Trial design
The SARAH trial was a pragmatic, multicentre RCT. The control arm (usual care) consisted of JP education, advice on simple mobility exercises for the whole body and, if appropriate, functional splinting. This was compared with an experimental intervention consisting of the same regimen supplemented with an optimised exercise programme. This programme consisted of strengthening and stretching exercises for the hand, wrist and upper limb delivered over six sessions with a therapist over a 12-week period.
Participants
Inclusion criteria
Summary of ACR clinical and immunological criteria Patients must meet four of the following seven criteria:
Exclusion criteria
Age less than 18 years.
Upper limb joint surgery, or fracture, in the previous 6 months.
Being on a waiting list for upper limb orthopaedic surgery.
Pregnancy.
Screening and recruitment
Patients were approached primarily by clinical staff during routine clinic visits. They were provided with a patient information sheet (see Appendix 1) and asked if they would consider participating. If they agreed, a research clinician (physiotherapist, occupational therapist or nurse) then contacted them to book an appointment at a research clinic. At this appointment, the trial information was discussed, the patient had an opportunity to ask questions, eligibility was rechecked and, if appropriate, patients were consented (see Appendix 2). Patients were asked to give written informed consent according to principles of Good Clinical Practice and the Declaration of Helsinki.36 At the time of consent, outcome assessors collected baseline measures.
In addition, patients on rheumatology consultant or therapy review lists without planned clinic appointments were approached using postal questionnaires. They were sent an information sheet and a brief questionnaire to return with a pre-paid envelope by the research team. The questionnaire asked whether or not they were currently experiencing problems with their hands and/or wrists and about other eligibility criteria. Patients who returned the questionnaire were contacted by the research clinician and, if appropriate, had a research clinic appointment made.
Settings and locations
The trial was run in the hospitals of 17 NHS trusts (names at time of participation):
Basingstoke and North Hampshire Hospitals NHS Foundation Trust (North Hampshire Hospital)
Derby Hospitals NHS Foundation Trust (Royal Derby Hospital)
Dorset Primary Care Trust (Victoria Hospital)
George Eliot Hospital NHS Trust (George Eliot Hospital)
Heart of England NHS Foundation Trust (Solihull Hospital)
Nuffield Orthopaedic Centre NHS Trust (Nuffield Orthopaedic Centre)
Poole Hospital NHS Foundation Trust (Poole General Hospital)
Portsmouth Hospitals NHS Trust (Queen Alexandra Hospital)
Royal Bournemouth NHS Foundation Trust (Christchurch Hospital)
Royal National Hospital for Rheumatic Diseases NHS Foundation Trust (Bath Royal National Hospital for Rheumatic Diseases)
South Warwickshire General Hospitals NHS Trust (Warwick Hospital and Stratford-Upon-Avon Hospital)
Sussex Community NHS Trust (Bognor Regis War Memorial Hospital)
Winchester and Eastleigh Health Care Trust (Royal Hampshire County Hospital)
Worcestershire Acute Hospitals NHS Trust (Worcestershire Royal Hospital; Alexandra Hospital, Redditch; Kidderminster Hospital and Treatment Centre)
Wrightington, Wigan and Leigh NHS Foundation Trust (Wrightington Hospital)
University Hospitals Coventry and Warwickshire [University Hospital (Walsgrave site) and Rugby St Cross Hospital]
University Hospitals of Leicester NHS Trust (Leicester Royal Infirmary and Leicester General Hospital).
Interventions
Full details of the interventions are provided in Chapter 3, but the interventions are described briefly here for continuity.
All interventions were delivered by physiotherapists or occupational therapists experienced in the treatment of hand and rheumatology conditions. They were independent of the recruitment and randomisation procedures, attended a training session delivered by the trial team and received ongoing support and guidance regarding the intervention to ensure quality standards were met. Therapists were trained to deliver both the experimental and control interventions. Contamination was minimised through a variety of quality control measures that included monitoring the treatment logs completed at each session, visiting therapists at the beginning of their time delivering interventions on the trial for quality assurance purposes and providing additional materials sufficient to cover only participants randomised to the experimental arm of the study. The rationale and protocol for the interventions were described in a training and reference manual. Both control and experimental interventions were developed using focus group meetings with stakeholder clinicians from across the UK.
