Collagenase injection versus limited fasciectomy surgery to treat Dupuytren's contracture in adult patients in the UK: DISC, a non-inferiority RCT and economic evaluation

Dupuytren’s contracture

Dupuytren’s contracture (DC) is a fibro-proliferative disease, characterised by the excessive accumulation of connective tissue in the hand. 1,2 The connective tissue organises into cords which shorten, causing the finger to bend. This condition interferes with hand function and dexterity over time as the affected individual loses the ability to straighten one or more affected fingers. It can have a multifaceted, detrimental effect on quality of life. Individuals living with DC need to adapt to or avoid certain daily activities, and experience embarrassment and anxiety regarding contracture recurrence. 3–5

About 2–2.5 million UK adults are affected, with greater prevalence in males aged over 50 years old, and of northern European descent. 6–8 Smoking and occupations that involve manual labour or forceful stretching of the fascia are associated with increased risk of developing DC. 6

Dupuytren’s contracture is first detected as firm nodules in the palm. These nodules can generate cords which span from the palm to the fingers. When this cord crosses the metacarpophalangeal (MCP) joint and/or proximal interphalangeal (PIP) joint and gradually shortens, the finger is pulled in towards the palm (see Appendix 1, Figure 31). 2,9 The cord contracts over a period of months or years. The most widely used method to establish the severity of this contracture is measurement of the affected joints in both extension (total active extension and passive extension deficit) and flexion (total active flexion), using a goniometer. 10 Medical treatment is usually offered following this, given the cord is mostly irreversible. 2

There is no cure for Dupuytren’s disease, even if the associated skin is radically excised and replaced with a graft, and so the cord can recur. Recurrence has been defined as a change in extension deficit of 6 degrees between 3 and 6 months after treatment or 20 degrees between 3 months and 1 year after correction of the contracture, again assessed using goniometric measurement. 11,12 The rate of recurrence depends on treatment given and certain risk factors. These include occurrence of DC before the age of 50, male gender, presence of Ledderhose disease (see Appendix 1, Figure 32), contractures in both hands (see Appendix 1, Figure 32) and a family history of Dupytren’s. 13–17

Treatments

A comprehensive range of options are used to treat DC, including surgical, pharmacological, and radiotherapy. Physical therapies such as splinting, massage, and ultrasound have no evidence of efficacy. Radiotherapy, which impairs the cells that create the contracting fibrous tissue,18 is occasionally used to manage early-stage DC or as concomitant therapy to other surgical or pharmacological interventions for more aggressive disease. 18

Surgical and pharmacological interventions, such as limited fasciectomy surgery (LF), collagenase injection, and percutaneous needle fasciotomy (PNF) aim to remove, break, or dissolve fibrous tissue of the cord to correct or improve the joint contracture. 2,9 The relative benefits and risks of each of these interventions relate to differences in recovery time, number and severity of treatment complications, and DC recurrence. However, current data on risks and benefits is derived from low-quality, non-randomised studies largely. 19

Inclusion criteria

Patients were eligible if they met the following criteria (and none of the exclusion criteria):

• Male or female and aged 18 years or over.

• Presence of discrete, palpable, contracted cord involving the MCP and/or PIP joint of a finger.

• Degree of contracture ≥ 30 degrees in either joint that is patient cannot put the palm of the hand flat on a table (Hueston’s tabletop test).

• Able to identify a predominant cord for treatment, which would not require more than one collagenase injection as treatment to the joint.

• Appropriate for LF surgery and collagenase injection for DC [i.e. cords suitable for collagenase injection and LF and not requiring skin grafting or PNF (e.g. discrete MCP joint cords in elderly)].

• Willing and able to give informed consent for participation in the study.

Exclusion criteria

• Severe contractures of the MCP and/or PIP joint (Tubiana Grade 4 – total extension deficit > 135 degrees). 41

• History of previous treatment for DC (e.g. surgery, collagenase injection or needle fasciectomy) to the study reference digit.

• History of any other pre-existing disorder of the hand causing significant restriction of movement and/or pain and affecting hand function, for example, post-traumatic stiffness, stiffness due to other causes, infection or arthritis.

• Non-English-speaking because of the need to complete multiple questionnaires which have not been validated in multiple languages.

• Resident in a location where attendance for follow-up at one of the recruiting centres would not be possible.

• Contraindicated for use of collagenase including:

  • Hypersensitivity to: collagenase, sucrose, ketorolac trometamol, hydrochloric acid, calcium chloride dehydrate, and sodium chloride.

• Diagnosis of a coagulation disorder.

• Any other significant disease or disorder (including autoimmune disorders) which, in the opinion of the investigator, may put the participant at risk because of participation in the study, or may influence the result of the study, or the participant’s ability to participate in the study.

• Participation in another research study involving an investigational product in the past 12 weeks.

• Female participants who were pregnant or breastfeeding.

Collagenase Clostridium histolyticum (intervention)

Collagenase was supplied through routine hospital stocks at participating study sites and stored in accordance with the current approved summary of product characteristics (SmPC) for CCH. 28 The investigator used a trial prescription to request this medication from the local site pharmacy department. Once selected for use with a study participant, the intervention was handled as an investigational medicinal product (IMP).

For treatment, either 0.25 ml (MCP joint) or 0.20 ml (PIP joint) of reconstituted solution (0.58 mg CCH) was injected as three aliquots at set anatomical points in accordance with the current approved SmPC. 28 After an interval of 1–7 days, the participant returned to clinic and, under local anaesthetic, the cord was snapped correcting the contracture. This extended period for manipulation, was approved by the MHRA.

When the DISC trial commenced, collagenase was manufactured by Auxilium and marketed by Sobi, in Sweden. However, marketing authorisation for collagenase use within Europe was withdrawn in March 2020 by the parent company (Endo) for commercial reasons. 28 The DISC trial team worked extensively with Sobi to facilitate availability of sufficient vials to enable completion of the study to the contracted target. These efforts were driven by a clear steer by clinicians [site principal investigators (PIs) and co-applicants] that results of this study would have an important bearing on treatment options offered to patients. These efforts were reviewed and supported by the funder (NIHR) and DISC trial oversight committees.

Most treatments were delivered in the same NHS hospital where participants were recruited. However, in a small number of cases treatment occurred elsewhere, in line with local and national guidance and treatment pathways during the COVID-19 pandemic. For example, in line with UK government guidance on non-emergency operations and COVID-19 ‘Green Patient Pathways’, a small number of participants were treated at other UK sites (NHS or private). 42 In addition, to maximise recruitment and treatment following marketing authorisation withdrawal, where supplies at the recruiting site were depleted, if necessary participants could be referred onto other trial sites for collagenase treatment subject to local NHS pathways and approval.

Limited fasciectomy (control)

Under anaesthesia, the diseased fascia, nodule and cord, or a part of it, are removed to correct the joint contracture. 25,43 Following LF, as determined by clinical need, the skin was left to heal by secondary intention, closed directly, closed with a Z-plasty, or using a full thickness skin graft.

Following LF, participants were reviewed at a routine wound check appointment.

Concomitant and care following the trial

Further assessment, interventions and treatments, including collagenase injections, and prescribing of concomitant medications were determined by clinical need and were recorded in follow-up CRFs (see Report Supplementary Material 17). Following completion of study follow-up, participants returned to the care of their treating healthcare professional for any re-intervention if required.

Primary outcome

The primary end point was the score obtained for the 11 items in part 2 of the Patient Evaluation Measure (PEM) 44 at 1 year after treatment (0–100, with higher scores indicating worse outcome).

The PEM is a validated 19-item patient-reported outcome measure comprised of three parts: Part 1 – treatment (5 items), part 2 – hand health profile (11 items) and part 3 – overall assessment (3 items). The score for part 2 (hand health profile) was used as the primary outcome, with the scores for part 1 (treatment) and parts 2 and 3 combined (hand health profile and overall assessment) serving as secondary outcomes The PEM was collected at baseline, just prior to treatment delivery and at 3 and 6 months, and 1 and 2 years (see Report Supplementary Materials 18 and 19).

The inclusion of a patient-reported primary outcome was stipulated by the NIHR commissioned funding call. The PEM was chosen as the primary end point as opposed to other validated measurement tools for the hand, given this can fully capture changes in a patient’s hand health after treatment. The other validated measures were included as secondary outcomes. 45 The primary outcome time point was set as 1 year after treatment to ensure that any associated complications had subsided sufficiently prior to outcome assessment.

Secondary outcomes

Both patient-reported and clinical outcomes were included as secondary outcomes. Patient representatives specifically noted the importance of a return to function as soon as possible following treatment and given the limited available evidence comparing recurrence rates following collagenase injection and LF, relevant measures for each were included and prioritised to allow treatment effectiveness in the context of these key elements to be assessed.

Unité Rhumatologique des Affections de la Main patient-rated outcome measure

The Unité Rhumatologique des Affections de la Main (URAM) was included as a validated, nine-item, six-interval disease-specific disability scale, with higher scores indicating greater difficulty. 46 This outcome was collected at baseline and at 3 and 6 months, and 1 and 2 years.

Michigan Hand Outcomes Questionnaire

The Michigan Hand Outcomes Questionnaire (MHQ), validated for use in this patient group, assesses each hand individually via 63 questions across 6 domains (overall hand function; activities of daily living; work performance; pain; aesthetics; and patient satisfaction with hand function). 47,48 The function and pain domains refer to patient symptoms while work and activities of daily living refer to disability and handicap. Higher scores indicate better overall functioning and satisfaction. This outcome was collected at baseline and at 1 and 2 years.

Objective measures (recurrence, extension deficit and total active movement)

Goniometric measurements of the joints of the reference digit were taken at baseline, immediately prior to treatment delivery and at 3 and 6 months, and 1 and 2 years by qualified NHS practitioners. At baseline and the time points after treatment, sites were asked to provide three repeated goniometric measurements of the active extension, passive extension, and flexion of the joints of the reference digit (i.e. 18 measurements at each time point if the reference digit was the thumb, and 27 measurements if the reference digit was not the thumb). The arithmetic means of the three repeated measurements were used for analysis. At treatment delivery, sites were asked to provide a single goniometric measurement of active and passive extension for each joint of the reference digit (i.e. four measurements in total if the reference digit was the thumb, and six measurements if the reference digit was not the thumb).

These goniometric measurements were used to derive several outcomes: recurrence, passive and active range of movement (RoM), and passive and active extension deficit and stiffness (maximal flexion). For the study, recurrence was defined as a change in extension deficit (as measured by passive extension) of 6 degrees between 3 and 6 months, or 20 degrees from 3 months to 1 year after treatment12,29 at the reference joint. At each time point, RoM for each joint was calculated as the difference between the mean total active flexion measurement and the mean total active extension or extension deficit measurement.

A study-specific manual for performing joint measurements was provided (see Report Supplementary Material 2) and goniometers were required to meet pre-specified criteria (permits measurement of up to 30 degrees of hyperextension; measures flexion to 120 degrees; measures in at least 2-degree increments), to ensure that assessments were standardised.

In addition to goniometric measurements, photographs of the participant’s reference hand (extension, flexion, and anterior-posterior views) were taken. A study-specific manual was used to standardise the images (see Report Supplementary Material 3). If willing, participants also took and returned a photograph of their hand. A study-specific procedure and video were provided to participants to assist with standardisation of these images (see Report Supplementary Materials 4 and 5). The photographs afforded the opportunity for further quality assurance of joint measurements. Clinical members of the team used these images to undertake validation measurements using OsiriX software (Geneva, Switzerland).

Further treatment

All relevant treatment required for the participants’ DC were collected and documented at each follow-up assessment.

Ongoing hand therapy and/or physiotherapy appointments to treat chronic regional pain syndrome, stiffness, swelling or scar problems were referred to as further care. Participants who underwent more than six outpatient follow-up visits for hand therapy and/or physiotherapy had the details of these appointments reviewed on an individual basis to determine whether this extended duration of therapy should be considered further care or routine care.

For the purposes of this report, if participants underwent further collagenase injection, LF, dermo-fasciectomy or PNF to the reference digit at any point during follow-up, then this was counted as further intervention for contracture correction and deemed re-intervention.

Complications

Complications relating to the intervention and control treatments were recorded. Expected complications for the collagenase group are listed in the SmPC. 28 Complications which were expected for the LF group are listed later in Chapter 2 in relation to adverse events (AE).

