The Ankle Injury Management (AIM) trial: a pragmatic, multicentre, equivalence randomised controlled trial and economic evaluation comparing close contact casting with open surgical reduction and internal fixation in the treatment of unstable ankle fractures in patients aged over 60 years

Article history

The research reported in this issue of the journal was funded by the HTA programme as project number 07/37/61. The contractual start date was in January 2010. The draft report began editorial review in March 2015 and was accepted for publication in December 2015. The authors have been wholly responsible for all data collection, analysis and interpretation, and for writing up their work. The HTA editors and publisher have tried to ensure the accuracy of the authors’ report and would like to thank the reviewers for their constructive comments on the draft document. However, they do not accept liability for damages or losses arising from material published in this report.

Declared competing interests of authors

Sarah E Lamb is the chairperson of the National Institute of Health Research Health Technology Assessment programme Clinical Evaluation and Trials Board and is on the Clinical Trials Unit Standing Advisory Committee. Keith Willett declares design royalties from Zimmer Biomet, outside the submitted work, for intramedullary bone fixation implants.

Copyright statement

© Queen’s Printer and Controller of HMSO 2016. This work was produced by Keene 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.

Background

Ankle fractures are a common injury in the adult population. In high-income countries, the burden from ankle fractures placed on health-care services is substantial, as 9–17% of all fractures seen in acute care hospitals are ankle fractures. 1,2 A Scottish study reported an incidence of 132 fractures per 100,000 in men and of 112 fractures per 100,000 in women per year, with the highest incidence of 248 per 100,000 per year occurring in women between the ages of 75 and 84 years. 3 A threefold increase in the number of ankle fractures per year is predicted from 2000 to 2030 as the population ages. 4

Ankle fractures often result in ankle pain, limitation in joint range of motion,5 muscle weakness6,7 and difficulties with weight-bearing tasks such as walking and climbing stairs. 5,8 Limitations in ankle-related function are especially pronounced in the 6 months after injury, but can persist for several years. 9

Physical function outcomes after ankle fracture are generally worse when the injury occurs in older age. 9,10 The injury may contribute to a loss of independence, have a substantial impact on quality of life and have higher health and social care costs. 11–13

In this introduction chapter, we provide a background to the current management of ankle fractures, explore the issues with existing treatment options in older-aged adults, summarise the existing evidence base and outline the rationale for the Ankle Injury Management (AIM) trial and the novel close contact casting (CCC) intervention.

Management of ankle fractures

Ankle fractures can be managed conservatively or surgically, depending on the extent and pattern of bone and ligament injury. The extent/pattern of injury is determined by direction and magnitude of traumatic forces on the ankle and resilience of the musculoskeletal tissues. Ankle fractures are usually graded on radiographs using the Lauge-Hansen classification based on mechanism of injury or the Danis–Weber classification extensions developed by the AO Foundation, which are based on anatomical injury characteristics (Figure 1). 14

FIGURE 1.

Comparison of the Lauge-Hansen (left column) and the AO Foundation and Orthopaedic Trauma Association (AO/OTA) (right column) systems of ankle fracture classification.© From Oxford Textbook of Trauma and Orthopaedics edited by Bulstrode (2011) Figure 12.59.4. By permission of Oxford University Press (www.oup.com).

Stable and undisplaced fractures are typically treated conservatively in an ankle splint or cast. Displaced ankle fractures resulting in anatomical misalignment are initially managed by manipulation of the limb to restore anatomical congruence. This is termed ‘fracture reduction’, as in reduction of the deformity. The limb is then held in a cast or splint until definitive management is implemented. Usually displaced and/or unstable ankle fractures are managed with open reduction and internal fixation (ORIF) surgery to both restore and maintain congruence of the ankle mortise. 14 During ORIF, bone fragments are repositioned and held in place by plates and/or screws until healing (union) occurs.

The rationale for ORIF surgery is to improve outcomes by reducing complications from malunion such as post-traumatic osteoarthritis. 15,16 Malunion of weight-bearing joints is postulated to lead to increased contact stresses on the articular surfaces,17,18 precipitating osteoarthritis that can result in persistent symptoms and disability and potentially the need for further surgery. 19,20 However, the aetiology of post-traumatic osteoarthritis may also be a result of genetic factors and direct damage caused by the initial trauma. 21

Non-union (incomplete healing) of the fracture can be considered to be less likely when electing to treat an ankle fracture with ORIF surgery than with conservative treatment. 22 However, there are limited randomised controlled trial comparisons on whether or not non-union is more likely with conservative casting or ORIF. 23

Research objectives

1. To determine if the application of CCC for unstable ankle fractures in older adults resulted in an equivalent clinical outcome compared with the standard care of open surgical reduction and internal fixation (ORIF).

2. To estimate the cost-effectiveness of the two treatments to the NHS, and the broader societal perspective, including to the individual and their family.

3. To explore the experiences of participants of the interventions and the impact of taking part in the trial.

Summary of study design

This study was a pragmatic, multicentre, equivalence randomised controlled trial with parallel prospective economic evaluation. Participants were randomised to receive ORIF or CCC after admission for surgery for displaced and/or unstable ankle fractures in the trauma and orthopaedic surgery departments of 24 hospitals from 22 NHS trusts. The final study protocol has been published. 38

Settings and locations

The trial was run in the following hospitals in England and Wales:

• Addenbrooke’s Hospital, Cambridge University Hospitals NHS Foundation Trust

• Darlington Memorial Hospital, County Durham and Darlington NHS Foundation Trust

• Derriford Hospital, Plymouth Hospitals NHS Trust

• Frenchay Hospital, North Bristol NHS Trust

• Great Western Hospital, Great Western Hospitals NHS Foundation Trust, Swindon

• Ipswich Hospital, Ipswich Hospital NHS Trust

• John Radcliffe Hospital, Oxford University Hospitals NHS Foundation Trust

• Leeds General Infirmary, Leeds Teaching Hospitals NHS Trust

• Leicester Royal Infirmary, University Hospitals of Leicester NHS Trust

• Morriston Hospital, Abertawe Bro Morgannwg University Health Board, Swansea

• Musgrove Park Hospital, Taunton and Somerset NHS Foundation Trust

• Newham University Hospital, Barts Health NHS Trust, London

• Norfolk and Norwich University Hospital, Norfolk and Norwich University Hospitals NHS Foundation Trust

• North Staffordshire Royal Infirmary, University Hospital of North Staffordshire NHS Trust, Stoke-on-Trent

• North Tyneside General Hospital, Northumbria Healthcare NHS Foundation Trust

• Poole Hospital, Poole Hospital NHS Foundation Trust

• Royal Berkshire Hospital, Royal Berkshire NHS Foundation Trust, Reading

• Royal Victoria Infirmary, Newcastle upon Tyne Hospitals NHS Foundation Trust

• Southport and Ormskirk Hospitals, Southport & Ormskirk Hospital NHS Trust

• Torbay Hospital, South Devon Healthcare NHS Foundation Trust

• University Hospital Coventry, University Hospitals Coventry and Warwickshire NHS Trust

• University Hospital of North Durham, County Durham and Darlington NHS Foundation Trust

• Wansbeck General Hospital, Northumbria Healthcare NHS Foundation Trust

• Yeovil District Hospital, Yeovil District Hospital NHS Foundation Trust.

Participants

Men or women aged over 60 years with an isolated displaced and/or unstable ankle fracture who met the following eligibility criteria.

Participant approach and recruitment

The treating surgical team undertook the initial approach to participants. If the participant was willing, a member of the research team explained the study in more detail and checked the eligibility criteria. A Mini Mental State Examination to assess cognitive function was undertaken prior to randomisation. Potential participants were given as long a time as possible to consider participation.

Interventions

All interventions were undertaken within a hospital and conformed to the NHS standard of being performed under consultant supervision. Both study interventions were applied in theatre under general or regional anaesthesia. We recorded time to treatment and the type of anaesthesia (regional, general or both). Surgeons were advised that talar tilt or shift resulting in significant joint incongruence would be considered unacceptable.

Intervention: manipulation under anaesthetic in theatre and application of a close contact casting

Standardisation of the casting materials, cast design and application, and moulding technique was achieved by surgeon instruction (training session and access to training videos and documentation). The training video was available to view prior to any application if a prompt was desired, which could be accessed online (www.youtube.com/playlist?list=PL2Gg_an4nwPfIUC9RQV54Y2lbD76HiWcV). The method of closed-fracture manipulative reduction of deformity was at the discretion of individual surgeons and this falls within the common contemporary skills set of senior surgical trainees and consultants. We advised that CCC should be applied once major swelling had subsided, at a similar time to when open surgery would be considered. We also issued guidance that a consultant grade surgeon who had attended a CCC training session was required to be involved, or at least directly observe, CCC because this was a new technique being introduced into clinical practice. The use of specific moulding points and correspondingly sited pressure pads within the cast (Figure 2) aimed to prevent fracture displacement while also minimising the risk of skin damage. The locations of the moulding points were specific to the participants fracture pattern, so the points of application required to hold the fracture were the decision of the treating surgeon.

FIGURE 2.

Outline of the shapes and locations of the foam pads used in the CCC.

Fracture position was checked by monitoring radiographs in the weeks following initial CCC application, and after any reapplication of the CCC in the clinic if the cast had loosened. Such cast changes did not require anaesthesia. The protocol specified that, if during clinical follow-up there was an unacceptable loss of fracture position prior to clinical union, it was the treating surgeon’s decision whether to reapply a CCC in the outpatient clinic, remanipulate and apply CCC in the operating theatre or convert to ORIF. It was advised that participants undergoing CCC should not do more than touch weight-bearing in the initial 4 weeks after application but then build up to full weight-bearing by 6–8 weeks, but as a pragmatic trial this progression was at the surgeon’s discretion.

A member of the trial team provided training on CCC for surgeons. The training involved a presentation on the AIM trial and the rationale and guidance related to CCC. A training video was then viewed, followed by a live interactive demonstration of the cast application.

The CCC construct was formed by (1) an underlayer of bandage stockinette (e.g. bandage stockinette jersey; BSN Medical GmbH, Hamburg, Germany), (2) small self-adhesive foam pads (e.g. fleecy foam 5 mm; Hapla, Oldbury, UK) over bony prominences (tibial crest, fibular head, calcaneum and metatarsal heads) and over the moulding points on the lower leg medially and laterally, (3) two self-adhesive strips running the full length of the cast (e.g. fleecy web roll 5 cm; Hapla, Oldbury, UK) where the plaster saw would be placed to aid safe removal of the cast, (4) a single non-overlapping layer of synthetic wool roll (e.g. Soffban^® Plus; BSN Medical GmbH, Hamburg, Germany), (5) plaster of Paris (e.g. Gypsona^®; BSN Medical GmbH, Hamburg, Germany) and (6) a topcoat over reinforcing synthetic casting material (e.g. 3M™Soft Cast Casting Tape; 3M Health Care Ltd, Loughborough, UK).

It was possible that in some cases, after randomisation, the intervention delivered would necessarily be changed. CCC was excluded as an option outside the group randomised to CCC. However, the study protocol expected and gave instruction on the following scenarios.

• After randomisation and allocation to ORIF, at the point of intervention with anaesthesia commenced, the temporary cast was removed and the ankle skin condition was such that the surgeon considered one or all of the necessary surgical incisions to be unsafe. An alternative treatment would be given (1) traditional plaster cast or (2) external fixation.

• After randomisation and allocated to CCC, at the point of intervention with anaesthesia commenced, a fracture may have proved irreducible by closed manipulation. The surgeon would necessarily have proceeded to ORIF.

• If there was an unacceptable loss of position by either treatment method prior to fracture healing, the surgeon was advised to adopt an alternative treatment approach best judged to achieve a favourable outcome.

• A combination of bone and skin fragility and gross joint instability may have excluded either intervention. The surgeon would then apply a temporising external fixator, and definitive treatment was at the surgeon’s discretion.

• After randomisation, there may have been a requirement to have a temporising treatment applied in theatre (manipulation and back slab cast or external fixator) until it is clinically appropriate to return to theatre to receive the allocated treatment.

