VenUS IV (Venous leg Ulcer Study IV): Compression hosiery versus compression bandaging in the treatment of venous leg ulcers: a randomised controlled trial, mixed treatment comparison and decision analytic model.

Multicomponent compression bandage systems

In the UK, the most widely used multicomponent compression bandage system is the four-layer bandage (4LB). The short-stretch bandage (SSB) is also widely used. Both systems were compared in VenUS (Venous Ulcer Study) I, which randomised 387 participants with venous leg ulcers. 7 The study found that the 4LB significantly reduced time to healing [median survival time 92 days in the 4LB group compared with 126 days in the SSB group: adjusted hazard ratio (HR) 1.33, 95% confidence interval (CI) 1.05 to 1.67]. This study finding was supported in an individual patient data (IPD) analysis of all relevant RCTs,24 in which the 4LB was more effective in terms of ulcer healing than the short-stretch system (adjusted HR 1.31, 95% CI 1.09 to 1.58). Thus, the 4LB is considered a gold standard treatment for venous leg ulcers. However, it is important to note that good clinical outcomes from bandaging (in terms of ulcer healing) rely heavily on nurses’ application skills and patient concordance in wearing compression continuously. Application and concordance are a particular issue for the 4LB, which consists of a wool component plus three subsequent bandage components, making the final compression system time-consuming to apply and bulky. This bulk may impact on concordance by reducing mobility, making shoes difficult to wear and causing discomfort. 25

Compression hosiery

Mixed-treatment comparisons

Mixed-treatment comparison meta-analysis (also known as network meta-analysis) extends meta-analytic methods by enabling the simultaneous comparison of multiple interventions in a single model. 38,39

Mixed-treatment comparisons allow eligible RCTs to be included when they are linked by one or more common comparator(s) in a network of evidence. 40 As an example, assume three relevant treatments (A, B and C) have been evaluated in two RCTs: one comparing A with B, and another B with C. A network can be defined linking A to B to C. By linking RCT evidence, a consistent evidence base is created and derived relative treatment effect estimates may be informed by direct, indirect or both direct and indirect data. Indirect data can produce effectiveness estimates for pairs of treatments that have not been compared in head-to-head trials. In our example this means inferences over A vs. C can be generated using the indirect evidence available through the network.

In practice, where multiple RCTs inform multiple network links, the synthesis can become complex, as direct and indirect evidence contribute to estimating relative effectiveness for specific comparisons. However, by simultaneously using all (direct and indirect) evidence forming the network, uncertainty is appropriately considered for the comparisons of interest. Additionally, this approach provides an opportunity for formal assessment of consistency between direct and indirect evidence. In all cases, the validity of these methods depends on certain assumptions being held – some are common to standard pairwise meta-analysis (e.g. ‘similarity’ between included trials) and others relate to the ‘consistency’ between the direct and the indirect estimates, which can be assessed within the network. In the presence of heterogeneity, incorporating information on patients’ characteristics in the synthesis model has been shown to improve estimates and, by doing so, possibly resolve inconsistencies. 37,41,42

Decision-analytic modelling

Although MTCs are able to synthesise all RCT data, beyond trials, the relative cost-effectiveness of all relevant treatments for a condition can be investigated in a decision-analytic model. Though within-trial economic analysis offers the opportunity to evaluate cost-effectiveness in participants who contribute resource-use data alongside effectiveness data, such analyses are also limited by the scope of the RCT in which they sit. Decision-analytic models provide a structure within which both RCT and non-RCT evidence from a range of sources can be synthesised to describe a specific problem, and through this framework overall costs and effects can be estimated. The advantage of using this framework is that cost-effectiveness results can be based on all available evidence, across the full range of possible alternative interventions and clinical strategies over a relevant time horizon, and specific patient groups can be analysed separately.

Inclusion criteria

People for whom all of the following criteria applied:

• At least one venous leg ulcer. A venous leg ulcer was defined as any break in the skin which had either (1) been present for > 6 weeks or (2) occurred in a person with a history of venous leg ulceration. Ulcers were considered purely venous if, clinically, no other aetiology was suspected. The ulcer was required to be venous in appearance (i.e. moist, shallow, of an irregular shape) and was to lie wholly or partially within the gaiter region of the leg.

• An ankle–brachial pressure index (ABPI) of ≥ 0.8, taken within the previous 3 months.