Complications were reviewed and graded on their severity using an 8-level ordinal classification system from 0 (no complications) to 8 (death). Two clinical observers independently completed grading, with conflicts in grading resolved through discussion.

EuroQol

The EuroQol-5 Dimensions, five-level version (EQ-5D-5L)49 assesses 5 dimensions of health (mobility, self-care, usual activities, pain/discomfort and anxiety/depression) on 5 severity levels (no problems, slight problems, moderate problems, severe problems and unable/extreme problems). A visual analogue scale (VAS) from 100 (best imaginable health) to 0 (worst imaginable health),49 also records participants’ overall evaluation of their health.

This validated, generic, patient-reported health status measure was included to enable assessment of health-related utility and quality-of-life outcomes as required for the cost-effectiveness evaluation (see Chapter 4).

Resource use

Resource use data were collected from hospital records and through participant self-report and documented in study-specific forms. Data collected included health resource use [treatment delivery, inpatient episodes, outpatient visits, emergency hospital admissions, and primary care visits (e.g. GP, nurse and physiotherapy)] in addition to return-to-work and out- of-pocket expenses. Resources were utilised for the cost-effectiveness evaluation detailed in Chapter 4.

Time to recovery of function

Time to recovery of function was assessed using a Single Assessment Numeric Evaluation (SANE) measure,50 a single-question, patient-reported measure, which assesses patient hand functionality. This outcome was collected at baseline, 2 and 6 weeks (see Report Supplementary Material 20), 3 and 6 months, and 1 and 2 years.

Overall hand assessment

At 3 and 6 months, and 1 and 2 years, participants were asked a single global question about the problems they experienced with the hand that was treated compared with the problems they experienced prior to treatment. Responses were given on a seven-item ordered scale from terrible to cured.

Randomisation

Participants were randomly allocated 1 : 1 to one of the two study arms (collagenase injection or LF) using blocked randomisation, with randomly varying block sizes (four or six allocations) and stratification by the designated reference joint (MCP or PIP). The randomisation sequence was amended, with effect from 21 January 2020, to include stratification by centre to account for the limited availability of collagenase following marketing authorisation withdrawal. The study statistician, independent of participating NHS hospitals, generated the randomisation sequence.

To ensure adequate allocation concealment for the study, randomisation was carried out, following completion of baseline assessments, via the internet using a secure, central service hosted by Sealed Envelope Ltd (London, UK). This centralised system recorded information to identify all potential participants and to confirm their eligibility to avoid inappropriate entry of participants into the trial.

Blinding

Given the pragmatic nature of the trial, and the surgical and injection interventions used, it was not possible to blind clinicians or participants to study allocation. It was also not possible to blind the analysing statistician to trial allocation due to the way in which data were collected. To mitigate any impact of this, a statistical analysis plan (available at: www.isrctn.com/ISRCTN18254597) pre-specified all analyses and any changes made to the data set prior to analysis were documented appropriately.

Internal pilot phase analysis

An internal 6-month pilot study was conducted at the start of the recruitment period to check the initial assumptions about recruitment and feasibility of the trial. A summary of the data were provided to the independent Data Monitoring and Ethics Committee (DMEC – described later in Chapter 2), which reviewed the pilot data and made a recommendation to the Trial Steering Committee (TSC – described later in Chapter 2) and Trial Management Group (TMG) to recommend any changes required to the study team and the funding body.

The success of the internal pilot phase was based on the following objectives:

• To set up 6 pilot sites, with a target to recruit 48 participants from these sites

• To ensure that set-up of a further nine sites (inclusive of pilot sites) had been completed

• To monitor closely operational aspects of the trial, including training, eligibility and time to consent, study activity, and participant adherence.

FIGURE 1.

Participant flow diagram.

Findings from the internal pilot study are detailed in Internal pilot.

Statistical methods

A detailed analysis plan was written and agreed with the DMEC prior to completion of recruitment. Brief details of the analyses undertaken are given below (please refer to the analysis plan available at www.isrctn.com/ISRCTN18254597 for further details), and any deviations from the planned analyses are detailed and justified below. All analyses were undertaken at the end of the follow-up period using all available data; hence no stopping rules or associated adjustments for multiplicity were required. All analyses included all participants with data available for the relevant outcome in the groups that they were allocated to, except for the imputed data analyses that included all randomised participants. Analyses were conducted using Stata/SE v17.0 (StataCorp LP, College Station, TX, USA).

Baseline data were summarised by allocation and overall, both as-randomised and as-analysed. The as-randomised set comprised all randomised participants, excluding any ineligible patients randomised in error. The as-analysed set comprised all participants included in the primary analysis (i.e. all participants with primary outcome data available for at least one time point after treatment).

For all outcomes/time points, continuous outcome data were summarised in terms of the non-missing sample size, mean, SD, median, interquartile range and range, and categorical data in terms of frequencies and proportions. A Consolidated Standards of Reporting Trials diagram was used to summarise participant flow and data completeness. 52

Data collection, case report form processing and data checks

A comprehensive data management plan was generated at the outset of the trial to document details of the data processing.

Receipt of completed CRFs at the United Kingdom Clinical Research Collaboration-registered University of York Trials Unit (YTU) were recorded in a bespoke research data management system.

Data from completed CRFs were entered using an automated, electronic system (Teleform, Waterloo, Canada) in accordance with the licence held by the YTU. Computerised data cleaning checks and validation rules were applied to review for completion and accuracy of key variables required for the statistical analysis, check for discrepancies, and ensure consistency of the data.

Where discrepancies were identified, these were raised with the local research teams and changes made in accordance with good clinical practice (GCP).

An electronic audit trail system was maintained to track all database data changes and regular backups of the electronic data were also performed.

Promoting participant retention and follow-up completion

Participants received £40 following completion of each of the 1- and 2-year participant outcome questionnaires given this has found to have an effect in improving participant retention and questionnaire response rates. 55,56

Participants were also sent a study newsletter 4 weeks before the 1-year time point to maintain trial engagement and to encourage attendance at the 1-year study visit. This was accompanied by a cover letter from the study chief investigator to thank them for their continued contribution to the study.

Trial sites were contacted by the YTU to ensure that visits for the 1-year primary outcome time point were arranged accordingly. Where a visit could not be arranged within 4 weeks of the visit due date, a postal questionnaire was sent to the participant. If there was still no response after a further 4 weeks, the participants were telephoned to collect their data.

Pressures on the NHS due to the COVID-19 pandemic and associated national restrictions and guidelines had significant implications on research. Cancelling/delaying of clinic appointments, patient concerns about COVID-19 risk and attending hospital, and NHS research capacity strains meant that remote data collection methods were implemented on DISC to ensure that follow-up data could be collected during this time. 57 Remote methods were used to collect site-reported (telephone or video consultations) and participant-reported (postal or telephone questionnaires) data. At sites where COVID-19 burden meant that research nurses were redeployed, follow-ups were temporarily supported by the YTU.

Two nested, randomised retention studies within a trial were undertaken to ascertain effectiveness of retention strategies: one evaluated the effectiveness of a thank you card, the other the effectiveness of a festive greetings card (implemented December 2019). Details of these studies will be reported separately.

Discontinuation or withdrawals of participants

Trial participants had the right to withdraw from the study at any time. In addition, the investigator could discontinue a participant from the study at any time if they considered it necessary for any reason. The reason for withdrawal was recorded in the relevant CRF (see Report Supplementary Material 14).

Participants who requested to withdraw during a study visit were asked if they were willing to complete the questionnaires prior to withdrawal. Where a participant requested to withdraw fully outside of a scheduled study visit, no further follow-up questionnaires were completed.

Unless the participant specifically withdrew consent for their data to be stored, all data collected from them continued to be stored as per the original participant consent. At a participant’s request, their data collected up to the point of withdrawal could however be withdrawn from the trial and would not be used in the final analysis.

Where participants requested remote follow-up (i.e. follow-up without clinic visits), the research nurse contacted the participant at each visit time point to complete a safety assessment (AE reporting).

Confidentiality and data protection

The DISC trial was conducted in accordance with the Declaration of Helsinki, relevant regulations, International Council for Harmonisation Guidelines for GCP, the Data Protection Act 2018, and the General Data Protection Regulations in place during the trial.

To ensure participant confidentiality, participants and associated data were identified by initials and a unique four-digit participant ID number.

All documents were stored securely, accessible only by delegated trial staff and authorised personnel.

For the photography substudy, photographs of participants’ hands were anonymised prior to electronic transfer between sites, the YTU and the University Hospitals of Leicester NHS Trust. Similar processes were also used for the qualitative substudy. Details on confidentiality for these components are available in Chapter 5 and Appendix 2.

In accordance with applicable regulations, authorised persons could review data at any time during the study to verify that the study was being carried out correctly. However, this was not required during the study.

All study documentation will be retained in accordance with UK law, following which data will be disposed of securely.

Adverse event reporting

Information regarding event description, onset and end date, relationship to study medication, and action taken were recorded on study-specific reporting forms (see Report Supplementary Material 21–23). NSAE were required to be reported within 5 days and SAE within 24 hours of being made aware of the event.

Events related to the study medication or procedures were followed up until resolution or the event was considered stable. Related events which resulted in participant withdrawal from the study were followed until a satisfactory resolution was achieved.

The relationship of SAEs/NSAEs to the study medication was assessed by a medically qualified investigator. For SAEs event details, causality and expectedness were reviewed by the chief investigator or other delegated medic.

Should any event have been deemed to be a SUSAR, then these would have been reported to the MHRA and REC within 7 days for fatal or life-threatening SUSARs, and 15 days for all other SUSARs.

All events were routinely reported to the TSC, DMEC and sponsor. Annually throughout the trial, a Developmental Safety Update Report was submitted to the MHRA and REC, detailing patient safety considerations.

Trial Management Group

The TMG met quarterly to oversee the management of the trial. The group included the chief investigator, co-investigators, and members of the YTU and the Academic Team of Musculoskeletal Surgery (AToMS) responsible for the day-to-day management of the study. A representative of the sponsor also attended when available.

The TMG meetings monitored the progress of the DISC trial in relation to recruitment (e.g. enrolment, consent, and eligibility), allocation to study groups, adherence of the trial interventions to the protocol, retention of trial participants, monitoring of (S)AEs and reasons for participant withdrawal.

Trial Steering Committee

An independent TSC was appointed by the funding body (NIHR) to provide overall supervision of the trial and to advise on its continuation. Meetings were held on a bi-annual basis during the study. The TSC comprised two independent members (one clinician and one methodologist) and a patient and public representative. Membership is detailed in the Additional information section.

Data Monitoring and Ethics Committee

An independent DMEC was appointed by the funding body (NIHR) to monitor trial data in respect of any ethical or safety reasons why the trial should not continue. Access to unblinded data to facilitate this review was provided if required.

Meetings were held on a bi-annual basis during the study. The DMEC comprised three independent members (two clinicians, one statistician) and a patient and public representative. Membership is detailed in the Additional information section.

Site monitoring

In accordance with regulatory approvals, regular monitoring was required for the DISC trial, and a monitoring plan was prepared accordingly.

A combination of onsite and remote SIVs, including comprehensive study training, were conducted prior to study activity commencing at each site. During onsite visits, the pharmacy department storage facilities were reviewed. Where visits were completed remotely, the pharmacy team provided confirmation of temperature monitoring arrangements and imaging of storage arrangements (if the IMP was to be held outside of the main site pharmacy).

An initial on-site monitoring visit was completed with all sites (except one due to COVID-19 restrictions) following recruitment of three participants, or once 8–12 weeks had elapsed since the study opened to recruitment at the site. Scheduling was amended once three participants had been recruited and treated or at 18 weeks after recruitment activity commencing (with an interim review at 8–12 weeks) to enable a review of pharmacy activity at the visit. Source data verification (about 20% of CRF data), a review of AE, pharmacy processes and accountability, and the investigator site file were completed.

A second monitoring visit was completed remotely using self-completed checklists to confirm investigator site file maintenance and provision of relevant information in participants’ medical records.

Centralised monitoring included CRF completion checks (with a specific focus on the confirmation of eligibility CRF) and a 100% check of consent.

Study closure was completed remotely using a close-out checklist to confirm investigator site file contents and documentation in participants’ medical records.

