UK DRAFFT - A Randomised Controlled Trial of Percutaneous Fixation with Kirschner Wires versus Volar Locking-Plate Fixation in the Treatment of Adult Patients with a Dorsally Displaced Fracture of the Distal Radius

Inclusion criteria

Patients were eligible for this study if:

• They sustained a dorsally displaced fracture of the distal radius, which was defined as a fracture within 3 cm of the radiocarpal joint.

• The treating consultant surgeon believed that they would benefit from operative fixation of the fracture.

• They were over the age of 18 years and able to give informed consent.

• The injury was < 2 weeks old.

Exclusion criteria

Patients were excluded from participation in this study if:

• The fracture extended > 3 cm from radiocarpal joint.

• The fracture was open with a Gustilo grading greater than 1. 10

• The articular surface of the fracture could not be reduced by indirect techniques. (In a small number of fractures, the joint surface is so badly disrupted that the surgeon will have to open up the fracture in order to restore the anatomy under ‘direct’ vision.)

• There were contraindications to anaesthetic.

• There was evidence that the patient would be unable to adhere to trial procedures or complete questionnaires, such as cognitive impairment or intravenous drug abuse.

Patients who sustained other injuries, which could potentially affect the primary outcome measure at 1 year post injury, for example disruption of the carpal ligaments, were documented but included in the analysis.

Screening and recruitment

Patients were recruited from the emergency departments and fracture clinics at the trial centres. Any patient with a fracture of the distal radius who, in the opinion of the treating surgeon, required an operative intervention was referred to the local research associate for screening. The research associate then presented the eligible patient with the participant information sheet. Patients were given the opportunity to discuss any issues related to the trial with the research associate, as well as with members of their family and friends.

Consent

Informed consent was obtained by the local research associate. In general, patients who were admitted with a fracture of the distal radius had their surgery on the following day, so there was sufficient time for the patients to consider taking part in the trial. Any new information that arose during the trial that affected participants’ willingness to take part was reviewed by the Trial Steering Committee (TSC); if necessary this was communicated to all participants by the trial co-ordinator. A revised consent form was completed if necessary.

Settings and locations

The trial was run in 18 trauma centres across the UK (Box 1).

Sequence generation

Randomisation was on a 1 : 1 basis, stratified by centre, intra-articular extension of the fracture and age of the patient (above or below 50 years).

Stratification by centre helped to ensure that any clustering effect related to the centre itself was equally distributed in the trial arms. The catchment area (the local population served by the hospital) was similar for all of the hospitals (c. 500,000), each hospital being a major orthopaedic unit dealing with these fractures on a daily basis. While it is possible that the surgeons at one centre may be more expert in one or other treatment than those at another centre, all of the recruiting hospitals were chosen on the basis that both techniques are currently routinely available at the centre, that is theatre staff and surgeons were already equally familiar with both forms of fixation. This could not eliminate the surgeon-specific effect of an individual at any one centre. 12 However, as the fixation of a fracture of the distal radius is a common procedure performed on routine trauma operating lists, many surgeons would be involved in the management of this group of patients: between 10 and 30 surgeons at each centre, including both consultants and trainees. Therefore, we expected that each individual surgeon would operate on only two or three patients enrolled in the trial, greatly reducing the risk of a surgeon-specific effect on the outcome in any one centre.

Stratification on the basis of intra-articular extension of the fracture (specifically involvement of the articular surface of the radiocarpal joint) eliminated a major potential confounder, since disruption of this articular surface may predispose to secondary osteoarthritis of the wrist. 13 Recent evidence14 suggests that other associated features of fractures of the distal radius which commonly appear in the classification systems, such as involvement of the ulna styloid process and distal radio-ulnar joint involvement, do not actually affect the functional outcome of the injury. Therefore, we did not include any other variables in the stratification of the randomisation sequence.

Stratification on the basis of age was used to discriminate between younger patients with normal bone quality sustaining high-energy fractures and older patients with low-energy (fragility) fractures related to osteoporosis. Empirically, both of these groups of patients could benefit from the fixed-angle stability provided by locking-plate fixation. However, the stratification would help to identify any effect related to the quality of the patient’s bone. The use of dual-energy X-ray absorptiometry (DEXA) is widely regarded as the gold standard for the assessment of bone density. However, such an investigation may be expensive and not routinely available at all centres. Therefore, we used age as a surrogate for bone density. In a large study in Norway involving 7600 participants, it was demonstrated that forearm bone mineral density remains stable up until the age of 50 years. After the age of 50 years, bone mineral density in males decreased steadily, while in females there was an initial decline between the ages of 50 and 65, with a further decline in the age groups thereafter. 15 A recent study by Court-Brown and Caeser16 assessed over 1000 patients with a fracture of the distal radius. This study confirmed that there is a clear bimodal distribution for this type of fracture according to the age of the patient. The crossover of the two peaks of incidence was around 50 years of age. These studies provide strong evidence that patients over the age of 50 years become increasingly vulnerable to fragility fractures of the distal radius. Therefore, we chose the age of 50 years as the cut-off for this trial. Furthermore, the study by Court-Brown and Caeser16 demonstrated that in the UK approximately 60% of patients sustaining a fracture of the distal radius are over 50 years, while 40% are younger than 50 years. The number of patients above and below this stratification age would, therefore, be similar (see Statistical analysis).

Blinding

In this surgical trial, the patients could not be blinded to their treatment. The treating surgeons were, of course, not blind to the treatment, but took no part in the postoperative assessment of the patients. The statistical analyses were not performed blind.

Postrandomisation withdrawals

Participants could withdraw from the trial at any time without prejudice. If patients decided to have the treatment to which they were not randomised, they were followed up wherever possible and data collected as per the protocol until the end of the trial. The primary analysis was on an intention-to-treat basis with a secondary per-protocol analysis.

Kirschner-wire fixation

The wires are passed through the skin over the dorsal aspect of the distal radius and into the bone in order to hold the fracture in the correct (anatomical) position. The size and number of wires, the insertion technique and the configuration of wires were left entirely to the discretion of the surgeon. A plaster cast was applied at the end of the procedure to supplement the wire fixation as per standard surgical practice. This cast holds the wrist still and is left on until the wires are removed at the 6-week follow-up appointment.

Volar locking plate

The locking plate is applied through an incision over the volar (palm) aspect of the wrist. Again, the details of the surgical approach, the type of plate, and the number and configuration of screws were left to the discretion of the surgeon. The screws in the distal portion of the bone were ‘fixed-angle’, that is screwed into the plate, but this is standard technique for the use of these plates. The type of proximal screw was left to the discretion of the surgeon; these may be locking or non-locking screws, as the bone in this area provides a much better purchase for the screws. Some surgeons used a temporary plaster cast to hold the patients’ wrist still but the fixed-angle stability provided by the locking plate was generally sufficient to allow early controlled range-of-movement exercises. The use of a cast or otherwise was again left to the discretion of the surgeon as per usual practice.

Risks and benefits

The risks associated with this study were predominantly the risks associated with the surgery: infection, bleeding and damage to the adjacent structures such as nerves, blood vessels and tendons. Participants in both groups underwent surgery and were potentially at risk from any/all of these complications. The application of the locking plates requires a surgical incision over the volar aspect of the wrist, which, empirically, may lead to a higher risk of injury to the adjacent structures. However, the evidence available to quantify this risk was limited. 13 Late attrition and rupture of the extensor tendons have been associated with volar locking plates,31 but, equally, late collapse and secondary deformity have been associated with K-wire fixation. 19 The risk of secondary (as a result of the injury) osteoarthritis of the wrist joint, particularly in those patients with intra-articular extensions of the fracture, is well recognised,13 but there were no data to suggest what that risk is or that the risk is greater in one group or another. We believed that the overall risk profile was similar for the two interventions but an assessment of the number of complications in each group was a secondary objective of this trial.