• Able and willing to tolerate high compression.

• Aged ≥ 18 years.

Exclusion criteria

Potential participants were excluded if they fulfilled any of the following exclusion criteria:

• An ABPI of > 1.20 (taken with the previous 3 months) and, in the treating nurses’ clinical judgement and/or according to local guidelines, the potential participant should not receive high compression.

• A leg ulcer of non-venous aetiology (i.e. arterial).

• Wound exudate levels too high for the use of HH (decision made according to the nurses’ clinical judgement).

• Unwilling to give informed consent to participate within this trial.

• Unable to give informed consent to participate within this trial.

• Currently in another study evaluating leg ulcer therapies.

• A known allergy to any trial product.

• Gross leg oedema.

• Previously been recruited into this trial.

• Another reason that excluded them from participating within this trial (decision made according to the nurses’ clinical judgement).

Participant details

Participants’ contact details (name, address, telephone numbers and e-mail address), date of birth and GP details (name of GP, name of GP surgery and address) were recorded.

Ulcer history and assessment

The most recent ABPI measurement and the date it was taken were recorded. Also recorded were the total number of ulcer episodes (n); time since development of the first ulcer (years and/or months); duration of the reference ulcer (years and/or months) and duration of oldest ulcer (years and/or months).

To measure ulcer area, a tracing of the reference ulcer (the largest eligible ulcer) and all other ulcers on the reference leg (the leg that contained the reference ulcer) was taken using a fine-nibbed marker pen on to a wound measurement grid composed of 1-cm² squares (P12v2, ConvaTec, Uxbridge, Middlesex, UK). Ulcer area was determined by totalling the number of squares and/or partial squares on the grid contained within the traced ulcer area. At the end of the trial, ulcer area was calculated from the wound grid by the use of a software program (Mouseyes, version 3.1, Dr Robert John Taylor, Salford, UK). 43

All ulcers on both legs were drawn onto a leg diagram and the reference ulcer was clearly labelled using an identification code. The reference ulcer was coded as either R1 (located on the right leg) or L1 (located on the left leg). If there was more than one ulcer, the ulcers were coded according to descending area – the next largest as R2 (if located on the right leg) or L2 (if located on the left leg) and so on.

Participant mobility, anthropometry and glycated haemoglobin measurements

The level of participant mobility (walking and ankle mobility) was noted as was the participant’s height (feet/inches or centimetres), weight (stones/pounds or kilograms) and ankle circumference (centimetres). Body mass index (BMI, kg/m²) was calculated using the formula ‘weight (kg) divided by [height (m)]²’. If the participant was diabetic then their glycated haemoglobin (%) was recorded with the measurement date. However, as the trial progressed, some centres found it difficult to obtain this measurement. As it was not considered necessary for the trial analyses, the collection of this measurement was stopped following a decision by the Trial Management Group (TMG).

Current treatments received

Participants’ current treatment(s) for their venous leg ulcer(s) were recorded.

Treatment preference

Each participant’s trial treatment preference (i.e. no treatment preference, prefer to receive HH or prefer to receive the 4LB) was recorded.

Digital photographs

Study research nurses took a digital photograph of the reference ulcer and reference leg using a Nikon digital camera (Coolpix L20 or Coolpix L22, Nikon Corporation, Tokyo, Japan) according to the protocol developed for taking digital photographs during the trial (see Appendix 7). These anonymised photographs were uploaded onto an online management system and sent securely to the York Trials Unit (University of York).

Health-related quality of life/health utility

Participants completed a baseline questionnaire consisting of 12-Item Short Form Health Survey (SF-12) (Short Form questionnaire-12 items, version 2, standard recall, QualityMetric, Lincoln, RI, USA) and the EQ-5D™ (European Quality of Life-5 dimensions questionnaire, EuroQol Group, Rotterdam, The Netherlands).

The SF-12 measures eight health areas, which can be used to calculate a Physical Component Summary (PCS) score (based upon physical functioning, role limitations due to physical health, bodily pain and perceptions of general health) and a Mental Component Summary (MCS) score (based upon vitality, social functioning and role limitations owing to emotional health and mental health). The SF-12 is well validated in a variety of UK populations, including older people and leg ulcer patients. 44,45

The EQ-5D questionnaire measures five domains of health and provides an assessment of mobility, individual’s ability to self care, ability to perform usual activities, and evaluation of pain/discomfort and anxiety/depression. It is a widely recognised and validated generic measure of health utility and has been assessed for acceptability and validity in patients with venous leg ulcers,44,46–48 as well as being validated in other patient groups. 49–57

Ulcer-related pain

Participants were asked to rate the intensity of any leg ulcer-related pain that they had experienced in the previous 24 hours using the 21-point Box Scale (BS-21). The BS-21 pain scale was divided into units of five, and ranged from a value of 0 (no pain) to 100 (the worst pain imaginable).