Monitoring intervention delivery

Routine data checks on theatre procedure forms facilitated monitoring of the intervention delivery as per the protocol. All sites were visited for delivery of training in the CCC intervention, as this was not part of standard care. All attending surgeons completed forms to provide details on their grade and experience of the interventions. The surgeon was then allocated a code which was required on the theatre procedure forms to monitor that the CCC interventions were being applied by those who had attended training and that a trial-trained consultant grade surgeon was involved or directly supervising. Any discrepancies in the grade of surgeon recorded at CCC training compared with their grade at primary theatre were because they had moved up a grade. Discrepancies were identified during routine data checks and sites were contacted if any occurred. Additional training was offered when there were issues regarding availability of trained surgeons.

Learning and expertise effects

It is common practice that surgeons have particular expertise in selected techniques, and for surgical teams to organise their workloads so that expertise is utilised to best effect. For each surgeon participating in the study, we collected the following information: historical experience and preferences for ORIF and casting, grade of surgeon and time since first operation on the study. Surgeon codes allocated at training were used to trace the surgeons’ activity during the trial in order to inform the learning effects analysis. However, some of the trained surgeons involved in the pilot study did not have an allocated code and were thus known exclusions at the outset.

Baseline assessments

Baseline assessments were undertaken by one of the research team following consent and prior to randomisation. None of the participants would have been ambulatory at the baseline phase, but we collected information about pre-injury mobility status using a retrospective report from the Olerud–Molander Ankle Score (OMAS)41 as well as health-related quality of life using the European Quality of Life 5-Dimensions (EQ-5D)42 and the Short Form questionnaire-12 items (SF-12). 43 Although not ideal, recall was the only method that we had for assessing pre-fracture abilities. As the recall period was relatively short (maximum of 2 weeks from injury), we did not expect problems. The type of residence in the month prior to admission was recorded, as was the level of support provided and whether or not the participants lived alone prior to the injury. Information on pre-injury mobility, medical history, smoking, alcohol intake, allergies, medication and care requirements were collected.

Outcome measures

Participants were asked to attend study assessments at 6 weeks and 6 months; those unable to were offered telephone or postal follow-up. The outcomes and time points for the study are outlined in Table 1. The primary outcome measure was the OMAS, an ankle function outcome questionnaire, at 6 months. 41 The OMAS is a rating scale from 0 (totally impaired) to 100 (completely unimpaired) based on nine items. The response to each of the nine items is assigned a score and the summation of these scores makes up the overall score. The items include pain, stiffness, swelling, stair climbing, running, jumping, squatting, supports and work/activities of daily living.

Secondary outcomes included health-related quality of life measured by the EQ-5D (three levels) and SF-12 (version 1). The EQ-5D questionnaire is used to compute a health utility score that typically ranges from 0 (death) to 1 (perfect health). Negative scores can also be obtained that are reflective of a patient’s quality of life being worse than death. The SF-12 questionnaire is used to compute a mental component summary score and physical component summary score. These scores are measured on a scale of 0–100, with a lower score indicating poorer physical or mental functioning. Pain outcomes were reported and analysed using the pain subscales of the OMAS and the EQ-5D separately.

Physical impairment of ankle joint range of motion and mobility were measured. The range of joint motion was assessed at 6 weeks, when clinically appropriate, and at 6 months after randomisation. Range of motion assessments were conducted using a goniometer. 44,45 The standardised technique for the trial was outlined in an illustrated guide issued to sites. Additional training was provided when requested. Participants were positioned on a plinth in long sitting, reclined to about 45°. A support was placed under the upper part of the lower legs to flex the knee to 20–30°, to reduce tension in the triceps surae muscle complex and to lift the heels off the surface of the plinth. Ankle dorsiflexion and plantar flexion range of motion were measured in degrees from the anatomical neutral position, otherwise known as plantar grade (Figure 3). Participants with insufficient dorsiflexion to move the ankle beyond neutral were recorded as having a negative score (e.g. if the participant was 5° from reaching neutral his or her score was –5°).

FIGURE 3.

Lateral aspect of the lower leg with a line indicating plantargrade at the ankle.

For dorsiflexion and plantar flexion the goniometer was placed on the lateral aspect of the ankle; the axis of the goniometer was placed approximately 1.5 cm inferior to the lateral malleolus; the stationary arm was placed parallel to the longitudinal axis of the fibula, lining up with the fibula head; and the moveable arm was placed parallel to the longitudinal axis of the fifth metatarsal (Figure 4). For inversion and eversion, the goniometer was placed on the anterior aspect of the ankle; the axis of the goniometer was placed over the middle of the ankle joint line, the stationary arm was placed along the longitudinal axis of the tibia lining up with the tibial tuberosity; and the movable arm was placed along the second metatarsal in line with the base of the second toe (Figure 5).

FIGURE 4.

Positioning of the goniometer for the assessment of ankle dorsiflexion and plantar flexion range of motion.

FIGURE 5.

Positioning of the goniometer for the assessment of ankle inversion and eversion range of motion.

Mobility, assessed using the timed up and go test,46 was included as an outcome at 6 months. The timed up and go test is a test specifically designed for frail older people. The test recorded the time taken to get up from a chair with armrests, walk 8.6 m (standardised distance for the AIM trial), turn at a mark on the floor, then walk back to the chair and sit down again. Participants were asked to walk safely as fast as possible and used a walking aid if they normally did so. Performance tests are a recognised standard for measuring mobility and associations with important end points including risk of falling, functional decline and institutionalisation. 47 We also noted the date that participants commenced partial weight-bearing. Patient satisfaction was measured on an ordered categorical scale.

Outcomes for the economic evaluation were the incremental quality-adjusted life-years (QALYs) and the incremental costs of CCC over ORIF. Incremental differences were the differences in totals. Total costs were the sum product of resource use with unit resource cost. Total QALYs were the integration of utility over time.

Randomisation

Randomisation took place following screening and baseline assessments. The unit of randomisation was the individual and assignment was in a 1 : 1 allocation ratio. We used the remote 24-hour telephone randomisation service available from the University of Aberdeen.

Allocation concealment was ensured by registering participants before computer generation of the allocation code. Sequence generation was by random block size and stratified by centre and fracture pattern, using trans-/infrasyndesmotic (type A/B) and suprasyndesmotic categories (type C) as stratification factors.

Blinding

At 6 months a health professional, who was blind to treatment assignment, completed the clinical measurements and ensured completion of the study questionnaires. The presence or absence of surgical incision(s) was obscured by opaque bandage(s) applied by a research nurse/therapist prior to the participant meeting the blinded assessor. Patients unable to attend were contacted by telephone or visited at home. We undertook an analysis of the success of the blinding strategy using outcomes from questions asking blinded assessors to indicate if they believe they knew the intervention allocated and/or received.

It was not possible to blind the surgeons or participants because of the nature of the interventions. It was also not possible to blind radiograph assessors or treating surgeons as the implants, or their absence, would be apparent on the radiographs, as would the surgical scars on examination. The 6-week follow-up was conducted in the clinic by researcher, who was not blinded to treatment allocation. At 6 weeks, the researcher needed to be aware of relevant precautions and contraindications to the clinical assessment of range of ankle motion (i.e. discuss this with the surgeon to ensure fracture was stable).

Sample size

Given the paucity of data in the published literature, pilot trial data were used as a primary source to inform estimates of variance and treatment effects measured using the OMAS, and a range of secondary outcomes.

Although the original sample size estimate was based on a difference in proportions, this was modified following advice from the Data Monitoring and Ethics Committee (DMEC), as data from the pilot study showed the data to be normally distributed and that analysis based on a continuous score would be more efficient and meaningful. Parameters for the sample size were informed by data from the pilot study, known only to the study statisticians and the DMEC. We utilised a sample size calculation with one-sided testing (p = 0.05), as we were not trying to prove that the new treatment was better than the standard and, therefore, gained considerable statistical efficiency. 49,50 The power of the study was set at 80% as is conventional in clinical trials. 50

Data from the pilot study informed the sample size; the standard deviation (SD) on the operative arm was 16.2 OMAS points. An equivallence margin of ± 6 points on the OMAS yielded the final sample size of 560 in total. 49 We inflated for loss to follow-up of nearly 10%, yielding a total sample size of 620. Published estimates to inform the selection of equivalence margins using the OMAS were non-existent. Using the pilot data to calculate standardised effects sizes, the equivalence margin included small differences (< 0.37), but excluded moderate or large treatment differences. This was consistent with clinical opinion supporting a 6-point margin excluding clinically important differences in this condition gathered in an informal survey of orthopaedic surgeons. It was also consistent with published data on the minimally clinically important differences for similar scores (Foot and Ankle Score and visual analogue pain scores in acute injury) that reported minimally clinically important differences > 10 points on a 100-point scale.

Pilot study

A pilot study was conducted in the trauma service at the Oxford University Hospitals NHS Foundation Trust. We recruited 95 participants and this informed the design and established the feasibility of the multicentre phase of the study. None of the participants of the pilot study completed the full health economic evaluation questionnaire used later in the multicentre phase of the AIM trial. The independent DMEC agreed that the pilot trial design was sufficiently similar to the multicentre phase for the data from these participants to be integrated.

Statistical methods

In equivalence testing, a maximum clinical difference (Δ_T) is prespecified at a level within which the two treatments can be considered not to differ in any clinically meaningful way. Therefore, the relevant null hypothesis was that a difference of greater than Δ_T exists in either direction (H0: Δ ≤ –Δ_T or Δ ≥ Δ_T) and the trial was targeted at disproving this in favour of the alternative that no clinical difference existed (HA: –Δ_T < Δ < Δ_T). The US Food and Drug Administration’s regulations recommend both a treatment received (per-protocol) and an intention-to-treat (ITT) analysis, aiming to demonstrate equivalence. 51 Use of an ITT approach in a superiority trial sometimes increases the chance of falsely claiming equivalence. 52,53 ITT is conservative when trying to detect a difference, and so the opposite is true when trying to show similarity. The effect of low protocol adherence would therefore make the treatment groups appear more similar that they actually are. Initially, a per-protocol analysis was undertaken where only the patients who received their allocated treatment were analysed and those patients who did not were excluded from the analysis. Following we carried out an ITT analysis was carried out in which all randomised patients were analysed according to the treatment to which they were randomised.

Participants were excluded from the per-protocol analysis if they did not receive their allocated treatment during either their initial inpatient stay (i.e. during their primary theatre procedure or as an additional theatre procedure) or during a readmission. The protocol allowed for the eventuality of participants requiring a visit to the operating theatre for a temporary intervention prior to receiving the allocated treatment (e.g. a temporary cast, back slab or external fixation). However, if the allocated intervention was received after a clinically definitive intervention, then the participant was not deemed to have received the allocated treatment.

The result of the analysis of the primary end point could be one of the following:

• The confidence interval (CI) for the difference between the two treatments lies entirely within the equivalence range, –Δ_T to Δ_T, so that equivalence may be concluded with only a small probability of error.

• The CI covers at least some points that lie outside the equivalence range, so that differences of potential clinical importance remain a real possibility. In other words, superiority of treatment cannot be ruled out and thus equivalence cannot safely be concluded.

• The CI is wholly outside the equivalence range (although this is likely to be rare). This is indicative of the trial being underpowered as the CI includes non-trivial effect sizes that are in favour of both arms.

As well as assessing if equivalence was demonstrated in either case, this also formed part of an additional sensitivity analysis to assess the range of potential biases that could have resulted from loss to follow-up, protocol deviations, withdrawal (and mortality). Numerical and graphical summaries of all the data were compiled, including descriptions of missing data at each level. Estimates of treatment effect were reported with 95% CIs and a figure showing CIs and margins of equivalence were also presented. Our main analytical methods were generalised linear models (GLMs), and all primary and secondary outcomes were adjusted for important baseline covariants [age, sex, centre (hospital), fracture pattern (Weber A and B or C), baseline score] to maximise precision.

The OMAS, at 6 months, was the primary outcome in this study and was compared between treatment groups as the dependent variable in a linear regression model for the primary analysis. The treatment difference was based on the estimates of the adjusted means and 95% CIs. The OMASs were also presented as an ordinal outcome in a secondary analysis using ordered logistic regression (proportional odds model). Secondary outcome measures were similarly analysed with logistic regression models being used for categorical data and linear regression models for continuous data. The EQ-5D measure was summarised both at an item level and an overall score as recommended by the EuroQol group. Time-to-event data (time from randomisation to discharge and time from randomisation to readmission) were analysed using a log-rank test. Any patients who did not experience an event at the time point of interest or who withdrew were censored. The p-value and a hazard ratio with its 95% CI from a Cox proportional hazards model were also presented. The proportional hazards assumption across treatment arms was checked graphically using a log-cumulative hazard plot. Outcomes that were not normally distributed were evaluated using a non-parametric Wilcoxon rank-sum test. A data analysis plan was agreed with the DMEC.