Primary outcome: completeness and timelines

Of the 672 participants randomised, 621 received treatment as part of the DISC trial, 295 (88%) in the LF group and 326 (97%) in the collagenase group. Of these, 531 (85.5%) had primary outcome data (PEM Hand Health Questionnaire) available at 3 months, 514 (82.8%) at 6 months, 534 (86.0%) at 1 year (the primary end point) and 426 (68.6%) at 2 years. Overall, there were 599 participants with primary outcome data available for at least one time point after treatment (89.1% of randomised, 96.5% of treated), all of which are included in the primary analysis model. The main difference between groups in terms of completeness of follow-up data was driven by participants in the surgery arm having only baseline data, primarily due to more of these participants not receiving treatment. Once participants were treated, the broad patterns of follow-up data completeness were similar across groups (see Appendix 6, Tables 73 and 74 for details regarding the baseline data completeness and Appendix 6, Tables 75–78 and Figures 76–79 for time frames for PEM completion at each time point). There is some evidence that rates of primary outcome completion were higher among participants that had complications reported than among participants that had no complications reported (see Appendix 6, Table 79). Reported of complications being associated with higher rates of retention is not surprising given that complications would have been reported only for participants attending follow-ups, during which they would also generally have provided a response for the primary outcome.

Primary outcome: descriptive summaries

Summaries of the available PEM scores (the primary outcome) at each time point are given in Table 13. The marginal distributions of the PEM scores at 1 year (the primary end point) in each group are shown in Figure 6. Finally, the (raw/unadjusted) mean PEM score trajectories by time point and allocation are shown in Figure 7.

FIGURE 6.

Patient Evaluation Measure Hand Health Questionnaire scores at 1 year by allocation. Kernel density estimates, available cases only.

FIGURE 7.

Patient Evaluation Measure mean score profiles by allocation. Vertical dashed line plotted at the median time between randomisation and treatment with subsequent follow-ups referenced to this time point.

Both Table 13 and Figure 6 clearly show the similarity of the mean/expected PEM scores at both baseline and immediately prior to treatment, albeit with some evidence of a slight increase (worsening) of PEM scores between these time points (in both groups). Following treatment, the means of the available scores in both groups are smaller (better) than at baseline and pre-treatment. In general, the scores at all follow-up time points are towards the lower (i.e. better) end of the PEM scoring scale, with approximately 75% of participants having scores between 0 and 20 (out of 100) at each time point after treatment and over 25% of available cases in the surgery group having scores of zero at 1 and 2 years. This abundance of scores at or near zero is evident as skew in the marginal distribution of the 1-year scores in Figure 6, and is indicative of poor discrimination of different levels of DC-related disability at this more favourable end of the scale (i.e. a ceiling effect).

Primary outcome: primary analysis

While the ceiling effects noted above mean the primary outcome data show some departure from the assumptions of the planned primary analysis, we believe the benefits of adhering to the pre-specified analysis outweigh the statistical limitations resulting from using a poorly specified analysis model. We therefore report the estimates and tests based on the planned primary analysis in the first instance. However, we also report results from a series of univariate semiparametric analyses (based on ordinal regression) that are more robust to the ceiling effects observed. These analyses were specified as contingencies in the Statistical Analysis Plan.

The point and interval estimates of the treatment effects at 3 and 6 months, and 1 and 2 years from the planned primary analysis are given in Table 14, together with two-sided p-values for tests of H0:δ=0 [where δ denotes the true, unknown difference (collagenase – LF) in expected PEM score at the relevant time point]. Also given is the p-value for the planned non-inferiority comparison of the primary end point – that is the p-value for a one-sided test of H0:δ≥6 at 1 year [with 6 points (6%) being the pre-specified non-inferiority margin used to plan the trial]. The point and interval estimates from the planned primary analysis, and the pre-specified non-inferiority margin are also illustrated in Figure 8. The estimates and relevant tests from the univariate semiparametric analyses are given in Appendix 6, Table 80.

FIGURE 8.

Treatment effect estimates at 3 months, 6 months, 1 year and 2 years obtained from the primary analysis. Vertical dashed line plotted at median time to treatment delivery.

The estimates in Table 14 and Figure 8 indicate substantial variation in treatment effect across the four follow-up time points. Broadly they suggest there was some initial benefit of collagenase over LF in the short term (up to about 3–6 months after treatment), but that LF provided greater benefit in the longer term. This aligns with the profiles plotted in Figure 7.

At 3 months there appears to have been a small-to-moderate treatment effect in favour of collagenase, with the participants allocated to this treatment having PEM scores about 1–6 points lower (‘better’) on average than those allocated to LF. The p-value of 0.003 for the test of H0:δ=0 suggests substantial discrepancy between the test hypothesis (i.e. that allocation had no impact on expected PEM score at 3 months) and the data observed. By 6 months, the initial additional benefit of collagenase waned, and there is little evidence of any important differences between the groups at this time point. The supplementary semiparametric analyses of the data from these time points yield similar results to those from the planned primary analysis.

The primary end point was the difference in PEM scores at 1 year after treatment, with this end point being used to test the hypothesis that collagenase is inferior to LF by a clinically relevant amount (≥ 6 points). The results in Table 14 suggest the data provide little evidence to reject the hypothesis that collagenase is inferior to LF (with regard to the primary end point). In other words, the observed data are compatible with collagenase being worse than LF (with respect to the primary end point) by a clinically relevant margin. For example, the point estimate for the difference in expected PEM score at 1 year is approximately equal to the non-inferiority margin of 6 points, and the upper tail of the interval estimate nearly covers values as large as 9 points, 1.5 times the magnitude at which differences are hypothesised to be of clinical importance. The p-value for the relevant test of inferiority (i.e. H0:δ≥6) is about 0.49, again showing that there is almost no information in the data to support rejection of the hypothesis that collagenase is inferior to LF (with respect to the primary end point). The estimates from the supplementary semiparametric analysis of the 1-year primary outcome data are slightly attenuated compared with those from the planned primary analysis, but the tests reach the same conclusions (at least at a 5% two-sided or 2.5% one-sided significance level), and the substantive findings are practically identical.

The apparent benefit of LF over collagenase appears to have continued to increase slightly between 1 and 2 years. By the later time point, participants allocated to LF appear to have scores that are about 4–11 points better on average than participants allocated to collagenase. This interval estimate again encompasses differences that are likely to be of clinical importance (e.g. the upper limit is close to double the non-inferiority margin used for the primary end point). Hence there is again little evidence to support rejection of the hypothesis that collagenase is inferior to LF with regard to PEM Hand Health Questionnaire scores at 2 years. The estimates and p-value from the supplementary semiparametric analysis of the 2-year primary outcome data are like those from the planned primary analysis, and the conclusions remain the same.

In conclusion, our primary analysis shows that there is little evidence to support rejection of the hypothesis that collagenase is inferior to LF at 1 and 2 years post treatment. Indeed, the observed data are highly compatible LF being superior to collagenase with regard to the primary outcome measure at both these time points.

Primary outcome: missing data sensitivity analyses

The primary analysis reported in the previous section is valid under the assumption that any missing primary outcome data (at any follow-up after treatment) is MAR with respect to the predictors included in the analysis model (i.e. allocation, study reference joint type, baseline score, and recruitment site) and any observed primary outcome data. In this section, we report the results from several planned analyses to investigate the sensitivity of results to variation in, and departures from, the MAR assumption outlined above.

A summary of the overall patterns of missing primary outcome data is provided in Appendix 6, Table 81. The main differences in primary outcome data completeness at follow-up are driven by differences in receipt of treatment (12% lost prior to treatment in the LF group vs. 3% in the collagenase group). Hence, allocation clearly affected the probability of receiving treatment. If there were other baseline characteristics X (e.g. baseline disease severity, sex, age, etc.) that also affected the probability of receiving treatment and affected outcomes after treatment, then analysing only those that were treated (as in the primary analysis) would result in confounding of the relationship between random allocation and outcome (through a back-door path between allocation and outcome via X induced by selection on treatment receipt). We investigated the possible impact of selection on treatment receipt by repeating the primary analysis, including additional baseline predictors of treatment receipt (as well as outcome missingness) as covariates. We also imputed primary outcome data for those that were not treated under different MAR and MNAR scenarios.

Empirical cumulative distribution functions (ECDF) of the baseline PEM scores are shown in Figure 9, stratified by primary analysis inclusion status (included or not included) and allocation.

FIGURE 9.

Empirical cumulative distribution functions for baseline PEM scores by inclusion/exclusion from primary analysis and allocation.

The left-hand panel of Figure 9 suggests that participants in the surgery group who were not included in the primary analysis had a similar distribution of baseline PEM scores to those that were. In contrast, the right-hand panel suggests that collagenase participants that were excluded from the primary analysis had slightly higher (worse) baseline PEM scores than those that were included. Given the sparse data (particularly with regard to collagenase participants excluded from the primary analysis), it is difficult to draw any firm conclusions about differing selection mechanisms across the randomised groups. However, overall these plots suggest that the LF participants included in the primary analysis are representative of those excluded in terms of their baseline disability/disease severity, whereas those excluded from the primary analysis in the collagenase group may be those with slightly poorer prognosis than those included.

Figures 10 and 11 illustrate the distributions of the available PEM scores at 3 and 6 months stratified by whether primary outcome data were available at 1 year (the primary end point) or not. In the collagenase group, both 3- and 6-month PEM scores were generally higher (worse) for participants who were missing 1-year PEM scores, whereas in the LF group the distributions of these earlier measurements were about the same across the two subgroups.

FIGURE 10.

Empirical cumulative distribution functions of the available PEM scores at 3 months by presence/absence of primary end-point data and allocation.

FIGURE 11.

Empirical cumulative distribution functions of the available PEM scores at 6 months by presence/absence of primary end-point data and allocation.

The following baseline variables were assessed for association with different patterns of missing data: age; number of digits affected by DC; alcohol consumption (binary); comorbidities (binary); sex (binary); smoking status (never, previous, current); previous treatment for DC using surgery or collagenase (binary); general health (as measured by the EQ-5D-5L VAS), MHQ score and URAM score. The following variables were found to be at least weakly associated with one or more of the missing data patterns assessed: age; number of digits affected; sex; smoking status and EQ-5D-5L general health VAS.

The estimates from a re-analysis of the primary outcome including these additional predictors of missingness are reported in Table 15. Compared with the planned primary analysis results (see Table 14), it is evident that the inclusion of these additional baseline covariates has made little difference to the estimated treatment effects (all are within about 0.1 of each other). The estimate for the primary end point is very slightly larger than obtained for the primary analysis, suggesting slightly greater additional benefit from LF, but the substantive conclusions are identical.

The results of the analyses using multiply imputed data are reported in Table 16. Again, these are broadly in line with those obtained for the primary analysis, albeit suggesting slightly greater additional benefit of surgery at 6 months, 1 and 2 years, and slightly smaller additional benefit from collagenase at 3 months. Overall, the substantive conclusions of the primary analysis seem insensitive to slightly weaker MAR assumptions, with these analyses suggesting even less discrepancy between the hypothesis of inferiority and the observed data.

To investigate the sensitivity of the primary analysis to various systematic departures from MAR at 1 year, we undertook a sensitivity analysis using a mean score and pattern mixture modelling approach. The estimated differences in PEM score at 1 year, under different departures from MAR (i.e. γ=E[Y|X,missing]−E[Y|X,notmissing]) are illustrated in Figure 12. All missing not at random (MNAR) scenarios considered assume that missingness is associated with poorer (i.e. higher) responses for the primary outcome (conditional on the covariates in the substantive analysis – reference joint and baseline PEM score). Figure 12 shows that for the range of γ considered, the impact of the primary outcome data being MNAR is limited if the MNAR assumption is across randomised groups (see dark blue line in figure). For example, the upper limit of the two-tailed 95% CI estimate is > 6 (the non-inferiority margin) across all scenarios considered, meaning the hypothesis that collagenase is inferior to LF (with regard to the primary end point) is not rejected even under quite extreme departures from MAR (provided these are identical across groups). The conclusions of the primary analysis are also relatively unaffected by the outcome data being MNAR in just the collagenase arm (for the range of δ considered) as would be expected. If the primary end-point data are assumed to be MNAR in the LF group, but MAR in the collagenase arm, then the outcome of the test of non-inferiority would only result in rejection of the null hypothesis (that collagenase is inferior to surgery with respect to the primary end point) if γ≥7.5. While a discrepancy in conditional expectation between those with/without missing primary end-point data is possible, it is highly implausible that this discrepancy would occur only in the surgery arm. Overall, the results in Figure 12 provide no compelling reason to seriously doubt the results of the primary analysis.

