Total ankle replacement versus ankle arthrodesis for patients aged 50 85 years with end-stage ankle osteoarthritis: the TARVA RCT

Article history

The research reported in this issue of the journal was funded by the HTA programme as project number 12/35/27. The contractual start date was in November 2013. The draft report began editorial review in April 2021 and was accepted for publication in February 2022. 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.

Copyright statement

Copyright © 2023 Goldberg et al. This work was produced by Goldberg et al. under the terms of a commissioning contract issued by the Secretary of State for Health and Social Care. This is an Open Access publication distributed under the terms of the Creative Commons Attribution CC BY 4.0 licence, which permits unrestricted use, distribution, reproduction and adaption in any medium and for any purpose provided that it is properly attributed. See: https://creativecommons.org/licenses/by/4.0/. For attribution the title, original author(s), the publication source – NIHR Journals Library, and the DOI of the publication must be cited.

2023 Goldberg et al.

Background

Ankle osteoarthritis is a condition in which the cartilage lining the ankle joint has worn away. The cartilage acts as a shock absorber and allows smooth, gliding motion. Absence of the cartilage and the resultant bone spurs that form (bony projections or osteophytes), and calcification and scarring of the capsule, lead to progressive pain and stiffness.

More than 29,000 patients in the UK present to specialists each year with symptomatic ankle osteoarthritis, a condition that causes major disability and has a similar impact on quality of life (QoL) as end-stage cardiac failure1 and end-stage hip arthritis. 2 The current demand incidence of ankle osteoarthritis has been estimated to be 47.7 per 100,000 per year. 3

The most common aetiological factor in the development of osteoarthritis of the ankle is previous trauma, often following fractures or severe sprains of the ankle. 4 The incidence of both of these is rising; hence, post-traumatic osteoarthritis of the ankle is likely to become an increasing health burden. Indeed, ankle sprains are one of the most common reasons for attendance at emergency departments. Other causes of ankle osteoarthritis include long-standing inflammatory arthropathies (e.g. rheumatoid arthritis, haemochromatosis and haemophiliac arthropathy).

In the early stages of disease, non-operative measures such as a change in activity levels, weight loss, physiotherapy, painkillers and ankle braces should be used. When these conservative management measures have failed for at least 6 months, and providing the surgeon confirms the diagnosis of osteoarthritis (now termed ‘end-stage osteoarthritis’) on the basis of radiological and clinical evidence (i.e. plain radiographs and unrelenting symptoms, respectively), surgery might then be considered.

Although ankle fusion is the most common surgical treatment for end-stage ankle osteoarthritis, surgeons are increasingly performing total ankle replacement (TAR), also known as arthroplasty, in response to patient demand. TAR started in the 1970s, with initial poor results. However, over the last 50 years, several new generations of implants have been developed with far improved results and its use is increasing globally. At least 4000 patients are treated with ankle fusion or TAR each year in the NHS. 5 Every TAR implanted in England, Wales, Northern Ireland, the Isle of Man and Guernsey is captured on the National Joint Registry, which has revision surgery as its only end point. No comprehensive outcome data are captured for ankle fusion patients. All studies comparing TAR with ankle fusion to date are observational and, to the best of our knowledge, there are no prospective randomised trials.

Many studies have shown that ankle fusion provides good short- and medium-term results. However, in the long term, it poses major risk (> 80%) of the development of adjacent joint arthritis owing to the transfer of stresses and motion to other joints. 6,7 Other complications following ankle fusion include pain, dysfunction, non-union and malalignment. 8

On the other hand, TAR can preserve the functional range of ankle motion, relieve pain and might avoid potential osteoarthritis in the adjacent joints. However, it may also result in revision surgery for aseptic loosening, intraoperative fracture, malalignment, impingement and heterotopic ossification. 9–11

To the best of our knowledge, there is no high-quality study comparing the two procedures, and the literature on this subject does not provide conclusive differentiation of the treatments, with varying length of follow-up, sample size and types of technique and implants. 12–19 The studies use a wide range of patient-reported outcome measures, without consistency of reporting or statistical analysis. Many of them have missing data, which makes the interpretation and comparison of results from individual studies next to impossible.

More recently, Daniels et al. 20 looked at 281 TARs and 107 ankle fusions and found comparable outcome scores between the two surgeries at a mean follow-up of 5.5 years. In their study, which was not randomised, patients who received ankle fusion were younger, more likely to be diabetic, less likely to have inflammatory arthritis and more likely to be smokers than those who received TAR. Veljkovic et al. 21 analysed 88 TARs and 150 ankle fusions at a follow-up of 3.6 years and found comparable clinical outcomes between ankle fusion and TAR in patients with non-deformed end-stage ankle arthritis.

In the National Joint Registry, which covers England, Wales, Northern Ireland, the Isle of Man and Guernsey, the most commonly used ankle replacement implants drastically changed between 2014 and 2019. 22 Prior to 2014, the majority of implants used in the UK were mobile bearing. In 2014, the Mobility™ Total Ankle System (DePuy Synthes Companies, Raynham, MA, USA) was withdrawn from the market. By 2019, the majority of implants used in the UK were fixed bearing. In 2019, the most commonly used implant was the Infinity™ Total Ankle System (Stryker, MI, USA) with the STAR™ (DJO, LLC, Vista, CA, USA) and Box^® Total Ankle Replacement (MatOrtho Limited, Leatherhead, UK) implants the second and third most popular, respectively. 22 With regard to ankle fusion, there were a heterogeneity of techniques used to perform the ankle fusion, including arthroscopic and open techniques.

Esparragoza et al. 17 conducted a 2-year follow-up study of 30 patients [ankle fusion, n = 16; TAR with Ankle Evolutive System (AES) prosthesis (Transystème JMT Implants SA, Nîmes, France), n = 14], comparing their QoL before and after the procedure. They showed that the third-generation TAR provided greater improvement in QoL (physical conditions, and perception of general health and QoL) at 2 years post surgery. 17 On the other hand, Krause et al. ,18 in their 3 year-follow up study of 161 patients [ankle fusion, n = 27; TAR with Agility™ (DePuy Synthes Companies), HINTEGRA^® (DT MedTech, LLC, TN, USA), STAR or Mobility Total Ankle System implants, n = 114], found no significant difference in the mean improvement between the two groups, although the rate of complication was significantly higher after TAR than after ankle fusion. In another short-term follow-up study, Slobogean et al. 16 assessed QoL 1 year after TAR or ankle fusion in 107 patients and demonstrated that preference-based QoL was improved following TAR and ankle fusion, but the improvement was not significantly different between the two procedures. 16

Two systematic reviews comparing outcomes from TAR with ankle fusion, using second-generation prostheses13 or third-generation three-component meniscal-bearing prostheses,12 showed no significant differences in short-, mid- or long-term outcomes between the two treatments. Haddad et al. 13 reported that ankle fusion resulted in a higher risk of lower limb amputation, although they did not include any studies that directly compared TAR with ankle fusion. A systematic review and meta-analysis of 7942 modern TARs by Zaidi et al. 23 reported that TAR has a positive impact on patients’ lives, with benefits lasting 10 years, as judged by improvement in pain and function, and improved gait and increased range of movement. Zaidi et al. 23 reported an overall survivorship at 10 years of 89%, with an annual failure rate of 1.2% [95% confidence interval (CI) 0.7% to 1.6%]. The same authors reported improvements in clinical scores, although the scores used were heterogeneous and without consistency. Radiolucency was identified in up to 23% of TARs after a mean of 4.4 years (95% CI 2.3 to 9.6 years).

Gougouilas et al. 15 also performed a systematic review of the outcome of seven TAR implants that are currently in use [Agility, STAR, Buechel-Pappas™ (Endotec, Inc., Orlando, FL, USA), HINTEGRA, Salto Talaris^® Total Ankle Prosthesis (Integra LifeSciences Corporation, Boston, MA, USA), TNK (Kyocera Corporation, Kyoto, Japan) and Mobility implants] and showed that most patients experienced significant improvement, as assessed by the clinical score. In contrast to Zaidi et al. ,23 Gougouilas et al. 15 suggested that the postoperative improvement in the range of ankle motion was relatively small (0–14°). A decision analysis using a Markov model showed that TAR was a better treatment than ankle fusion, as assessed by the quality well-being index score. 19 These systematic reviews have exposed significant bias and a lack of prospective controlled data for either procedure.