The success of the blinding strategy was assessed using James’ Blinding Index, which ranges from 0 (total lack of blinding) to 1 (complete blinding). 54 The learning effects were assessed by fitting a longitudinal random model. For each surgeon, operation time was ordered sequentially by date and a time variable was created. This time variable was fitted as a random effect into a longitudinal model, with the operation time as the response variable. The surgeon was also included as a random effect. Another similar model was fitted nesting surgeon within hospital and treating these as random effects. This assessed the learning effect allowing for surgeons within their hospitals.

The clinical results were analysed using Stata version 13.1 (StataCorp, College Station, TX, USA) and SAS version 9.3 (SAS Institute Inc., Cary, NC, USA).

Economic analyses

An evaluation of cost-effectiveness of CCC compared with ORIF was conducted as part of the AIM trial. The economic evaluation presented in this chapter was conducted for both the NHS and societal perspectives. We conducted analyses for both the per-protocol and ITT populations. Consistent with the evaluation of cost-effectiveness, the primary analysis was per protocol.

Analysis methods

The economic evaluation was conducted for both the NHS and societal perspectives. The evaluation was also then conducted for both the per-protocol and ITT populations. The primary analysis used the per-protocol population. The overall flow of analysis is depicted in Figure 6.

FIGURE 6.

Conceptual diagram of relevant economic inputs and their relationship to intermediate and final economic outcomes. Starting on the left are the resource use and utility information from the trial, which, when applied to their cost and time weights, give total costs and total QALYs, respectively. Incremental differences can then be calculated and synthesised into an incremental cost-effectiveness ratio, if appropriate. Sampling uncertainty, parameter uncertainty and heterogeneity are incorporated or controlled for by means of resampling/bootstrapping and adjustment via statistical modelling. ICER, incremental cost-effectiveness ratio; LOS, length of stay.

Missing data

In the case of missing data, multiple imputation with chained equations was used, in which evidence supported the conclusion that data were missing at random. Individuals who had zero information on all outcomes were removed from the analysis. This applied to individuals with no utility information at presentation, 6 weeks and 6 months.

Multiple imputation with chained equations is the dominant method for complex incomplete data across multiple variables. 65 It accounts for the process that created the missing data, preserves the relations in the data and preserves the uncertainty about these relations. 65 Potential predictors included baseline characteristics as well as all resource use and health outcome items. Downstream variables were removed as predictors of upstream variables. Multiple imputation was conducted using 40 iterations and five imputations. Density and convergence plots were inspected to confirm comparable distribution shapes to observed data and healthy convergence.

Costs and quality-adjusted life-years

Mean resource use was tabulated by category and type of health-care service for ORIF and CCC. After consideration of normality using visual inspection and Shapiro–Wilk tests, differences in resource use were estimated using the Wilcoxon’s signed-rank test, Fisher’s exact test or chi-squared tests, depending on the variable. Total NHS and societal costs for each participant were calculated by adding the cost of each resource use.

Utility weights at presentation, and at 6 weeks and 6 months after randomisation, and total QALYs are presented for the ORIF and CCC groups. After consideration of normality using visual inspection and Shapiro–Wilk tests, differences in utility and QALYs were estimated using the Wilcoxon’s signed-rank test. QALYs were estimated by integrating the length of life by the utility weights as an ‘area under the curve’. We assumed monotonic changes in utility between any two measurement points.

We presented 95% confidence limits using a bootstrap approach. Total NHS cost, total societal cost and total QALYs were resampled 10,000 times, stratified by treatment with ORIF or CCC to ensure equal sample sizes to that of the original number of participants in each group.

Total NHS and societal costs, as well as total QALYs, were estimated using a regression approach. The regression included a treatment term in addition to other relevant baseline characteristics. As total costs and QALYs were skewed (as they generally are), we used a GLM. The appropriate family and link functions were determined using the modified Park’s and Hosmer–Lemeshow tests, respectively. We used the best-fitting GLM and the generic ordinary least squares (OLS) model to present results. Baseline covariates of interest were selected by clinical reasoning, independent association with the dependent variable at the p < 0.25 level and prevalence > 10%.

The developed statistical models were then used to estimate the mean total NHS and societal costs as well as mean total QALYs using marginal prediction. Marginal prediction eliminates known covariate imbalance between the intervention groups before estimating total costs and QALYs. Any observed comparison of two groups, whether randomised or not, is likely to show imbalance in baseline covariates, regardless of statistical significance. These imbalances should be regarded as sources of bias in the estimation of mean costs and QALYs, and by extension, differences in costs and QALYs between the intervention groups. Marginal prediction effectively holds all else equal, save for the predictor of interest – treatment.

To appropriately incorporate sampling uncertainty, unit cost uncertainty and statistical model uncertainty, we conducted the entire process within a bootstrap framework. The process consisted of three steps. First, a sample was drawn from the original trial sample, stratified by treatment group. Second, unit costs were randomly drawn from their probability distributions and applied to the resource uses in the sample while QALYs were estimated as the area under the curve. Third, and finally, statistical models were assembled and marginal estimation was conducted. Thus, each of the 10,000 iterations of the bootstrap produced a new resample, a new set of unit costs and a new statistical model for marginal estimation and a new set of total NHS cost, total societal cost and total QALYs.

Lifetime extrapolation

Lifetime extrapolation or modelling may be warranted when there is reason to expect differences in long-term costs or QALYs. Factors that affect either include life expectancy, quality of life or events; however, only incremental differences in these factors would affect long-term costs or QALYs.

Life expectancy is not expected to be any different after ORIF or CCC. Quality of life and the rate of its recovery are not expected to differ in the long term. However, if there were clear differences by the end of the trial period, their extensions into the lifetime would be explored.

The AIM trial incorporated a 6-month follow-up. Under normal circumstances, clinical union takes 6 to 12 weeks. The 6-month follow-up of the AIM trial was well suited to capture failure of closed reduction or loss of reduction resulting in reoperation/rehospitalisation. Beyond 6 months, reoperation/rehospitalisation because of excess complications may be possible. However, the probability of this is likely to be very low. In an analysis of 57,183 patients with ORIF, the 6-year reoperation rate was < 1% while complications were < 2%. 27 More specifically in elderly patients, Koval et al. 66 found in 33,704 elderly patients the incidence of complications to be ≤ 2%. Importantly, there was no difference between those treated operatively and those treated non-operatively. These findings are supported by a Cochrane review that compared surgical and conservative management of ankle fractures with follow-up from 20 weeks to, on average, 7 years. 23 A randomised trial in elderly patients with a mean follow-up of 27 months also found negligible incidence beyond 3 months. 29 As published evidence of the long-term complications shows low incidence and no difference between operative and non-operative management, we did not plan to model any long-term differences in complications between ORIF and CCC. Extended follow-up of the AIM trial will provide an opportunity to investigate long-term complication rates following ORIF or CCC.

Excess implant removal because of local irritation is also possible beyond 6 months. Estimates in the literature may be biased where removal is a paid part of a surgeon’s practice. Based on the experience of the trial group, removal for irritation was estimated to be very low, at around 2–3%. We assumed the removal cost to include a theatre time and length of stay similar to that observed for an additional procedure/readmission from the trial data. This was later determined to be 1.3 theatre hours and 5 days in hospital. Thus, the total cost was £2648 per removal. After weighting by a 3% removal rate, this amounted to £79 for the population receiving implants. We attributed this removal during the index year. No impact on QALYs was modelled, as general health domain functioning is independent of whether or not hardware is removed in those experiencing pain/discomfort. 67

Incremental analysis

Using the estimated total NHS cost, total societal cost and total QALYs, we calculated the incremental differences between the CCC and ORIF groups. We presented the incremental cost-effectiveness ratio (ICER) for the NHS and societal perspectives, when appropriate. The mean ICER is the mean incremental cost divided by the mean incremental QALYs.

There are four potential combinations of incremental costs and QALYs:

1. CCC is more effective and more costly.

2. CCC is less effective and less costly.

3. CCC is more effective and less costly.

4. CCC is less effective and more costly.

These four potential combinations of incremental costs and QALYs were visualised by a cost-effectiveness plane, which is a scatterplot of incremental costs on the y-axis and incremental QALYs on the x-axis.

We used the net benefit statistic to estimate the probability that CCC is cost-effective. CCC is cost-effective when the net benefit is greater than zero. The net benefit was the incremental benefit minus the incremental cost. The incremental benefit was the incremental QALYs multiplied by the willingness to pay. There were two incremental costs: (1) the incremental total NHS cost and (2) incremental total societal cost. We estimated two net benefits for the NHS and societal perspectives.

The probability of cost-effectiveness changes as a function of the willingness to pay. Using all 10,000 bootstrap samples, we estimated the overall probability of cost-effectiveness for all willingness to pay over a common range (£0/QALY to £30,000/QALY). The probability of cost-effectiveness was plotted on a cost-effectiveness acceptability curve.

Sensitivity

The bootstrapped model already incorporates sampling uncertainty, unit cost uncertainty and statistical model uncertainty.

We explored structural uncertainty in the statistical models by employing the second best-fitted GLM as well as the generic OLS model (a GLM with Gaussian family and identity link). The entire economic evaluation was repeated for the ITT population.

Pilot patients did not have any utility information at presentation, and at 6 weeks and 6 months after randomisation. They also did not complete health economic questionnaires at 6 weeks or 6 months after randomisation. They therefore had no information valid for an economic evaluation. Patients with no valid information were excluded from the analysis. However, we explored including these patients with completely missing information after imputing their values on the basis of baseline characteristics and primary procedure resource use.

Identification of best-fit GLMs was conducted in Stata version 12 (StataCorp, College Station, TX, USA). All other statistical and economic analyses, as well as data handling, were conducted in R version 3.1.2 (R Foundation for Statistical Computing, Vienna, Austria).

Database and data processing

The return of case report forms (CRFs) to the study office was monitored daily to proactively request any documentation that was expected for arrival and to track any forms that were yet to be returned. On arrival into the study office, each CRF was checked through central monitoring systems to ensure completeness of the data and to monitor protocol compliance. These systems facilitated immediate identification of missing or contradictory data. Sites were contacted in a timely manner to resolve these queries. Data queries were reviewed on a weekly basis and followed through to completion to ensure maximum data capture. Items of missing data and the reason not obtained were logged for reference. The statistician, prior to Trial Steering Committee (TSC) and DMEC meetings, reviewed data completion and follow-up rates.

Data were single-entered into the database by study personnel. For data quality assurance, 10% of baseline, 6-week and 6-month follow-up questionnaires were randomly selected for data checking. Any discrepancies identified when checking were corrected and recorded. The data were held on Warwick Clinical Trial Unit’s SQL Server (Microsoft, Redmond, WA, USA) system that facilitated data validation, range checks and flagging of missing data.

Qualitative study

In order to explore patient experience of their treatment, recovery and what it was like to be in the trial, a purposive sample of 36 patients was interviewed between 6 and 10 weeks post treatment. An estimated sample of up to 40 patients was identified in the study protocol, but data saturation (when interviews stop revealing new data) was achieved with a sample of 36 participants. The sample covered patients from both treatments, two study sites, and a range of ages and sexes. Participants provided informed written consent. The interviews were conversational in style to allow patients to identify their experiences and the issues that concerned them. The research questions were (1) what are participants’ experiences of ankle fracture in the first 6–10 weeks after treatment? and (2) what are participants’ experiences of being in a trial? Further information on methodology and findings for the qualitative study is presented in Chapter 5.

Patient and public involvement

The study design, patient information sheets and consent forms for the main trial and qualitative substudy were discussed with the Oxford Trauma User Group who provided feedback and support for the study. The user group are former users of the Oxford Trauma Service and/or previous participants in clinical studies conducted by the Trauma Research Group based in the Kadoorie Centre for Critical Care Research and Education, all based at the John Radcliffe Hospital, Oxford, UK. A user representative was a member of the TSC and as part of this role made a suggestion for the additional secondary analysis of the primary outcome as an ordered categorical outcome, as outlined in Statistical Methods.

Ethical approval and monitoring

The Oxfordshire Research Ethics Committee A gave approval for this study (09/H0604/129). We complied with the Medical Research Council: Guidelines for Good Clinical Practice in Clinical Trials,68 and the trial was run using the standard operating procedures of Warwick University Clinical Trials Unit and the Oxford Clinical Trials Research Unit.