Resource use

Participants provided details of any care received from the NHS within the past 3 months, recording the number of consultations the participants had with health professionals at different locations. Participants were asked to record details according to whether the consultation was related to their leg ulcer or a different reason.

Control group

Participants in the control group were allocated to receive the 4LB (standard/usual care). Nurses were allowed to choose which 4LB system the participant should receive, as long as it was designed to deliver 40 mmHg at the ankle and adhered to recommendations shown in Table 2. Examples of key 4LB components are shown in Table 3. The treatment was applied according to standard practice, either by the participant’s usual treatment nurse or by a study research nurse.

Intervention group

Participants in the intervention group were allocated to receive HH. This was a two-layer compression hosiery kit, consisting of an understocking (or liner) and an overstocking, worn over the understocking. The two stockings were designed to deliver sustained, graduated compression of up to 40 mmHg (and no less than 35 mmHg) at the ankle (Table 4).

The size of the HH kit used depended upon the participant’s ankle circumference (cm), calf circumference (cm) and/or foot length (cm); nurses consulted product-specific measurement tables to ensure that the participants received the correctly sized kit. Made-to-measure HH was obtained for participants who require this. HH was applied and worn according to manufacturer’s instructions. Participants were told they could remove the overstocking only at night, if deemed necessary by the treating nurse.

Participants were expected to receive their allocated trial treatment immediately after randomisation. However, where a treatment was not available for immediate application (e.g. if a participant required a made-to-measure HH kit), the participant was treated with an appropriate treatment in the interim, as dictated by the treating nurse, until the allocated trial treatment became available. The interim treatment received was recorded in a treatment log (see Appendix 5, Treatment log: trial dressing log booklet).

Time to healing of the reference ulcer (blinded)

The primary outcome measure of this study was time to healing of the reference ulcer. Healing was defined as complete epithelial cover in the absence of a scab (eschar) with no dressing required. When the treating nurse considered the reference ulcer to have healed, a digital photograph was taken of the healed reference ulcer site. Additional photographs were then taken of the same site, once per week for 4 weeks; photographs were taken regardless of whether the reference ulcer was considered to have recurred during this time period. Digital photographs were uploaded onto a secure server at the York Trials Unit using a secure online management system.

All digital images were assessed by two experienced Tissue Viability Specialist Nurses, who were blinded to treatment allocation. Independently, assessors determined whether they considered a participant’s reference ulcer to have healed, not healed or were unsure whether healing had occurred. When an ulcer was deemed to have healed, the appropriate date (based on date of photograph in which healing was recorded) was then taken to be the date of healing for the reference ulcer. A set of decision rules were made for resolving disagreements between assessors (see Appendix 9). In cases when one assessor considered the ulcer to have healed and the second assessor considered the ulcer unhealed, a third assessor was consulted, who determined whether healing had occurred.

Time to healing of the reference ulcer (unblinded)

Although the primary outcome measure for this trial was time to healing of the reference ulcer (as determined by assessors blinded to treatment allocation), time to healing of the reference ulcer as determined by the treating nurse was used as a secondary outcome measure. When the treating nurse considered the reference ulcer to have healed, she/he telephoned the York Trials Unit randomisation line to report the date of healing.

Time to ulcer-free reference leg

Nurses reported the date on which they thought the reference leg was ulcer free, and this was reported by telephone to the York Trials Unit randomisation line; this information was also captured via the ulcer healed form (see Appendix 5).

Health-related quality of life/utility and resource use

Post randomisation, participants’ health-related quality of life and resource use were measured at 3, 6, 9 and 12 months using postal questionnaires. Each quarterly questionnaire was identical in content and contained the same health-related quality-of-life/utility instruments as the baseline questionnaire (EQ-5D and SF-12) and the same tools for measuring ulcer-related pain (BS-21 and verbal descriptor pain scales) and resource use (see Appendix 5).