Participants randomised to the control intervention (usual care) had, depending on the clinical need, between one and three sessions of outpatient therapy with a maximum total contact time of 1.5 hours. Treatments in these sessions consisted of JP advice, general mobility exercises and, if appropriate, functional splinting. Participants were provided with information sheets 29,37,38 and were encouraged to remain active. The participant was not reviewed again by the treating therapist after the 1.5 hours unless there were additional splinting requirements.
Participants randomised to the exercise programme received the usual care package plus an additional five sessions with a therapist over a 12-week period. The aim was to increase hand function using exercises to stretch and strengthen the muscles and tendons, as well as mobilise the joints of the hand and wrist and improve dexterity. This was supported by a home exercise programme (reinforced by a behaviour change approach and exercise diary) to be performed daily.
Monitoring the intervention delivery
Attendance rates and content of treatment appointments were recorded by the therapist using treatment logs. Logs were completed for each session for all participants and returned to the study co-ordinating centre. Close contact was maintained between the clinical research fellow and the therapy departments to address any problems that were highlighted by the treatment logs.
All sites were visited to ensure smooth implementation of the interventions within the trial. This quality control process involved the same clinical research fellow auditing treatment logs and notes and observing experimental arm intervention sessions.
Outcomes
Follow-up data collection was by face-to-face clinical assessments at 4 and 12 months following randomisation. Where face-to-face assessment was not possible, postal and telephone data collection methods were used to obtain core data. The outcome measures are detailed in and Appendix 3.
The primary outcome measure was the overall hand function subscale of the Michigan Hand Outcome Questionnaire (MHQ), which has shown to be a reliable, valid and responsive measure for a RA population.39–41 The overall MHQ contains six domains:
overall hand function
activities of daily living (ADLs)
pain
work performance
aesthetics
patient satisfaction with hand function.
Scores range from 0 to 100, with higher scores indicating better performance, except for the pain scale. For the pain scale, a higher score indicates more pain.
In the initial version of the protocol, the Arthritis Impact Measurement Scale 2 (AIMS2) finger and hand function subscale was documented as the primary outcome measure. However, prior to the full trial commencing, we conducted a pilot study involving 16 participants, and as part of this evaluated both the AIMS2 and MHQ. Based on this the MHQ was substituted as the primary outcome measure. This substantial amendment was approved by the Multicentre Research Ethics Committee and full trial protocol subsequently published.42
Secondary outcome measures consisted of measures of pain troublesomeness,43 self-efficacy,44 the European Quality of Life-5 Dimensions-3 Levels (EQ-5D-3L),45 the Short Form questionnaire-12 items (SF-12),46 socioeconomic questions (employment status, sick days in past month due to RA in wrists/hands, benefits, highest educational qualification, household income), treatment preference and healthcare use and costs. Blood test results [C-reactive protein (CRP), erythrocyte sedimentation rate (ESR), serum rheumatoid factor] and current medication (prescribed and as required in last 7 days) were taken from hospital and prescription records. The outcome assessor was present to answer any questions regarding the measures but was trained not to influence the participant’s responses. Following completion of the case report form, a physical assessment was performed in a standardised order and standardised positions. This included the measurement of joint deformity (MCP ulnar/radial deviation in maximum pronation, where ulnar deviation is recorded as a positive value) and active range of motion (wrist flexion and extension from the neutral position with a goniometer,47 combined finger flexion according to Ellis and Bruton,48 combined finger extension and thumb opposition according to Kapandji).49 A modified swollen and tender joint count (22 joints of hand and wrist)50 was taken, along with a test of upper limb dexterity (nine-hole peg test according to Mathiowetz et al.).51 Finally, two forms of grip strength (full-hand and tripod pinch) were measured using the MIE Digital Grip Analyser (MIE Medical Research Ltd, Leeds, UK).52 The standard test position recommended by the American Society of Hand Therapists was used.53 Patients were sat in a straight-backed chair without arm rests, feet flat on floor, the shoulder of the assessed limb relaxed by the side, the elbow flexed to 90°, the wrist extended and in ulnar deviation between 0° and 15° and the forearm rotated to neutral pronation/supination. The mean of three maximal 3-second grips was calculated for each hand, with 60-second rests between repetitions.