Reporting

The trial was reported in accordance with the Consolidated Standards of Reporting Trials (CONSORT) statement69 and its extensions relating to non-pharmacological,70 pragmatic71 and equivalence trials. 53 Reporting of the intervention was also based on the template for intervention description and replication (TIDieR) guidance. 72

Summary of changes to the project protocol

The changes to the project protocol are summarised in Table 3.

Participant flow

Figure 7 presents the CONSORT flow diagram summarising the different phases of the trial from the recruitment stages up to the point of final analyses. Table 4 provides details of the study flow. In summary, 620 patients were recruited and randomised. There were 309 participants randomised to ORIF, of whom 302 received their allocated treatment. There were 311 participants randomised to CCC, of whom 277 received their allocated treatment. At baseline, 618 out of 620 (99.7%) participants provided baseline data, 608 out of 620 (98.1%) participants provided data at 6 weeks and 593 out of 620 (95.6%) participants provided data at 6 months. Baseline data for two of the participants were missing as they were randomised in error, that is they were ineligible, and thus were withdrawn. Fifteen out of the 620 (2.4%) participants withdrew from the trial and 10 out of the 620 (1.6%) participants had died. At the 6-month primary end point, 558 out of 620 (90%) participants were analysed for the primary outcome in the per-protocol analysis and 593 out of 620 (96%) participants were analysed for the primary outcome in the ITT analysis. The details of each of the stages in the flow of the trial will now be described in more detail.

FIGURE 7.

The CONSORT diagram for the AIM trial. a, Number of participants (n) reported is cumulative.

Baseline data

The baseline demographic and clinical characteristics of all randomised participants have been summarised in Tables 13 and 14. The randomised participants were, on average, around 70 years old and nearly 74% of them were female. Both groups reported similar OMASs (before injury), SF-12 mental component summary scores, SF-12 physical component summary scores and EQ-5D scores at baseline. The baseline EQ-5D score and EQ-5D visual analogue scale (VAS) score for the ‘day before injury’ and on the ‘day of assessment’ clearly suggests that the participants’ quality of life deteriorated as a result of their ankle fracture. Around 30% of all participants suffered from allergies. Most participants had either never smoked (50.7%) or were ex-smokers (40.1%). The comorbidities reported by the recruited sample are presented in Table 15. The social circumstances of all randomised participants have been summarised in Table 16. The vast majority of patients (96.9%) were admitted from their own home, most of whom were living either alone or with someone and did not use walking aids.

Numbers analysed

All of the results from the primary per-protocol analyses are presented from here onwards using the per-protocol sample. The ITT analyses displayed similar results and thus are not presented in this chapter (apart from the analysis of the primary outcome). The results of the ITT analyses can be found elsewhere (see Appendix 2) where the tables have been assigned the same numbering as those presented in this chapter for ease of reference. The per-protocol analyses of the 6-week data comprised data from 573 participants (n = 298 for ORIF and n = 275 for CCC) and the 6-month analyses comprised data from 558 participants (n = 291 for ORIF and n = 267 for CCC). The ITT analyses included all participants that provided follow-up assessment data. In total, 608 participants were included in the ITT analyses at the 6-week time point (n = 305 for ORIF and n = 303 for CCC) and 593 were included in the analyses at the 6-month time point (n = 298 for ORIF and n = 295 for CCC). In both the per-protocol analyses and the ITT analyses, it was assumed that the participants received the treatment to which they were initially allocated. The quality of the primary and secondary outcomes data for both the per-protocol analyses and ITT analyses have been presented in Tables 17 and 18, respectively.

Per-protocol analyses results

Primary outcome measure (per protocol)

Table 19 presents the unadjusted and adjusted per-protocol analyses for the OMASs at 6 weeks and 6 months. The 95% CIs for the difference between the two treatment groups at both 6 weeks and 6 months lie totally within the equivalence margin of ± 6 points. Therefore, there is evidence to suggest that the two treatments are equivalent. In particular, at 6 months (the primary end point), the mean treatment effect from the adjusted analyses was estimated to be –0.65 (95% CI –3.98 to 2.68) which corresponds to a standardised effect size of –0.04 (95% CI –0.23 to 0.15). Figure 8 clearly illustrates this result in which the unadjusted and adjusted 95% CIs at 6 months can be seen to lie totally within the equivalence margin. The results of the ITT analyses of the primary outcome also provided evidence that the treatments are equivalent, as shown in Table 20 and Figure 9. A plot of the OMASs over time (from pre-injury to 6-month follow-up) has been presented in Figure 10. From this plot it is evident that the ankle functionality of participants in both groups follows a similar pattern over time. The same pattern of recovery was also evident in the ITT sample.

TABLE 19 - Primary outcome (OMAS) at 6 weeks and 6 months, summarised by treatment group (per protocol)

Treatment effect estimate using linear regression model adjusting for age, sex, hospital, baseline OMAS and AO classification.

Note

A negative value implies that the treatment effect is in favour of ORIF.

OMAS	Treatment group		Total	Unadjusted statistics (95% CI)	Adjusted statistics (95% CI)a
	ORIF	CCC
6 weeks
Mean	37.9	38.1	38.0	0.20 (–2.31 to 2.71)	0.26 (–2.20 to 2.72)
n	297	273	570
SD	14.7	15.8	15.2
Median	35	40	35
Minimum	0	0	0
Maximum	95	90	95
Missing	1	2	3
6 months
Mean	66.0	64.5	65.3	–1.57 (–5.19 to 2.05)	–0.65 (–3.98 to 2.68)
n	291	267	558
SD	21.1	22.4	21.8
Median	70	70	70
Minimum	0	5	0
Maximum	100	100	100
Missing	0	0	0

FIGURE 8.

Observed treatment difference plot at 6 months with 95% CIs (per protocol).

TABLE 20 - Primary outcome (OMAS) at 6 weeks and 6 months, summarised by treatment group (ITT)

Treatment effect estimate using linear regression model adjusting for age, sex, hospital, baseline OMAS and AO classification.

Note

A negative value implies that the treatment effect is in favour of ORIF.

OMAS	Treatment group		Total	Unadjusted statistics (95% CI)	Adjusted statistics (95% CI)a
	ORIF	CCC
6 weeks
Mean	38.0	38.0	38.0	0.03 (–2.41 to 2.47)	0.12 (–2.56 to 2.79)
n	304	301	605
SD	14.6	16.0	15.29
Median	35.0	40.0	35.0
Minimum	0	0	0
Maximum	95	90	95
Missing	1	2	3
6 months
Mean	66.0	64.7	65.3	–1.33 (–4.85 to 2.19)	–0.17 (–3.36 to 3.02)
n	298	294	592
SD	21.3	22.4	21.81
Median	70.0	70.0	70.0
Minimum	0	5	0
Maximum	100	100	100
Missing	0	1	1

FIGURE 9.

Observed treatment difference plot at 6 months with 95% CIs (ITT).

FIGURE 10.

Plot of mean scores and 95% CIs for the OMASs over time (per protocol).

Secondary outcome measures (per protocol)

Short Form questionnaire-12 items mental component summary subscale (per protocol)

The analyses of the SF-12 mental component summary subscale scores have been presented in Table 21. There is no evidence of a statistically significant difference between the two treatment groups at either 6 weeks or 6 months. Overall, the improvement from 6 weeks to 6 months is small. The difference between ORIF and CCC, on average, was estimated to be 0.06 (95% CI –1.64 to 1.75) at 6 weeks and –0.21 (95% CI –1.71 to 1.28) at 6 months. Participants in both groups followed a similar pattern of recovery over time, with their mental function at 6 months being similar to what it was pre-injury (Figure 11).

TABLE 21 - Summary of SF-12 and EQ-5D outcomes at the 6-week and 6-month assessments (per protocol)

Note

A negative value implies that the treatment effect is in favour of ORIF.

Outcome	Treatment group		Unadjusted statistics (95% CI); p-value	Adjusted statistics (95% CI); p-value
	ORIF	CCC
SF-12 mental component summary score
6 weeks
Mean	48.3	48.6	0.35 (–1.45 to 2.15); 0.700	0.06 (–1.64 to 1.75); 0.946
n	297	274
SD	11.3	10.6
Median	48.9	49.5
Minimum	16.5	19.2
Maximum	70.2	68.3
Missing	1	1
6 months
Mean	52.1	52.2	0.10 (–1.57 to 1.78); 0.902	–0.21 (–1.71 to 1.28); 0.781
n	291	267
SD	10.3	9.7
Median	55.4	55.7
Minimum	18.9	19.9
Maximum	66.9	68.4
Missing	0	0
SF-12 physical component summary score
6 weeks
Mean	33.3	33.4	0.07 (–1.02 to 1.17); 0.895	0.24 (–0.82 to 1.31); 0.654
n	297	274
SD	6.9	6.5
Median	32.4	32.7
Minimum	18.4	18.3
Maximum	57.1	54.9
Missing	1	1
6 months
Mean	45.6	44.0	–1.56 (–3.28 to 0.16); 0.076	–0.83 (–2.39 to 0.74); 0.299
n	291	267
SD	10.0	10.7
Median	47.0	44.5
Minimum	22.1	15.6
Maximum	66.2	64.3
Missing	0	0
EQ-5D mobility score
6 weeks
Level 1, n (%)	10 (3.4)	5 (1.8)	0.316	–
Level 2, n (%)	245 (82.2)	221 (80.4)
Level 3, n (%)	17 (5.7)	22 (8.0)
Missing, n (%)	26 (8.7)	27 (9.8)
6 months
Level 1, n (%)	134 (46.1)	106 (39.7)	0.333	–
Level 2, n (%)	128 (44.0)	132 (49.4)
Level 3, n (%)	3 (1.0)	3 (1.1)
Missing, n (%)	26 (8.9)	26 (9.8)
EQ-5D self-care score
6 weeks
Level 1, n (%)	164 (55.1)	136 (49.5)	0.026	–
Level 2, n (%)	107 (35.9)	104 (37.8)
Level 3, n (%)	1 (0.3)	8 (2.9)
Missing, n (%)	26 (8.7)	27 (9.8)
6 months
Level 1, n (%)	240 (82.5)	221 (82.8)	0.530	–
Level 2, n (%)	21 (7.2)	19 (7.1)
Level 3, n (%)	4 (1.4)	1 (0.4)
Missing, n (%)	26 (8.9)	26 (9.7)
EQ-5D usual activities score
6 weeks
Level 1, n (%)	10 (3.3)	10 (3.6)	0.037	–
Level 2, n (%)	150 (50.3)	109 (39.6)
Level 3, n (%)	112 (37.6)	129 (46.9)
Missing, n (%)	26 (8.8)	27 (9.9)
6 months
Level 1, n (%)	159 (54.6)	125 (46.8)	0.179	–
Level 2, n (%)	96 (33.0)	104 (39.0)
Level 3, n (%)	10 (3.4)	12 (4.5)
Missing, n (%)	26 (9.0)	26 (9.7)
EQ-5D pain/discomfort score
6 weeks
Level 1, n (%)	130 (43.6)	124 (45.1)	0.421	–
Level 2, n (%)	136 (45.6)	122 (44.4)
Level 3, n (%)	6 (2.0)	2 (0.7)
Missing, n (%)	26 (8.8)	27 (9.8)
6 months
Level 1, n (%)	118 (40.6)	105 (39.3)	0.589	–
Level 2, n (%)	139 (47.8)	132 (49.5)
Level 3, n (%)	8 (2.7)	4 (1.5)
Missing, n (%)	26 (8.9)	26 (9.7)
EQ-5D anxiety/depression score
6 weeks
Level 1, n (%)	183 (61.4)	162 (58.9)	0.084	–
Level 2, n (%)	84 (28.2)	86 (31.3)
Level 3, n (%)	5 (1.7)	0
Missing, n (%)	26 (8.7)	27 (9.8)
6 months
Level 1, n (%)	199 (68.4)	189 (70.8)	0.706	–
Level 2, n (%)	59 (20.3)	48 (18.0)
Level 3, n (%)	6 (2.1)	4 (1.5)
Missing, n (%)	27 (9.2)	26 (9.7)
EQ-5D score
6 weeks
Mean	0.52	0.48	–0.04 (–0.08 to 0.002); 0.065	–0.04 (–0.08 to 0.007); 0.100
n	272	248
SD	0.3	0.2
Median	0.59	0.49
Minimum	–0.48	–0.06
Maximum	1	1
Missing	26	27
6 months
Mean	0.76	0.76	–0.006 (–0.05 to 0.03); 0.754	–0.004 (–0.04 to 0.04); 0.861
n	264	241
SD	0.2	0.2
Median	0.80	0.78
Minimum	–0.36	–0.07
Maximum	1	1
Missing	27	26
EQ-5D VAS score
6 weeks
Mean	72.8	72.1	–0.67 (–3.70 to 2.36); 0.663	–0.37 (–3.18 to 2.44); 0.796
n	272	247
SD	17.1	18.1
Median	75	75
Minimum	0	10
Maximum	100	100
Missing	26	28
6 months
Mean	77.5	77.3	–0.16 (–3.35 to 3.03); 0.923	–0.09 (–3.19 to 3.02); 0.956
n	265	241
SD	18.5	18.0
Median	81	80
Minimum	0	1
Maximum	100	100
Missing	26	26

FIGURE 11.