A cost-effectiveness evaluation conducted in the USA concluded that TAR has the potential to be a cost-effective alternative to ankle fusion, but reaffirmed the poor quality of the supporting evidence. 24 To the best of our knowledge, there have been no level 1 RCTs to inform this important subject.

Objectives

The total ankle replacement versus ankle arthrodesis (TARVA) trial was a pragmatic, multicentre, parallel-group, non-blinded randomised controlled trial (RCT) that compared the two existing NHS treatment options: TAR and ankle fusion. The trial compared any current TAR implant with any isolated tibiotalar ankle fusion procedure. As a pragmatic trial should reflect the real-world situation, procedures varied in terms of technique owing to the specific requirements of each case and the preference of the operating surgeon. Thus, no surgical technique or type of ankle fusion was specified, although details were captured. Surgeons performing TAR were free to adopt their usual technique within each treatment arm, allowing the results of the trial to be extrapolated across the NHS. All surgeons included in this trial used implants and prostheses commonly used in the NHS only.

The trial assessed the comparative efficacy of the two main surgical treatments for end-stage ankle osteoarthritis: TAR and ankle fusion. It investigated the clinical effectiveness and complication rates of the two procedures in patients aged 50–85 years, measured through self-reported pain-free function using a standardised questionnaire of walking and standing ability at 52 weeks after the surgical intervention. It also aimed to determine whether or not there was a difference in physical function [measured using the Foot and Ankle Ability Measure – Activities of Daily Living (FAAM-ADL)], QoL [measured using the EuroQol 5-Dimensions, five-level version (EQ-5D-5L)] and range of motion (ROM) at 26 and 52 weeks post surgery. Last, we investigated the cost-effectiveness and cost–utility of TAR and ankle fusion. The adoption of a pragmatic trial design with broad entry criteria for the comparison of the two topical therapies means that the results can be generalised to the large number of patients presenting with ankle osteoarthritis who are treated each year.

Design

The TARVA trial was a randomised, multicentre, non-blinded, prospective, parallel-group trial of TAR versus ankle fusion in patients with end-stage ankle osteoarthritis who were aged between 50 and 85 years, comparing clinical outcomes (i.e. pain-free function, QoL, ROM and rate of postoperative complications) and cost-effectiveness.

The trial incorporated an internal feasibility phase to ensure the surgeons’ willingness to randomise and the patients’ willingness to be randomised. The feasibility phase involved four centres and took place over a 6-month period following randomisation of the first patient (24 cumulative months across all centres) to closely monitor eligibility, consent and randomisation rates and ensure that they were adequate. In addition, this provided 5 months’ information on whether or not patients accepted their randomised surgery, and whether or not the surgery took place.

The final protocol has been published previously. 25 All trial analyses were performed in accordance with a predefined SAP. 26

Ethics

London Bloomsbury Research Ethics Committee (REC) reviewed and approved (14/LO/0807) the trial protocol and all material given to prospective participants, including the informed consent forms (ICFs). Subsequent amendments to these documents were submitted for further approval. Before initiation of the trial at each additional clinical site, the same/amended documents were reviewed and approved by local Research and Development.

Patient and public involvement

Patients and the public were involved at all stages of the trial, from the development of the research questions and protocol to the running of the trial. This was important to ensure the salience of the research question and that the methods proposed were acceptable to potential participants, including the frequency of visits and relevance of outcome measures. One patient representative was part of the TARVA Trial Management Group, and one patient and public representative sat on the Trial Steering Committee. We will involve patient organisations and charities such as Versus Arthritis (Chesterfield, UK) and the Arthritis and Musculoskeletal Alliance (London, UK) in the dissemination of the findings to a wider audience, both professionals and patients, through their newsletters, at their annual members’ meetings and on their websites.

Setting

The trial was carried out in 17 UK hospitals, in a mixture of district general hospitals, university teaching hospitals and specialist orthopaedic hospitals (including their adjoining private hospitals) with adequate facilities to carry out the surgical procedures and trial assessments (see Acknowledgements for a list of participating sites).

Participants

The eligibility criteria for participation were patients with end-stage ankle osteoarthritis, aged 50–85 years, who the surgeon believed to be suitable for both TAR and ankle fusion (having considered various patient factors including deformity, stability, bone quality, soft tissue envelope and neurovascular status). The patients had to be able to read and understand the patient information sheet (PIS) and provide written informed consent. Eligible patients were randomised to a surgery type. ‘End-stage’ osteoarthritis is defined as a combination of severe unrelenting symptoms sufficient to make the patient consider surgical intervention, radiological changes consistent with osteoarthritis and failure of at least 6 months of non-operative measures, necessitating a definitive surgical procedure.

Exclusion criteria included patients with previous ipsilateral talonavicular, subtalar or calcaneocuboid fusion or surgery planned within 1 year of index procedure; those with more than four lower-limb joints fused; and those who were unable to undergo magnetic resonance imaging (MRI) or computerised tomography (CT). Those with a history of local bone or joint infection and those who had severe osteoporosis (T-score of < –2.5) with recent fracture (< 12 months previously) were not included in the trial. Patients with any comorbidity that, in the opinion of the investigator, was severe enough to interfere with the patient’s ability to complete the trial assessments or present an unacceptable risk to the patient’s safety were also excluded from the trial.

Interventions

In the UK, two broad types of prostheses are currently used in TAR: a two-component fixed-bearing prosthesis and a three-component mobile-bearing prosthesis. As both are commonly used, no restriction on the type of prosthesis used was stipulated, although data on prosthesis type were captured. The surgical technique followed the standard operative procedure, which involved an anterior approach to the ankle joint, protection of the neurovascular bundle, and talar and tibial preparation according to the prosthesis used and its instrumentation. Intraoperative fluoroscopy was used as required to confirm position, and final implantation used an uncemented technique. Thorough washout was followed by wound closure using the surgeon’s standard technique. Details of the surgeon’s technique were captured on a case report form (CRF). The surgeon’s usual postoperative protocol was followed with respect to method of immobilisation (plaster or walking boot) and weight-bearing status.

Ankle fusion was performed either as an open procedure or arthroscopically, depending on the surgeon’s preference. Tibial and talar joint surfaces were prepared to avoid bleeding from the cancellous bone, any deformity correction was addressed, and the surfaces were opposed and held with screws and/or plates as required to ensure that the foot was plantigrade and appropriately positioned to match the contralateral ankle in axial orientation. If performed arthroscopically, two portals were made, one anteromedially and one anterolaterally, over the ankle joint for access. If arthroscopic access was not favourable, the operation was performed using an open procedure, which involved either a standard anterior approach, two mini anterior incisions or a lateral approach. The surgical technique and implants used were captured on the CRF. The surgeon’s usual postoperative protocol was followed with respect to use of plaster or walking boot and weight-bearing status, and the specific details of these were captured for each patient on the CRF.

Magnetic resonance imaging

Each participant was booked to undergo MRI of the affected ankle, if this had not already been performed as part of routine care, once they had given written informed consent to take part in the trial. If the participant was ineligible for MRI, CT was booked instead. The grade of MRI/CT was determined by an independent radiologist using a methodology published by our group,27 the report of which was sent to the local principal investigator, and a preoperative assessment appointment was scheduled for the participant.

Randomisation

The randomisation process was based on a minimisation algorithm. The algorithm gave an overall chance of 85% of allocating the patient to the treatment arm that was under-represented with respect to three stratifying variables: surgeon, presence of osteoarthritis in subtalar joint and presence of osteoarthritis in talonavicular joint (as determined by preoperative MRI). The research nurse or delegated individual logged on to the Sealed Envelope randomisation service and provided patient information (including information on minimisation variables), and the surgical treatment to be received was supplied immediately. Patients were allocated in a 1 : 1 ratio to the TAR and ankle fusion arms. To protect against allocation bias, the person recruiting the patient to the trial was not aware of the allocation to be assigned prior to contacting the randomisation service. All surgeons were proficient in both surgical procedures, having independently performed ≥ 10 procedures of each type prior to participation.

Blinding

The trial was open (i.e. non-blinded). It was not possible to blind patients, surgeons, radiologists and clinical assessors for the following reasons:

• Surgeons would have known which procedure they were performing.