FIGURE 12.

Estimates of the difference in expected PEM score at 1 year under various departures from MAR.

Primary outcome: further sensitivity analyses

We undertook several pre-specified further analyses of the primary outcome to investigate the sensitivity of the primary analysis results to differences in time to treatment between groups and departures from the planned follow-up schedule.

Departures from planned follow-up schedule

To assess the sensitivity of the results of the primary analysis to departures from the planned follow-up schedule, we undertook 2 additional analyses of the primary outcome data, 1 pre-specified and 1 post hoc. The pre-specified analysis was to repeat the primary analysis including only primary outcome data collected within the protocol specified windows for completion. The results of this analysis are reported in Table 18. The post hoc analysis (see Appendix 6, Table 82) aimed to directly model the effects of time from treatment using generalised least squares (with a random intercept for recruitment site). This analysis modelled the effects of time from treatment using a four-knot restricted cubic spline (knots placed at 3 and 6 months and 1 and 2 years), and within patient correlation using an exponential covariance structure for the residual errors. This model was used to derive point and interval estimates at 3 and 6 months and 1 and 2 years to facilitate comparison with the results of the primary analysis (see Table 14), as well as estimates of the treatment effect over all time points up to 27 months (see Figure 13).

FIGURE 13.

Point estimates and pointwise two-sided 95% confidence band for the difference in expected PEM score. Positive values indicate greater benefit from LF than collagenase. The red diamond denotes the non-inferiority margin for the primary end point.

TABLE 18 - Treatment effect estimates at 3 and 6 months and 1 and 2 years obtained from an analysis including only primary outcome data collected within the protocol specified follow-up windows

Two-sided, based on t-test with degrees of freedom calculated using the Kenward–Roger method.

One-sided, based on t-test with degrees of freedom calculated using the Kenward–Roger method.

NoteBoldface denotes the primary endpoint.

	Estimated difference (collagenase – LF) (95% CIa)	p-valuea H0: δ = 0	p-valueb H0: δ ≥ 6
Month 3	−4.30 (−7.08 to −1.51)	0.0025	–
Month 6	2.41 (−0.61 to 5.43)	0.1174	–
1 year	6.09 (3.29 to 8.89)	< 0.00005	0.5252
2 years	8.33 (5.00 to 11.66)	< 0.00005	–

The treatment effect estimates in Table 18 (i.e. those obtained from the analysis excluding primary outcome data collected out of window) are similar to those obtained from the primary analysis (see Table 14), albeit slightly larger in absolute value at all time points, and the various tests all result in the same inferences. The estimates in Figure 13 and Appendix 6, Table 82 (from the model treating time from treatment as a continuous predictor) are again pretty similar to those obtained from the primary analysis. Estimates from a similar analysis with a different placement of the spline knots gave similar results. The extrapolated plot in Figure 13 suggests substantial short-term benefit from collagenase that quickly disappears and is eventually reversed at about 5–6 months after treatment. In fact, the upper limit of the two-sided 95% confidence band exceeds the pre-specified non-inferiority margin of 6 points from about 7 months after treatment onwards, with the difference in favour of LF continuing to steadily grow until the end of the follow-up period. Overall, these analyses suggest that the impacts of mistimed measurements of the primary outcome are minor and are unlikely to be important in driving the substantive conclusions of the primary analysis.

Primary outcome: further analyses

Primary outcome: subgroup analyses

We undertook two subgroup analyses to investigate the presence and extent of treatment effect heterogeneity associated with baseline treatment preference (preferred collagenase, preferred surgery, no preference), and designated study reference joint (MCP or PIP). Brief summaries of these baseline variables for the participants with primary outcome data for at least one time point after treatment are given in Appendix 6, Tables 87 and 88.

Treatment effects by subgroup and time point were estimated via addition of: the main effects of the subgroup variable to the primary analysis model (if not already present), the two-way interactions between subgroup and treatment group, and the three-way interactions between subgroup, treatment group and time point. Estimates by preference subgroup (and time point) are given in Appendix 6, Table 89 and Figure 80 and estimates by study reference joint subgroup (and time point) are given in Appendix 6, Table 90 and Figure 81.

The data provide little information regarding the effects of allocation in the subgroup of participants that preferred LF at baseline (due the small number of participants that reported preference for surgery at baseline). There are no clear differences in the treatment effects observed in this group compared with the other two baseline preference subgroups, but at the same time the observed data do little to rule out anything but large effects in the opposite direction to those of the other subgroups. The estimates for the two larger subgroups (prefers collagenase and no preference) appeared consistent with one another, despite the effects in the group that preferred collagenase being consistently smaller (i.e. slightly more in favour of collagenase) than those in the no preference subgroup. However, the interval estimates show considerable uncertainty regarding the subgroup specific effects even in these larger subgroups, and important differences between these groups cannot be ruled out based on the data alone.

The data are most compatible with the treatment effects in each reference joint subgroup being similar across all time points (and following a similar pattern across time points to those obtained on average for the whole sample) (see Appendix 6, Figure 81 and Table 90). However, the interval estimates cover a wide range of treatment effects (particularly those for the PIP stratum), and overall these data do little to rule out small-to-moderate variation in treatment effects across reference joint strata.

Patient Evaluation Measure Overall Assessment Questionnaire

Summaries of the available PEM Overall Assessment scores at 3 and 6 months and 1 and 2 years are given in Appendix 7, Figure 82 and Table 91. These show a similar pattern to the PEM Hand Health Questionnaire (primary outcome) scores with a small difference in favour of collagenase at 3 months, a small difference in favour of LF at 6 months, and moderate-to-large differences in favour of LF at 1 and 2 years.

Patient Evaluation Measure Treatment Questionnaire

Summaries of the available PEM Treatment Questionnaire scores at 3 and 6 months and 1 and 2 years are given in Appendix 7, Figure 83 and Table 92. These show that almost all participants reported that their overall treatment process was broadly positive with regard to the constructs measured, with almost all participants having a score < 5 (equivalent to giving the best or second-best response for all items). As such, there are no obvious or important trends or differences across groups, suggesting participants had a similarly positive experience regardless of the treatment they were allocated.

Unité Rheumatologique des Affections de la Main

Summaries of the available URAM scores are reported in Table 19 by time point and allocation. The approximate distributions of the available URAM scores are plotted in Figure 14 by time point and allocation. The profiles of the mean PEM, URAM, MHQ, and SANE scores (based on the available responses) over time are plotted in Figure 15.

FIGURE 14.

Kernel density estimates – URAM scores.

FIGURE 15.

Mean PEM, URAM, MHQ and SANE score profiles by allocation. Vertical dashed line plotted at the median delay between baseline and treatment, with all subsequent time points referenced to this. For the PEM (a), higher scores indicate greater disability/poorer hand health; URAM (b), higher scores indicate greater difficulties/poorer function; MHQ (c), higher scores indicate better function, less pain and greater satisfaction; and SANE (d) higher scores indicate better function.

Estimates from the planned analysis of the URAM are given in Table 20 and illustrated in Figure 16. The point estimates suggest greater benefit (on average) from LF than from collagenase at all time points, with this additional benefit steadily increasing over time.

FIGURE 16.

Treatment effect estimates (point estimates and two-sided 95% CIs) at 3 and 6 months and 1 and 2 years. Values above zero indicate greater benefit from LF than from collagenase. Vertical dashed line plotted at the median time between randomisation and treatment delivery with subsequent time points referenced to this.

At 3 months there is relatively weak evidence of any important differences between groups. The upper limit of the interval estimate (i.e. 1.84) is considerably smaller than the published point estimate of the minimal clinically important difference for this instrument of 2.9 points (± 2.6)46 and the p-value suggests the observed difference would be relatively unsurprising even if there were truly no difference between groups. At 6 months there is reasonably strong evidence that LF is superior to collagenase, with the hypothesis of no difference being rejected at around the 0.01% level. However, there is no clear evidence of a difference of clinically important magnitude, with much of the interval estimate falling below the estimated minimal clinically important difference of 2.9. That said, hypotheses positing differences as large as 2.9 are reasonably compatible with the observed data, and important differences at 6 months cannot be ruled out. By 1 year there is strong evidence of greater benefit from surgery, with hypotheses positing no difference being highly incompatible with the observed data. In addition, much of the 95% CI estimate lies above or around 2.9, suggesting little discrepancy between the observed data and true differences of clinically important magnitude. By 2 years there is clear evidence of the superiority of surgery over collagenase, in terms of both statistical criteria (i.e. differences of zero are strongly rejected) and substantive criteria (i.e. differences of a clinically immaterial magnitude are strongly rejected).

The results from the semiparametric analyses (reported in Appendix 7, Table 93). The estimates and tests of H0:δ=0 [where δ denotes the true difference (collagenase – LF) in expected score] from these analyses broadly align with those from the planned analysis.

Michigan Hand Outcomes Questionnaire

Summaries of the available MHQ scores (total scores obtained for the reference hand) are reported in Table 21 by time point and allocation. Further information regarding the distribution of the 1- and 2-year follow-up scores is illustrated in Figure 17. The profiles of the mean MHQ scores (based on the available responses) over time are plotted in Figure 15. See Appendix 7, Tables 94–96 for summaries of the MHQ data at each time point broken down by subscale.

FIGURE 17.

Kernel density estimates by allocation – MHQ scores at 1 and 2 years post treatment.

As with the URAM scores, the MHQ scores at baseline are similar across groups for both the randomised and treated populations, and as with the other PROMs, the overall post-treatment measurements are considerably better than the baseline measurements. At 1 year post treatment the scores are generally higher (better) in the LF group compared to the collagenase group. Like both the PEM and the URAM, this gap appears to grow between 1 and 2 years, with the mean score in the collagenase group decreasing slightly between 1 and 2 years, while remaining stable in the LF group. The ceiling effects seen for the PEM and URAM are also evident for the MHQ scores.

Estimates from the planned analysis of the MHQ scores are reported in Table 22 and illustrated in Figure 18. These show evidence of better outcomes in the LF group at 1 and 2 years post treatment, with mean scores in the LF group being about 2–7 points higher at 1 year, and about 4–10 points higher at 2 years, compared with the mean scores of the collagenase group. The p-values suggest substantial discrepancy between the observed data and hypotheses posting no difference. Previous studies suggest the minimum clinically important difference for the MHQ is about 1–2 points. 60 Hence the estimates in Table 21 suggest little discrepancy between the observed data and hypotheses positing clinically important differences in outcomes at 1 and 2 years post treatment (i.e. the data do little to refute claims that there exists a clinically important difference between treatment with regard to these outcomes).

FIGURE 18.

Treatment effect estimates (point estimates and two-sided 95% CIs) at 1 and 2 years post treatment. Values below zero indicate greater benefit from LF than from collagenase. Vertical dashed line plotted at the median time between randomisation and treatment delivery with subsequent time points referenced to this.

The results from the contingency semiparametric analysis (reported in Appendix 7, Table 97). These results paint a similar picture to those of the planned analysis, albeit with slightly larger point estimates at 2 years, and the overall conclusions remain consistent under this alternative approach/analysis.

Single assessment numeric evaluation

Summaries of the available SANE scores are given in Appendix 7, Table 98 by time point and allocation.

The estimates from the planned analysis of the SANE outcome are reported in Table 23 and plotted in Figure 19. Functional outcomes are better in the short term among those allocated to collagenase, with this initial benefit quickly waning and reversing at about 3–6 months. From about 6 months post treatment onwards, the apparent additional benefit of LF over collagenase continues to grow, resulting in estimates at 1 and 2 years that strongly suggest LF to be the superior treatment with regard to longer-term functional outcomes.

FIGURE 19.

Treatment effect estimates (point estimates and two-sided 95% CIs) at 2 and 6 weeks, 3 and 6 months, and 1 and 2 years post treatment. Positive values indicate greater benefit from collagenase than from LF. The vertical dashed line is plotted at the median time between randomisation and treatment delivery with subsequent time points referenced to this.

The estimates at 2 and 6 weeks provide strong evidence that treatment with collagenase results in more rapid recovery of function compared with LF. However, the additional benefit of collagenase over LF is not sustained beyond about 3 months post treatment, and important differences between groups are evident from about 1 year post treatment onwards. Similar conclusions applied when semiparametric analyses were conducted (see Appendix 7, Table 99).