Plot of mean scores and 95% CIs for the SF-12 mental component summary score over time (per protocol).

Short Form questionnaire-12 items physical component summary subscale (per protocol)

The analyses of the SF-12 physical subscale have been presented in Table 21. There is no evidence of a statistically significant difference between the two arms at either 6 weeks or 6 months. Overall, the SF-12 physical score improved from a mean score of 33.4 at 6 weeks to a mean score of around 45.0 at 6 months. The difference between ORIF and CCC on average was estimated to be 0.24 (95% CI –0.82 to 1.31) at 6 weeks and –0.83 (95% CI –2.39 to 0.74) at 6 months. Participants in both groups followed a similar pattern of recovery over time, with their physical function at 6 months being slightly less than what it was pre-injury (Figure 12).

FIGURE 12.

Plot of mean scores and 95% CIs for the SF-12 physical component summary score over time (per protocol).

European Quality of Life 5-Dimensions score (per protocol)

The analyses of the EQ-5D score have been presented in Table 21. There is no evidence of a statistically significant difference between the two arms at both 6 weeks and 6 months. Overall, the EQ-5D score improved from a mean score of around 0.51 at 6 weeks to a mean score of 0.76 at 6 months. The difference between ORIF and CCC on average was estimated to be –0.04 (95% CI –0.08 to 0.007) at 6 weeks and –0.004 (95% CI –0.04 to 0.04) at 6 months. Participants in both groups followed a similar pattern of recovery over time, with their quality of life at 6 months being slightly less than what it was the day before their injury (Figure 13).

FIGURE 13.

Plot of mean scores and 95% CIs for the EQ-5D over time (per protocol).

Ancillary analyses (per protocol)

Adverse events (per protocol)

No serious adverse events were reporting during the trial. A breakdown of the adverse event complications reported is shown in Table 38. The number of participants who experienced an infection and/or wound problem for the surgery group was 29/298 (10%) compared with 4/275 (1%) for CCC. The participants with wound breakdown or infection in the CCC arm initially received CCC as their primary theatre procedure but later converted to ORIF either as an additional procedure in theatre or during a readmission. The number of plaster sores and pain from casts was similar between groups, but there were more plaster saw lacerations in the CCC group [5/275 (1.8%) vs. 1/298 (0.3%)]. Less common but serious complications, such as deep-vein thrombosis (DVT), pulmonary embolism (PE) and wound infections, occurred in both treatment groups but were infrequent overall.

Additional procedures in theatre (per protocol)

A detailed breakdown of the additional procedures in theatre has been detailed in Table 39. The number of participants requiring one or more additional operating theatre procedures, for surgery-related complications, was 17 out of 298 (5.7%) in the ORIF group and 3 out of 275 (1.1%) in the CCC group.

Summary of key findings

• The results from the analyses of the primary end point allow us to conclude that the ORIF and CCC treatments are equivalent for the treatment of unstable ankle fracture in patients over 60 years.

• The functionality of the injury ankle in terms of the range of movement was similar in both groups at 6 months.

• The mental functioning, physical functioning and quality of life of participants in both groups follow a similar pattern of recovery up to 6 months, with their scores at 6 months being similar to their pre-injury scores.

• Malunion and non-union of the ankle were more likely after CCC than ORIF.

• Participants receiving CCC were more likely to be readmitted post hospital discharge; one in five had a later loss of fracture reduction resulting in conversion to ORIF or remanipulation and CCC applied in theatre.

• Complications directly relating to ORIF, wound problems or infections, occurred in around 1 in 10 participants. Within 6 months about 1 in 20 participants who initially received ORIF required an additional surgical procedure.

Introduction

An evaluation of cost-effectiveness of CCC compared with ORIF was conducted as part of the AIM trial. The economic evaluation presented in this chapter was conducted for both the NHS and societal perspective. Consistent with the evaluation of cost-effectiveness in Chapter 3, the primary analysis was per protocol.

Results

Participant flow

Participant flow is shown in Figure 14. The per-protocol population included 573 participants. The number of participants for whom information was completely missing and who thus were ineligible for inclusion in the economic evaluation, was 53 out of 573 (9%). Therefore, there were 520 participants with valid information for an economic evaluation. This constituted the primary analysis population.

FIGURE 14.

Flow diagram of per-protocol and ITT population: withdrawals, with exclusions because of insufficient information and final analysis populations.

The ITT population included 618 participants. The number of participants for whom information was completely missing and who thus were ineligible for inclusion in the economic evaluation was 67 out of 618 (10%). Therefore, there were 551 participants with valid information for an economic evaluation. This constituted the population used for the ITT analysis.

A full trial per-protocol or ITT population included those with completely missing information (after multiple imputation of their entire history). This corresponded to 573 and 618 participants for the per-protocol and ITT populations, respectively. Sensitivity analyses were conducted using these populations (see Appendix 3).

Per-protocol analysis

Health outcomes (per protocol)

The raw, unadjusted utilities at presentation, 6 weeks and 6 months after randomisation are presented in Table 41. At presentation, the utility in the CCC group trended higher than that in the ORIF group (0.07 vs. 0.04), although this difference was clinically small and not statistically significant (p = 0.11). Compared with the ORIF group, utility in the CCC group was comparable at 6 weeks (0.36 vs. 0.37; p = 0.15) and 6 months (0.49 vs. 0.49; p = 0.56). Total QALYs over the trial period were also comparable between the CCC and ORIF groups (0.30 vs. 0.30; p = 0.96). Figure 15 shows the mean utility scores; the QALYs are indicated by the area under the curve. QALYs were estimated from presentation time to 6 months.

TABLE 41 - Per-protocol raw, unadjusted utility and QALYs over the 6-month follow-up. 95% CIs obtained using the bootstrap method

Health outcomes	Treatment group		Difference, mean (95% CI)	p-value
	ORIF, mean (95% CI)	CCC, mean (95% CI)
Utility (baseline)	0.04 (0.02 to 0.07)	0.07 (0.04 to 0.10)	0.03 (–0.01 to 0.07)	0.11
Utility (6 weeks)	0.37 (0.35 to 0.39)	0.36 (0.34 to 0.38)	–0.01 (–0.04 to 0.01)	0.15
Utility (6 months)	0.49 (0.47 to 0.51)	0.49 (0.47 to 0.50)	0.00 (–0.03 to 0.03)	0.56
QALYs	0.30 (0.28 to 0.31)	0.30 (0.29 to 0.31)	0.00 (–0.01 to 0.02)	0.96

FIGURE 15.

Graph showing per-protocol utility (quality of life) over time (length of life). Area is the QALYs.

Costs (per protocol)

The mean raw, unadjusted costs of the resource uses using mean unit cost parameters for each group are presented in Table 42. Compared with the ORIF group, the CCC group had a large initial cost savings (–£1144), driven by savings in theatre time use (–£880), as well as a modest savings in care homes (–£137), that were eroded with costs in increased readmissions (+£448) and health services (+£435). Mean cost differences were negligible for additional procedure (+£8), casts (+£45) and medications (+£2).

TABLE 42 - Per-protocol raw, unadjusted costs using mean unit costs over the 6-month trial period. 95% CIs obtained using the bootstrap method

LOS, length of stay.

Resource	Treatment group		Difference, mean, £ (95% CI)
	ORIF, mean, £ (95% CI)	CCC, mean, £ (95% CI)
Index procedure
Theatre time	1602 (1547 to 1657)	722 (701 to 745)	–880 (–938 to –820)
Implant	79 (72 to 87)	0 (0 to 0)	–79 (–87 to –72)
LOS	2667 (2134 to 3541)	2482 (2165 to 2827)	–185 (–1120 to 493)
Total	4348 (3807 to 5225)	3204 (2886 to 3552)	–1144 (–2081 to –460)
Additional procedure
Theatre time	86 (43 to 136)	95 (48 to 147)	8 (–60 to 77)
Implant	4 (2 to 6)	4 (2 to 7)	0 (–3 to 4)
Total	90 (45 to 142)	99 (50 to 154)	8 (–63 to 81)
Readmission
Theatre time	37 (15 to 64)	278 (197 to 365)	241 (155 to 332)
Implant	0 (0 to 1)	14 (9 to 20)	14 (9 to 20)
LOS	122 (46 to 218)	315 (170 to 518)	193 (15 to 413)
Total	159 (70 to 272)	607 (408 to 851)	448 (219 to 713)
Casts
Number	2 (1 to 3)	47 (43 to 50)	45 (41 to 49)
Care homes
Community hospital (days)	547 (253 to 902)	455 (270 to 665)	–92 (–489 to 270)
Intermediate care (days)	49 (2 to 129)	88 (19 to 184)	39 (–69 to 151)
Nursing home (NHS) (days)	96 (25 to 190)	38 (1 to 87)	–57 (–160 to 27)
Nursing home (private) (days)	60 (0 to 141)	33 (0 to 82)	–27 (–118 to 51)
Total	752 (401 to 1165)	615 (393 to 862)	–137 (–606 to 289)
Health services	1,352 (1,075 to 1,670)	1787 (1460 to 2164)	435 (–27 to 904)
Health services (private)	94 (52 to 150)	68 (44 to 95)	–25 (–87 to 27)
Medications	5 (4 to 6)	7 (4 to 12)	2 (–1 to 7)
Medications (private)	4 (3 to 6)	4 (3 to 5)	–1 (–3 to 1)
Work-days	1058 (765 to 1381)	1091 (695 to 1558)	33 (–482 to 576)
Total societal cost	7864 (6926 to 9027)	7528 (6662 to 8453)	–335 (–1767 to 1005)
Total NHS cost	6648 (5744 to 7777)	6,332 (5599 to 7127)	–316 (–1661 to 919)

Societal cost differences between the CCC and ORIF groups were negligible across private nursing homes (–£27), private health services (–£25), out-of-pocket medications (–£1) and work days (+£33).

Over the 6-month trial period, the CCC group had a mean cost savings to the NHS of –£316. The mean cost saving to society was –£335. Figure 16 presents the cost burden for the CCC and ORIF groups for the cost categories. The cost burdens, from largest to least, were the primary procedure, health services/medications, care homes, readmission, additional procedure and then casts. The figure represents the total cost burden; 360° represents around £6600. The cost savings in the CCC group is represented by the white space, where a full circle fails to fill.

FIGURE 16.

Pie chart of per-protocol NHS costs for (a) ORIF and (b) CCC, grouped by primary procedure, health services/medications, care homes, readmission, additional procedure and casts. CCC displays cost savings represented by the white space.

Cost-effectiveness: 6 months (per protocol)

NHS perspective

After resampling the trial population and unit costs, refitting the GLMs and using marginal estimation, mean total NHS cost and total QALYs for the CCC and ORIF groups were generated. The results of the cost-effectiveness outcomes are presented in Table 46. The mean total QALYs for CCC and ORIF were 0.30 and 0.29, respectively, while the mean total NHS costs were £6050 and £6694, respectively.

TABLE 46 - Cost-effectiveness results: 6 months (per-protocol)

Outcome	Treatment group		Difference, mean (95% CI)
	ORIF, mean (95% CI)	CCC, mean (95% CI)
NHS costs (£)	6694 (5285 to 8465)	6050 (4711 to 7744)	–644 (–1390 to 76)
Societal costs (£)	8003 (6483 to 9906)	7320 (5815 to 9167)	–683 (–1851 to 536)
QALYs	0.2928 (0.2714 to 0.3116)	0.3034 (0.2853 to 0.3209)	0.0106 (–0.0158 to 0.0381)

The incremental differences favoured the CCC group. The CCC group had higher mean total QALYs (+0.01) and lower mean total NHS cost (–£644). At the mean for an ICER of £25,000/QALY, as CCC was both more effective and less costly, it dominated ORIF.