• Radiologists and patients would be able to identify which procedure had taken place from the radiographs.

• Patients who received ankle fusion would tend to have stiffer ankles and the incisions may also provide clues to the surgery type.

Recruitment and consent

All patients with ankle osteoarthritis who were considering surgery were screened prospectively by principal investigators at 17 UK hospitals. Potentially eligible participants were identified during routine clinic appointments to assess treatment need, or through screening of referral letters/clinic lists. Those identified through screening were sent a study information pack in the post prior to their appointment.

If considered eligible for the TARVA trial, the patient watched a bespoke trial video, and read the PIS and a generic factsheet about ankle arthritis and its treatment options. Participants either consented at that stage or received a follow-up telephone call from the research team to discuss participation. Reasons for non-enrolment in the trial (including lack of equipoise) were recorded. All participants provided written informed consent using an ICF.

Following consent, participants underwent MRI (or, where contraindicated, CT) (if this had not been performed as part of standard care within the previous 6 months), followed by a preoperative assessment 14–30 days prior to surgery. If declared fit for surgery, participants were randomised to one of the two surgical treatments. Participants who were found to be unsuitable for surgery at the preoperative assessment appointment were passed back to their general practitioner (GP) to be re-referred for surgery when they were considered fit.

Baseline visit

Baseline measures were recorded at the point of randomisation, once the participant had been found to be fit for surgery, at their preoperative assessment. Baseline measures included the EQ-5D-5L, MOXFQ, Foot and Ankle Ability Measure (FAAM), Client Service Receipt Inventory (CSRI), ROM and concomitant medication.

Follow-up assessments and treatment

All participants attended routine follow-up, which consisted of visits at 2, 6, 12, 26 and 52 weeks post surgery. These visits were standard care. Patients underwent routine clinical review at 2 weeks, during which the stitches were removed and plaster casts changed. Trial-specific outcome measures, including adverse events (AEs) and postprocedural complications, were recorded at 6, 12, 26 and 52 weeks. Concomitant medications were recorded from preoperative assessment to the 52-week visit. Participants underwent routine physical examination, as per standard care. Participants completed additional questionnaires (the MOXFQ, EQ-5D-5L and FAAM) at 26- and 52-week routine follow-up visits, with the EQ-5D-5L and CSRI additionally completed at 12 weeks. ROM (total floor to tibial shaft plantarflexion and dorsiflexion) was assessed using a goniometer at the preoperative assessment visit and 52 weeks post surgery. 28

To avoid bias, operating surgeons were not involved in measuring ROM. Preoperative and postoperative hindfoot deformity was measured using weight-bearing anteroposterior and lateral radiographs of the ankle and tibia at baseline, and on a postoperative radiograph (between 0 and 26 weeks post surgery) using the methods described by Knupp et al. 29 Plain radiographs were sent to the Royal National Orthopaedic Hospital via the NHS Image Exchange Portal (Sectra Ltd, Stevenage, UK) in one batch (containing preoperative and postoperative radiographs) after the second (postoperative) radiograph was taken. Investigators were blinded to participant treatment allocation when reviewing the preoperative radiographs. Each participant was in the trial from consent until the final 52-week follow-up visit, although long-term follow-up at 2, 5 and 10 years post surgery was part of their informed consent.

Safety

All medical device deficiencies, AEs and serious adverse events (SAEs) occurring during the trial that were observed by the investigator or reported by the participant (whether or not they were attributed to the surgery, surgery-related medications, device or other trial-specific procedures) were recorded in the participants’ medical records. Related AEs over and above what would normally be expected after ankle surgery were recorded on the relevant CRFs. SAEs were reported in line with procedures set out in the protocol. 25

The severity of all AEs (serious and non-serious) was graded using the TARVA trial safety management plan for expected AEs, in conjunction with the most recent version of the Common Terminology Criteria for Adverse Events (at the time the protocol was written, this was version 4.030) for other (unexpected) AEs. The ‘expectedness’ was determined by the list of expected events in the TARVA trial safety management plan.

Outcomes

Sample size

The sample size calculation for the primary outcome (change in MOXFQ walking/standing domain score by 52 weeks post surgery) was performed using Stata/IC^®, version 12.1 (StataCorp LP, College Station, TX, USA). It was based on achieving 90% power to detect the minimal important difference (MID) in the primary outcome at the 5% level of significance, accounting for expected loss to follow-up.

Dawson et al. 46 previously defined the MID in the MOXFQ when evaluating outcomes following surgery for hallux valgus as the mean change in MOXFQ score of those patients who reported feeling at least ‘slightly better’. They found the MID to be 16, 12 and 24 for the walking/standing, pain and social interaction domains, respectively. 46

A later paper by Dawson et al. 47 discussed the minimal detectable change, which is the smallest change for an individual that is beyond the measurement of error of a given instrument and therefore likely to represent a true change. Although Dawson et al. ’s 2007 paper46 looked at hallux valgus, their later paper47 specifically studied ankle procedures as a subgroup and estimated the MID to be 10.67.

For this trial, we determined that it was important to detect a difference of 12 in the change in the MOXFQ walking/standing domain scores from baseline between the two treatment arms.

The standard deviation (SD) of the MOXFQ walking/standing domain score was estimated to be 27. 46 We took into account an anticipated 10% dropout rate (attrition in orthopaedic trials is about 5–7%, as shown by other similar UK RCTs48). Based on these quantities, the required sample size was estimated to be 118 patients per arm.

However, the trial was multicentre and the outcome was assumed to vary by surgeon, so the sample size was increased to account for clustering by surgeon. The intraclass correlation coefficient (ICC) was estimated from the median of 10 previous surgical studies reporting patient-reported disease-specific measures 12 months post surgery,49 and the initial sample size estimate was inflated by a factor of f = 1 + (m – 1) × ICC. Assuming an average cluster size (m) of 14 (patients per surgeon) and an ICC of 0.03, an inflation factor of f = 1.39 was estimated, leading to a final required sample size of 164 patients per arm or 328 patients in total.

Data collection and management

A member of the research team captured data from patients on paper using the TARVA trial CRFs. The data were entered onto the main trial database (MACRO v4.1; Elsevier, Amsterdam, the Netherlands) by a delegated member of site staff.

The site retained the original paper copies of patient CRFs to allow monitoring and audit by the University College London Comprehensive Clinical Trials Unit (UCL CCTU) trial team. All data queries were resolved prior to trial closure and analysis.

At sites where electronic records were available, the site may have captured some of the data electronically, which were then transcribed onto the paper CRFs to ensure a complete record.

Statistical methods

All trial analyses were performed according to a predefined SAP. 26 All efficacy analyses were conducted following the intention-to-treat (ITT) principle, in which all randomised patients were analysed according to their randomised surgical procedure, irrespective of the type of surgery they received.

In addition, a per-protocol analysis was carried out for the primary outcome, which included only the outcome data that were collected within the protocol-specified time window from patients who underwent surgery according to their randomised surgical procedure, excluding crossover patients.

The baseline characteristics were summarised by randomised treatment arm. The categorical variables were summarised by number and percentage in each category; continuous variables were summarised by mean and SD, or median and interquartile range, as appropriate. No statistical tests of differences in baseline characteristics between arms were undertaken, as in a randomised trial any differences between treatment arms must be due to chance.

Study oversight

A Trial Steering Committee (TSC) was established, comprising seven independent members, including a patient and public representative, the chief investigator and representatives from among the principal investigators. The trial health economist and senior trial statistician attended meetings as observers. The committee provided advice to the chief investigator, UCL CCTU, the funder and the sponsor on all aspects of the trial.

The UCL CCTU was responsible for the day-to-day management of the trial, with oversight from a Trial Management Group on the design, co-ordination and strategic management of the trial. The Trial Management Group was chaired by the chief investigator.

An Independent Data Monitoring Committee (IDMC) monitored the accumulating data and made recommendations to the TSC on whether or not the trial should continue as planned. The committee consisted of three independent members: a professor of medical statistics, a professor of rehabilitation sciences and a professor of orthopaedic surgery (the chairperson).

All oversight committees had agreed terms of reference.