Analysis plan

The statistical analysis plan was agreed with the DMC at the start of the study. Any subsequent amendments to this initial plan are clearly stated and justified. Interim analyses were performed only where directed by the DMC. The trial registration details were updated soon after the start of the trial; this was not a change to the statistical analysis plan, but merely the inclusion of extra detail in the registration document, which was added at the suggestion of the DMC.

Software

The routine statistical analysis was carried out using R version 3.0.1 (The R Foundation for Statistical Computing, Vienna, Austria, www.r-project.org/). When any analyses were required, data were retrieved from the trial database by the trial statistician. The statistician imported data directly into the statistical package R for analysis and reporting. The version numbers of all software used, data files and all R scripts were made available to the DMC on request at any stage of the trial. Statistical results were reported in accordance with Consolidated Standards of Reporting Trials (CONSORT) guidelines (www.consort-statement.org/).

Data validation

Prior to formal analysis, data were checked for outliers and missing values and validated using the defined score ranges for all outcome measures. Queries were reported to the trial co-ordinator and investigated. Standard statistical summaries (e.g. medians and ranges or means and variances, dependent on the distribution of the outcome) and graphical plots showing correlations were presented for the primary outcome measure and all secondary outcome measures. Baseline data were summarised to check comparability between treatment arms, and to highlight any characteristic differences between those individuals in the study, those ineligible for the study, and those eligible but withholding consent.

Missing data

As seemed likely at the outset, some data were not available because of the voluntary withdrawal of patients, lack of completion of individual data items or general loss to follow-up. Where possible, the reasons for data ‘missingness’ were ascertained and reported. Although missing data were not expected to be a problem for this study, the nature and pattern of the missingness were carefully considered – including, in particular, if data could be treated as missing completely at random (MCAR). If judged appropriate, we planned to impute missing data using the multiple imputation facilities (mice package) available in R. Any imputation methods used for scores and other derived variables were carefully considered and justified. We planned that if the degree of missingness was relatively low, as expected, the primary analysis would be based on complete cases only (complete case analysis), with analysis of imputed data sets used to assess the sensitivity of the analysis to the missing data. Reasons for ineligibility, non-compliance, withdrawal or other protocol violations would be stated and any patterns summarised. More formal analysis, for example using logistic regression with ‘protocol violation’ as a response, was also considered if appropriate and if it aided interpretation.

Interim analyses

Interim analyses were performed only where directed by the DMC. Any interim analyses would follow the same procedure as the final analyses.

Final statistical analyses

Reporting

Wherever possible, the results of all analyses were presented in a simple and easy-to-follow manner and related any observed differences to their clinical importance, so that they could be clearly understood by those with only rudimentary statistical knowledge. Open and confidential reports of the statistical analyses were planned, as required, by the trial statistician and, where appropriate, results were to be disseminated through peer-reviewed journals, conference presentations and local mechanisms.

Objectives

The economic evaluation was designed to estimate costs of distal radial fractures treated by locking-plate fixation and by K-wire fixation. If appropriate, the primary objective was to evaluate the incremental cost-effectiveness of distal radial fractures treated by locking-plate fixation versus K-wire fixation.

A within-trial analysis comparing the outcomes and costs up to 12 months’ follow-up using trial data was undertaken for two age groups: (1) patients < 50 years of age and (2) patients ≥ 50 years of age.

The evaluation followed the reference case guidance for technology appraisals set out by the National Institute for Health and Care Excellence (NICE). 33

In the longer term, we will develop a cost-effectiveness analysis modelling outcomes and costs up to 10 years post surgery. The methods for this second analysis will be presented in due course. However, the results for this analysis are not presented in the present report; this will require longer follow-up data, which are not currently available.

Measurement of outcomes

The economic analysis used quality-adjusted life-years (QALYs) to measure health outcomes. Health-related quality of life was estimated using responses from the EuroQol – Five Dimensions 3 Level. 26 Patients completed the EQ-5D questionnaire at baseline, 3 months, 6 months and 12 months after randomisation as secondary outcomes of the trial. Standard UK tariff values were applied to these responses at each time point to obtain utility. QALYs were calculated as an ‘area under the curve’ and formed the main outcome measure of the study. In addition, patients self-completed EQ-5D assessing their pre-injury quality of life retrospectively.

Measurement of costs

Patient-reported data on resource usage were collected within the trial at 3 months, 6 months and 12 months. For the 3-month data, the recall period was since discharge from hospital. For the other cases, it was since the last questionnaire was due to be completed. The questionnaires included number of further inpatient stays following the initial operation (including specialty and length of stay), number of outpatient, primary and community care visits, use of aids and adaptations, and medication use. Patients also reported the use of personal social services related to their treatment, as well as private costs such as treatments within private settings and time off work. Personal social services included number of weeks of frozen/hot meals on wheels, laundry services and number of visits of carers and social workers.

Costs were estimated combining patient-reported resource usage with unit cost data obtained from national databases such as the British National Formulary (BNF) and Personal Social Services Research Unit (PSSRU) Costs of Health and Social Care. 34 Any unit cost that was not available was estimated in consultation with the UHCW finance department.

The analyses initially took the perspective of the NHS including the costs of health and social care. Subsequent analyses adopted a societal perspective taking into account productivity costs (time away from work) and out-of-pocket expenditures incurred by patients in relation with their treatment. The currency used is pounds sterling (£).

Missing data

Although missing data were not expected to be a problem in this study, the economic analyses followed the nature and pattern of the missingness reported by the statisticians. This was illustrated as MCAR, missing at random or not missing at random.

A base-case analysis of the complete data set was conducted including only patients with complete cost and QALY data.

We then explored the use of the last observation carried forward (LOCF) approach to deal with missing values. Costs and EQ-5D values were carried forward for only one successive missing observation.

Finally, we imputed missing data through multiple imputations. 35,36 This method, assuming that data are missing at random, replaces the missing values with plausible substitutes based on the distribution of observed data and, by using iterative multivariable regression techniques, includes randomness to reflect uncertainty. All baseline variables were used to impute missing data. This approach is recommended for economic analyses alongside clinical trials, as it reflects the uncertainty inherent in replacing missing data. 12 We considered the impact on the cost-effectiveness results using both imputation methods in the sensitivity analyses.

Within-trial analysis

Longer-term analysis

For the longer term, we will use decision-analytic modelling to compare locking-plate fixation and K-wire fixation for distal radial fractures over a longer time frame. We will use a 10-year Markov decision model based upon within clinical trial results to estimate the long-term effectiveness of fixations and the proportional hazards of complications and also revision in both people < 50 years of age and people ≥ 50 years of age.

The model will include four possible states: (1) successful fixation; (2) complications; (3) revised fixation/surgery; and (4) death and its structure (Figure 3).

FIGURE 3.

States and allowable state transitions corresponding to use of a distal radius acute fracture fixation.

The impact of complications and subsequent revision surgery will be assessed using information from various sources: the extrapolation of within-trial data and the longer complication follow-up data which is currently ongoing, the published literature and/or expert opinions to create the most accurate model to compare locking-plate fixation and K-wire fixation for distal radial fractures. We will conduct a systematic review of cost-effectiveness models in the treatment distal radius fracture. This will act as the main resource of transition probabilities for the model. Where possible, transition probabilities will be based on the latest UK studies reflecting current care pathways and associated expectations relating to long-term patient functioning.