To assess the variability of the estimates, we plotted the incremental costs and incremental QALYs on the cost-effectiveness plane (Figure 17). The cost-effectiveness plane summarises all 10,000 bootstrap estimates. The cost-effectiveness plane shows that the large majority of incremental costs are negative; the incremental QALYs, however, show more uncertainty with a modest proportion being negative.

FIGURE 17.

Per-protocol cost-effectiveness plane showing 6-month incremental total NHS costs along y-space and incremental QALYs along x-space for 10,000 bootstrap replicates. Origin is ORIF.

The variability across the range of willingness-to-pay thresholds was summarised with a cost-effectiveness acceptability curve (Figure 18). Over a common willingness to pay, CCC displayed a high probability of cost-effectiveness (> 95%).

FIGURE 18.

Per-protocol 6-month cost-effectiveness acceptability curve displaying the probability of cost-effectiveness (y-axis) as a function of the willingness-to-pay threshold (x-axis) under the NHS perspective. The dotted line presents the asymptote where the willingness to pay is infinite.

Societal perspective

The mean total QALYs remained the same for both the NHS and societal perspectives. Costs, however, differed. After resampling the trial population and unit costs; refitting the GLMs and using marginal estimation, mean total societal cost for the CCC and ORIF groups was £7320 and £8003, respectively (see Table 46).

The incremental differences favoured the CCC group. The CCC group had a lower mean total societal cost (–£683) and unchanged higher mean total QALYs (+0.01). At the mean incremental cost and QALYs for an ICER of £25,000/QALY, CCC dominated ORIF as it was both more effective and less costly.

The cost-effectiveness plane displayed modest variability in incremental costs (Figure 19). However, the large majority of incremental costs were negative. The incremental QALYs also displayed a modest proportion of negative estimates.

FIGURE 19.

Per-protocol cost-effectiveness plane showing incremental total societal costs along y-space and incremental QALYs along x-space for 10,000 bootstrap replicates. Origin is ORIF.

Over a common willingness to pay, CCC displayed a high probability of cost-effectiveness (around 85%) (Figure 20).

FIGURE 20.

Per-protocol cost-effectiveness acceptability curve displaying the probability of cost-effectiveness (y-axis) as a function of the willingness-to-pay threshold (x-axis) under the societal perspective. The dotted line presents the asymptote where the willingness to pay is infinite.

Discussion

An economic evaluation of CCC compared with ORIF for ankle injury management was conducted. Over the trial period, CCC showed evidence of a modest mean cost savings (around £400–700) to both the NHS and society. The cost savings were driven largely by savings in index procedure theatre time, but were eroded over time by increased readmission and health services costs. Incremental QALYs following CCC were no different from those following ORIF, with evidence leaning towards a slight advantage with CCC. Over common willingness-to-pay thresholds, the probability that CCC was cost-effective was very high (> 95% for NHS and 85% for societal perspective). Results were robust to the use of per-protocol or ITT populations, the use of standard OLS models instead of best-fitted GLM models and the inclusion of completely missing cases via multiple imputation.

The AIM trial was a randomised controlled trial. Despite randomisation, there was evidence of differences in utility prior to injury. The CCC group had a lower utility prior to injury than the ORIF group. This suggests a lower baseline quality of life. Other differences such as utility at presentation and baseline walking distance, albeit not statistically significant, represented imbalances that effect the estimation of mean total and incremental costs and QALYs.

The bootstrapped, adjusted results showed greater mean cost savings than the raw, unadjusted trial results. Without adjusting for imbalances in baseline characteristics, whether or not statistically significant, the estimated total and incremental costs can be biased. The GLM model results show that small differences in baseline characteristics can contribute to important differences in total cost. This was particularly the case for baseline walking distance and walking aids statuses, among others.

The NHS and societal perspectives showed comparable incremental differences in total cost at the mean. Uncertainty was slightly greater using the societal perspective. This led to slightly greater decision uncertainty as visualised by the cost-effectiveness acceptability curve (see Figures 18 and 20). However, incremental costs remained largely negative and the probability of cost-effectiveness for CCC remained very high.

Incremental total costs and QALYs were robust to whether the per-protocol or ITT population was used. Results with the ITT population showed cost savings to the NHS and society at similar magnitudes to the per-protocol population. The probability of cost-effectiveness of CCC was also similarly very high. The best-fit models also showed comparable results to the OLS models for the ITT population, as they did for the per-protocol population.

Incremental total costs and QALYs were also robust to whether completely missing cases were excluded or included using multiple imputation. CCC continued to show a high probability of cost-effectiveness over the common willingness to pay.

On average, CCC was followed by more rehospitalisation, because of loss of fracture reduction and conversion to ORIF, as previously shown in the clinical results. Cost increases following CCC also occurred outside hospitalisations. A substantial proportion of cost increases related to increased health services use in the follow-up period. CCC appeared to require additional health services following discharge.

We did not find justification to extrapolate trial outcomes or incorporate modelling of new outcomes to a lifetime. Utility was the same at each time point. This suggested that quality of life, its rate of recovery and QALYs were the same. There was no clinical reasoning for a difference in life expectancy between CCC and ORIF management. Implant removal because of local irritation may be an incremental cost to ORIF; however, it did not affect conclusions of incremental costs or QALYs. Given the evidence on functional assessments, quality of life, life expectancy and implant removal, there was not enough justification to model long-term differences.

Radiological evidence did identify a possible signal of increased malunion associated with CCC. Malunion has been suggested as a factor in the development of long-term osteoarthritis. 74 However, there is uncertainty regarding the rates and severity of post-traumatic osteoarthritis in relation to malunion after fracture because of a lack of clinical data. A recent review highlights the poor quality of the available evidence, with limited basic scientific knowledge and a dependence on retrospective case series. 75 We are conducting extended follow-up of the AIM trial cohort to address some of these uncertainties. When completed, this additional phase of follow-up will allow us to conduct modelling for any explicit long-term differences between ORIF and CCC regarding quality-of-life recovery, functional recovery, development of osteoarthritis or whether increased health service use with CCC is sustained or increases.

There are two other relevant cost-effectiveness studies. 76,77 Slobogean et al. 76 conducted a cost-effectiveness analysis of ORIF versus non-operative treatment for unstable ankle fractures in a Canadian setting. There are several differences in the design. Although Slobogean et al. 76 also used a trial-based evaluation, the trial randomised far fewer patients (81 vs. 620). In addition, the non-operative intervention did not include CCC. Slobogean et al. also considered a narrower perspective, only the procedure and rehospitalisations for reoperation or, in the case of the non-operative arm, secondary ORIF. In contrast, the AIM trial also included care home stay, health services, medication and productivity. Owing to their exclusion of other important resource issues, Slobogean et al. found a much larger short-term (index year) cost savings with non-operative treatment [CAD4512 (2014, GBP2718)]. This is still several times larger than just the hospitalisation costs of the present analysis. This may be because of the underlying differences in unit resource costs between Canada and the UK, as well as the use of summary unit cost estimates of different hospitalisations in contrast to the present study, which microcosted actual resource use.

Slobogean et al. 76 found mean incremental QALYs to be higher with ORIF (+0.02). They did not report uncertainty on incremental estimates or cost-effectiveness planes. However, the SDs of the total QALYs suggests imprecision nearly six times greater than the present study. Thus, incremental QALYs of +0.02 were likely highly imprecise across negative and positive space. In addition, QALYs were not similarly adjusted to baseline utility, which is important when estimating differences for cost-effectiveness (see Chapter 2). A simple plot using their Short-Form questionnaire 36-item physical component summary scores would show ORIF with consistently higher scores from time at presentation all the way through 1 year, compared with non-operative treatment. 78 This is, however, just one domain of health-related quality of life. Utility at each time point was not reported and thus it is not possible to explain the nature of the (imprecise) QALY difference in the Slobogean et al. analysis.

Slobogean et al. 76 concluded that ORIF was cost-effective in the long term (lifetime). The long-term model included only osteoarthritis and the assumed reconstruction following this event. ORIF became cost-effective because non-operative treatment gave way to osteoarthritis on the basis of higher malunion. However, osteoarthritis was modelled expressly on malunions not on ankles in which joint congruence had been achieved. In this way, it was assumed osteoarthritis could only follow malunion. However, normal anatomical alignment does not preclude osteoarthritis, neither does malunion mandate arthritis. The evidence does not support a non-significant absolute risk difference on the order of 50% between malunited and congruent ankles (50% vs. 0%), which Slobogean used. Moreover, the study which was used to derive this estimate was not a comparison of malunited and congruent ankles; it used a small sample of malunited ankles that underwent surgical reconstruction. 79 Slobogean also assumed a stable QALY benefit over the lifetime. The highly imprecise difference in QALYs at the end of the index year was simply assumed the same for all ensuing years in those without arthritic development. It is not clear that this is the most appropriate base case, given the difference was very imprecise and reasoning to justify how differences are sustained was lacking.

Michelson et al. 77 conducted a model-based evaluation of ORIF and traditional casting separately for unstable and stable ankle fractures in a US setting. There were a number of limitations. The only outcomes were infection following ORIF and osteoarthritis following either ORIF or casting; all removals, reoperations and rehospitalisations were not considered. In addition, as infection was just 1%, the driving force in the model was solely the development of arthritis. However, its cost was excluded from the analysis as well as everything outside of the index ORIF/casting hospitalisation cost. Michelson et al. estimated a 25% higher absolute risk of development of arthritis following casting than with ORIF (30% vs. 5%). However, the reasoning for this very large difference was not provided and neither was the source for this estimate. This contrasts to the evidence provided by Koval et al. 66 in 33,704 elderly patients with ankle fracture; the incidence of arthroplasty or arthrodesis (procedures assumed to be performed for osteoarthritis) was less than 1% for both operative and non-operative treatment. The Michelson et al. analysis assumed that quality of life was different in the presence of arthritis (and infection, although that branch’s effect is negligible to begin with) but otherwise equivalent between ORIF and casting. However, it is not clear the utilities employed were valid as they were not on the (0,1) range; no source was provided. The large difference in arthritis gave way to a large difference in utility and quality of life. These were applied at baseline and extended over the lifetime, this led to 0.96 greater QALYs attributed to ORIF for a 70-year-old with unstable fracture. Incremental costs for unstable fractures were not presented. Decision uncertainty was likewise not presented for unstable fractures. The Michelson et al. analysis presents evidence of low comparability because of limitations in its design and evidence base.

Summary of key findings

• The primary economic analysis used the per-protocol population, excluding cases without any valid information.

• There were no differences in QALYs between CCC and ORIF. However, management with CCC led to a mean cost savings greater than £600 under both NHS and societal perspectives (total £1327). The probability that CCC is cost-effective was very high (NHS > 95% and societal 85%) at common willingness-to-pay thresholds.

• Results were robust to the use of the ITT population, OLS models in replace of best-fit GLM models and the inclusion of completely missing cases via imputation.

Introduction

A qualitative study was embedded into the AIM trial to investigate patients’ experiences of living with a fractured ankle and experiences of being in a trial. This chapter outlines the background of this element of the trial, describes the approach and methods used before presenting and discussing the findings.

Background

Researching patient experience is crucial to understanding how patients make sense of injury and recovery and how interventions impact on their daily lives. As a form of evidence, experience has utility when translating interventions into practice. 80 For example, McInnes et al. ,81 through a qualitative meta-synthesis of older people’s experience of falls interventions, demonstrated that participants had complex and varied stories with many factors influencing acceptance of risk status and preference for intervention along with a strong need to maintain control and fit interventions into everyday life. Experience evidence is also valued as an integral part of NICE guidance. Staniszewska et al. 82 from development of the Warwick Patient Experience Framework have generated quality standards around knowing the patient, tailoring care, continuity of care, participation and essential care that require consideration alongside other sources of evidence. Interventions for orthopaedic trauma are increasingly incorporating patient experience within studies of clinical effectiveness and Gooberman-Hill and Fox83 suggest this will benefit both trial design and the suitability of interventions from provider and service user perspectives.