During the trial, the TSC and IDMC each met six times between August 2014 and July 2019, one of which was a joint meeting of the two committees. The joint meeting led to the abbreviation of the exclusion criteria so that the surgeons’ checklist was shorter. The committees also reviewed the impact of the withdrawal of the Mobility TAR implant, which occurred after the study began but prior to any recruitment. The IDMC and TSC also advised on a recovery plan for slow recruitment, including increasing the number of recruitment sites, extending the recruitment period and a qualitative study to provide insight into recruitment difficulties.

Recruitment

Participants were randomised between 6 March 2015 and 10 January 2019. A total of 1604 patients were screened for eligibility, of whom 303 were randomised: 152 to TAR and 151 to ankle fusion. The numbers of participants recruited and included in the ITT analysis are summarised in the Consolidated Standards of Reporting Trials (CONSORT) flow diagram in Figure 1. Of the 303 patients randomised, 21 withdrew from the trial before receiving surgery, one withdrew before the 26-week follow-up and a further five withdrew/had missing data at week 52. Six of those who received surgery were missing primary outcome measure data at 52 weeks. All patients who received surgery and attended either the 26-week or 52-week visit were included in the ITT analysis. Four patients randomised to arthrodesis did not receive their allocated surgery and crossed over to the TAR arm. All observed outcome data from these patients were analysed according to their randomised surgical procedure.

FIGURE 1.

Trial profile: CONSORT flow diagram.

Of the 282 patients who received surgery, one patient who withdrew before the 26-week follow-up could not contribute data to the primary outcome but was included in the baseline characteristics table. All 281 patients who received surgery and attended at least one follow-up were included in the mixed model for the primary outcome analysis (ITT analysis).

Table 1 lists the 17 sites in order of the date the site opened to recruitment. The first site to open was the Royal National Orthopaedic Hospital in December 2014. This site randomised the largest number of patients (24% of the total randomised).

Table 2 summarises the losses and exclusions after randomisation, with reasons, for each arm. There have been no losses to follow-up in the trial.

A total of 21 randomised patients withdrew from the trial prior to surgery – 14 (9%) in the TAR arm and seven (5%) in the ankle fusion arm. Of the patients who received surgery, four (two in each arm) withdrew from the trial prior to the 52-week follow-up. One patient in the TAR arm died after their 52-week follow-up visit.

Baseline characteristics of participants

The baseline characteristics of participants are presented in Table 3.

The mean (SD) age of the participants was similar in each treatment arm: 68.0 (8.1) years in the TAR arm and 67.7 (8.0) in the ankle fusion arm. In total, 81 (29%) participants were female. The rate of obesity (body mass index of ≥ 30 kg/m²) was 37% in the TAR arm and 51% in the ankle fusion arm.

The proportion of patients with respiratory pathology, diabetes or obesity was lower in the ankle fusion arm than in the TAR arm at baseline. However, there was more deformity in the TAR arm than in the ankle fusion arm (see Table 3). Overall, the two randomised arms were considered generally similar with regard to medical history factors and smoking habits. In terms of American Society of Anesthesiologists (ASA) grade (Table 4), there were slightly more ASA grade 3 patients (severe systemic disease) in the ankle fusion arm than in the TAR arm (17.4% vs. 14.5%, respectively).

Participants appeared to be equally distributed between treatment arms with regard to the minimisation factors, that is the presence of osteoarthritis in the two adjacent joints (subtalar and talonavicular). A total of 122 patients (34%) had osteoarthritis in the adjacent joints. For 25 (9%) of these patients, the osteoarthritis was in both adjacent joints.

Prior to their surgery, 44% of patients reported that they used assistive devices. The majority of those using an assistive device used a stick or cane. Patients also reported using other forms of assistive devices such as crutches (9%) and ankle braces (8%).

The majority of patients (77%) did not express a treatment preference, 17% of patients stated a preference for TAR and 6% expressed a preference for ankle fusion.

The baseline mean (SD) MOXFQ walking/standing score was 82 (16.6) in TAR and 82 (16.8) in ankle fusion patients. The baseline values for the outcome measures were similar in the two treatment arms.

Surgery details

The duration of the procedure was slightly longer for TAR (121 minutes) than ankle fusion (103 minutes). Patients were immobilised for longer in the ankle fusion arm than in the TAR arm: 26 (19%) patients in the TAR arm compared with seven (5%) in the ankle fusion arm were allowed to weight bear within 2 weeks of the surgery.

The arms were broadly similar in terms of previous surgery, although the TAR arm had slightly more patients who had previously had internal fixation on the index joint than the ankle fusion arm (16% vs. 12.5%, respectively).

For those patients who underwent TAR, 76 (53.5%) had a fixed-bearing TAR and 66 (46.5%) had a mobile-bearing TAR (Table 5). In the ankle fusion arm, 60% of the procedures were performed arthroscopically. For those patients who underwent an open ankle fusion, seven (14%) received a lateral approach (Table 6).

The proportion of patients who had an associated procedure was higher in the TAR arm than in the ankle fusion arm (35% vs. 18%, respectively). The most common procedure was Achilles tendon lengthening, which was undertaken in 17 (12.3%) patients in the TAR arm (Table 7). Six patients (4.3%) in the TAR arm required a lateral ligament repair. No patients in the ankle fusion arm underwent ligament repair.

The distribution of procedures by surgeon is shown in Table 8. Recruitment ended at the Royal National Orthopaedic Hospital in June 2018, 6 months prior to closure of recruitment.

Numbers analysed

Owing to the nature of the model used in the analysis of primary and secondary continuous outcomes (i.e. a mixed model), all patients with a baseline visit and at least one follow-up visit were included in the analysis. Therefore, the final primary outcome analysis was based on 281 patients: 137 in TAR and 144 in ankle fusion (Table 9).

Primary outcome

Findings for the primary outcome, MOXFQ (walking/standing domain), are presented in Table 10.

The mean (SD) MOXFQ walking/standing domain score at 52 weeks was 31 (30.4) in the TAR arm and 37 (30.6) in the ankle fusion arm. The mean (SD) change in scores between pre-surgery baseline and 52 weeks was –50 (30.7) in the TAR arm and –44 (31.9) in the ankle fusion arm. The adjusted difference in change score of –5.56 (95% CI –12.49 to 1.37) suggests that, on average, the improvement in the MOXFQ score from baseline to 52 weeks post surgery was 5.56 points greater in TAR patients than in ankle fusion patients (p = 0.12). The 95% CI for this difference includes both a difference of zero and the MID of 12. The MOXFQ scores improved following surgery in both arms (TAR: mean change –50, 95% CI –55 to –45; ankle fusion: mean change –44, 95% CI –50 to –39), but there was not a statistically significantly greater improvement in the TAR arm than in the ankle fusion arm. The proportion of patients with a reduction in MOXFQ score of at least 12 points from baseline at 52 weeks was very similar in the two arms: 82% of TAR patients compared with 80% of ankle fusion patients.

Four patients crossed over from ankle fusion to TAR after randomisation. Three patients had their 52-week visit outside of the protocol window and an additional five patients had missing 52-week scores. We carried out a per-protocol analysis as a sensitivity analysis for the primary outcome by excluding these 12 patients. The per-protocol analysis did not change the ITT conclusions.

Secondary outcomes

Findings for the secondary outcomes are presented in Table 11.

On average, patients in both arms reported an improvement in the MOXFQ pain and social interaction domains at 26 weeks, and on all MOXFQ domains at 52 weeks, but improvement was not greater in the TAR arm than in the ankle fusion arm. The difference between the TAR and ankle fusion arms in the change in MOXFQ walking/standing domain score at 26 weeks was statistically significant (p = 0.02).

The difference between the TAR and ankle fusion arms in the change in FAAM-ADL scores at 52 weeks was statistically significant (p = 0.01). There were substantial improvements from baseline in both arms, with a difference of 6.16 (95% CI 1.54 to 10.78) between the two arms.

The change in EQ-5D-5L index values between the two treatment arms was not significantly different at 26 weeks (p = 0.08) or 52 weeks (p = 0.32). The change in EQ-5D-5L VAS was statistically significant at 26 weeks (p = 0.03), but not at 52 weeks (p = 0.07).

Changes from baseline in ROM dorsiflexion and ROM plantarflexion were greater in the ankle fusion arm than in the TAR arm. Although ROM (dorsiflexion and plantarflexion) improved from baseline to 52 weeks in the TAR arm, it decreased in ankle fusion patients and the difference between arms was statistically significant (p < 0.001).