Harms/adverse events and reasons for withdrawals are tabulated in Chapter 4.
While aware of the Outcome Measures in Rheumatology (OMERACT) core set of disease activity measures, we prioritised outcomes that were likely to be sufficiently sensitive to detect any effect resulting from the interventions.
Sample size
A standardised mean difference of 0.3 is reported to represent a clinically important difference in hand function in this group.54 A previous small study using a similar intervention found a mean benefit of 0.7 in the AIMS2 with a standard deviation of 1.81 and a standardised effect size of 0.39.23 This suggested that in this larger, more pragmatic, multicentre trial, a standardised effect size of 0.3 in the similar function score using the MHQ would be realistic and meaningful. To show this difference with 80% power at the 5% significance level, we required data on at least 352 participants [using Statistical Analysis Software (V9.2, SAS Institute Inc., Cary, NC, USA) procedure GLMPOWER]. Assuming a 25% loss to follow-up would require at least 469 participants. Over 15 months we expected 1200 people with hand RA to be referred at our participating centres. If half of these were assessed for study entry and 80% of them joined the study, we would have 480 participants (1200 × 0.5 × 0.8). This was our target sample size. The assumptions underlying the sample size calculation were monitored by the Data Monitoring and Ethics Committee (DMEC) throughout the study.
Randomisation
Randomisation to the exercise programme or usual care was via a central telephone randomisation service at Warwick Clinical Trials Unit, University of Warwick.
Sequence generation
The variable block randomisation schedule was prepared by the trial statistician (CM). Randomisation was stratified by centre to control for any confounding factors evident at local recruitment sites, such as therapist effects or local contamination of intervention.
Allocation concealment
The research clinician telephoned the randomisation service, and only once the participant was registered in the trial was the random allocation generated. Hence allocation was concealed. Allocation was faxed or emailed direct to the therapist, and the participant was told either when their first appointment was made or at the appointment itself.
Blinding
The outcome assessor was blind to the group allocation of the participant and was independent of intervention delivery. Participants were requested not to disclose group allocation to the outcome assessor. If an outcome assessor was unblinded, this was recorded. If they remained unblinded, the assessors were asked to guess which allocation they thought the participant was given. The participants and therapists could not be blinded to the group allocation.
Statistical methods
Participants were analysed according to the treatment group to which they were randomised, regardless of the treatment that they actually received (intention-to-treat analysis).55 Analyses were guided by an analysis plan prepared before data were available. Demographic, clinical characteristics and baseline measurements are presented to evaluate the comparability of intervention groups and generalisability to clinical settings (see Chapter 4, Baseline data). A Consolidated Standards of Reporting Trials (CONSORT) diagram was produced (see ).56
The difference between the intervention groups in mean score of MHQ overall hand function score from baseline to 4 and 12 months was analysed by a linear model, adjusted for the baseline score. Means and 95% confidence intervals (CIs) are given.
The secondary outcome measures of change in pain, impairment measures, disease activity, quality of life, self-efficacy, satisfaction and global change scores were analysed in a similar manner to the primary outcome measure.
Analyses also took account of age, sex and current drug regimens by including baseline medication covariates on outcome analyses. The therapy groups used were biologic DMARD, combination non-biologic DMARDs, single non-biologic DMARD and no DMARD. Patient-rated improvement was compared using the Wilcoxon test. Other ordinal testing for patient-rated benefit/harm and treatment satisfaction used the Mantel–Haenszel chi-squared test for linear association and tests for differences in other categorical data including medication and performing exercises used Pearson’s chi-squared. Standardised mean differences are calculated as the unadjusted mean difference between treatments divided by the pooled baseline standard deviation.
A pre-specified complier average causal effect (CACE) analysis was conducted to determine if the level of compliance with the exercise programme affected participant outcome as measured by the overall hand function subscale of the MHQ (primary outcome). The CACE estimates were calculated using the instrumental variable method described in Dunn et al.57 The analysis was performed using a threshold of six or more sessions (i.e. attended all sessions) as defining compliance. An additional analysis investigated a threshold of three or more sessions.
The impact of therapist effects was evaluated by including a random therapist effect nested within centre in the main outcome model. The effect of missing data was investigated using a Markov chain Monte Carlo multiple imputation analysis.