Understanding how participants make sense of trials within their acute hospital care is important evidence for the design of future clinical trials in trauma. A study exploring participant experiences from a range of trials suggested that they take part in trials both for altruistic reasons and because of perceptions of personal benefit; however, personal benefit may take precedence. 84 Acceptance and understanding of the principles underlying trials in a clinical setting can be problematic, leading to therapeutic misconceptions. 84,85 Robinson et al. 86 identify the difficulty lay participants have viewing randomisation or equipoise as acceptable in a clinical setting in which they would prefer an interaction based on what is best for them. Other trial participants experience a sense of indifference to the interventions, particularly if they were satisfied with their treatment, and did not perceive any active harm as a consequence of the intervention. 85 Rejection of randomisation and uncertainty over outcome were reasons for declining to take part in a trial,85 as were perceptions of burden and concern over side effects of interventions. 84 How participants are informed about trials is therefore crucial. Based on their trust in clinicians, participants may pick up cues that subtly coerce them into consenting to take part in trials. 85 In this study, participants’ experience in relation to consenting to the trial and being randomised to CCC and ORIF was explored.

Current research evidence on patients’ experience of traumatic orthopaedic injury presents a broad range of emotional, physical, social and financial impacts on participants’ lives over prolonged periods of time. The experience of being hospitalised as a result of a traumatic orthopaedic injury is emotionally and physically demanding, where patients cope with feelings generated from the incident, the impact on their bodies and live with the uncertainty of full recovery. 87 Patients struggle to feel hopeful, often oscillating between hope and hopelessness as they focus on the immediate future aiming to progress their recovery and return home. 88 Evidence from patients who have experienced a broad range of injuries, the majority of whom had been hospitalised, reveals a range of experience that includes omissions in care, a struggle to cope at home with little social support, continuing pain, low emotional states, a general loss of confidence and a more cautious approach to life; physiotherapy is seen as beneficial for those who received it. 89 Patients interviewed up to 2 months after lower limb injury struggled with pain, fear and anxiety with a feeling of loss of control over their lives but had a strong desire to return to normal. 90 Trickett et al. 91 demonstrated that patients with open tibial fracture interviewed at least 15 months post injury suffered from pain, stiffness and discomfort, had not regained their normal mobility, felt fearful of many aspects of their recovery including falling, were concerned about their appearance and were anxious about their employment and finances as a result of being injured. Forsberg et al. 92 included ankle fracture in their sample of participants with surgery for lower limb fracture interviewed 1 month to 1 year after surgery. The participants identified a range of strong emotions such as anger, frustration and remorse in relation to the incident; anxiety, vulnerability and pain through the surgical period; and helplessness, fear of falling, and insecurity with use of equipment once discharged from hospital.

McPhail et al. 12 focused specifically on adults under 60 years of age from 6 to over 24 months post ankle fracture in order to develop an outcome measure. The findings suggest that participants had or were suffering from a range of impacts: reduced mobility, increased anxiety, a reduction in participation in activities and social life, disruption and sometimes change of work life, loss of earnings, weight gain and loss of ability to wear preferred shoes and some need for medication. Overall, patients experienced traumatic orthopaedic injury as a struggle to return to normal life. Evidence pertaining directly to ankle fracture is limited and there is a gap in relation to the older participant and different forms of treatment. The quality of the studies regarding description of their methodology is often limited; the studies cover a range of time frames since injury and do not take into account change in perceptions over time. This study aimed to explore older participants’ experience of ankle fracture and the acceptability of two interventions, CCC and ORIF.

Methodology

The lack of understanding around patients’ experiences of living with a fractured ankle and treatment led to the research questions.

1. What are participants’ experiences of ankle fracture in the first 6–10 weeks after treatment?

2. What are participants’ experiences of being in a clinical trial?

In order to answer these two questions a naturalistic approach drawing on the principles of Heideggerian phenomenology was utilised. Phenomenology is commonly used in health sciences to facilitate exploration of how participants come to know and understand the world through their embodied experience of what it is like to be in the world. It focuses on gaining ‘a deeper understanding of the nature of meaning in our everyday experiences’ (p. 9). 93 Todres and Wheeler94 suggest it grounds research in the reality of human experience harnessing the reflective skills of the researcher. Key tenets are (1) being in the world, (2) forestructures, (3) temporality and (4) space.

1. Being in the world is conveyed through the notion of dasein (being or presence) in which subject and object are inseparable and human beings exist within the framework of the world. Being ‘is to be shown as it is proximally and for the most part – in its average everydayness’ (pp. 16–17). 95 Research focuses on what it is like in the life-world of the other; or the ‘understandings, feelings and perceived relationships i.e. all the everyday phenomena of common experience’ (p. 3). 94

2. Forestructures reflect the notion that knowledge and meaning is founded on the social, cultural and historical context of the person. The researcher through description and interpretation uncovers what is already in existence and thus understanding leads to meaning. ‘Our investigation itself will show that the meaning of phenomenological description lies in interpretation’ (p. 61). 95 Meaning is further enhanced by a cycle of understanding, from the part to the whole, often called the hermeneutic circle by researchers. 96

3. For Heidegger, temporality is an integral part of being human as the past, present and future are part of the life world ‘time must be brought to light – and genuinely conceived – as the horizon for all understanding of being and for any way of interpreting it’ (p. 39). 95

4. How people are situated in relation to space is also important; this may be spatially but also what is brought to the fore in relation to ‘concern for’ or what merges into the background. This involves listening to what participants say to identify what is important to them in relation to the phenomena. 96 Madjar and Walton97 refer to the listening gaze as a phenomenological way of being: openness to what is unknown or taken for granted in order to develop a better understanding of the other.

Findings

The participants’ experience was conveyed through a core theme of struggling to live with a fractured ankle. This was further divided into four themes and nine categories:

1. Theme 1: suffering – categories (1) being vulnerable and (2) renegotiating roles and relationships.

2. Theme 2: getting on with daily life – categories (1) finding ways of doing things and (2) finding ways of keeping busy.

3. Theme 3: struggling to move – categories (1) living with hopping and (2) moving forward.

4. Theme 4: treatment and being in a trial – categories (1) living with a CCC, (2) living with metalwork and (3) decision-making.

Discussion

The findings reflect that the participants experienced the themes of suffering, getting on with daily life, struggling to move and treatment and being in the trial. The theme of suffering reflected the emotional work participants undertook in relation to feeling vulnerable and being old and to emotional containment as a way of enduring but also maintaining relationships for the future. The theme of getting on with daily life demonstrated how participants reframed their lives by being proactive, drawing on their own and others resources to recreate meaningful lives within existing limitations. The theme of struggling to move encompassed new ways of being, as participants worked hard to move their fragile bodies around a confined life space and moved forward with a heightened sense of emotional fragility. The theme of treatment and being in the trial identified the different experiences of the two treatments; how participants actively managed their symptoms and made sense of being in a trial.

Conclusions

The findings of this study suggest that participants require professional support for their new way of being to make it easier for them to endure this period of recovery. Supportive activities that focus on the experience of injury in relation to being old, the vulnerable and fragile body, changing relationships, skills required for mobility and support for active management of living with the prescribed treatment are required. Supportive activities would require research evidence but could focus on co-ordinated packages of care, accessible learning materials, specialist-based emotion-focused intervention, consistent allocation of helpful equipment and educational materials around lived experience for patients and staff. Further research is required to explore how this experience, in relation to biographical disruption, is processed regarding healthy ageing and future health-care experiences. A longitudinal study would provide valuable insights into the impact of injury over time and implications for rehabilitation. This study has clearly identified the similarities and differences between the lived experiences of both treatments; both of which were considered acceptable. Further exploration is required to understand how patients make decisions about preferred treatments and process the consequences of these decisions. The experience/stocks of knowledge participants draw on, the way they make sense of risks, the way they process changes in treatment because of a lack of bone healing and how the treatment choice impacts on future treatment decisions require further exploration.

Aim and overview of study findings

The decision to treat an ankle fracture by surgery in older adults, compared with younger adults, is complicated by a higher prevalence of comorbidities, increasing the risk of infection, surgical wound problems and inadequate fixation as a result of poor bone quality. Existing trial data are of poor quality but overall indicate that traditional casting methods may lead to a higher risk of malunion because of a loss of fracture reduction. Traditional surgical teaching equates malunion (post-traumatic joint deformity) with poor patient outcome. Therefore, surgery is usually the preferred recommended treatment, with the aim of optimising alignment and congruence (best fit) of the ankle joint, assuming this will translate into better ankle function for patients. In the AIM trial, we aimed to estimate the clinical effectiveness and cost-effectiveness of a new casting intervention, CCC, compared with ORIF surgery.

Our primary objective was to estimate if there was equivalence in patient-reported ankle function outcome between CCC and ORIF, as measured by the OMAS, at 6 months. A qualitative study of participants’ experiences of the intervention and of being in a randomised controlled trial was also embedded into the trial in order to build the context and frame of reference for interpretation of the study findings by clinicians and patients.

This study was a definitive large-scale equivalence randomised controlled trial; assessors were blinded and it was undertaken in a pragmatic way across 24 hospital orthopaedic trauma departments in the UK. The trial has demonstrated ankle function outcomes were equivalent between CCC and ORIF at 6 weeks and at the primary end point of 6 months. There were also no differences between treatments in the secondary outcomes of quality of life (mental and physical), range of ankle motion, ankle pain, mobility and patient satisfaction.

The most common treatment-related issue in the CCC arm was a loss of closed reduction. As a result about 1 in 5 patients who had CCC later returned to theatre to receive ORIF. A later conversion to ORIF was an expected and allowable event in the study protocol. Fewer participants in the ORIF group, about 1 in 20, required additional surgery within 6 months; these surgeries were usually for surgical wound- or implant-related complications, but some of which were clinically serious. At 6 months, CCC was associated with higher rates of radiological malunion and non-union.

The experiences of the treatments were similar as both groups endured the impact of ankle fracture and uncertainty regarding future function and the necessity for further interventions. However, those who had CCC experienced concern regarding healing and the possibility of further damage to their ankle. Returning to theatre for ORIF in the first 6 to 10 weeks after treatment impacted on how they felt and how they made sense of their treatment. Participants who received ORIF had concerns about the use of metalwork and the potential for infection.

Close contact casting resulted in a modest mean cost savings to the NHS and society. The cost savings were driven largely by savings in theatre time, which were partially eroded over time because of increased readmission and health services costs. Consistent with the clinical results, incremental QALYs following CCC were no different from ORIF, with evidence leaning towards a slight advantage with CCC. Over common willingness-to-pay thresholds, the probability that CCC was cost-effective was very high.

The conclusions for the clinical and health economic evaluations were consistent for the per-protocol or ITT analyses.

Internal validity and methodology

The study was powered to be a definitive equivalence trial and had lower than expected loss to follow-up. Blinding of outcome assessors and very low levels of missing data across the outcome measures strengthen the trial results. Allocation concealment was ensured through random computer allocation, stratified by site and fracture pattern, on participant registration via a remote telephone randomisation service. Analyses were preplanned and agreed by the DMEC. The consistency between the conclusions of the per protocol and ITT analyses improves confidence in the findings of the AIM trial.

The OMAS is one of the most commonly used outcome measures in ankle fracture research,23,110 but has been criticised for methodological shortcomings in its development. 111 We found no published minimal clinically important differences for the OMAS and, therefore, had to develop the sample size drawing on wider literature and in consultation with clinical experts. The OMAS has evidence supporting its construct validity and there are few alternative condition-specific measures. 112 More recently, additional evidence supporting the validity and reliability of the OMAS has been published. 113 In the context of the AIM trial, the conclusions drawn from the OMASs are supported by the high precision of estimates and the consistency with the secondary outcomes [SF-12 (mental and physical), EQ-5D and EQ-5D VAS, timed up and go test, ankle range of motion and pain].

On formal analysis the success of the outcome assessor blinding at 6 months was adequate. For the 6-week assessment the assessor was aware of relevant precautions and contraindications to the clinical assessment and so was not blinded. The 6-week results did not have a strong influence on the analysis or conclusions for the clinical and heath economic evaluations, so we feel this limitation does not detract from the overall findings.

External validity and generalisability

We consider the external validity of the AIM trial to be good for several key reasons. The study was conducted in a range of acute hospitals including major trauma centres through to smaller district general hospitals and, therefore, is representative of the range of settings for ankle fracture interventions in the UK. The CCC intervention was delivered by NHS surgical teams with relatively little training and the trial was pragmatic, which allowed other aspects of care except the intervention to continue unchanged.

