trained member of the research team completed and returned the CRFs to the ARTISAN trial office. The co-ordinating team checked and entered the data onto a secure trial database held at WCTU, in accordance with the WCTU SOPs.

Various methods were used to chase missing data/unreturned questionnaires, including post, phone, text, mobile app and e-mail. Appropriate consent was sought to contact participants.

Database

The database was developed by the Programming Team at WCTU and all specifications (i.e. database variables, validation checks, screens) were agreed between the programmer and appropriate trial staff.

Data storage

All essential documentation and trial records were stored at WCTU in conformance with the applicable regulatory requirements, and access to stored information (paper and electronic) was restricted to authorised personnel. All data were stored in a designated storage facility within the University Hospitals Coventry and Warwickshire and/or WCTU. Electronic data were stored on password-protected university computers in a restricted-access building.

Data access and quality assurance

All data collected were pseudonymised after the collection of the baseline demographic data for each participant. Confidentiality was strictly maintained, and names or addresses were not disclosed to anyone other than the staff involved in running the trial and collecting follow-up information when necessary. Participants were identified by ID number, initials and date of birth only where necessary. Identifiable participant data were held in a locked filing cabinet and coded with the trial number to tag identifiable data to the outcome data. Direct access to source data/documents was available for trial-related monitoring or audit.

Summary of changes to the trial protocol

Changes to the protocol via substantial amendments are summarised in Table 4.