Overall hand assessment

Descriptive summaries of the responses by time point and allocation are given in Appendix 7, Table 100 and in Figure 20.

FIGURE 20.

Overall hand assessment responses by time point and allocation.

We used a proportional odds model to estimate the effect of allocation on the log-odds of being in the ‘higher’ (worse) category for each natural (i.e. order respecting) dichotomisation of the ordinal scale. We used this model to estimate the absolute difference (between randomised groups) in the probability of providing a response of ‘Cured’ or ‘Much better’ at 1 year. The results of the planned analyses of this outcome at 1 year post treatment are reported in Table 24. There was limited evidence of any important departures from proportional odds with regard to the treatment effect (p = 0.22), hence just the results in Table 24 are reported for this outcome.

As expected, based on the raw summary data (see Appendix 7, Table 100), the results in Table 24 show strong evidence that participants allocated to LF were more likely to respond with more positive response than similar participants allocated to collagenase. The estimates of the odds ratio (OR) suggest that for any natural (i.e. order preserving) dichotomisation of the ordinal scale, allocation to collagenase increases the odds of being in the worse category (of the dichotomy) by about two- to fourfold, compared to the odds of the same outcome following allocation to LF.

This apparent shift toward poorer outcomes/less benefit following allocation to collagenase results in substantial differences in the ‘risk’ of a response of ‘Cured’ or ‘Much better’ when mapped to the absolute risk scale. For example, a participant with MCP joint designated as the study reference joint, and baseline PEM score equal to the mean of the observed baseline PEM scores, that is allocated to LF, is estimated to be about 9–18% more likely (in absolute terms) to provide a response of ‘Cured’ or ‘Much better’ at 1 year post treatment than a similar participant allocated to collagenase. Even larger treatment effects are apparent for participants with the PIP joint designated as the study reference joint (and baseline PEM score equal to the mean of the observed baseline PEM scores). For such a participant, this analysis suggests that allocation to collagenase would reduce their probability of providing a response of ‘Cured’ or ‘Much better’ at 1 year by 17–31% (in absolute terms). Hence these results provide strong evidence that allocation to collagenase reduced the probability of more positive overall hand assessments at 1 year post treatment compared with allocation to LF.

Passive extension

Passive extension deficit of the reference joint

Summaries of the passive extension measurements of the reference joint are given in Table 25 by time point and allocation for the subset of 621 participants who were treated as part of the trial. Mean profiles for the available measurements are plotted by randomised group in Figure 21. One feature of the summaries in Table 25 is the number of missing cases (across both groups) at the post-treatment time points (particularly from 6 months onwards). Aside from the losses to follow-up and administrative censoring that affected collection of all outcomes, the collection of the passive extension measurements was particularly impacted by the COVID-19 pandemic. Passive extension can only be measured in clinic by research staff, and therefore could not be collected when participants were followed-up remotely (as was necessitated by lockdowns and/or precautions regarding participant safety).

TABLE 25 - Available passive extension deficit measurements by time point and allocation (treated participants)a

Positive values indicate extension deficit and negative values indicate hyperextension.

	LF N = 295	Collagenase N = 326	Total N = 621
Baseline – passive extension deficit of reference joint (°)
N (% of treated)	288	318	606
Mean (SD)	46.0 (16.7)	45.6 (17.2)	45.8 (17.0)
Median (Q1, Q3)	44.7 (34.0, 59.0)	45.2 (32.0, 58.3)	44.7 (32.7, 58.7)
Minimum, maximum	0.0, 90.0	−10.0, 84.7	−10.0, 90.0
Pre-treatment – passive extension of reference joint (°)
N	265	309	574
Mean (SD)	49.5 (17.8)	48.3 (17.1)	48.9 (17.4)
Median (Q1, Q3)	50.0 (38.0, 62.0)	46.0 (38.0, 60.0)	48.0 (38.0, 60.0)
Minimum, maximum	0.0, 91.0	−10.0, 92.0	−10.0, 92.0
Month 3 – passive extension deficit of reference joint (°)
N	210	226	436
Mean (SD)	5.6 (19.9)	9.5 (21.0)	7.6 (20.5)
Median (Q1, Q3)	0.7 (−7.3, 18.3)	2.7 (−3.3, 23.3)	1.7 (−5.2, 20.0)
Minimum, maximum	−35.0, 70.7	−42.7, 74.0	−42.7, 74.0
Month 6 – passive extension of reference joint (°)
N	171	203	374
Mean (SD)	5.1 (22.8)	11.4 (22.5)	8.5 (22.8)
Median (Q1, Q3)	0.0 (−10.0, 16.0)	6.0 (−0.7, 25.0)	2.0 (−5.0, 20.7)
Minimum, maximum	−30.7, 78.0	−48.0, 80.0	−48.0, 80.0
1 year – passive extension deficit of reference joint (°)
N	148	172	320
Mean (SD)	4.0 (23.6)	13.0 (25.3)	8.8 (24.9)
Median (Q1, Q3)	0.0 (−11.3, 18.7)	7.7 (−5.2, 29.3)	1.0 (−7.7, 25.5)
Minimum, maximum	−72.0, 73.3	−38.0, 78.7	−72.0, 78.7
2 years – passive extension deficit of reference joint (°)
N	116	142	258
Mean (SD)	5.4 (26.6)	16.5 (27.9)	11.5 (27.9)
Median (Q1, Q3)	0.0 (−12.3, 18.8)	11.0 (−3.3, 38.7)	6.0 (−9.3, 30.7)
Minimum, maximum	−40.0, 90.0	−50.0, 88.7	−50.0, 90.0

FIGURE 21.

Mean extension deficit profiles by allocation. Passive extension deficit is plotted in (a) and active extension deficit in (b). For both plots, larger values indicate poorer extension. Vertical dashed line plotted at median time elapsed between randomisation and treatment with subsequent time points referenced to this.

On average, the measurements are similar across groups at baseline (even when considering just treated participants), with the majority having about 30‒60 degrees passive extension deficit. The mean and median of the available pre-treatment measurements are slightly larger than the analogous summaries of the baseline data, but there is little evidence that this increase in passive extension deficit between baseline and treatment differed by group. Following treatment, both groups have reduced passive extension deficit (i.e. greater passive extension) on average, with most participants having measurements between about −5 degrees (i.e. 5 degrees hyperextension) and 20 degrees at 3 and 6 months post treatment (based on the available cases). Despite the improvements following treatment in both groups, the data in Table 25 and profiles in Figure 21 suggest that even at these early time points passive extension deficit was greater (i.e. passive extension was poorer) on average in the collagenase group compared to the LF group. This difference appears to have grown by 1 and 2 years, with mean extension deficit remaining at < 5 degrees in the surgery group, but gradually increasing in the collagenase group from about 11 degrees at 6 months, to about 17 degrees at 2 years post treatment.

Estimated differences in passive extension deficit from the planned analysis of the available data are reported in Table 26. We also repeated this analysis using multiply imputed data. The results from this analysis are reported in Table 27. Note the estimates (from both analyses) can be interpreted as the difference in expected outcome (at each time point), or differences in expected change from baseline (at each time point).

TABLE 26 - Passive extension deficit (available cases) – treatment effect estimates at 3 and 6 months and 1 and 2 years post treatment

Positive values indicate greater benefit from LF, that is greater passive extension than from collagenase.

Two-sided, based on t-test with degrees of freedom calculated using the Kenward–Roger method.

	Estimated differencea (collagenase – LF) (95% CIb)	p-valueb H0: δ = 0
Month 3	5.73 (2.88 to 8.59)	0.0001
Month 6	9.48 (6.19 to 12.77)	< 0.00005
1 year	10.10 (6.46 to 13.73)	< 0.00005
2 years	12.92 (8.38 to 17.46)	< 0.00005

TABLE 27 - Recurrence by allocation including missing cases (treated participants only)

	LF N = 295	Collagenase N = 326	Total N = 621
Recurrence, n (%)
No	137 (46.4)	154 (47.2)	291 (46.9)
Yes	22 (7.5)	32 (9.8)	54 (8.7)
Missing	136 (46.1)	140 (42.9)	276 (44.4)

The apparent additional benefits of surgery over collagenase discussed above is reflected in the estimates in Table 26. Extension deficit appears to be worse (on average) following allocation to collagenase at all post-treatment time points. In particular, the p-values at each time point suggest significant differences between groups at the 0.1% level or less. Hence there is strong evidence (on statistical grounds) that LF results in superior outcomes for passive extension deficit in the short term and even more so in the longer term. The results for the imputed data (see Appendix 8, Table 101) are generally similar (as expected given the absence of any a priori reason to seriously doubt the MAR assumption of the planned analysis), and inference is unaffected.

Active extension

Treatment complications

Overall, there were 177 treatment complications reported for participants in the LF group (0.60 complications per treated participant), and 267 for participants in the collagenase group (0.82 complications per treated participant). Of the 295 participants allocated to LF that received treatment as part of the trial, 106 (36%) had at least one treatment complication reported. Of the 326 participants allocated to collagenase (and treated as part of the trial) 135 (41%) had at least one treatment complication reported.

A detailed summary of the complications reported among the 621 treated participants is provided in Table 29.

The number and proportion of the 621 treated participants that experienced a complication at each grade (or not at all) are reported by allocation in Table 30. Brief details of the complications graded as ‘Moderate’ or ‘Severe’ are provided (see Appendix 8, Table 117). From Table 36, it is apparent that a larger proportion of participants in the collagenase group experienced at least one complication (36% in the LF group vs. 41% in the collagenase group). However, we again see that a higher proportion of participants in the LF group experienced ‘Moderate’ or ‘Severe’ complications (5% in the LF group vs. 2% in the collagenase group), and a lower proportion experienced ‘Very minor’ complications (7% in the LF group vs. 18% in the collagenase group). Together these tables suggest that while incidence of complications was higher in the collagenase group, they tended to be somewhat less severe than those in the LF group.

A breakdown of the severity of the worst single complication experienced by each participant is given in Table 31. This restricted set of complications exhibits similar patterns to the full set, with more complications overall among participants allocated to collagenase, but more complications deemed to be ‘Moderate’ or ‘Severe’ among those allocated to LF.

The planned analysis of this outcome (i.e. most severe single complication) was using a proportional odds model (i.e. ordinal logistic regression with a single term for the treatment effect to capture overall shifts in outcomes across all levels of the ordinal scale), with a secondary analysis using a partial proportional odds model (with the proportional odds assumption relaxed for the effect of allocation but retained for all other variables in the model). However, Table 31 suggests any effects of allocation show considerable departure from proportional odds, with allocation to collagenase seemingly shifting participants toward worse grades at the lower end of the scale (compared to LF), while simultaneously shifting participants toward less-severe grades at the top end of the scale. We therefore consider only the results of analyses based on the partial proportional odds model (since any single-number summary of treatment effect is unlikely to be appropriate for the data at hand).

Table 32 provides treatment effect estimates (conditional on covariates X) on both the RD (RR; collagenase – LF) and risk ratio (collagenase/LF) scales based on the fitted partial proportional odds model. The risks being contrasted in each case are the risk of a complication at least as bad as grade y, where y takes the values ‘Very minor’, ‘Mild’ or ‘Moderate’. It should be noted that the effect of allocation on experiencing a complication graded ‘Severe’ (or worse) cannot be reasonably estimated based on the data alone due to the sparseness of the cells at this upper end of the scale (three events in the LF group and zero events in the collagenase group). Hence based on the data alone, there is little evidence to refute any plausible hypotheses regarding the effect of allocation on the incidence of severe (or worse) complications, except those that posit any strong favourable effects of allocation to LF.

As expected, based on the frequencies and proportions given in Table 31, the data are most compatible with hypotheses positing that treatment with collagenase causes a small increase in the risk of having a complication (of any severity). This appears to apply similarly to participants in both reference joint strata (at least for those with average baseline passive extension deficit). However considerable uncertainty remains, with the data providing little refutational information against hypotheses positing both small reductions in risk of any complication following allocation to collagenase, as well as substantial increases. There appears to be little impact of allocation on the incidence of complications graded ‘Mild’ or worse, although again there is little information in the data to refute important differences (e.g. 7% on the absolute risk scale) in either direction. The data alone are again relatively uninformative with regard to the effect of allocation on the incidence of complications graded ‘Moderate’ or worse, due to the small number of participants that had complications assigned these grades. The data are most compatible with allocation to collagenase resulting in a lower incidence of complications graded ‘Moderate’ or worse (e.g. 7% reduction in absolute risk and 8–9% reduction in relative risk), and reasonably compatible with large reductions in the risk of these more serious complications (e.g. 18–20% reduction in absolute risk and 20% reduction in relative risk). However, the data are also not particularly at odds with the frequencies that might occur if allocation to collagenase truly had no effect on the incidence of complications of moderate severity or worse, or even slightly increased the incidence of such complications (compared to LF).