Two different base case scenarios will be considered, < 50 years of age cohort and ≥ 50 years of age cohort, both followed for 10 years.

Following the NICE reference case,33 costs and benefits will be discounted at 3.5% and a NHS perspective adopted.

Deterministic one-way sensitivity analysis will be performed to check the results over a plausible range of prior distributions placed on the time-varying model parameters. Multiway deterministic sensitivity analyses will be undertaken by modelling optimistic (‘best case’) and pessimistic (‘worse case’) scenarios. 38

Reporting

The results of the within-trial analyses are presented such that people with basic cost-effectiveness analysis knowledge can understand it.

It was planned that results would be disseminated through peer-reviewed journals, conference presentations and other types of research work dissemination. We expected to provide the within-trial analysis results for the trial main paper and we aimed at producing one extra paper focusing on the full health economics analysis with more mature data in the longer term.

Ethics committee approval

The DRAFFT trial was approved by the Coventry Research Ethics Committee in February 2010 (REC reference 10/H1210/10) and by the research and development department of each participating centre. The final approved study protocol has been published.

Trial management group

The day-to-day management of the trial was the responsibility of the trial co-ordinator, based at Warwick Clinical Trials Unit (CTU) and supported by the CTU administrative staff. This was overseen by the TMG, which met monthly to assess progress. It was also the responsibility of the trial co-ordinator to undertake training of the research associates at each of the trial centres. The trial statistician and health economist were closely involved in the setting up of data capture systems and the design of databases and clinical reporting forms.

Trial Steering Committee

A TSC was responsible for monitoring and supervising the progress of the DRAFFT trial. The TSC consisted of four independent experts, a lay member and leading members of the TMG. Membership of the TSC is given in Acknowledgements.

Data Monitoring Committee

The DMC was independent of the trial and was tasked with monitoring ethical, safety and data integrity. The trial statistician provided data and analyses requested by the DMC at each of the meetings. Membership of the DMC is given in Acknowledgements.

Overall

The overall flow of patients in the trial is shown in Figure 4. The total number of patients screened was 12,162; Table 3 shows the age and gender split.

FIGURE 4.

Consolidated Standards of Reporting Trials (CONSORT) plot: overall flow of patients in the study. MUA, manipulation under anaesthetic.

The overall median age of screened patients, where data were available, was 53 years [interquartile range (IQR) 18–72 years]; for females it was 63 years (IQR 39–77 years) and for males 23 years (IQR 12–48 years).

There was clear bimodality in the age distributions for each gender (Figure 5), particularly for females. A Gamma mixture model, gammamixEM function in R package mixtools, was fitted to data from each gender to estimate the proportionate split and median ages in each group; bootstrapping (n = 100) was used to estimate 95% CIs for these statistics.

FIGURE 5.

Age distribution of screened patient bars and fitted distributions for (a) females (n = 7123) and (b) males (n = 3628).

The modelling suggests that, for females, the ‘young’ group has a median age of 25.1 years (95% CI 24.1 to 25.8 years) and the ‘old’ group has a median age of 71.1 years (95% CI 70.6 to 71.5 years), with the split by group (young : old) given by 32.3% : 67.7% (95% CI 30.8 to 33.5). For males, the ‘young’ group has a median age of 19.7 years (95% CI 18.3 to 20.6 years) and the ‘old’ group has a median age of 63.7 years (95% CI 61.3 to 65.1 years), with the split by group (young : old) given by 73.5% : 26.5% (95% CI 69.3 to 76.0).

A recent study by Court-Brown and Caeser16 assessed over 1000 patients with a fracture of the distal radius. This study suggested that there is a clear bimodal distribution for this type of fracture according to the age of the patient; with a crossover of the two peaks of incidence at around 50 years of age. This was also the case for our data, for both males and females, with Figure 5 showing that the approximate division between the two age groups was also at around 50 years. Stratification on the basis of age was used to discriminate between younger patients with normal bone quality sustaining high-energy fractures and older patients with low-energy (fragility) fractures related to osteoporosis.

Eligibility

Figure 4 shows that the total number eligible for the study was 639, that is the total number screened (12,162) minus the total number ineligible (11,523). The main reasons for patient ineligibility were that the fracture did not require fixation (32.1%), the fracture was dorsally displaced > 3 cm from the radiocarpal joint (26.1%) and the patient was aged < 18 years (27.8%).

Willingness

Of the 639 eligible patients, 461 (72.1%) were willing and 178 were unwilling to take part in the trial. The main reason for being unwilling to participate was a simple unwillingness to be part of a research project (55 patients; 30.9%). Of those patients who expressed a preference, there was a balance between the treatment groups, with 26 patients stating a preference for wires (14.6%) and 29 stating a preference for a plate (16.3%).

In total, 461 patients were willing to participate in the DRAFFT study. The gender and age of the individuals who were willing and unwilling to participate in the study are shown in Table 4 and the distribution of ages by gender and group in Figure 6.

FIGURE 6.

Box plot showing age distribution by gender and willingness to participate in trial. F, female; M, male.

A t-test showed that ages did not differ significantly between groups (p = 0.088) and a chi-squared test showed no difference in gender split between groups (p = 0.301).

Overall

Overall trends in recruitment are shown in Figure 7; the final number of participants recruited to the study was 461 (Table 6).

FIGURE 7.

Overall DRAFFT recruitment (thick black line) and recruitment by centre (January 2011 to June 2012); logarithmic scale used to enhance detail for individual centres. Original target = 390 patients.

The final number of recruited participants was above the initial target of 390.

Recruitment proceeded at a faster rate than expected; therefore, the decision was taken to allow the sample size to increase above the target of 390 patients.

Population characteristics

The randomisation was stratified by hospital, age group and intra-articular extension. Table 7 shows recruitment numbers by treatment group, age group and intra-articular extension.

The breakdown by age, gender and treatment group is shown in Table 8.

The majority (304 out of 461) of patients recruited to the trial were female and ≥ 50 years of age, as expected.

Surgeons

Treatments were undertaken by 244 different surgeons; the median number of operations per surgeon was 1 (IQR 1–2). Figure 8 shows a histogram of the number of procedures undertaken by each surgeon.

FIGURE 8.

Histogram of numbers of operation for each surgeon.

As expected, any individual surgeon operated only on a small number of patients (n = 2 or 3) enrolled in the study; 88% of surgeons (215 out of 244) treated fewer than three study participants. This greatly reduces the likelihood of a surgeon-specific effect on the outcome at any one centre, that is one particularly good or bad surgeon dominating the other surgeons in the study.

Operations

Details of the surgeon grade and experience and methods employed for each procedure are shown in Table 9 and operation times by surgeon grade are shown in Figure 9.

FIGURE 9.

Operation times by surgeon grade and procedure.

There was no evidence to suggest that operation times differed between surgeon grades.

Treatment allocation

Table 10 shows the treatment allocation and the treatments received by treatment group. In the K-wire group 90.4% of participants received their allocated intervention and in the locking-plate group 92.2% received their allocated intervention.

Unless stated otherwise, all analyses reported here will be on an intention-to-treat basis, that is by allocated treatment.

Baseline characteristics

The baseline characteristics are well balanced between treatment groups (Table 11).

Imaging outcomes

Dorsal angle (°) and ulnar variance (mm) data were extracted from available images from all study participants before their operation (pre-op), and 6 weeks and 12 months postoperatively. In addition to the principal clinician extracting these images (assessor 1), an additional clinician (assessor 2) also assessed and extracted data independently. Table 12 shows estimates of intraclass correlation coefficients (ICCs), with 95% CIs based on 1000 bootstrapped samples, for dorsal angle and ulnar variance data at the preoperative assessment, and at 6 weeks and 12 months postoperatively. The ICC provides a measure of agreement between the two assessors and can be interpreted as follows: 0 to 0.2 poor, 0.2 to 0.4 fair, 0.4 to 0.6 moderate, 0.6 to 0.8 substantial and 0.8 to 1.0 almost perfect.