Subgroup analyses

A total of 45 patients (33%) in the TAR arm and 50 (36%) in the ankle fusion arm had osteoarthritis in one adjacent joint at baseline; 11 patients (8%) in the TAR arm and 12 patients (9%) in the ankle fusion arm had osteoarthritis in both the subtalar and talonavicular joints. Adjusted MOXFQ scores were lower in the TAR arm than in the ankle fusion arm at 52 weeks in the subgroup analyses (Table 12). However, we did not find a significant interaction caused by this factor. There was also no evidence to suggest that the effect of treatment was moderated by age. The subgroup analyses are presented in Figure 2.

FIGURE 2.

Forest plot showing subgroup analysis by treatment arm.

Adverse events

A total of 20.8% of randomised patients experienced at least one SAE and 53.5% of patients experienced at least one AE during the course of their trial pathway (Table 13). The risks of patients experiencing a SAE or an AE were not significantly different between the two arms (p = 0.19 and p = 0.84, respectively).

All the AEs and SAEs reported during the trial have been summarised as postoperative complications in Table 14. One patient in the ankle fusion arm died during the follow-up period and one patient in the TAR arm died after the 52-week visit (not presented in Table 14). Both events were unrelated to surgery.

There were 20 wound-healing problems in 19 (13.4%) patients in the TAR arm and eight wound-healing problems in eight (5.7%) patients in the ankle fusion arm. One patient in the ankle fusion arm and none in the TAR arm required debridement for the infection, although the majority of TAR patients were administered prophylactic antibiotics to treat the superficial wound infections.

There were eight nerve injury events reported in six patients (4.2%) in the TAR arm and one nerve injury event reported in one (< 1%) patient in the ankle fusion arm. Two events were reported twice.

There were 17 non-unions (12.1%) in the ankle fusion arm, which were diagnosed through the presence of a lucent line on plain radiographs at the 52-week follow-up. Of the 17 patients, eight were expected to be revised (47%), two (12%) were symptomatic but not planning to be revised due to serious comorbidities and seven (41%) were completely asymptomatic. Hence, 10 (7.1%) of 140 patients who received ankle fusion went on to symptomatic non-union.

There were 13 thromboembolic events in 11 patients: four (2.9%) patients in the TAR arm and seven (4.9%) in the ankle fusion arm. Two (1.4%) patients in the TAR arm experienced deep-vein thrombosis events and there were five events in four (2.8%) patients in the ankle fusion arm. Two (1.4%) patients experienced pulmonary embolism events in the TAR arm and there were four events in three (2.1%) patients in the ankle fusion arm. None of these events was fatal.

At 52 weeks’ follow-up, nine patients (3.2%) required further unplanned reoperation other than revisions: five in the TAR arm and four in the ankle fusion arm. In the TAR arm, one revision took place within the 52-week window. This was due to a traumatic fall, leading to a fracture and conversion to a tibiotalocalcaneal fusion. We are aware of several other revisions that will take place outside the 52-week window and these data will be reported in the 2-year results. Table 15 lists reoperations and revisions.

Post hoc analysis

The baseline characteristics of each of the subtypes of TAR (fixed and mobile bearing) and ankle fusion (open and arthroscopic) are reported in Appendix 3. Of those who received TAR, 53.5% received fixed-bearing TAR and 46.5% received mobile-bearing TAR. Of the ankle fusion patients, 61% received arthroscopic ankle fusion and 39% received open ankle fusion. Overall, all four subtypes of patients appeared similar with respect to baseline factors and baseline outcome measures. We carried out a post hoc comparison of each TAR subtype (those who received fixed-bearing TAR and those who received mobile-bearing TAR) with the ankle fusion arm (including both open and arthroscopic ankle fusion patients).

The mean (SD) MOXFQ walking/standing domain score at 52 weeks was 25.9 (28.3) in the fixed-bearing TAR group, compared with 36.8 (30.6) in the ankle fusion arm (Table 16). The adjusted difference of –11.1 (95% CI –19.3 to –2.9) suggests that, on average, the MOXFQ score at 52 weeks post surgery was 11.1 points lower in those who received fixed-bearing TAR than in those who underwent ankle fusion. This difference is statistically significant (p = 0.008). Comparing mobile-bearing TAR patients with ankle fusion patients, the adjusted difference is 2.1 (95% CI –6.6 to 10.8). This difference is not statistically significant (p = 0.64). There was no difference in the change in MOXFQ score at 52 weeks (p = 0.83) between open and arthroscopic patients in the ankle fusion arm (see Appendix 4).

The subgroup analyses by subtype of TAR patients (fixed and mobile bearing) compared with ankle fusion patients are shown in Figure 3.

FIGURE 3.

Forest plots showing subgroup analysis by treatment arm and TAR subtype. (a) Fixed-bearing TAR vs. ankle fusion; and (b) mobile-bearing TAR vs. ankle fusion.

Overview

The main objective of the economic evaluation was to assess the cost-effectiveness of TAR compared with ankle fusion for patients with end-stage osteoarthritis. We compared the costs and outcomes in the TAR and ankle fusion arms over a 52-week time horizon. The primary analysis was conducted in accordance with the ITT principle from the NHS and personal social services (PSS) perspective. Sensitivity analysis included per-protocol analysis and analysis from a societal perspective. The per-protocol analysis included only patients who received the surgery to which they were randomised. The societal perspective included out-of-pocket costs incurred by participants and productivity loss. The analytical approach took the form of cost–utility analysis. The main result of the analysis was the mean incremental cost per QALY gained. Costs and outcomes were not discounted because of the short time horizon (i.e. 52 weeks) of the within-trial economic evaluation.

We estimated a long-term cost-effectiveness of TAR compared with ankle fusion using decision-analytic modelling. The effect of TAR on participants’ QoL is expected to last longer than the time horizon of the trial and is affected by the rate of future revisions. We constructed a simple Markov model that simulated patients’ pathways after TAR and ankle fusion. The costs were taken from the trial data. The transition probabilities, cost of revision and EQ-5D-5L index values were obtained from published sources detailed in the subsequent section. The rate of revision was based on the clinicians’ opinion.

Methods

Long-term economic modelling

We used a modelling approach to extrapolate to a lifetime horizon. Our literature search identified two relevant cost-effectiveness studies comparing TAR with ankle fusion. 24,60 Based on these studies, we constructed a simple Markov model that simulated patients’ pathways after TAR or ankle fusion. The structure of the model is shown in Figure 4.

FIGURE 4.

Model structure.

The Markov model, based on 1-year cycles, was used to simulate the impact of TAR and ankle fusion on patients’ health for the lifetime horizon. There are 17 cycles in the model, as average life expectancy for a cohort aged 50–85 years is 17 years. 61 After surgery (TAR or ankle fusion) in year 1, patients stay in good health, move to revision or die. A patient can be in a revision state for 1 year only and then they can move to the good health or death state. We specify the ‘good health after revision’ state as we assume that revision reduces the QoL of a patient. Transition probabilities are reported in Tables 18 and 19.

TABLE 18 - Transition probabilities for TAR

Transition probability (%)	Good health	Revision	Good health after revision	Death
Good health	95.8	1.2	0.000	3.00
Revision	0.00	0.00	97.00	3.00
Good health after revision	0.00	0.00	97.00	3.00
Death	0	0	0	100

TABLE 19 - Transition probabilities for ankle fusion

Transition probability (%)	Good health	Revision	Good health after revision	Death
Good health	92.0	5.0	0.00	3.00
Revision	0.00	0	97	3.00
Good health after revision	0.00	0	97	3.00
Death	0	0	0	100

Revision rates are based on clinicians’ opinion. They are comparable to the previous cost-effectiveness model of TAR compared with ankle fusion. 60,62 The revision rate for ankle fusion was assumed to be 5% in the first 3 years (see Table 19) and 0% thereafter. The death rate is based on Public Health England’s Life Expectancy Calculator 2021. 63 Each health state in the model was assigned a cost and a QALY outcome. These are reported in Table 20.