On the suggestion of the trial data monitoring committee, we pre-specified and undertook an analysis of the complications data excluding skin tears sustained during manipulation following treatment using collagenase. The rationale for this supplementary analysis is that skin tears are an expected and easily managed aspect of the procedure, and therefore could justifiably be discounted as treatment complications (see Appendix 8, Figure 90 and Table 118 for results of these analyses).

Further treatment

Overall summaries relating to the delivery of further treatments (i.e. further care and/or re-intervention) by 1 and 2 years post treatment and are given in Tables 33 and 34, respectively.

From Table 33 we see that about 35% of the 295 treated participants in the LF group and about 30% of the 326 treated participants in the collagenase group are missing these data due to having wholly or partially incomplete follow-up to 1 year post treatment. Most of these participants have incomplete measurement histories due to dropping out before 1 year post treatment or having their follow-up histories administratively censored before 1 year, although there are a minority of participants that have been followed up at 1 year, but have interval censored follow-up histories (e.g. they have a 1-year follow-up, but are missing data at 3 and 6 months). Among the available cases, we see that the vast majority (92%) had not undergone any further treatment by 1 year, with an even higher proportion (95%) not having undergone any re-intervention by 1 year. Overall, the most common type of further treatment received by 1 year was an extended course of physiotherapy (approximately 3% of those with these data available), followed closely by LF (approximately 3% of available cases). A handful of participants underwent re-intervention of the reference digit using collagenase and PNF by 1 year, and none received dermo-fasciectomy by this time point. Extended physiotherapy to the reference digit predominantly occurred in the LF group, while other further care generally occurred more among those allocated to collagenase. Hence the proportions of participants receiving any further treatment by 1 year are similar across groups, while the proportion of participants receiving a re-intervention is higher in the collagenase group.

From Table 34 we see that about 45–50% of participants were missing data relating to further treatments received between treatment and 2 years post treatment. This increase in proportion of missing cases is primarily driven by administrative censoring of follow-up times between 1 and 2 years post treatment, as well as a small amount of loss to follow-up between these time points. The overall patterns of further treatment in the observed cases are fairly similar to those at 1 year, with most participants not receiving further treatment to the reference digit. However, LF had overtaken extended physiotherapy as the leading type of further treatment by 2 years, with 10 additional cases occurring in the collagenase arm (and no additional cases in the LF group). It follows that the proportion receiving any sort of further treatment by 2 years is lower in the LF group, in contrast with the proportions seen for the 1-year end point. Furthermore, the proportion receiving re-intervention by 2 years in the LF group is appreciably lower than the proportion in the collagenase group.

The results from the analyses of the 1- and 2-year binary further treatment outcomes are given in Table 35. In line with the frequencies and proportions in Table 33, the results in Table 35 suggest that the data provide some evidence to refute the hypothesis that the incidence of further treatments and re-intervention during the first year following treatment is unaffected by allocation. For example, the lower limit of the 95% CI estimate is about 1.2, which (given that the event is relatively rare in all relevant strata) suggests allocation to collagenase likely caused at least a 20% increase in the relative risk of re-intervention by 1 year compared to allocation to LF. However, the substantial differences on the relative odds (or risk) of re-intervention translate into small differences in absolute terms due to the overall rarity of the event. For participants allocated to collagenase, the slightly higher rate of re-intervention reflects slightly greater rates of early recurrence in this group, whereas the similar rates of further treatment (including extended physio for the reference digit) reflects a greater need for after-care/rehabilitation following LF.

The results for these outcomes at the 2-year time point show similar trends as those for the 1-year time point. There is a reasonably clear signal of superiority (in favour of LF) with regard to the incidence of re-intervention, but a less clear signal with regard to further treatment overall (i.e. including post-treatment rehabilitation and physiotherapy). The stronger signal in favour of LF with regard to incidence of re-intervention by 2 years is likely driven by higher rates of short- to medium-term recurrence of contracture in the collagenase group. However, for the overall further treatment outcome, any influence that higher rates of recurrence might be having is being offset to some degree by higher rates of post-treatment rehabilitation following LF.

Kaplan–Meier failure estimates for the proportion of at-risk participants that had received further treatment over time are given by allocation in Figure 22. These clearly suggest a higher rate of re-intervention in the collagenase group, and suggest that by 2 years post treatment, nearly 10% of collagenase participants (that were still being followed-up) had undergone a re-intervention, compared to just 2.5% of LF participants (corroborating the findings of the analyses of the binary incidence of re-intervention outcomes above).

FIGURE 22.

Time to re-intervention (following initial correction) by allocation (Kaplan–Meier failure estimates).

Estimates of the conditional hazard ratio (HR) for allocation are given in Table 36, together with a p-value for a two-sided test of H₀ : HR = 1. Estimates of the difference (collagenase – LF) in cumulative incidence of re-intervention by 2 years post treatment (conditional on each level of reference joint and baseline PEM score equal to the mean of the observed scores) are given in Table 37.

The estimated HR and associated test of H₀ : HR = 1 suggest the data show considerable departure from what would be expected if the hazard of re-intervention across groups were equal. Similarly, to the analyses above, the apparent impact of LF on the HR scale does not translate to huge impact on the absolute risk scale due to the relatively small risk of re-intervention by 2 years. For example, true differences in absolute risk of 0–10% (in favour of LF) are reasonably compatible with the observed data for both reference joint strata (although the interval estimates also suggest relatively limited incompatibility between the observed data and true differences of 15–20%). Overall, these analyses offer strong evidence to refute hypotheses positing that rates of re-intervention are lower or the same following treatment with collagenase, compared to treatment with LF.

Overview

The main question addressed in this economic evaluation is the cost-effectiveness of collagenase compared with LF. The within-trial analysis covered a 2-year time horizon post treatment and followed the principle of intention to treat (ITT), with participants analysed as randomised. An NHS and PSS perspective was adopted in the CEA. Costs were expressed in Great British pounds (£) at a 2019–20 price base, and costs from other years were adjusted to the 2019–20 level using the NHS Cost Inflation Index. 62 Analyses were conducted following the manual of National Institute for Health and Care Excellence (NICE) health technology evaluations63 and in accordance with the detailed health economic analysis plan (available at: www.isrctn.com/ISRCTN18254597). Within-trial CEA was conducted in Stata 17, and economic modelling in Microsoft Excel 365 (Microsoft Corporation, Redmond, WA, USA). Differences are reported as collagenase – LF unless otherwise stated.

Intervention cost

A micro-costing approach was used to generate cost estimations by multiplying the quantity of detailed resource utilisation and corresponding unit cost. 64 Table 38 lists all the unit costs used.

The cost of LF included staff, theatre, anaesthesia, medication, and wound clinic appointment costs. Several healthcare professionals are involved in LF surgery (e.g. consultant surgeons, trainee surgeons, nurses, healthcare assistants, and operating-department practitioners) and costs were calculated by multiplying the PSS Research Unit (PSSRU) unit cost by the procedure duration (time from participants’ entry to the anaesthetic room until the time they left the operation theatre). 62 An additional cost was estimated by combining the unit cost of the operation theatre (£15 per minute at the 2019–20 price) with the time that the theatre was in use. 66 The cost of inpatient nights was included if inpatient admission was required after the operation. 65,67 The cost of subsequent wound-clinic appointments after LF were estimated based on the staff types and duration collected in the treatment delivery CRF.

The intervention cost of collagenase consisted of two parts: collagenase administration and joint manipulation. Based on the British National Formulary, the price of collagenase was £650 per vial prior to September 2018 and £572 per vial thereafter. 69,70 The staff cost was estimated using the staff members present during the procedure and the duration of the procedure. For joint manipulation, staff, anaesthesia and medication costs were calculated using the data collected alongside the appointments. Premises and service cost for procedures that took place in clinic rather than operation theatre were assumed to be covered in the PSSRU unit cost. 62

Healthcare resource costs

Community health and social care visits were valued by attaching unit costs from PSSRU. Outpatient attendance without specified treatment information were treated as a 10-minute average reassurance or observation appointment. Inpatient admissions during the follow-up periods were recorded along with the treatment received and length of stay. The cost of accident and emergency (A&E) attendance was also included if the number of visits was available. The number of prescription items was counted and multiplied by the national average cost per item (£8.55 in 2020). 68

Health-related quality of life

The primary health outcome measure for the CEA was QALYs. QALYs integrate both the length and health-related quality of life (HRQoL) and was recommended in the manual of NICE health technology evaluations. 63 The results collected in EQ-5D-5L were converted into a single index value (health utility) via mapping onto the EuroQol-5 Dimensions, three-level version UK population valuation set. 71 The utility scores at baseline and each follow-up appointment were then used to calculate QALYs using the ‘area under the curve’ method. 72 In addition, overall self-rated health status was measured via the EuroQol VAS. 73 Descriptive statistics of the utility scores, EQ-VAS at each time point and the overall QALYs for the two trial groups were presented and compared in the analysis.

Incremental cost-effectiveness analysis

The incremental CEA was conducted for both within-trial analysis and the long-term model-based analysis. 64

The calculated ICERs were compared to the NICE-recommended maximum acceptable ICERs [or willingness-to-pay (WTP) thresholds] of £20,000–£30,000 per QALY to determine whether collagenase is cost-effective. 63 Results were also presented in terms of NHB,64 which is particularly informative when the intervention generates fewer health benefits but, at the same time, is less costly; that is, it generates cost savings. 63,64 Positive NHB indicates an increase in overall population health and the new intervention should be considered to be funded.

To control for possible baseline imbalance, the differences in cost and QALYs were adjusted for the baseline cost and EuroQol-5 Dimensions (EQ-5D) score. 74 The costs of prescribed medications recorded on the baseline CRF, were utilised to represent baseline healthcare costs. The mean differences in costs and QALYs were estimated using seemingly unrelated regression equations with 95% confidence intervals (CIs) estimated using bootstrap methods. 75 The variables included in the regression model were selected based on a combination of clinical theories, empirical evidence, and statistical considerations. The model adjusted for the following covariates: baseline utilities and costs, treatment factor (intervention group), demographic factor (age), and clinical factor (PEM). For hypothesis testing, differences were considered as statistically significant at p < 0.05.

Analysis of uncertainty

To assess the degree of uncertainty surrounding the estimated incremental costs and QALYs, a non-parametric self-resampling bootstrap technique was adopted. 76,77 The results of the 5000 replicated samples generated by bootstrapping were depicted in CEPs and CEACs to illustrate the probability of collagenase being cost-effective over different maximum acceptable ICERs.

Missing data

The level of missing data in trial-based economic evaluation is usually high. 78 The base-case analysis was conducted using an imputed data set, under the assumption that data were MAR and was imputed using Rubin’s MI method. 79 Although the MAR assumption theoretically cannot be tested due to its dependence on unobserved data, we employed certain strategies to make this assumption more plausible. This included checking missing data patterns and carrying out a series of logistic regressions that analysed the missingness of costs and QALYs against baseline variables, such as age, gender, baseline PEM score, and baseline EQ-5D utility scores. 75

Multiple imputation by chained equation was performed to impute missing outcomes with 25 imputations. 80 The selection of variables for the imputation model was based on expert knowledge, prior research, and current data set observations. All variables used in the CEA were included in the imputation model, along with several additional auxiliary variables to provide further information about the missing values. These auxiliary variables included variables that were found to be predictive of missingness and those that were correlated with the variable that had missing values. The imputation model included primary outcomes (costs and utility scores at baseline and each follow-up point), demographics (age, gender), condition severity (number of cords affected), participant trial status (e.g. participation, withdrawal or deceased) and hand-related outcomes (PEM). In the CEA, for deceased participants where the date of death was prior to the scheduled follow-up appointment, a zero utility score and zero cost were imputed by definition. 81 To examine the impact of missing data, sensitivity analyses were conducted using only complete cases, where data on both costs and QALYs were available simultaneously.