Post-randomisation questionnaires were posted to participants from the York Trials Unit, along with a pre-addressed and prepaid envelope. Where necessary, reminder letters were sent by post to participants at 2 and 4 weeks had questionnaires not been returned. A systematic review assessing ways to increase participant response to postal questionnaires59 reported that response rates could be almost doubled by the use of monetary incentives (odds ratio 1.99, 95% CI 1.81 to 2.18) and could also be increased if the incentive was unconditional. Participants were therefore sent £5 with their final questionnaire and were informed that this was an unconditional token of appreciation for the time they had taken to complete questionnaires throughout the study.

Briefly, we also collected disease-specific health-related quality-of-life data using the VEINES quality-of-life questionnaire (VEINES-QOL) instrument at baseline and 4 months. These data were collected as part of a substudy that was aiming to conduct a validation study of this measure in patients with venous leg ulcers. This substudy is reported separately; however, we report summary data in this report for reference.

Participant concordance to treatment

In order to monitor treatment concordance and patterns of compression use, study nurses recorded in a treatment log when a participant was no longer receiving their allocated trial treatment but rather was receiving a non-trial treatment, i.e. change to a leg ulcer treatment that was different from that to which the participant was allocated at randomisation and the reason for this change (see Appendix 5, Trial dressing log booklet). Participants were also asked about their opinions and use of their allocated trial treatment via a treatment-specific 1-month participant questionnaire (see Appendix 6).

Ulcer recurrence

One month after the reference leg was reported as healed, nurses undertook a monthly assessment for ulcer recurrence (see Appendix 5, Monthly nurse assessment form); this monthly assessment continued until the participant had exited the trial. Where there had been a recurrence of venous leg ulceration on the reference leg, the date of recurrence was also recorded. To maximise data collection participants with a healed reference leg were also provided with a reference leg-specific postcard (see Appendix 6) to complete if they had a recurrence. This was then returned in a pre-addressed and postage-paid envelope to the York Trials Unit.

Adverse events

An adverse event is defined as any undesirable medical experience occurring to a participant, whether or not considered related to the trial treatment. 60 Adverse events were classified as serious (death, life- or limb-threatening event, hospitalisation required/prolonged, persistent or significant disability/incapacity or other medically important condition) or non-serious (all other adverse events). 60

Nurses were asked to report adverse events (see Appendix 5). The treating nurse was also required to make an assessment regarding the relationship between the adverse event and trial treatment (HH or 4LB) assessed by a set of decision rules (see Appendix 10). All adverse event forms were returned to the York Trials Unit. Serious adverse events (SAEs) were also reported directly to the York Trials Unit via a telephone call to the trial coordinator and/or via fax to the York Trials Unit. Nurses reviewed adverse events until they had resolved (see Appendix 5).

All SAEs were reported to the Trial Steering Committee (TSC), trial centre research and development offices and the North East Yorkshire and Northern Lincolnshire Comprehensive Local Research Network. The local REC was notified of any unexpected SAEs that were considered to be related to trial treatment. Expected events in this study population were defined using previous research data. Participants in previous VenUS trials have been generally elderly (mean ages of 71, 74 and 69 years, respectively, in VenUS I–III7,13,61) and VenUS IV participants were expected to be similar. As one might expect, people with venous leg ulceration have been shown to present with other co-morbidities, including hypertension, congestive heart failure and osteoarthritis;62 these and other medical conditions may require hospitalisation. A small percentage of participants would also be expected to die during the study, as reported in previous VenUS trials. 7,13,61

For analyses, all adverse events were reviewed by the TMG (blinded to allocation), which made the final decision regarding the relationship between an adverse event and trial group. Furthermore, all non-serious adverse events (NSAEs) that were considered to be definitely or probably related to trial treatment were reviewed by a tissue viability research nurse and categorised for analysis (Box 1).

Baseline data

All categorical baseline variables were summarised as a frequency (n, %) by treatment group. Continuous variables were summarised using descriptive statistics [n, mean, standard deviation (SD), minimum, maximum, interquartile range (IQR) and median]. No formal statistical analyses were conducted.

Trial completion

Details of trial exit were taken from the Trial Exit form. When an exit form had not been completed, data were derived from the length of time a participant was in the study from randomisation to their last clinical follow-up documenting healing status, together with data on deaths.