It is clear from Table 12 that for both measures there was almost perfect agreement between both assessors. This is displayed visually by the Bland–Altman plots in Figure 10, where the difference between data from the two assessors is plotted against the mean with 95% CIs. We conclude from the analysis presented in Table 12 and Figure 10 that there was excellent agreement between assessors, indicating a high level of confidence in the measurements made by the principal assessor (assessor 1). The following comparative analysis (between intervention groups) is based purely on the data extracted from the images by this assessor.

FIGURE 10.

Bland–Altman plots showing agreement between assessors 1 and 2 for dorsal angle (°) preoperatively (a), 6 weeks (c) and 1 year (e); and for ulnar variance (mm) preoperatively (b), 6 weeks (d) and 1 year (f). Plots show differences plotted against mean data, with 95% CIs indicated by dashed lines.

Image data were available from all 461 study participants before their operation (pre-op), and 6 weeks and 12 months postoperatively. However, some data were not available for analysis because of lack of collection, corrupted data files (which could not be opened) and poor-quality images that could not be analysed. In summary, only six study participants (1%) had no image data available at any occasion.

Metaphyseal comminution was reported at the preoperative assessment, in addition to dorsal angle and ulnar variance, and appeared to be well balanced across intervention groups, with 70.0% (152 out of 217) and 70.1% (155 out of 221) of study participants being assessed affirmatively for this characteristic for the K-wire and locking-plate groups respectively. Table 13 shows the mean and SD for imaging outcomes at baseline, 6 weeks and 12 months postoperatively, and estimated treatment effects after adjustment at 6 weeks and 12 months. Adjusted effects were from mixed-effects regression analysis based on complete cases with treatment group, age group, sex and intra-articular extension as covariates (fixed effects) and recruiting centre as a random effect.

In summary, these results indicate that there was a significantly larger (more positive) dorsal angle in the K-wire group than the locking-plate group, at both 6 weeks and 12 months post operation. In addition, the variability in dorsal angle was significantly less in the locking-plate group than in the K-wire group at both 6 weeks and 12 months (F-test to compare the variances of two samples from normal populations; p = 0.020 and p < 0.001 and ratio of variances 0.73 and 0.45, at 6 weeks and 12 months, respectively). There was a significantly larger (more positive) ulnar variance in the K-wire group than in the locking-plate group, at both 6 weeks and 12 months post operation. However, there was no difference in variability between groups for ulnar variance at 6 weeks and 12 months (p = 0.615 and p = 0.081).

The image data are summarised further in the box plots of Figure 11, which clearly show the temporal trends in data and the difference in variance between dorsal angle data between groups at 12 months post operation.

FIGURE 11.

Box plots of baseline and postoperative scores and trends in means (filled circles) with 95% CIs for (a) dorsal angle and (b) ulnar variance.

Functional and quality of life outcomes

The primary outcome for this study is the PRWE questionnaire score23 at 12 months after surgery. Early functional status was also assessed and reported at 3 months and 6 months. The DASH questionnaire25 was used as a more general measure of physical function and symptoms in people with musculoskeletal disorders of the upper limb. Quality of life was assessed using EQ-5D. Data are summarised for each outcome measure, at each assessment occasion, in Tables 14–16. Baseline data are retrospective measurements made by study participants to assess pre-injury function and quality of life at recruitment. For EQ-5D, participants also made a post-injury assessment.

Figures 12–14 show trends in means and box plots of distribution of scores for PRWE, DASH and EQ-5D by treatment group.

FIGURE 12.

(a) Box plots of postoperative scores and (b) trends in means with 95% CIs for the PRWE questionnaire scores.

FIGURE 13.

(a) Box plots of postoperative scores and (b) trends in means with 95% CIs for the DASH questionnaire scores.

FIGURE 14.

(a) Box plots of postoperative scores and (b) trends in means with 95% CIs for EQ-5D scores.

There appears to be, from a visual inspection of the plots, very little evidence for treatment group difference for the PRWE questionnaire score but some evidence for marginally better function in the locking-plate group than the K-wire group for the DASH questionnaire scores. Scores recover in the postoperative period in both groups but function is still worse than before the injury (based on retrospective assessment). Both PRWE and DASH questionnaire scores are left-skewed at the 12-month assessment, as they approach the score minimum. There is also very little evidence for treatment group difference for the EQ-5D score.

Table 17 shows means and SDs by treatment group. A mixed-effects linear regression model is used to estimate treatment differences (and 95% CIs). The full (planned) model incorporated terms that allowed for possible heterogeneity in responses for patients owing to the recruiting centre and the surgeon, in addition to the fixed effects of the treatment groups, sex, patient age and intra-articular extension. However, as we foresaw in the study protocol, the overwhelming majority of surgeons operated on fewer than three patients (see Chapter 3, Surgeons and operations), so a simplified model is used that implements a single random effect to account for the recruiting centre. Although baseline PRWE questionnaire scores were recorded using a retrospective assessment by each patient post injury, these were not considered sufficiently reliable for formal baseline adjustment in the multilevel model. The interactions between the stratifying variables (age and intra-articular extension) and the main treatment effect were not expected to be large; therefore, for practical reasons, the study was not powered with subgroup analysis in mind. However, formal tests of interaction between each stratifying variable and the treatment factor are reported together with appropriate 95% CIs and p-values in Table 17.

The adjusted estimate of the treatment effect for the PRWE questionnaire score for the full population is –1.3 (95% CI –4.5 to 1.8) in favour of the locking-plate group; the p-value of 0.398 indicates that there is no evidence for a statistically significant difference in PRWE questionnaire scores between the treatment groups at 12 months post operation. The MCID for the PRWE questionnaire score is 6 points; therefore, we conclude, based on our estimated CIs, that if there is a difference in functional scores between treatment groups it is likely to be sufficiently small as to be clinically unimportant.

The ≥ 50 years age group formed a sizeable subgroup of patients; therefore, this was identified a priori as being of particular interest. The adjusted estimate of the treatment effect for the full population is –2.2 (95% CI –5.8 to 1.4) in favour of the locking-plate group; the p-value of 0.338 indicates there is no evidence for a significant interaction term between age group and treatment group.

There was some evidence for a small and marginally significant (p = 0.051) treatment effect for the DASH score at 12 months in favour of the locking-plate group, although the effect size was small, at –3.2 (95% CI –6.5 to 0.0). As for the PRWE questionnaire scores, an interaction test provided no evidence that the treatment effect for DASH differed between age groups (p = 0.309) and the small improvement in function for the plate group is evident in the over-50-years age group only. EQ-5D failed to show any significant differences between treatment groups.

A likelihood ratio test provided no strong evidence to support the inclusion of a random-effect to model hospital centre (p = 0.119; chi-squared test) for the primary outcome measure, PRWE questionnaire score. In addition, the effects of age group, sex and intra-articular extension were not significant (at the 5% level) in the fitted model for PRWE questionnaire scores [p-values 0.764, 0.090 and 0.462; F-tests from analysis of variance (ANOVA)]. Therefore, the inferences for the PRWE questionnaire scores would be approximately equivalent to the above if a simple t-test had been used to compare treatment groups (t-test; p = 0.396 and 95% CI for raw difference –4.5 to 1.8). The above analyses are based on assumed approximate normality of the residuals for the PRWE questionnaire score at 12 months; the box plots in Figure 12 suggest that this approximation is poor. The median 12-month PRWE questionnaire scores were 9.0 for the K-wire group and 7.5 for the locking-plate group, and a Mann–Whitney test also provided no evidence to suggest that the treatment groups differed significantly (0.127). Inferences for the DASH questionnaire score were also equivalent between unadjusted and adjusted analysis (t-test; p = 0.055 and 95% CI for raw difference –6.6 to 0.1); median DASH questionnaire scores were 8.3 for the K-wire group and 6.7 for the locking-plate group, and a Mann–Whitney test also provided some evidence to suggest that the treatment groups differed significantly (0.071).