Based on previous studies in the UK,1,24 the demographics of the trial participants were representative of the age and sex expected in older adults with ankle fractures, with the majority being female and approximately 70 years of age. The exclusion of patients with insulin-dependent diabetes mellitus and established peripheral vessel disease limits generalisability. However, the results of this study open up the option to investigate the applicability of CCC to patients with these diseases, as there is a high risk of harms after surgical treatment. 27,114,115

The primary end point at 6 months could be criticised as being too early for definitive assessment of outcome. A recent systematic review and meta-analysis,9 however, demonstrated that improvement of patient outcomes from all previously published studies of ankle function outcomes after ankle fractures showed that improvement plateaued by 6 months and, therefore, supports this as the definitive assessment point.

There were differences in experience level of the operating surgeon for ORIF and CCC. The operating surgeon for CCC was consultant grade for approximately 85% of procedures. For ORIF, approximately 39% of operating surgeons were consultants. CCC and ORIF procedures had high levels of consultant supervision, 97% and 76%, respectively. This difference was driven by the protocol as it was felt it was necessary to introduce the standard of direct consultant supervision for CCC because of the uncertainty of it as a new intervention and the likelihood of clinical issues early in the introduction of the technique. We did not identify a learning effect within the trial, although we might not expect to detect it given that just under half of surgeons who conducted CCC in the trial performed only two applications and that these were not necessarily conducted within a close enough time period to reinforce learning. Nevertheless, high consultant involvement is consistent with a new intervention and contemporary NHS practice, further supporting its applicability in practice without concern of unattributed inexperience or learning effects.

Interpretation and implications for clinical practice and policy

The implications of the results of this study to the orthopaedic trauma surgical community are significant and challenge what has been orthopaedic surgical dogma and teaching for several decades. For patients, who are increasingly being involved in decision-making about their treatments, these findings are an important contribution. The qualitative study within this trial elucidated the patient view that the opportunity to avoid undergoing surgery was appealing. The tight equivalence around the point estimate of the primary outcome and the almost identical patient-reported secondary outcomes strongly support the validity of the conclusion that CCC is clinically equivalent to ORIF. The previous extrapolation of witnessed outcomes for this injury in younger patients to patients 60 years of age and above, in whom there are higher incidences of poor bone quality, diminished circulation and skin resilience, now seems unjustified. The study protocol identified that both CCC and ORIF would be associated with treatment-related issues and complications, but that the type would be different for each (e.g. loss of reduction with CCC and surgical wound problems with ORIF). Such treatment-related issues and adverse events that were known to occur and therefore ‘expected’ were monitored for absolute numbers and as a proportion, and expected ranges were agreed with the DMEC. This information is now important for surgeons explaining the equivalence and for patients to understand the advantages and disadvantages of the two treatment options. In general, ORIF has the established risks of surgery and if problems occur they are usually more serious and 6% of patients require a second surgery within 6 months. Approximately 20% of patients who start off by receiving CCC later require an additional visit to theatre to undergo ORIF for unacceptable loss of position, although this may be seen as a positive outcome with 80% avoiding surgery to no detriment. Patients may value the opportunity to try to avoid surgery, in the confidence that there will be an equivalent outcome in terms of ankle function at 6 months. However, the qualitative study would suggest that support for the group that converted to ORIF would be required to help them process the change in their treatment and endure another period of non-weight-bearing.

A significant and consistent finding across the outcomes in the AIM trial was the major impairment in function and quality of life resulting from an ankle fracture, irrespective of the treatment received. This observation was supported by the qualitative study that identified how participants struggled to live with a fractured ankle. They experienced feelings of vulnerability and were challenged in their roles, relationships, daily life and in their ability to mobilise. This finding highlights the implications of ankle fracture for older adults to stakeholders in health- and social-care and policy in the UK.

There has been a deficiency in the published literature relating to the treatment of ankle fractures in older people. Surgeons recognise that they have their own preferences, formed over time from clinical and training experiences, and perhaps modified by the limited published research. Those preferences were borne out in our study by enquiry made of surgeons after randomisation. Of patients allocated ORIF, for 85% of their surgeons that would have been preferred treatment, whereas CCC was the surgeon’s preference in only 28% of those allocated to CCC. This has considerable implications for the difficulty in adoption of this study’s findings and translating them into shared decision-making with patients.

There are also implications in interpreting the results of this trial, as it can be assumed participating surgeon preferences may have influenced perseverance with CCC and willingness to redo the manipulation and cast in the presence of loss of fracture reduction. One key aspect of follow-up when using CCC, as with any conservative treatment for unstable fractures, is close monitoring of the fracture position in the first few weeks after injury with the opportunity to reapply the cast. The willingness to reapply CCC may be important. We recently conducted a laboratory investigation of the effects of reduction in lower leg swelling. We found that CCC was susceptible to a loss of ankle stabilisation as swelling resolves, highlighting the importance of monitoring and reapplication of the cast as necessary in the first few weeks after injury. 116

As a pragmatic trial, the decision to reapply CCC or convert to ORIF was left to the treating surgeon’s judgement. The decision to convert 20% of CCCs to ORIF may, we assume, have been based on the same preconceived opinions that were responsible for the preference of surgeons for ORIF, that is that they needed to achieve exact anatomical restoration of bony architecture of the ankle joint components and that implants were necessary to maintain that reduction of deformity and hold fracture fragments until healing. Those judgements have been central to surgical methods teaching for fractures involving joints for several decades. They are now challenged by this definitive study in which we move from process measures, complications rates and radiological outcomes to patient-reported outcomes in an arguably well-conducted adequately powered and valid randomised blinded trial.

The higher rates of malunion/non-union associated with CCC compared with ORIF indicate that achieving alignment may be more challenging with conservative treatment. However, the equivalence in clinical outcome challenges the importance of restoring anatomical alignment of the joint as part of ankle fracture management in older adults. There will be justifiable reservations of whether or not these differences in radiological outcomes manifest as clinical issues in the longer term. Pain and disability as a result of later post-traumatic osteoarthritis will be a concern. 20 However, the aetiology of post-traumatic osteoarthritis may also be because of genetic factors and direct damage caused by the initial trauma. 21 Lateral malleolar non-union is often thought to be more of a clinical issue than medial malleolar non-union. However, a 2012 systematic review found low-quality evidence regarding lateral malleolar non-unions, reporting these as an uncommon complication and one that can often be asymptomatic. 117 These uncertainties regarding the longer term will be addressed by the extended follow-up of the AIM trial cohort that is under way. We aim to establish if there are any differences in later ankle function, quality of life and resource use [e.g. need for ankle arthrodesis (fusion) or arthroplasty (replacement)] at least 2 years after injury.

Implications for training and implementation

The novel intervention (CCC) in this trial was successfully taught to 317 orthopaedic surgeons and the technique demonstrated to other staff (plaster nurses, technicians and operating room practitioners). Training was achieved through a short instructional video and a face-to-face demonstration. Plaster cast application is within the skill set of orthopaedic surgeons, although modern surgical practice has deprived many younger surgeons of the manual skills of fracture reduction and cast-moulding techniques of former generations. The CCC treatment method draws on elements of both the total contact cast technique32–34 for managing diabetic leg ulcers and also those traditional three-point moulding and gravity fracture-reduction techniques applicable to Potts (ankle) fractures. 118 For the CCC intervention to be adopted there will also need to be a change in the scrutiny of interpretation and surveillance of follow-up radiographs in orthopaedic fracture clinics.

Cast-related complications such as plaster sores and pain from the cast were common in both treatment arms. The most clinically serious technical casting complication was plaster saw lacerations on removal of casts, occurring in 5 out of 275 (1.8%) participants in the CCC group compared with 1 out of 298 (0.3%) in the ORIF group. It seems plausible that there may have been some technical challenges removing the CCC with the plaster saw compared with traditional casts and there are several potential reasons. The operator of the saw may have (1) not been experienced in using the saw on a minimally padded cast, (2) not known the cast was minimally padded or (3) been unaware of the location of the foam cutting strips for protecting the skin, especially if not adequately marked on the exterior of the cast. In addition, the foam cutting strips may have offered insufficient protection or were not applied during CCC application. As a pragmatic trial we did not monitor the application and removal of CCC, therefore it is recommended that when implemented in clinical practice that steps are taken locally to reduce the risk of plaster saw injuries.

When the AIM study was proposed there was concern raised regarding return to weight-bearing. The assumption was that ORIF gave patients the advantage of earlier weight-bearing compared with conservative treatment. As identified in our qualitative study, participants found coping with non-weight-bearing challenging and so an earlier return to weight-bearing may be advantageous during recovery. In the AIM study, hospitals followed their own postoperative weight-bearing protocols for ORIF and we advised no more than touch weight-bearing in the first 4 weeks after CCC. However, we found no difference in the time to start partial weight-bearing between the two interventions and so those concerns prior to the trial were not substantiated.

In addition to being of interest to the UK NHS, the findings of the AIM trial are also likely to be of interest to health services in other comparable high-income countries. The finding of clinical equivalence for a conservative treatment option for older patients with unstable ankle fracture is also important for low- and middle-income countries,119 where availability of specialist surgeons and surgical implants can be limited. 120

Comparison with other literature

A clinical trial from Canada has been published since the start of the AIM trial and subsequent to the 2012 Cochrane review, comparing ORIF with conservative treatment with a brace or cast. 78 The study included 81 participants who had an isolated, undisplaced, unstable, type B fracture of the distal fibula. Participants were younger than those in the AIM trial, aged, on average, 41 years. There were no differences in patient-reported ankle function or quality-of-life scores during the 1-year follow-up period. In the conservative treatment group, 8 out of 40 (20%) developed radiographic joint misalignment. In the operative group, 6 out of 41 (15%) experienced a surgical site infection, one of which was deep and required an additional operation to wash out the joint. Five participants in the operative treatment group also required an additional operation to remove the metalwork. The Canadian study included a health economic analysis. 76 The economic analysis was from a health service cost perspective, which estimated that within 1 year operative treatment was not cost-effective. From the lifetime horizon analysis, these conclusions were different, if it was assumed that conservative treatment results in much higher rates of post-traumatic osteoarthritis. These health economic findings cannot be easily translated to the UK NHS, as costs of rehabilitation were not included. There is also a lack of high-quality prospective research to support the rationale that conservative treatment necessarily results in higher rates of post-traumatic osteoarthritis. 21

Further research

As mentioned previously, in accordance with the IDEAL (Idea, Development, Exploration, Assessment, Long-term study) recommendations for long-term follow-up in the development of surgical innovation,36 we are conducting an extended follow-up of the AIM trial participants beyond 2 years, funded by the National Institute for Health Research Health Technology Assessment programme. The extended follow-up should be completed by November 2016. Longer-term follow-up was deemed important because of concerns from clinicians over potential later complications or additional procedures (e.g. intolerance of metalwork resulting in removal, post-traumatic osteoarthritis and possible ankle joint fusion or replacement) and their potential to impact on ankle function. The extended follow-up will also inform a longer-term projection of the cost-effectiveness analysis.

As part of the implementation of CCC in NHS practice, it is also important to examine how patients develop preferences for treatments and process the consequences of decisions made about treatments in the light of their recovery. In addition, the need to develop support strategies for patients dealing with the uncertainties faced when recovering after CCC and ORIF has been identified.

We plan to build on the findings from the AIM trial cohort. We will undertake predictive modelling of the anatomical fracture patterns, variants of radiological malunion and joint incongruence that predispose to loss of reduction with CCC (malunion or convert to ORIF). The results of the AIM trial directly challenge the assumption that the best outcomes for unstable ankle fractures can only be achieved through surgery. The equivalence of outcome with CCC still begs the question, given the overall significant impairment of function at 6 months, whether or not it is possible to identify the patients who would benefit most from one or other intervention. It is possible that certain patient or fracture injury characteristics may be associated with better outcomes or lower complication rates with one treatment. The substantial decline in function and quality of life after ankle fracture for those aged over 60 years also indicates a need to further optimise the recovery and rehabilitation of older patients.

Finally, the findings of the AIM trial, and the recent trial from Canada,78 also raise questions of whether or not CCC could provide an alternative surgery to adults under 60 years of age; further studies are now warranted.

Contributions of authors

David J Keene (Research Physiotherapist in Trauma and Rehabilitation) was clinical co-ordinator, designed the study, delivered intervention training and wrote and reviewed the report.

Dipesh Mistry (Research Fellow in Statistics) conducted the clinical quantitative analysis and wrote and reviewed the report.