In the original protocol, the primary analysis was stated as a repeated measures mixed model, but this was changed to an adjusted linear model to make the results more interpretable by a wide audience. It was also decided that survival analysis methods applied to surgical and serious adverse (SAE) event data were not appropriate. These changes were approved by the DMEC and Multicentre Research Ethics Committee (MREC) in the statistical analysis plan and Version 2 of the protocol which was subsequently published.42
Statistical analyses were performed using SAS V9.2 software.
Database and data processing
The database was held on Warwick Clinical Trial Unit’s Microsoft SQL Server system (Microsoft Inc., Bellevue, WA, USA) and imposed rules for data entry which include valid range for responses, linked dates and patient identification numbers.
Data were single entered into the database by study personnel. The trial statistician carried out checks of plausibility of values, missing data and form return rates to enable further queries to be resolved prior to freezing data for scheduled DMEC reports and analysis.
For data quality assessment, 10% of all forms at all time points were randomly selected for data checking on a 2-monthly basis. All disagreements found when checking were corrected and any systematic faults found as a result of the checks were also corrected. The levels of accuracy specified were 1% for primary end point data and 5% for secondary end point data. Five time periods were covered and disagreement levels were always below 1% apart from one occasion when, in one 2-month period, a 2% level was found for primary data. Corrections were made and the problem did not recur.
Scoring and missing items
Michigan Hand Outcome Questionnaire scales were scored as described by Chung et al.58 Individual item scores are summed to give a raw score for each scale. These are converted to a score between 0 and 100 using the MHQ scoring algorithms. Averages of right and left hand are used for those scales where both hands are measured. Scales were calculated if at least 50% of their items were completed. An overall score can be obtained by averaging the score for all scales (with pain scale reversed). At least four scales need to be completed to calculate the overall MHQ score.
The SF-12 was scored as described by Ware et al.59 We analysed mean component scores for the Mental (Mental Component Score; MCS) and Physical (Physical Component Score; PCS) subscales. Missing responses were dealt with using an imputation algorithm.60
European Quality of Life-5 Dimensions was used mainly for economic analyses evaluation; methods and results for the economic analyses evaluations are described in Chapter 6.
Further details on the scoring used for impairment measures is given in Appendix 4.
Subgroup analyses
Statistical tests of interaction were used to perform pre-specified subgroup analysis on baseline drug regimen (no DMARD, single DMARD, combination DMARD or biologic DMARD) and disease duration (< 5 years or ≥ 5 years). Post hoc exploratory analyses were also conducted for baseline MHQ overall hand function (< 52.5 points or ≥ 52.5 points), age (< 60 years or ≥ 60 years), type of referral (clinic or review list mail out), gender, baseline ESR (< 16 mm/hour or ≥ 16 mm/hour), baseline CRP (< 6 mg/l or ≥ 6 mg/l), SF-12 PCS (< 34 points or ≥ 34 points) and SF-12 MCS (< 50 points or ≥ 50 points).
Sensitivity analyses
Sensitivity analyses were planned to explore effects of results of adjustment for age, sex, site and any imbalance in baseline characteristics. A repeated measures analysis was also performed.
Ethical approval and monitoring
Ethics committee approval
The SARAH trial was approved by the Oxford C Multicentre Research Ethics Committee in June 2008 (Research Ethics Committee reference 08/H0606/47) and by the research and development department of each participating centre. The final approved study protocol has been published.42
Trial Steering Committee
A Trial Steering Committee (TSC) was responsible for monitoring and supervising the progress of the SARAH trial. The TSC consisted of three independent experts, a lay member and leading members of the Trial Management Group (TMG). Membership of the TSC is given in Acknowledgements.
Data Monitoring and Ethics Committee
The DMEC was independent of the trial and was tasked with monitoring ethical, safety and data integrity. The trial statistician provided data and analyses requested by the DMEC at each of the meetings. Membership of the DMEC is given in Acknowledgements.
Trial Management Group
A TMG was responsible for the day-to-day management of the trial, consisting of the chief investigator, research fellows, statistician, trial co-ordinator, research nurse and data-entry clerk. The role of the group was to monitor all aspects of the conduct and progress of the trial, to ensure that the protocol was adhered to and to take appropriate action to safeguard participants and the quality of the trial itself.