An adjusted per-protocol analysis of the PRWE questionnaire gave an adjusted treatment effect estimate of –1.0 (95% CI –4.2 to 2.2) and p-value equal to 0.530. An adjusted per-protocol analysis of the DASH questionnaire gave an adjusted treatment effect estimate of –3.1 (95% CI –6.3 to 0.2) and p-value equal to 0.066.

Missing data

There are 46 study participants with missing data at the 12-month study end point; the data are 90.0% (415 out of 461) complete. A summary of missing values, by age group and sex, for the PRWE questionnaire at 12 months is shown in Table 18 and the full pattern of missing values is shown in Tables 19 and 20 for K-wire and locking-plate groups, respectively. Baseline data, age group, hospital, treatment allocation, intra-articular extension and sex were all complete.

Tables 19 and 20 show that 336 study participants (n = 168 + n = 168) had complete score data on all occasions. There is no evidence that missingness patterns differed between treatment groups. Forty-five study participants did not provide PRWE questionnaire score data at 12 months: 19 in the K-wire group and 26 in the locking-plate group. The data were 90% complete for the primary outcome measure. In total, only 19 study participants had no post-baseline data: eight in the wire group and 11 in the plate group.

Logistic regression, with missing data coded as 1 and complete data as 0, indicated that age group and sex were predictive of the PRWE questionnaire score missingness at 12 months; p-values from chi-squared tests for including age and sex in the logistic regression model were 0.017 and 0.022 respectively. Males were more likely not to complete questionnaires than females, and those < 50 years were less likely to complete questionnaires than those ≥ 50 years of age. This is clearly apparent from an inspection of Table 19.

Study participants aged < 50 years and males were less likely to provide complete data; experience suggests this is not an unexpected result. Assuming that missingness is fully explained by these variables (for which we have complete data), that is assuming that data are missing at random, our approach to the analysis in which we adjust for the variables that are predictive of missingness is sensible and should yield unbiased estimates of the treatment effect.

Imputing the missing data and rerunning the analysis on fully complete data provides a useful sensitivity analysis. Missing data were imputed using the mice package in R (multiple imputation by chained equations; http://cran.r-project.org/web/packages/mice/) and pooled estimates35 of model parameters based on 50 data imputations were obtained for the mixed-effects regression models reported in Table 9. The pooled estimates of the treatment group effect were –1.2 (95% CI –4.2 to 1.9) for the PRWE and –3.2 (95% CI –6.2 to –0.1) for the DASH questionnaire scores, with the percentage of the variability attributable to the uncertainty caused by the missing data estimated at 5.8% and 5.5% for the two outcome measures respectively. In summary, the inferences based on the complete data, after imputation, are not markedly different from those reported from the complete case analysis in Table 18.

Complications

Complications were reported by participants at the 6-week follow-up assessment, and also at any point during the first 12 months of follow-up using the study adverse event reporting procedures. Table 21 shows numbers of patients reporting complications by treatment group.

There was no evidence to suggest that rates for any of the reported complications (e.g. wound infections) differed between treatment groups, based on comparing counts in groups using Fisher’s exact test. The proportion of participants given a cast after the operation differed between treatment groups, as expected, as did, consequently, the number of plaster changes required.

Unit cost of resource use

Individual-level resource use was combined with unit costs to calculate the total health-care use cost for each patient. In order to convert resource usage figures into costs, unit cost figures were assigned from national databases such as the PSSRU Costs of Health and Social Care 201234 and the Department of Health’s National Schedule of Reference Costs. 39 Further inpatient care following the initial operation was costed as Minimal Elbow and Lower Arm Procedures for Non-Trauma (HRG codes HB79Z and HB73Z) except if surgical hand complications such as metal removal or debridement were reported at 6-week follow-up, where inpatient stay was costed as Minor Elbow and Lower Arm Procedures for Trauma (HRG codes HA73B and HAB73C). All costs were adjusted to 2012 prices using the Campbell and Cochrane Economics Methods Group and Evidence for Policy and Practice Information and Coordinating Centre Cost Converter. 40 Table 22 presents the summary of unit costs.

Unit cost of medications

Unit costs for medications were obtained from the BNF No. 67 (September 2013)41 and the NHS Electronic Drug Tariff (October 2013). 42 Patients reported details for medications that were taken within the three budgetary periods (discharge to 3 months, 3–6 months, 6–12 months), including the quantity taken per day and the number of days. We calculated the total medication costs using the average cost per dose for each product and the mean quantity taken per day during the reported number of days. Where a dose range reported ‘as required’, we included the cost of one box of the drug. We assumed that all medications were in tablet form unless stated. If the dose of the drug was not recorded, we assumed the patient received the same dosage as other patients who reported the same drugs. If the quantity was not recorded, we applied the average quantity for that drug as reported in the data. Table 23 shows all the unit costs for the drugs that were reported in the trial.

Unit cost of aids and adaptations

Patients reported any equipment that they had used to protect their injury or make their daily life easier to manage and the number of those. A number of aids and adaption tools were suggested and patients could add other equipment that was not in the list. Unit costs for aids and adaptations were taken from the website MobilitySmart (www.mobilitysmart.cc/), accessed in October 2013, which supplies the NHS and the PSSRU Costs of Health and Social Care 2012. 34 We calculated the total aids costs using the number of aids and the cost for each product. If the quantity was not recorded, we applied the average quantity for that aid as reported in the data. Reported aids and adaptations and unit prices are available in Table 24. Those costs were incurred by the NHS.

Societal costs

Costs from the societal perspective were calculated by combining productivity loss and loss of earnings due to work absence, private costs such as treatments within private settings and out-of-pocket expenses incurred as a result of the wrist surgery, and reported use of Personal Social Services related to the treatment. We assumed that the Personal Social Services costs were unlikely to be directly supported by the NHS and this is why they were included in the societal perspective sensitivity analysis.

Personal social services included number of weeks of frozen/hot meals on wheels, laundry services per load, and number of visits of carers and social workers. Unit costs were assigned using PSSRU and Information Centre of Personal Social Services in councils; they are reported in Table 26.

Patients reported their time off work in days or in terms of lost income because of their wrist fracture at every collection point. We employed a human capital approach using the gross median weekly pay rate for full-time employees from the Office for National Statistics (£506, April 2012)44 and divided this by 5 to generate the cost of lost productivity per day.

Utility and quality-adjusted life-years

Patient health-related quality of life was assessed using the EQ-5D,43 which was included along with the patient resource-use questionnaires. Changes in EQ-5D scores at baseline, 3 months, 6 months and 12 months were evaluated using two-sample t-tests to explore any important differences in these end points within the time frame of the trial. They are useful to understand the impact of the length of the follow-up on repeated measurements of the health outcome in which insufficiently long follow-up periods may potentially introduce biases in the subsequent cost-effectiveness analysis.