TABLE 20 - State rewards

State reward	Good health	Revision	Good health after revision	Death
Cost (£)
TAR	275	7218	275	0
Ankle fusion	316	7218	316	0
EQ-5D-5L index value
TAR	0.74	0.59	0.59	0
Ankle fusion	0.71	0.66	0.66	0

We assume that the resource use estimated at baseline is a good estimate of what patients would use in ‘good health’ and ‘good health after revision’ states. The cost of revision is based on the cost of ankle fusion. We assumed that patients continue to have the same EQ-5D-5L index values in the subsequent years after the surgery while they are in good health. Decrements in index values after revision are based on estimates in SooHoo and Kominski62 as this is the only available source of these data. 62 The QoL decreases after revision surgery and patients have the same QoL for the rest of their life. We discount costs and QALYs at the rate of 3.5% recommended by NICE. 64

A cost-per-QALY ratio was calculated using the data from the trial for year 1 and data from the Markov model for years 2–17. Probabilistic sensitivity analysis (PSA) was conducted to account for parameter uncertainty. We assigned probability distributions to parameters in the model and then used Monte Carlo simulations to obtain ICERs. We plotted the incremental costs and QALYs on a cost-effectiveness plane. We also estimated the probability of the intervention being cost-effective at a range of cost-effectiveness thresholds. The probabilities were plotted against the thresholds on a CEAC.

Results

This section presents the results of the health economic analysis of TAR compared with ankle fusion. We compare the cost of surgery, cost of health-care resource use, out-of-pocket costs, total costs, QALYs and cost–utility results of the base-case and sensitivity analyses.

Cost–utility analysis

Primary within-trial analysis

Although orthopaedic surgery lasts patients for many years, if we analyse the data over the 52 weeks of the study, the mean incremental cost per QALY gained was £127,931.50 from the NHS and PSS perspective. Using the bootstrapping technique, we generated an empirical distribution of ICERs and presented them on the cost-effectiveness plane (Figure 5).

FIGURE 5.

Cost-effectiveness plane: ITT, NHS and PSS perspective.

Virtually all ICERs are above the x-axis; therefore, TAR is more expensive than ankle fusion. However, with respect to QALYs, most ICERs suggest that TAR generates more QALYs than ankle fusion. However, some ICERs are on the negative side of the x-axis, implying that TAR generates fewer QALYs than ankle fusion. Hence, there is a high degree of uncertainty in the data. In total, 95% of iterations fall between –£253,647.40 and £182,814.20, implying that TAR may be more expensive or cost-saving.

We also used the bootstrapping results to estimate the probability of TAR being cost-effective compared with ankle fusion at various cost-effectiveness thresholds. The probability is low: 1.3% at the threshold of £30,000 per QALY gained. The probability increases and reaches 37.6% at the threshold of £100,000 per QALY gained (Figure 6).

FIGURE 6.

Cost-effectiveness acceptability curve: ITT, NHS and PSS perspective.

It is important to note that the benefits of the surgery begin after the 52-week window. Therefore, the results need to be interpreted with caution and we conducted long-term economic modelling to account for this.

Sensitivity analysis

The total societal costs were £15,142.95 (SD £7820.42) for TAR and £14,961.09 (SD £9978.75) for ankle fusion. The unadjusted mean difference was £181.86. This difference was not statistically significant. When we adjusted for baseline values and minimisation factors, the mean difference was £198.55. The difference in costs between the two arms was lower than that in the NHS and PSS perspective and it lost significance.

Over 52 weeks, the mean incremental cost per QALY gained was £9927.50 from the societal perspective. This ICER is considerably lower than the base-case result and would be recommended under NICE’s threshold of £20,000–30,000 per QALY gained. 65 However, costs from the societal perspective introduced considerable uncertainty as they were difficult to estimate. Using the bootstrapping technique, 95% of iterations fall between –£41,024.23 and £184,899.90. The cost-effectiveness plane shows that the ICERs can be in any quadrant of the plane, which implies a high degree of uncertainty (Figure 7).

FIGURE 7.

Cost-effectiveness plane: ITT, societal perspective.

The CEAC shows that the probability of TAR being cost-effective compared with ankle fusion at £30,000 per QALY gained is 61.6% and increases to 80.6% when the threshold increases to £100,000. The CEAC is shown in Figure 8.

FIGURE 8.

Cost-effectiveness acceptability curve: ITT, societal perspective.

The probability is equal to about 60% when the cost-effectiveness threshold is zero because 60% of ICERs on the cost-effectiveness plane suggest that TAR is cost-saving compared with ankle fusion.

When we included the cost of replacing an employee, total societal costs were £15,616.05 (SD £8854.13) for TAR and £15,622.46 (SD £11,071.48) for ankle fusion. The unadjusted difference between arms was small (£6.42) and not statistically significant (p-value 0.996). Adjusting for the baseline values and minimisation factors did not change the result. We applied this cost to patients who retired after the surgery and reported that they retired because of their ankle problem. There were two such patients in the TAR arm and three in the ankle fusion arm.

The per-protocol analysis resulted in very minor differences in total costs and QALYs. The ICER was £127,154.60 per QALY gained, which is a lot higher than the threshold used by NICE. 65 In total, 95% of bootstrap values fall between –£165,764.40 and £654,921.60. The cost-effectiveness plane shows that TAR is more expensive than ankle fusion, as all ICERs are on the positive side of the y-axis (Figure 9).

FIGURE 9.

Cost-effectiveness plane: per-protocol, NHS and PSS perspective.

Therefore, the within-trial probability of TAR being cost-effective at 52 weeks was 1% at the cost-effectiveness threshold of £30,000 per QALY gained, increasing to 35.7% if the threshold increases to £100,000 per QALY gained. The CEAC is shown in Figure 10. The results suggest that it is important to account for societal costs when comparing TAR and ankle fusion, as these have a large impact on the ICER. However, societal costs also introduce a high degree of uncertainty. When considering implants and definitive surgery, 52-week data need to be interpreted with caution, as the benefits begin after the 52-week window; hence, the more important analysis relates to longer-term modelling.

FIGURE 10.

Cost-effectiveness acceptability curve: per-protocol, NHS and PSS perspective.

Long-term economic modelling

The model-based analysis suggested that TAR is more expensive than ankle fusion, but generates more QALYs when extrapolated to a lifetime horizon. The ICER was estimated to be £4201.81. Cost and QALY differences are presented in Table 28.

TABLE 28 - Model-based total cost and QALYs per treatment arm

Measure	TAR arm (n = 129)	Ankle fusion arm (n = 134)	Difference
Total cost (£)	2,138,343	1,878,140	260,202
Total QALYs	1110	1048	61

The result of Monte Carlo simulation (n = 5000) is presented in the cost-effectiveness plane in Figure 11.

FIGURE 11.

Cost-effectiveness plane: lifetime horizon.

The mean ICER is in the north-east quadrant. This means that TAR is more expensive and also generates more QALYs than ankle fusion. When we varied cost and QoL parameters in the model, we observed that most points still lay in the north quadrant. Therefore, TAR is likely to be more expensive than ankle fusion over the lifetime horizon. However, there was uncertainty regarding the number of QALYs attained as some ICERs were in the north-west quadrant, implying that TAR may generate fewer QALYs and be more expensive than ankle fusion. Hence, there is considerable uncertainty around the lifetime ICER, and longer-term data are required to obtain a more robust result.

Although there is uncertainty, over the lifetime horizon there was a 69% probability that TAR is cost-effective compared with ankle fusion at the cost-effectiveness threshold of £20,000 per QALY gained (Figure 12).

FIGURE 12.

Cost-effectiveness acceptability curve: lifetime horizon.

As seen in the statistical and economic analysis, the fixed-bearing TAR group performed better than the mobile-bearing TAR group and the ankle fusion arm. If we assume that all patients receive fixed-bearing TAR and assign their QoL to all TAR patients, the difference in QALYs between the two arms increases (Table 29).

TABLE 29 - Costs and QALYs: fixed-bearing TAR vs. ankle fusion

Measure	TAR arm: fixed-bearing group (n = 129)	Ankle fusion arm (n = 134)	Difference
Total cost (£)	2,138,343	1,878,140	260,202
Total QALYs	1150	1048	102

Total ankle replacement is still more expensive and generates more QALYs than ankle fusion. The ICER was £2535.32. When we conducted a PSA for this result, the probability of TAR being cost-effective compared with ankle fusion was 72.2% at the cost-effectiveness threshold of £20,000 per QALY gained.