Secondary analysis

Evidence indicates that more than half of participants diagnosed with DC have functional limitations impacting both their work and leisure activities. 82–84 To explore the impact of lost time from work and normal activities on the cost-effectiveness results, we conducted a secondary analysis from a wider societal perspective. Information about sick days and lost time from normal activities (e.g. housework, caring duties or voluntary work) was obtained from participant CRFs. Applying a human capital approach, the time off work and away from normal activities were calculated using the national mean weekly wage of £710 for full-time work and £248 for part-time work in 2020. 85 For participants not in paid employment, the national minimum wage of £8.72 per hour was applied. Participants’ private treatments received during the trial period were also considered in the secondary analysis.

Sensitivity analyses

In addition to the complete-case analysis, several additional sensitivity analyses were performed to test the robustness of the results obtained in the base-case analysis.

Several published cost-effectiveness studies on treatments for DC use standardised costs of procedures instead of the micro-costing method. In the UK setting, the Health Resource Group (HRG) reference cost has often been adopted. 86,87 HRG codes for the procedures included in the DISC trial were identified using the HRG4+ code to group workbook. 88 To facilitate the comparison of our results with previous studies, a sensitivity analysis was performed using the HRG unit cost to compute the intervention cost. The HRG codes used for LF, collagenase injection, wound care, and manipulation were HN43B (unit cost £2936), HN45A (£1143), HN46Z (£196) and HN46Z (£196), respectively. 65

Compared to LF, collagenase administration has the advantage that instead of occupying a highly costly operating theatre, treatment can take place in a clinic or recovery room. 66 The location of the treatment can also impact the staffing required as well as the waiting time for patients. 89 To examine whether treatment location is a determining factor in the cost-effectiveness of treatment with collagenase, a sensitivity analysis was conducted by assuming that either 50% or 100% of collagenase injections were performed in an operating theatre.

The unit cost of trainee surgeons is less than half that of consultant surgeons, at £50 per hour versus £114 per hour. 62 Training less experienced trainee surgeons to deliver LF could potentially reduce the overall cost of this expensive surgery. 66,90 Thus, a further sensitivity analysis was conducted, assuming all LFs were delivered by trainee surgeons.

Threshold analysis

Baltzer et al. and Chen et al. reported that the price of collagenase is a determinant in CEA, and collagenase would only be considered cost-effective at a price lower than a certain threshold. 91,92 A threshold analysis was performed to explore the maximum acceptable price that would result in collagenase being considered cost-effective. Given collagenase has been withdrawn from the UK market since March 2020, this additional analysis provides information for further decision-making should collagenase be reintroduced into the UK market in the future.

Long-term Markov decision analytic model

The within-trial time horizon was not long enough to capture the entire costs and benefits of collagenase and LF especially since the trial period included the COVID-19 pandemic which meant that some participants who had recurrent issues did not have the chance for re-treatment. 57 Therefore, a decision analytical model was constructed (Figure 23 and Table 39) to project the lifetime cost-effectiveness of collagenase compared with the LF by incorporating recurrence, further treatments, long-term costs and QALYs.

FIGURE 23.

Long-term Markov model structure.

Intervention cost

A summary of the breakdown of the costs by treatment delivered is presented in Table 40 for the 288 participants who underwent LF and the 332 who received collagenase. Only one participant in the LF group received PNF, which was costed using the reference cost rather than micro-costing. The 41 LF participants and 10 collagenase participants who did not receive treatment as part of the trial had associated intervention costs of zero. The average cost per participant for LF was £2510 (SD £818), with the operating theatre being the most expensive component. In contrast, the average cost for the collagenase treatment was £1008 (SD £94), with the cost of the collagenase itself being the largest expense (see Table 40).

Appendix 9, Tables 119 and 120 provide a summary of the staff costs associated with the trial interventions. On average, LF involved 5.5 staff members and took about 109 minutes, costing £630 per patient in staff time. The collagenase treatment needed fewer staff, about three for both the injection and joint manipulation, and took about 90 minutes total. This resulted in a lower average staff cost of £310 per patient for the collagenase treatment. Proportions of each staff cost by intervention component are illustrated in Appendix 9, Figure 91.

Healthcare resource use cost

Appendix 9, Tables 121–124 summarise the mean costs associated with healthcare resource use among available cases by cost item at 3 and 6 months and 1 and 2 years post treatment. Appendix 9, Figure 92 shows the overall mean costs for the five main categories of healthcare resource use.

For each time point, costs of resource use were considered missing if none of the cost categories were captured. For accumulative costs over various time points, for instance, 1- and 2-year costs used in the following CEA, costs of resource use were considered missing for a participant if any of the time points were missing. The estimated costs of medication at baseline were estimated at £23.0 (SD £23.7) in the LF group and £22.8 (SD £22.3) in the collagenase group (mean difference: −0.2, 95% CI −£3.7 to £3.3).

For the three categories of secondary care, more details were gathered concerning whether the resources used were related to the reference hand and treatment complications or AE. Appendix 9, Tables 125–127 summarise the responses for each cost category to these questions.

The overall mean healthcare cost was higher in the collagenase group (£313, SD £502) compared with the LF group £248 (£832; mean difference: −£65, 95% CI −£181 to £52) during the first 3 months post treatment (see Appendix 9, Table 121). A significantly higher frequency of primary care and outpatient visits was observed after receiving LF than collagenase. The number of visits the patients paid to both community and outpatient physiotherapists as well as occupational therapists in the LF group was at least 2.5 times larger than the collagenase group. Participants in the LF group also reported a higher use of wound care and splinting. The level of utilisation of these physical therapies was reduced to a more similar level between the two groups at the 6-month and 1-year follow-up (see Appendix 9, Tables 122 and 123). For the 2-year follow-up, higher costs can be observed in these cost items in the collagenase group during the second year post intervention, but the differences were not statistically significant (see Appendix 9, Table 124).

The results also show that the proportion of outpatient visits related to the designated reference hand reduced in the LF group with time: from 98% at 3 months follow-up to 29% at 2 years follow-up (see Appendix 9, Table 125). However, in the collagenase group, the proportion related to the reference hand dropped from 92% at 3 months to 61% at 6 months and stayed at that level until the end of the trial. Consequently, compared to the LF group, we detected a significantly lower proportion (92% vs. 98%, p < 0.01) of reference hand-related outpatient visits in the collagenase group at 3 months and a significantly higher proportion (61% vs. 29%, p < 0.01) at 6 months. No significant differences in costs were detected in A&E attendance throughout the trial period (see Appendix 9, Table 126).

The number of hospital inpatient admissions is also reflected in Appendix 9, Table 127. During the second year post treatment, 21 and 18 hospital admissions were recorded in the LF and collagenase groups, respectively. The average length of stay was longer in the LF group, with four participants staying in the hospital more than five nights. The longest stay for LF participants was 20 nights, whereas the longest stay was only four nights in the collagenase group. The shortened stay contributed to a lower overall mean healthcare cost in the collagenase group compared to the LF group at the 2-year follow-up (£79 vs. £332, mean difference: −£253, 95% CI −£489 to −£17). Approximately 80% of the participants were admitted for reasons unrelated to their hands, such as laparotomy surgery, aortic valve replacement, or knee replacement. Although we did not detect lengthy hospital stays in the collagenase group, more deaths were recorded than in the LF group (8 vs. 1).

During the 2-year trial period, some participants received further surgical and pharmacological interventions for DC after the initial trial intervention. Appendix 9, Table 128 shows the number of further treatments in both outpatient and inpatient settings for all available cases. To capture the total cost of further treatments from an NHS perspective, all interventions were included in the costing regardless of whether they were related to the reference digit.

In general, participants in the collagenase group received more treatments for DC during the trial period. Three sessions of further collagenase injection were recorded in the collagenase group during the first year following treatment, but none were found in the LF group. No injections were recorded in both groups during the second year post treatment, mostly because collagenase was withdrawn from the UK market since March 2020, and most participants (95 out of 504) had their 1-year visit thereafter. Roughly four times more (31 vs. 8) LFs were given to participants in the collagenase group during the 2-year follow-up period. This increased number of treatments was a major contributor to hospital inpatient costs, which resulted in a much higher inpatient cost in the collagenase group in all follow-ups, except the 2-year costs.

Health-related quality of life

Baseline utility scores were slightly higher in the LF group as compared to the collagenase group (mean of 0.794 and SD of 0.170 vs. mean of 0.791 and SD of 0.174, respectively), but the difference was not statistically significant (95% CI −0.029 to 0.024) (Table 41). In the following analyses, the baseline utility scores were used as pre-treatment values, as we did not collect EQ-5D-5L in the pre-treatment CRF. It is assumed that the participants’ HRQoL does not change much without receiving treatment.

The mean EQ-5D utility scores and relative marginal distributions at each time point by allocation are illustrated in Figure 24 and Appendix 9, Figure 93. Soon after treatment for DC, participants in both groups experienced a decline in utility scores, especially those who received LF. Two weeks after the treatment, the EQ-5D utility scores reduced from 0.794 to 0.715 in the LF group and from 0.791 to 0.776 in the collagenase group. Six weeks after the treatment, the utility scores started to increase back as the participants recovered. At both the 2- and 6-week follow-ups, the mean utility scores were significantly higher in the collagenase group than in the LF group (mean difference at 2 weeks: 0.061, 95% CI 0.037 to 0.085; at 6 weeks: 0.03, 95% CI 0.005 to 0.055).

FIGURE 24.

Mean EQ-5D scores at baseline and follow-up points.

Three months after the treatment, the utility scores in both groups had reverted to the baseline level, and participants in both groups reported improved HRQoL as compared to before treatment. By the end of 2 years, the mean difference in utility scores between the groups was estimated to be −0.044 (95% CI −0.077 to −0.010). It should be noted that the figures reported in this section are unadjusted EQ-5D utility scores generated from all the available cases. In the following CEA, the utilities adjusted for the baseline are used instead.

The percentages of participant responses to each EQ-5D-5L dimension at each timepoint are presented in Appendix 9, Figures 94–98. Among all five EQ dimensions, the proportion of participants who reported ‘no problems’ was the lowest (ranging from 34.3% to 55.1%) in the pain/discomfort dimension throughout the trial. Furthermore, the largest change of responses was detected at 2 weeks after treatment, especially for the dimensions of self-care, usual activities, and pain/discomfort. In contrast, neither DC nor the interventions exhibited much impact on participants’ mobility and mental health (anxiety/depression dimension). For most EQ-5D dimensions, the proportion of participants who reported ‘no problems’ returned to a similar or even higher level than the baseline at 6 weeks’ follow-up. The only exception was the pain/discomfort domain, where the proportion of participants without problems was lower at 6 weeks than the baseline in both groups. In addition, a significant difference was detected between the LF group and the collagenase group in terms of the percentage of participants without any problems in pain/discomfort at 6 weeks (18.6% vs. 29.8%, p = 0.03). This indicates that compared to the collagenase injection, participants who received LF took a longer time to recover from the associated pain and discomfort.

At 2 weeks, a significantly higher proportion of participants in the LF group reported having problems with self-care as compared to the collagenase group (57.1% vs. 23.9%, p < 0.001) and usual activities (79.1% vs. 50.3%, p < 0.001; Appendix 9, Figures 94 and 95). From 3 months and after, the percentage of reported problems had reduced, and participants experienced fewer issues in self-care, usual activities and pain/discomfort as compared to their pre-treatment status.

At the baseline, the mean EQ-VAS scores were 83.8 (SD 14.8) and 85.3 (SD 15.1) in the LF and collagenase groups, respectively (see Appendix 9, Figure 99 and Table 129); however, the difference between the two groups was not statistically significant (1.45, 95% CI −0.85 to 3.75). At 2 weeks, LF participants reported statistically significantly lower (worse) EQ-VAS scores than collagenase participants (83.4 vs. 86.3, mean difference: 2.98, 95% CI 0.75 to 5.12). At 6 weeks, the EQ-VAS score was still higher in the collagenase group (mean difference: 1.8, 95% CI −0.43 to 4.03), but it was not statistically significant. There was little difference in the EQ-VAS scores between the two trial groups at any of the time points from 3 months onwards.