In line with the NICE reference case,33 the primary outcome for the economic evaluation was QALY. Patient responses to the EQ-5D questionnaire at each time point were converted to QALYs using the standard UK tariff values45 and an area under the curve approach. QALYs were calculated by multiplying these values with the time spent in each state, with quality of life linearly interpolated for the periods between the four observations provided in the trial data. QALYs represent a quality-weighted survival value in which 1 QALY is the equivalent of 1 year of full health. Average QALYs between adjacent time points were calculated to generate smoothed estimates between those time points.

Missing data

We calculated the mean total costs per patient from a NHS perspective adding the cost of inpatient stay, outpatient visit, consultations, medication, equipment, and applicable intervention costs for all patients where response data were available. Respondents who fail to complete individual items of the EQ-5D are not allocated a utility index score. From the overall sample, missing data represented 7.07%. Importantly, the health economics criteria for inclusion were slightly more restrictive than those for the statistical analysis. The complete data analysis was based on 278 patients.

For those cases in which neither resource usage nor quality-of-life data were available, these figures could not be calculated. We addressed missingness using two alternative methods.

We initially used the LOCF approach. Costs and EQ-5D values were carried forward for only one successive missing observation. Patients with more than one consecutive missing observation were omitted from the analyses. Observations were not carried backwards. The LOCF imputation analysis was then conducted on a sample of 367 patients. The advantage of the LOCF method is that it minimises the number of subjects who are eliminated from the analysis. However, the LOCF method has also been found to give a biased estimate of the treatment effect and underestimate the variability of the estimated result. 46,47

We then used multiple imputations via chained equations35,36 to complete missing data assuming missing at random and using Stata 12 (StataCorp LP, College Station, TX, USA). Missing cost data were predicted in terms of QALYs, treatment received, length of stay, age, gender, job status, patient’s self-reported health status, PRWE questionnaire score and DASH questionnaire score. Missing QALY data were predicted in terms of this same list (excluding QALYs), plus each of the cost items. In order to remove implausible data, missing cost data were constrained to be positive. A total of 10 imputations were created to stabilise the result. The reported cost-effectiveness results were synthesised based on all imputed data sets.

Cost-effectiveness results

Resource use and QALY data were available for 278 patients, with 137 patients receiving volar locking-plate fixation and 141 patients receiving K-wire fixation.

Health-care resource use

Table 27 presents the details of health-care resource used over the 12 months after discharge for complete case data. Resource use is broadly comparable between the two treatments. Patients were frequent users of physiotherapy outpatient visits (6.9 visits over the 12 months) and reported on average 3.6 radiology visits and 3.6 visits in orthopaedics outpatient visits. Use of general practitioner (GP) visits and nurse visits was infrequent. Patients reported using a wrist brace or splint. In terms of medications, we note that paracetamol tablets were the most reported medications in the questionnaires. No significant differences were observed between the two groups. Patients receiving K-wires appear to have used more bathing aids than patients receiving volar locking plates at a 10% significance level (p = 0.087).

Costs

The total costs associated with resource use during the trial among complete case data are shown in Table 28. The mean total NHS resource use costs are £3385 for K-wires and £4288 for volar locking plates and were significantly higher for volar locking plates (by £903). Lost earnings and productivity losses to employers through sickness absences appeared higher in the K-wire arm, but the difference does not appear to be significant. Overall, societal costs are on average £48 higher in the K-wire arm. However, there were no significant differences in NHS resource use costs and societal costs between the treatment arms according to the t-tests.

Health outcomes

Table 29 details the EQ-5D scores at pre injury, baseline, 3 months, 6 months and 12 months in complete case data and when missing values are imputed. Both groups showed increasing EQ-5D scores from baseline to the last follow-up point; the most important increase is observed between baseline and 3-month follow-up point for both treatments, doubling the baseline EQ-5D score. It is noted that patients at 12 months have not recovered an EQ-5D score equivalent to their pre-injury, retrospectively reported, EQ-5D score.

On average, patients receiving a volar locking plate had higher EQ-5D index scores at baseline and at each follow-up point, but independent-sample t-tests indicated that the changes in EQ-5D score over time were not statistically significant. The average total number of QALYs over the 12 months was marginally higher in the volar locking-plate arm (0.745) than in the K-wire arm (0.731). Tables 30 and 31 provide the mean EQ-5D change scores between pre-injury and the follow-up points for the LOCF imputed analysis and multiple imputed data respectively. The same comments applied to the EQ-5D scores. The statistical tests indicated that there were no significant differences between EQ-5D scores at baseline for the two treatment arms, and, for this reason, adjustments on EQ-5D at baseline were not required.

Cost-effectiveness results within the NHS perspective

Table 32 shows the total costs and QALY gains for each of the treatment arms. The differences in QALY gains between groups, 0.008 QALYs, were minimal and suggested only small health benefits of volar locking plates over K-wires. The volar locking-plate group had the highest QALY gains over the trial period. The mean total costs were higher for the volar locking-plate group; the high SD for the cost estimates indicates the presence of a few outlying individuals who incurred significant health service costs.

Table 33 provides the cost-effectiveness results, showing the incremental costs and benefits as well as the ICER. Interpretation should be tempered by considering the small between-group differences in QALY gains and the high level of uncertainty surrounding the QALY estimates. The results suggest that volar locking plates were more expensive than K-wires, with an incremental cost of £726.46, and had higher, albeit small, QALY gains.

We graphically represented the uncertainty of these cost-effectiveness estimates in a cost-effectiveness plane (Figure 15) using bootstrapping with 10,000 iterations. We sampled, at random with replacement, each of our 10 imputed data sets, producing 10,000 incremental cost and incremental QALY estimates. Figure 15 shows that all the points are above the x-axis, indicating that the volar locking plate was more costly than K-wires, and more points are to the right of the y-axis, indicating that the volar locking plate produced more QALYs than K-wires.

FIGURE 15.

Cost-effectiveness plane generated from bootstrapped mean cost and QALY differences over 12 months (NHS perspective, all patients).

The points lay in the north-east quadrant of the cost-effectiveness plane. In this area there is a trade-off between effect and cost: additional health benefit can be obtained but at a higher cost. The question that then arises is whether or not the trade-off is acceptable, that is if the health gain is worth the additional cost. 48 This decision is based on the ICER and what the decision-makers are willing to pay for the additional benefit. The guideline rule is the NICE threshold of £20,000–30,000 per QALY, and the ICER of volar locking plates versus K-wires equals £89,322 per QALY. Under these circumstances, volar locking plates cannot be considered cost-effective, as its ICER is above the £20,000–30,000 per QALY range.

The CEAC showing the probability that volar locking-plate fixation is cost-effective is presented in Figure 16 with a range of cost-effectiveness WTP threshold values. The probability of being cost-effective at a WTP threshold of £20,000 per QALYs was nil and at a threshold of £30,000 was 3%.

FIGURE 16.

Cost-effectiveness acceptability curves at 12 months (NHS perspective, all patients).

Sensitivity analyses

We undertook a number of sensitivity analyses to test the robustness of the base-case results, which are presented in Table 34.

The first sensitivity analysis was based on complete-case data. This resulted in an incremental cost of £903 and an incremental QALY gain of 0.014, yielding an ICER of £64,026 per QALY.

The sensitivity analyses also expanded the cost-effectiveness analysis to the societal perspective. The incremental societal cost was £581 and the QALY gain was very comparable from the base-case analysis.

All the sensitivity analyses showed a higher mean cost and a higher but marginal QALY gain per patient receiving a volar locking plate. The ICERs of volar locking-plate fixation compared with K-wires were always more than £44,000 per QALY.