Primary analyses

Secondary analyses

Sensitivity analyses and handling of missing data

Survival data were assumed to be completely known (data collection was designed to be complete). Adverse event reporting was also assumed to be complete. Missing data on covariates included as adjusting factors in the statistical models were to be assumed missing at random. Sensitivity analyses were carried out:

1. for comparison of results with and without a centre random effect

2. where there were missing data on the covariates, comparison of results from the best model chosen in (1) with and without multiple imputation.

Resource use and unit costs

The cost of resources used for each trial participant was calculated as the product of resources used during the trial follow-up period and the relevant unit costs. Three different elements of resource use were considered in the estimation of ulcer-related treatment costs:

• use of allocated compression treatments during trial (compression hosiery, 4LB and any other compression treatments used, e.g. while waiting for the trial treatments to become available)

• use of non-trial treatments (for participants who changed to a non-trial treatment)

• health-care consultations (visits to/from health-care provider for ulcer-related reasons).

Other treatments, such as primary and secondary contact dressings or skin preparations, were assumed to be used equally across treatment groups and these resources were not included in the economic analyses. Unit costs for the compression treatments were based on the NHS prescription cost data for 2010–11,73 and all values are in British pound sterling (£).

Health benefits

Statistical methods for within-trial economic analysis

Cost-effectiveness analysis

Although the regression models above estimate the difference in costs and health benefits between treatment groups (and the statistical significance of this difference), for decision-making it is the expected value of this difference that is of interest. To assess incremental cost-effectiveness and cost utility, we compared the expected value of the mean difference in costs between trial groups to the expected value of the mean difference in the number of ulcer-free days and QALYs, respectively. There are four possible scenarios from such an analysis (shown using a cost-effectiveness plane in Figure 1). HH may be estimated to be (on average) more expensive and bring fewer health benefits than the 4LB (HH is dominated by the 4LB). In this case, the decision regarding preference of 4LB over HH seems straightforward, as HH is more costly and refers less benefit to patients.

FIGURE 1.

Cost-effectiveness plane showing cost and benefit differences between alternative treatments.

A similarly straightforward scenario is the reverse, i.e. if HH is expected to be less costly and more beneficial than the 4LB (HH dominates the 4LB). However, if HH were expected to be more costly and more beneficial than the 4LB, or less costly and less beneficial, it is necessary to evaluate whether the increased cost of the new intervention is worth the increased benefit, or if the reduced benefit associated with the new treatment is justified by the reduced costs.

To ascertain the cost-effectiveness of a health-care intervention relative to another in the absence of dominance, one needs to combines costs and health benefit in a single measure to which a decision rule regarding cost-effectiveness can be applied. The incremental cost-effectiveness ratio (ICER) is commonly used, and is the ratio of the mean difference in cost between alternative treatments being compared to the mean difference in health benefits:

⁽¹⁾ ICER = C 1 − C 0 B 1 − B 0

where:

• C₁ = mean cost associated with the use of HH

• C₀ = mean cost associated with the use of the 4LB

• B₁ = mean health benefit associated with HH

• B₀ = mean health benefit associated with the 4LB.

The decision rule for cost-effectiveness on the basis of the ICER indicates that a treatment strategy can be considered cost-effective only if the decision-maker’s willingness to pay for an additional unit of health benefit (QALY, ulcer-free day) is greater (or equal) to the ICER. According to NICE, the willingness to pay for an additional QALY ranges between £20,000 and £30,000. 83 Therefore, if the result of cost–utility analysis (the estimated cost per QALY) is below this threshold, the intervention will be considered cost-effective. Caution is required when interpreting the cost-effectiveness results, as there is no established threshold for cost per ulcer-free day gained. Without this information we cannot determine whether or not the new intervention is cost-effective – a decision-maker interested in the results will be responsible for establishing the threshold.