Recruitment

Our aim was to recruit one patient per centre per month. Overall, the recruitment rate achieved was 0.46 patients per centre per month. The lead site achieved a recruitment rate of 1.7 patients per month and the other 16 sites achieved an average recruitment of 0.38 patients per month. There were several challenges to recruitment, which is not unusual for surgical trials. A qualitative study was conducted that identified four common obstacles: (1) patient preference for an intervention, (2) a complex recruitment pathway, (3) logistical issues and (4) lack of equipoise and role conflicts. Clinicians in the study felt that they could predict that specific patients may achieve better outcomes with either TAR or ankle fusion.

A total of 22 (7.3%) randomised patients were excluded from our ITT analysis, which was well within the anticipated 10% drop-out rate from our power calculation. Our trial attrition is similar to that of other reported orthopaedic trials, which had attrition rates of between 5.3% and 18.2%. 73–75 The original sample size calculation for the TARVA trial made a number of assumptions. Based on the data available now, loss to follow-up at 52 weeks was slightly lower than predicted: 9% rather than 10%. The number of recruiting surgeons was 34 rather than 17, so the average number of patients per surgeon was nine rather than 14. Therefore, the power achieved with our 276 patients who had data available at 52 weeks for the ITT analysis was > 88%, very close to our desired power of 90%. The slightly lower power achieved is unlikely to have influenced our conclusions.

Economic evaluation

Over the first 52 weeks following primary surgery, TAR was more expensive than ankle fusion, which was expected owing to the higher prices of the implants and longer duration of the surgery. However, after accounting for other costs associated with the surgery, including mobility aids and home adaptation, productivity loss and transportation cost, the difference between the two arms was no longer statistically significant and the ICER reduced considerably. This suggests that TAR is likely to have a wider impact on patients’ lives that is not accounted for in the effectiveness and QoL measures.

To the best of our knowledge, there is sparse published health economic data regarding the cost-effectiveness of ankle surgery. Slobogean et al. 16 estimated index values in patients after TAR and ankle fusion using a prospective non-randomised cohort of TAR and ankle fusion patients. Their baseline values were higher than those of the TARVA trial for both TAR (0.67, 95% CI 0.64 to 0.69) and ankle fusion (0.66, 95% CI 0.63 to 0.68). At 52 weeks, their index values (TAR 0.73, 95% 0.71 to 0.76; ankle fusion 0.73, 95% 0.70 to 0.76) were comparable with those of the TARVA trial (TAR 0.74, 95% CI 0.70 to 0.77; ankle fusion 0.71, 95% CI 0.67 to 0.74). In their cohort, the authors were unable to account for medical comorbidities. 16

Extrapolating the results further than 1 year after the surgery is common in the orthopaedic literature. 76–78 Two models have been used to explore the cost-effectiveness of TAR compared with ankle fusion. 60,62 SooHoo and Kominski62 implemented a simple decision-tree model, which suggested that TAR had the potential to be cost-effective compared with ankle fusion if the implant survived more than 7 years, but these data were obtained when TAR surgery was in its infancy. Courville et al. 60 built a Markov model and, at the lifetime horizon, showed the cost-effectiveness of TAR compared with ankle fusion in a similar hypothetical cohort of patients aged 60 years with end-stage ankle osteoarthritis. The researchers highlighted the lack of data on the QoL of these patients and the requirement for more detailed estimates of both direct and indirect medical costs.

Our study provided the index values using the EQ-5D-5L at baseline, and at 26 and 52 weeks, which allowed us to estimate QALYs and detailed cost estimates from both health-care and PSS, and societal perspectives. Therefore, the results of the model-based analysis are more robust than those of existing studies and we estimate a 69% probability of TAR being cost-effective compared with ankle fusion at the NICE cost-effectiveness threshold of £20,000 per QALY gained. 65 This increases to 72.2% probability when comparing fixed-bearing TAR implants with ankle fusion.

Patient and public involvement

This study had a significant impact on the patients and members of the public who were involved at all stages of the trial. Almost all patients asked to be kept informed, and high-quality newsletters were developed and sent out at regular intervals, summarising recruitment updates and featuring interviews with the research and oversight team and educational insights. More than 1350 people followed the TARVA trial’s Twitter account (@TARVA_Trial, twitter.com; Twitter, Inc., San Francisco, CA, USA). One patient recorded a video log (vlog) for their own blog channel and many patients remain in communication with the trial team. Following publication, the authors intend to present the results of the study widely and work closely with relevant charities to relay the findings to their members.

Limitations

The limitations relate to the pragmatic nature of this study. There is always a conflict between pragmatic studies and perceived robustness. It could be argued that the arms were too heterogenous because surgeons were allowed to use any implant for TAR and any technique for ankle fusion. However, a design in which surgeons used only one implant and one ankle fusion technique would be logistically difficult, especially across sites, and far less generalisable.

A further limitation relates to the use of a patient-reported outcome as the primary outcome measure, which may be insensitive to a clinically meaningful outcome even if one were present. There are many methods used to assess patient-reported outcome measures. In anchor-based methodology, the outcome of interest is ‘anchored’ to someone’s clinical judgement, typically that of a patient or clinician, to define the important difference. In a distribution-based methodology, two approaches are used. The first looks at measurement error and tries to find a consistent difference that patients would consider meaningful and that is also greater than the imprecision of the measurement. The second distribution approach uses a ‘rule of thumb’; for example, a 10% change may be considered important.

The magnitude of the target difference on a standardised scale (standardised effect size) is commonly used to infer the value of detecting this difference in comparison with other possible standardised effects. 79,80 Cohen’s d has been used as de facto justification for this, using a standardised effect size of 0.2, 0.5 and 0.8 for small, medium and large effects, respectively. 81

For our primary outcome measure, the MOXFQ walking/standing domain, there had been no previous RCTs and, hence, the literature used pertained to studies published by the author of the tool. Dawson et al. evaluated the utility of the measure in several cohorts of patients with forefoot, midfoot and hindfoot disorders. 31,32,46,82–84 These studies looked at the change in score from baseline to post surgery. Two main approaches were used to estimate the smallest change on the measure that was likely to be meaningful or important. The first approach was distribution based, that is based on the statistical characteristics of the sample under study. Examples include the effect size, the standard error of measurement and the minimal detectable change. This approach aimed to identify the smallest change for an individual that is beyond the measurement of error of a given instrument and therefore likely to represent a true change. Although Dawson et al. ’s 2007 paper46 looked at hallux valgus, later papers31,32,82–84 specifically studied surgical ankle procedures as a subgroup, estimating the MID to be 10.67 for the MOXFQ walking/standing domain.

Based on this information, we determined that it was important to detect a difference of 12 in the change in MOXFQ walking/standing domain score from baseline between the two treatment arms; it was on this premise that our power calculation was performed. Cohen’s d for a small, medium and large effect size would be between 6 and 24 points based on the SD in our series, which is not too different from the MID determined by Dawson et al. 31,32,46,82–84 More than 80% of patients in our study achieved the MID when comparing their pre-surgery scores with their postsurgery scores. In fact, they exceeded their MID severalfold, with a mean (SD) improvement for TAR patients of 49.9 (30.7) and 44.3 (31.9) for the ankle fusion patients. However, the difference in the changes between the TAR and ankle fusion arms was 5.56 points, with TAR having, on average, an improvement of 5.56 points more than ankle fusion (because a negative score is better). Our CI for the difference in the improvement was –1.37 to 12.49, which included both the 10.67- and 12-point differences defined by Dawson et al. 31,32,46,82–84 Hence, we cannot rule out this being meaningful. It is important to be aware that the MID of 12 was an estimate only.

Another method for determining clinical importance involves opinion-seeking. A value, a range of plausible values or a prior distribution for the target difference is sought by asking one or more ‘experts’ to state their opinion on what would be an important and/or realistic value for a difference. It is possible that once patients’ scores have improved by > 40 points from baseline to 52 weeks an additional 5.56 points may still be clinically relevant. However, on the basis of our estimated MID of 12, overall, the current study showed no significant difference between the groups in our primary outcome measure at 52 weeks post surgery.