We also carried out cost-effectiveness analyses adjusting for the following baseline covariates including age, gender and EQ-5D scores. The adjusted analysis on the multiple imputations showed a lower incremental cost for a similar incremental QALYs gain both in the NHS and the societal perspectives; however, both ICERs remained above the cost-effectiveness threshold of £30,000 per QALY.

Subgroup analyses

The same analysis was then undertaken for patients < 50 years of age and patients ≥ 50 years of age, as the two age groups are assumed to present different types of fractures (high- and low-energy fractures) and so the impact of the fixation is expected to differ by age.

Table 35 shows the total costs and QALY gains for each of the treatment arms in each subgroup. Differences in QALY gains between groups were minimal in both age groups. The volar locking-plate group had the highest QALY gains in patients aged ≥ 50 years and K-wires had the highest QALY gains in younger patients. The mean total costs were higher for the volar locking-plate arm for both age subgroups. Older patients showed higher total costs than younger patients.

Table 36 provides the cost-effectiveness results, showing the incremental costs and benefits as well as the ICER. Interpretation should be tempered by considering the small between-group differences in QALY gains and the high level of uncertainty surrounding the QALY estimates.

In the sample of patients aged ≥ 50 years, the evaluation suggests that patients treated with a volar locking-plate fixation gained 0.014 QALYs more than those treated with a K-wire fixation, at an increased cost of £752 per patient, yielding an ICER of £54,218 per QALY. Given a WTP of £20,000 for an additional QALY, the probability of volar locking plates being cost-effective is 46% based on the trial data, as shown in Figure 17.

FIGURE 17.

Cost-effectiveness acceptability curves by age subgroups at 12 months (NHS perspective).

In the sample of patients aged < 50 years, the evaluation suggests that volar locking-plate fixation is associated with both lower benefits and higher costs than if they are treated with K-wires. The K-wire fixation appears to dominate for this group of individuals.

Sensitivity analyses

We undertook a number of sensitivity analyses to test the robustness of the subgroup results, which are presented in Table 37.

The results based on the complete cases are the same as the main analysis for patients aged ≥ 50 years, showing an incremental QALY gain of approximately 0.02 and an incremental cost between £629 and £945, yielding ICERs higher than £35,323 per QALY. With regard to younger patients, the QALY gain was close to nil for patients aged < 50 years with an additional cost of £778, and K-wires would be a preferred option for cost-minimisation reasons.

Declined to participate

Only 178 potentially eligible patients (28%) declined to take part. This is reassuring in terms of the trial paperwork and in particular the state of equipoise presented in the patient information sheet, but also in terms of the generalisability of the trial to the broader population. The most common reason (55 patients, 31% of those who declined to participate) was the ‘patient did not want to take part in a research project’.

Importantly, the number of patients declining to take part because they had a strong preference for one intervention was both small and equally divided between the treatment arms: 26 patients (15%) ‘definitely wants K-wires’ and 29 patients (16%) ‘definitely wants plate’. Therefore, patients had equipoise and patient preference should not have influenced the result of the trial.

Treatment according to allocation

This was clearly an important issue because either surgeon or patient preference could have led to crossover from one group to the other post randomisation. However, > 90% of both patient groups had their allocated treatment according to protocol, 208 out of 231 patients allocated to K-wires (90%) and 213 out of 230 patients allocated to locking plates (92%). This ensured that the primary intention-to-treat analysis was not different from the planned secondary per-treatment analysis. Interestingly, in the case of nine patients, the surgeon decided not to use either form of fixation, that is the patient had a manipulation under anaesthetic and application of plaster cast without ‘metalwork’.

Recruitment by centre

Each of the 18 centres successfully recruited to the trial. As expected, the centres that opened earliest in the trial recruited the most patients overall. Most centres achieved a recruitment rate of 1–2 patients per month, indicating that the pre-trial recruitment rate estimate was accurate.

The lead centre recruited the most patients, as may be expected. However, there was also a noticeable difference in the rate of recruitment in some other centres. This may reflect the size of the catchment area (number of patients seen) at that centre or the amount of snow and ice, and hence falls, in that area, for example Newcastle and North Tyneside, but may also reflect a different ‘clinical threshold’ for offering patients surgery following a fracture of the distal radius. This warrants further investigation.

Patients

There were slightly more fractures of the left wrist than the right, but these were also evenly balanced between treatment groups, as were injuries to the ‘dominant hand’. This may reflect the fact that people generally carry things in their dominant hand, and, therefore, may be more likely to put out their non-dominant hand if they fall.

The number of patients with ‘pre-existing wrist problems’ was low (14% K-wires, 17% locking plates) and this was reflected in the pre-injury wrist scores, which were near ‘normal’ in both groups (PRWE questionnaire 3/100, where 0 is the best score).

As expected, the most common mechanism of injury was ‘low-energy fall’ at 82% in the K-wire group and 83% in the locking-plate group. All other patient characteristics were also evenly balanced between groups.

Fractures

As expected, the great majority of fractures were either displaced extra-articular (AO type A) or complete articular (AO type C) fractures. Partial articular (AO type B) fractures are most commonly palmar fractures or styloid fractures and would, therefore, not be eligible for this trial of dorsally displaced fractures of the distal radius. The type of fracture was evenly distributed between groups.

The other feature of note is that the proportion of AO type C3 fractures – the most severe intra-articular fractures – was low in both groups. This was also expected as these are the fractures where the surgeon would most commonly need ‘to open the fracture site to achieve reduction of the articular surface’ and were also, therefore, not eligible for the trial.

Surgeons

In this pragmatic trial, we expected that a large number of surgeons, reflecting the true breadth of expertise and experience provided within the NHS, would deliver the trial interventions. This was indeed the case, with 244 surgeons delivering a median of one operation each. This greatly reduces the likelihood of a surgeon-specific effect on the outcome at any one centre, that is one particularly good or bad surgeon dominating the other surgeons in the study.

However, it was also important to look at the balance of surgeons delivering the two different interventions. Interestingly, this was also well balanced between the groups.

Before the trial, some senior surgeons may have considered the K-wire intervention ‘simpler’ and therefore allowed more junior colleagues to perform this operation while the more experienced surgeons performed the more ‘complicated’ locking-plate fixation. This may have led to a ‘surgeon-effect’ within the trial. However, only marginally more of the plate fixations were performed by consultant (attending) surgeons: 31% for locking plates compared with 26% for K-wires. Similarly, the surgeon’s experience, as defined by the number of operations performed previously, was also well balanced between the groups. Therefore, a surgeon effect is very unlikely.

Just under half of the procedures in both groups (44% K-wires, 46% locking plates) were performed by ‘specialist registrars’, that is, trainees under supervision, which again reflects the reality of the delivery of trauma care in the NHS and is a testament to good training in this area.

Surgery

Although the principles of the fixation are inherent in the technique for both K-wires and locking plates, the details of the surgery were left to the discretion of the treating surgeon in this pragmatic trial.

Duration of surgery

A key finding of this trial is that the K-wire surgery took less than half the time of the locking-plate fixation, 31 minutes (24 to 45 minutes) for K-wires compared with 66 minutes (50 to 85 minutes) for locking plates. There was no evidence that there was a difference in the duration of surgery according to the seniority of the lead surgeon.

As access to trauma operating theatre space is a key limiting factor in the delivery of trauma care across the NHS, and indeed has secondary effects on the delivery of elective (planned) surgery as well, this finding is of particular importance to policy-makers and NHS managers. It also has financial implications, as theatre operating time is an expensive resource. However, it is difficult to ‘cost’ this time accurately, as many of the costs of theatre time – staff, cleaning, maintenance – are present even if the theatre is not being used at 100% capacity. Theatre time was, therefore, not included in the health economics analysis, which may have led to some underestimation of the cost-effectiveness of K-wire fixation.