Base-case cost-effectiveness analysis

Tables 42 and 43 combine the mean cost and QALY gains in the two trial groups and the results of the base-case CEA based on the imputed data. Appendix 9, Table 130 summarises the level of missing data at each time point, categorised by treatment group. The pattern of missing data is non-monotonic, where patients with missing data at one follow-up may have data in the subsequent follow-ups. Apart from age at randomisation, all logistic regressions exploring the correlation between missingness and baseline characteristics yielded statistically insignificant results (p-value > 0.05). These results imply that the MAR assumption is reasonably justified for proceeding with our analysis.

The results indicate that treating participants with collagenase resulted in a decrease in total cost and QALY gains at both 1- and 2-year follow-ups compared with LF. Instead of presenting ICERs for a less-costly and less-effective intervention, NHBs were calculated to demonstrate the cost-effectiveness of collagenase compared to LF.

After adjustment for baseline costs and utilities, participants assigned to receive collagenase showed a statistically insignificant decrease in QALY gains of −0.003 QALYs (95% CI −0.006 to 0.0004 QALYs) and experienced a significantly reduced total cost of −£1090 (95% CI −£1139 to −£1042) compared to the LF group in the first year post treatment (Table 42). The cost difference was mainly attributable to significantly higher intervention cost in the LF group, whereas the estimated healthcare costs were similar in the two groups. The NHB of collagenase were estimated to be 0.052 and 0.033 at the maximum acceptable ICERs of £20,000 and £30,000 per QALY gained.

Bootstrapping was used to quantify uncertainty around the cost-effectiveness estimates. The majority of the 5000 bootstrap iterations were in the South-West quadrant of the CEP (see Figure 25), indicating collagenase was cheaper and less effective than LF. The CEACs (see Figure 26) showed that the probability of collagenase being decrementally cost-effective exceeded 99% at WTP thresholds ranging from £20,000 to £30,000 per QALY at 1 year.

FIGURE 25.

Cost-effectiveness analysis plane (MI) at 1 year. The x-axis represents the 1-year difference in QALY gains between the collagenase and LF groups (collagenase – LF), while the y-axis represents the difference in cost. The lines represent the £20,000 (in red) and £30,000 (in blue) per QALY WTP thresholds. Points in the North-East quadrant suggest that collagenase is more effective but also more costly than LF. Points located below the WTP threshold lines in this quadrant indicate that collagenase is incrementally cost-effective compared to LF. In the South-East quadrant, collagenase is both more effective and less costly than LF, making it the clearly cost-effective intervention. Similarly, the North-West quadrant is where LF is the cost-effective choice. However, in the South-West quadrant, where most points are located in this figure, collagenase is less effective and less costly than LF. Points that lie below the WTP lines in this quadrant indicate that collagenase is decrementally cost-effective compared to LF.

FIGURE 26.

Cost-effectiveness acceptability curve (MI) at 1 year. The CEAC plots the probability that collagenase (in blue) or LF (in red) is cost-effective (y-axis) against various WTP thresholds (x-axis). In the WTP thresholds range of £20,000–£30,000 per QALY, the curve indicates a 99% probability that collagenase is decrementally cost-effective.

At 2 years post treatment, collagenase continued to be significantly less costly (mean difference: −£1212, 95% CI −£1276 to −£1147) and became significantly less effective (mean difference: −0.048, 95% CI −0.055 to −0.040) (Table 43). Compared to the 1-year results, the overall cost-effectiveness of collagenase was reduced given the consistently worsened HRQoL in the collagenase group compared to the LF group after 3 months post treatment (see Figure 24). The NHBs associated with the collagenase group were estimated to be 0.013 at £20,000 per QALY gained. Figures 27 and 28 demonstrate that the probability of collagenase being cost-effective compared to LF was 72% at a WTP of £20,000 per QALY. At a WTP threshold of £30,000 per QALY, the results were more in favour of more expensive alternative (LF) and the NHBs gained in the collagenase group was lower than the LF group with a negative estimate of −0.007 QALY. In this case, the probability that collagenase would be considered cost-effective was only 37% and LF became the optimal treatment strategy. The intersection point of the two curves in Figure 27 represents the decision-making turning point, that is, LF became the cost-effective treatment strategy when the WTP threshold is higher than £25,488.

FIGURE 27.

Cost-effectiveness analysis plane (MI) at 2 years.

FIGURE 28.

Cost-effectiveness acceptability curve (MI) at 2 years.

Secondary analysis

The mean costs of sick leave and time lost from normal activities reported by participants at each follow-up point on all available cases are listed in Table 44. Participants allocated to the LF group took more time off from work and normal activities in the first 3 months after treatment. The productivity loss due to absence from work is significantly higher in the LF group compared to the collagenase group (£487 vs. £138, mean difference: −£350, 95% CI −£537 to −£162). No significant differences were detected in productivity lost from normal activities or in the cost of private health care.

The estimated mean total societal costs based on MI were £607 per participant in the LF group and £363 in the collagenase group at 2 years (mean difference: −£245, 95% CI −£293 to −£196). The probability that collagenase would be considered cost-effective was higher than the base-case analysis at both £20,000 and £30,000 WTP thresholds (72% vs. 92% and 37% vs. 64%, respectively) (Table 45). This means that collagenase is the optimal treatment for DC at both 1- and 2-year time horizons.

Sensitivity analyses

The results of all sensitivity analyses showed that the collagenase group incurred lower costs and generated less QALY compared to the LF group both 1 year and 2 years after treatment. The cost-effectiveness analysis indicated that treatment with collagenase was consistently associated with positive NHBs, and the probability of decremental cost-effectiveness was always over 95% more cost-effective at 1 year, and this was robust for all the sensitivity analyses performed.

By contrast, the 2-year results were more sensitive to the varying assumptions.

The distributions of both 2-year costs and QALYs before and after imputation were plotted in Appendix 9, Figure 100. The first complete-case analysis comprised deceased participants and the results were consistent with the base-case cost-effectiveness findings (Table 45). A second sensitivity analysis treated deaths as missing, which is consistent with the statistical analysis. However, in this case, the QALYs for the collagenase group were overestimated as deceased participants with extremely low utility score (zero) were not included. As a result, the 2-year probability of collagenase being cost-effective increased from 37% in the base analysis at a WTP threshold of £30,000 per QALY to 66%.

The sensitivity analysis based on the HRG reference cost impacted both the difference in cost between the two groups and the cost-effectiveness conclusions. The adjusted mean cost difference changed from −£1212 to −£1434 making collagenase even more cost-saving. Consequently, the probability of collagenase being decrementally cost-effective increased to 92% (WTP threshold £20,000), and 61% (£30,000) at 2 years.

We investigated the potential impact of alternative locations for collagenase administration in a sensitivity analysis. We tested the assumptions where either 50% or 100% of injections were performed in an operating theatre. The results indicate that if all collagenase injections were performed in theatre, the cost advantage of the intervention would no longer be enough to cover the benefit lost, given the high cost of running a theatre. As a result, collagenase treatment becomes less cost-effective at 2 years. However, if only half of the injections were conducted in an operating theatre, the conclusions would remain unchanged from the base-case analysis.

We performed another sensitivity analysis to determine if replacing consultant surgeons with trainees would increase the cost-effectiveness of LF. However, changing the staff category did not affect the results, and so the findings remained unchanged.

Threshold analysis

The results of the threshold analysis exploring the impact of the price of collagenase on the cost-effectiveness of collagenase compared to LF over a 2-year time horizon are shown in Figure 29.

FIGURE 29.

Threshold analysis of price of collagenase at 2 years. A positive NHB indicates an improvement in overall population health achieved by adopting collagenase over LF, whereas a negative incremental NHB suggests that replacing LF with collagenase would reduce overall population health.

The base-case analysis demonstrates that, compared to LF, collagenase is a less-costly and less-effective intervention in treating DC. Therefore, for the given amount of effects derived from the treatment, the advantage of collagenase stems from its relatively low cost. The result of the threshold analysis suggests that, at 2 years, collagenase should be considered as cost-effective when the price of collagenase is < £910 per vial at the maximum acceptable ICERs of £20,000 per QALY. If decision-makers are willing to pay £30,000 for an additional QALY, collagenase would be a cost-effective option only if the drug price is < £300 per vial.

Long-term Markov decision analytic model

To investigate the cost-effectiveness over different time horizons, a Markov model was used based on the 345 participants who had 1-year recurrent status recorded. The mean costs, QALY gains, NHB, and probability of collagenase being cost-effective over 1-year, 2-year, 3-year, 4-year and lifetime time horizons are listed in Table 46. Consistent with the within-trial findings, 1 year after the initial intervention, collagenase was associated with a positive incremental NHB compared to LF, with a high probability (> 99%) of being decrementally cost-effective.

Among the participants with recurrent data available, 54 were recurrent; a recurrence rate of 15.7% (13.8% LF group and 17.2% collagenase group). Of these participants, 40 had healthcare resources use data available at the 2-year follow-up point, and only 12 participants received re-intervention for DC. Only one participant received re-intervention for the recurrent digit within the second year of treatment; one participant received a PNF (LF group). This is much lower compared to the revision rate reported in the literature, and the reduced numbers of re-interventions were primarily attributed to the impact of the COVID-19 pandemic.

Therefore, in the Markov model, re-intervention options were introduced for recurrent participants 1 year after the initial intervention. Offering timely re-intervention to 40% of recurrent participants in both groups led to an increase in the probability of collagenase being the optimal strategy over 2 years, rising from 72% (within-trial results) to 94% (model-based results) at a WTP threshold of £20,000 per QALY (see Figure 30). At a WTP threshold of £30,000 per QALY, the probability of collagenase being cost-effective also increased, from 37% to 69%, resulting in a shift in the optimal strategy from LF to collagenase (Figure 30).

FIGURE 30.

Cost-effectiveness acceptability curve (lifetime).

In line with the within-trial analysis, the incremental NHB of collagenase and the probability of it being cost-effective decreased over time, as shown in the last four columns of Table 46. Over the lifetime horizon, collagenase was associated with an estimated total cost savings of £2968 per participant compared with LF, and this corresponded to a mean QALY loss of −0.484. At WTP thresholds of £20,000 and £30,000 per QALY, the probability that collagenase is cost-effective compared with LF at the lifetime horizon fell below 22% and 16%, respectively.

The results of the two-way sensitivity analyses, which tested the impact of recurrence rates and re-intervention rates on cost-effectiveness at different time point, are shown in Appendix 9, Figure 101. The results indicate that, at 1 year, collagenase was always the optimal strategy, regardless of how the recurrence or re-intervention rates changed. Similarly, when the time horizon extended beyond 4 years, LF consistently remained the cost-effective treatment. The decision however changed in response to the variation of recurrence or re-intervention rates for time horizons with higher levels of uncertainty in the base-case analysis, that is 2–3 years after the initial intervention. The results in Appendix 9, Figure 102 indicate that the 2-year within-trial cost-effectiveness of collagenase might have changed if not for the impact of the COVID-19 pandemic, which affected the opportunities for re-intervention.

Study aims

The aim of this nested, qualitative study was to explore participants’ experiences of DC treatment and to consider factors which may shape preference for various treatment modalities.

Study objectives

• To explore participants’ experience of DC.

• To explore participants’ experience of surgical and non-surgical treatment of DC.

• To consider patients’ preference for treatment options, and those factors which might be pertinent in this.

Participants

Participants were selected purposively121 from those DISC participants who indicated a willingness to take part in a qualitative interview.

Initial sampling focused on ‘typical cases’ where treatment had been successful and uncomplicated (interview batch 1, n = 6–10 interviewees). Subsequent sampling, in batches of five interviews, sought to include more working age and female participants, and to balance instances of PIP and MCP joint being affected. Sampling ceased when data saturation had been achieved, that is no new emergent issues were evident in the data. Saturation was established by the two qualitative interviewers and confirmed with the DISC TMG.

Written informed consent was obtained for all participants at the initial DISC consent stage and was reconfirmed prior to interview.

Data collection

Participants were invited to take part in a single, standalone interview between 3 and 6 months post treatment to ensure that experience immediately post treatment could be captured. Interviews could be face to face or via telephone dependent on participant preference and were carried out by two experienced qualitative researchers (PL and MA).

Interviews were semistructured and organised to allow participants to focus on areas that they felt most important, and to introduce new topics that they felt relevant. 122,123 As a general foundation four broad topic areas were discussed: (1) Lived experience of DC; (2) Knowledge about DC treatments; (3) Experience of collagenase treatment and (4) Expectations for the future (recovery/recurrence).