Total ankle replacement is more expensive than ankle fusion at 52 weeks. Resource use for these costs were collected from patients during the trial and hence bias due to missing data and loss to follow-up is limited. We considered the societal perspective and the analysis showed that the difference in costs between TAR and ankle fusion may be lower and not statistically significant. Estimating costs from the societal perspective requires more assumptions, such as the length of time off work for those patients who were employed. Because it was considered that most patients would not be at work, patients were not asked this question directly; therefore, the value was estimated based on the average duration of immobilisation after surgery. The cost of lost productivity was calculated using national average gross hourly earnings,55 accounting for sex. The cost of informal care, which is based on time spent taking care of the patient by family and friends, is difficult to estimate and different approaches are used in the literature. 85 We used national average gross hourly earnings as unit costs.

As joint replacements last for several years, not just 52 weeks, we extrapolated our results to the patients’ lifetime horizon. In this situation, TAR was still more expensive than ankle fusion, but the ICER was low, at £4401.81. The model structure was based on existing literature. However, it was simplified and did not account for, for example, possible below-knee amputation or developing ipsilateral arthritis. Important assumptions were made regarding the revision rate in the ankle fusion arm as data on this in the literature are scarce. Decrements in EQ-5D-5L index values after revision are based on estimates in SooHoo and Kominski,62 as these estimates are the only available source of these data. 62 The model has parameter uncertainty, which we accounted for by conducting a PSA. We made the parameters probabilistic by randomly selecting them from appropriate distributions. The results of the PSA show that there is uncertainty in the QALY estimates, but there is a 69% probability of TAR being cost-effective at the NICE threshold of £20,000 per QALY gained. 65 The model is a simplification of reality and the results may change as new evidence becomes available. However, based on the sensitivity analysis, current results are fairly robust.

Generalisability

We designed the trial with the aim of ensuring that the decision-making streams reflected the usual standard of care as closely as possible. Our 17 centres were widely dispersed across the NHS, including district general hospitals, university teaching hospitals and specialist orthopaedic hospitals. Recruitment was performed by experienced surgeons who chose to use the specific technique that they also used in regular NHS practice. Other than MRI, which is invariably part of standard of care, there were no requirements for extra tests or hospital visits. The use of fixed-bearing TAR implants is now dominant in the NHS, with the National Joint Registry (NJR), which covers England, Wales, Northern Ireland, the Isle of Man and Guernsey, showing that these implants were used in over 70% of cases in 2019. 22 The results should therefore be generalisable to standard NHS care. Although one centre recruited 24% of the total patients, several other centres also recruited well, which enhances the generalisability of the results.

Interpretation

Both TAR and ankle fusion improve patients’ QoL at 1 year, but we have not shown one group to be superior in terms of clinical scores at 52 weeks using either ITT or per-protocol analysis. The TARVA trial is inconclusive in terms of the superiority of TAR, as the 95% CI for the adjusted treatment effect includes both a difference of 0 and the MID of 12, but it can rule out superiority of ankle fusion. Both operations appear to be safe, but there were more wound-healing problems and nerve injuries in the TAR arm than in the ankle fusion arm. Seven per cent of patients in the ankle fusion arm went on to symptomatic non-union and are likely to require revision surgery in the future.

When we excluded mobile-bearing TAR and assessed the most common type of implant in the UK (fixed-bearing TAR, representing a 70% market share), we showed a statistically significant improvement of TAR over ankle fusion, suggesting that fixed-bearing TAR may outperform ankle fusion. However, it is important to point out that this is a post hoc analysis and may be inadequately powered. The reason for using post hoc analysis is that, at the time the study began, fixed-bearing TAR was not used in the UK. In 2014, the study onset was delayed owing to the withdrawal of the most commonly used implant in the UK (the Mobility mobile-bearing implant). Therefore, no Mobility implants were used in this study. Between 2014 and 2019, fixed-bearing implants became the dominant implant used and hence this post hoc analysis was considered essential by the investigators.

Using long-term economic modelling, we estimate that there is a 69% probability of TAR being cost-effective compared with ankle fusion at the NICE cost-effectiveness threshold of £20,000 per QALY gained over a patient’s lifetime. 65 This increases to a 72% probability when analysing fixed-bearing implants against ankle fusion.

Recommendations for research

To the best of our knowledge, this is the first level 1 RCT in this area and we would recommend longer-term follow-up of this important cohort of patients with end-stage ankle arthritis. There is a strong case for continuing follow-up, in particular to study the radiological and clinical progress of these patients, and the need for revision surgery.

Although there is a focus on selecting outcome measures that matter to patients, it is clear that studies such as these have to select MIDs based on observations between baseline and a postsurgery time point. Researchers have assumed that the MIDs within groups are the same as the MIDs between groups when both groups have already improved significantly from their baseline scores. We would recommend that studies explore the sensitivity of clinically important differences to patients in this situation.

Contributions of authors

Andrew J Goldberg (https://orcid.org/0000-0002-8650-4503) (Consultant Orthopaedic Surgeon and Visiting Professor) contributed to the study conception and design, the analysis and interpretation of results and draft manuscript preparation.

Kashfia Chowdhury (https://orcid.org/0000-0002-8185-5152) (Medical Statistician) contributed to the analysis and interpretation of results and draft manuscript preparation.

Ekaterina Bordea (https://orcid.org/0000-0002-3772-7049) (Health Economist) contributed to the analysis and interpretation of results and draft manuscript preparation.

James Blackstone (https://orcid.org/0000-0003-4335-5269) (Clinical Project Manager, UCL CTU) contributed to the data collection and draft manuscript preparation.

Deirdre Brooking (R&D Manager, Royal National Orthopaedic Hospital) contributed to the data collection.

Elizabeth L Deane (https://orcid.org/0000-0002-1503-7768) (Clinical Project Manager) contributed to the data collection.

Iva Hauptmannova (Director for R&D, Royal National Orthopaedic Hospital) contributed to the analysis and interpretation of results and draft manuscript preparation.

Paul Cooke (Consultant Orthopaedic Surgeon, Nuffield Orthopaedic Centre) contributed to the study conception and design.

Marion Cumbers (Patient Representative) contributed to the analysis and interpretation of results and draft manuscript preparation.

Simon S Skene (https://orcid.org/0000-0002-7828-3122) (Professor of Medical Statistics) contributed to the study conception and design.

Caroline J Doré (https://orcid.org/0000-0001-9796-4970) (Professor of Clinical Trials and Statistics) contributed to the study conception and design, the analysis and interpretation of results and draft manuscript preparation.

All authors reviewed the results and approved the final version of the manuscript.

Publications

Goldberg AJ, Zaidi R, Thomson C, Doré CJ, Skene SS, Cro S, et al. Total ankle replacement versus ankle fusion (TARVA): protocol for a multicentre randomised controlled trial. BMJ Open 2016;6:e012716.

Thornton J, Sabah S, Segaren N, Cullen N, Singh D, Goldberg A. Validated method for measuring functional range of motion in patients with ankle arthritis. Foot Ankle Int 2016;37:868–73.

Muller P, Skene SS, Chowdhury K, Cro S, Goldberg AJ, Doré CJ, on behalf of the TARVA Study Group. A randomised, multi-centre trial of total ankle replacement versus ankle arthrodesis in the treatment of patients with end stage ankle osteoarthritis (TARVA): statistical analysis plan. Trials 2020;21:197.

Zaidi R, Hargunani R, Calleja M, Foley J, Goldberg A. MRI classification of subtalar joint osteoarthritis using a novel scoring system. Open J Radiol 2020;10:69–78.

Data-sharing statement

All data requests should be submitted to the corresponding author for consideration. Access to anonymised data may be granted following review.

Patient data

This work uses data provided by patients and collected by the NHS as part of their care and support. Using patient data is vital to improve health and care for everyone. There is huge potential to make better use of information from people’s patient records, to understand more about disease, develop new treatments, monitor safety, and plan NHS services. Patient data should be kept safe and secure, to protect everyone’s privacy, and it’s important that there are safeguards to make sure that it is stored and used responsibly. Everyone should be able to find out about how patient data are used. #datasaveslives You can find out more about the background to this citation here: https://understandingpatientdata.org.uk/data-citation.

Disclaimers

This report presents independent research funded by the National Institute for Health and Care Research (NIHR). The views and opinions expressed by authors in this publication are those of the authors and do not necessarily reflect those of the NHS, the NIHR, the HTA programme or the Department of Health and Social Care. If there are verbatim quotations included in this publication the views and opinions expressed by the interviewees are those of the interviewees and do not necessarily reflect those of the authors, those of the NHS, the NIHR, the HTA programme or the Department of Health and Social Care.