Primary outcome

The main finding of this trial is that there is no difference in the primary outcome PRWE questionnaire score at 3 months, 6 months or 12 months (difference in PRWE questionnaire score at 12 months –1.3, 95% CI –4.5 to 1.8; p = 0.398) between patients treated with K-wire fixation versus locking-plate fixation.

As the CIs exclude the MCID for the PRWE questionnaire, we conclude that any difference in functional scores between treatment groups is unlikely to be important to patients.

The size of the treatment effect seen in this trial is consistent with that seen in other studies of recovery following fracture of the distal radius. MacDermid et al. 30 conducted a prospective cohort study of 129 patients with a fracture of the distal radius; all patients completed the PRWE questionnaire at their baseline clinic visit and at 2 months, 3 months, 6 months and 12 months following their fracture. They reported a mean PRWE questionnaire score of 20.0 (SD 20.6) at 6 months and a score of 13.5 (SD 17.0) at 12 months. This is consistent with our reported values of 21.5 (SD 18.1) and 14.6 (SD 16.4) for the full population at 6 months and 12 months postoperatively.

Secondary outcomes

Pre-planned subgroup and secondary analyses

At the beginning of the trial, three variables were identified as being of potential importance in terms of their influence on the recovery of patients following fracture of the distal radius. The randomisation sequence was therefore stratified on the basis of trial centre/hospital, age above and below 50 years as a surrogate for osteoporosis, and intra-articular extension of the fracture into the radiocarpal joint.

A single random effect was used to take account of the trial centre in the regression model used in the adjusted analysis; this made very little difference to the mean difference in PRWE questionnaire score between the groups (all participants: –1.3 adjusted, –1.4 raw).

The trial was not powered specifically for subgroup analysis. However, as three-quarters (314 out of 461) of the patients were in the ‘over 50 years’ subgroup, the estimate of the mean difference at 12 months is reasonably precise. The adjusted estimate of the mean difference between the treatments for the ‘over 50 subgroup’ is –2.2 points in favour of the plate fixation, but as the CIs exclude the MCID of 6 points on the PRWE questionnaire score (95% CI –5.8 to 1.4) this is unlikely to be clinically relevant.

The adjusted analysis for the ‘under 50 years’ subgroup was 1.4 in favour of the K-wire fixation. As the number of patients in this subgroup was smaller, the CIs are broader and do include the MCID (95% CI –5.0 to 7.8).

The number of patients analysed at 12 months with an intra-articular extension of the fracture (n = 200) was similar to those without (n = 215). Again, there was no evidence of a difference between the treatment groups. The mean difference was 0.7 (–3.8 to 5.2) for those patients with an intra-articular extension and –3.3 (–7.6 to 1.1) for those without.

Therefore, this trial provided no evidence of a clinically important difference between the effect of K-wire fixation and that of locking-plate fixation for patients under or over 50 years or for patients with and without an intra-articular extension of their distal radius fracture.

Finally, we also pre-planned a per-treatment (per protocol) secondary analysis of the data. As compliance with the trial was good, with > 90% of patients receiving their allocated treatment, the analysis by treatment given did not alter the result.

Health economic evaluation

Cost-effectiveness

There were no significant differences in health-related quality of life during the 12-month follow-up period and hence very little difference in average total QALY between the two groups: 0.08 difference, volar locking plate 0.745 versus K-wire 0.731.

As the locking-plate intervention was more expensive, with a total incremental cost of £726, the ICER of volar locking plates versus K-wires was £89,322 per QALY. Under these circumstances, volar locking plates cannot be considered cost-effective. The probability of being cost-effective at a WTP threshold of £20,000 per QALYs was nil and at a threshold of £30,000 was 3%.

Limitations

The main limitation of the trial is that it was not possible to blind either the surgeons or the patients to the study treatments. However, it could be argued that this is a positive feature in a pragmatic trial, as patients would know about their proposed treatment before surgery during routine care. 50

It should be noted that this study excluded patients whose fracture could not be reduced by indirect means, that is the results should not be generalised to the minority of patients whose fracture requires that the surgeon expose the individual bone fragments in order to restore the congruity of the wrist joint. The low number of patients in the trial with C3 fractures indicates that, as expected, the patients with the most complex intra-articular fractures were the ones who were excluded. None of the patients with extra-articular fractures should have been excluded, as this eligibility criterion refers only to the reduction of the ‘joint surface’. While it is possible that a small number of patients were excluded in error, we think this unlikely given the high quality of the screening data which we collected throughout the trial. Of course, we have no way of checking the radiographs of those patients whom the surgeon considered to be ineligible, but variation in the interpretation of fractures and fracture patterns is inherent in clinical practice and is reflected in this trial.

Compliance with the trial was good, with > 90% of patients receiving their allocated treatment. Some patients did cross over to have a different form of fixation, but an analysis by treatment given did not alter the result.

It is also important to underline that the health economics analysis used a conservative estimate of the cost of initial surgery. This did not take into account the extra theatre operating time required for the locking-plate fixation at 31 minutes for K-wires and 66 minutes for locking plates. It is likely that a microcosting analysis of the initial surgery in each arm could lead to even higher cost differences with regard to the cost of the initial surgery.

The long-term outcome of patients with a fracture of the distal radius is not known. Our data suggest that patients’ wrist function and quality of life improve over the 12 months following their surgery but do not return to pre-injury levels. The patients in this trial will be followed up annually, to determine the prevalence of late complications such as arthritis. In addition, this will provide us with data on the longer term that will allow a decision-analytic model where extrapolation will rely on more robust data on the benefits.

Trial management group

Professor Matthew L Costa, chief investigator; Dr Juul Achten, senior research fellow; Miss Jaclyn Brown, trial co-ordinator; Dr Nick R Parsons, trial statistician; Professor Sarah E Lamb, director of Clinical Trials Unit; Dr Caroline Plant, clinical input; Mrs Susie Hennings, senior project manager; Mrs Isabel Wall, trial administrator/acting trial co-ordinator; Miss Kate Packard, data clerk; and Mr Rhys Mant, data clerk.

Trial applicants

Professor Matthew L Costa, chief investigator; Professor Sarah E Lamb, co-applicant; Professor Damian Griffin, co-applicant; Dr Juul Achten, co-applicant; Dr Nick R Parsons, co-applicant; Dr Richard Edlin, co-applicant; and Mr Amar Rangan, co-applicant.

Trial Steering Committee

Professor David Torgerson, Director of York Trial Unit (chair); Jonathon Hill, Lecturer in Physiotherapy (independent member); Professor Ashley Blom, Professor of Orthopaedic Surgery (independent member); Professor Matthew L Costa, chief investigator; Professor Sarah E Lamb, director of Clinical Trials Unit; Professor Amar Rangan, principal investigator; Dr Juul Achten, senior research fellow; Miss Jaclyn Brown, trial co-ordinator; Dr Nick R Parsons, trial statistician; Dr Richard Edlin, trial health economist; Mrs Ceri Jones, manager, research and development, UHCW (sponsor representative); and Mrs Carole Beamish, lay member, quality assurance (independent member).

Data Monitoring Committee

Professor Lee Shepstone, Professor of Medical Statistics (chair); Mr Adrian Chojnowski, Consultant Orthopaedic Surgeon; and Mr David Ford, Consultant Orthopaedic Surgeon.

Statistician

Dr Nick Parsons.

Health economists

Dr Sandy Tubeuf and Dr Ge Yu.

Radiological evaluation

Dr Caroline Plant and Dr Salil Karkanis.

Research team: Distal Radius Acute Fracture Fixation Trial Study Group