Decision uncertainty assessment

All presented analyses are based on sampled data collected within this RCT. Thus, if this trial were repeated using a different study sample we would expect different mean incremental cost and benefit values to been observed, resulting in a different ICER estimate. Thus to fully understand inferences made from our data, the expected costs and benefits calculated must be estimated with uncertainty. Furthermore, although decision-makers can then decide on the provision of services, using expected cost-effectiveness findings in the presence of uncertainty, it is vital that the consequences of this uncertainty, and the extent to which it impacts on the adoption decision, should be investigated to inform whether further research is needed. 84

Subgroup analysis

A post hoc subgroup analysis was conducted to investigate whether total cost varied based on participant baseline characteristics; hence, cost per participant was evaluated based on age (≤ 60, 61–70, 71–80, > 80 years), BMI [underweight (< 18.5 kg/m²), normal healthy weight (18.5–24.9 kg/m²), overweight (25–29.9 kg/m²), obese classes I–II (30–39.9 kg/m²) obese class III (40+ kg/m²)86] and mobility (walks freely vs. walks with difficulty or immobile). To investigate potential interaction between treatment and baseline characteristics in the cost regression, separate interaction terms were included for treatment and age, BMI or mobility in the model used in the primary analysis (see Estimating mean time to healing, costs and quality-adjusted life-years, above). As in the primary analysis, an IPW-weighted mixed-effects model was used to evaluate costs at all quarterly time points at which cost data were collected. This post hoc analysis was initiated after unblinding of the primary results to the investigators and should therefore be regarded as exploratory in nature.

Sensitivity analysis

Duration of treatment with allocated compression treatment

Time to treatment change was investigated in post hoc analyses and was defined as time from randomisation until date treatment change was recorded. There were 16 participants (6 HH, 10 4LB) noted to have had a change of treatment on the day of randomisation (11 participants never received their allocated treatment and five tried their allocated treatment but changed to another treatment on the same day of randomisation). Time to treatment change was investigated using a CPH regression, in which the 16 participants who changed treatment on the day of randomisation were given an arbitrary time to treatment change of 0.1 day (one-tenth of a day) and data were right censored in participants who (1) withdrew from the study; (2) were lost to follow-up; (3) died; (4) healed; or (5) reached the end of the trial – whichever came first. Median time to treatment change was 133 days in the HH group (95% CI 91 days, upper limit not estimable) and 213 days in the 4LB group (95% CI 182 days, upper limit not estimable). Treatment allocation, age and whether or not a participant had experienced a NSAE were included as covariates in this analysis. All of these factors were found to be associated with time to treatment change, with evidence of a shorter time to treatment change (from allocated trial treatment to non-trial treatment) in the HH group (HR 1.59, 95% CI 1.14 to 2.21; p = 0.005), those who were older (HR 1.02, 95% CI 1.00 to 1.03; p = 0.003) and those with at least one NSAE (HR 1.75, 95% CI 1.18 to 2.59; p = 0.005). The model was repeated including baseline ulcer area, duration, participant mobility and centre, but there was no evidence to suggest that these were associated with time to treatment change, and the associations between time to treatment change allocated treatment, age and NSAEs remained.

Serious adverse events

For SAEs, the numbers, classification and relationship to treatment overall were fairly well balanced between the treatment groups, with a slightly higher number of deaths reported in the HH group (see Table 22). There were no SAEs that were definitely related to treatment: only one was probably related (4LB group) and 16.5% of SAEs (14.0% HH, 19.1% 4LB) were possibly related. No participant had more than two SAEs reported during the course of study.

Non-serious adverse events

A higher proportion of participants in the HH group experienced one or more NSAEs compared with those allocated to the 4LB group [70.0% compared with 58.0%, χ² = 3.86 (1 df); p = 0.050]. We also compared the total number of events experienced by participants (HH vs. 4LB) adjusting for the prognostic factors (baseline ulcer area, duration, participant mobility) and using robust SEs for centre effects in a zero-inflated negative binomial regression. There was no statistically significant difference between groups with a RR of 1.12 (95% CI 0.95 to 1.32). The proportion of events classified (by blinded assessors) as probably or definitely related to trial treatment was 41.7% in the HH group and 37.8% in the 4LB group. These probably or definitely related events were also classified into types of event, blind to treatment allocation. Of these events a higher number were classified as failure in the HH group, but more new ulcer occurrences were recorded in the 4LB group.

Compression treatments

Participants were allocated to their compression treatment at randomisation (the ‘trial treatment’) and received this treatment until they changed to another treatment (designated the ‘non-trial treatment’); their reference ulcer leg healed (i.e. treatment was not needed any more); or they were lost to follow-up or died.

Number of trial treatments received

During each treatment consultation, the existing compression treatment (if any) was checked, and if needed, new compression treatment/s were applied or given to the participant. The mean number of treatment applications per participant was 3.0 (SD = 3.87) in the HH group and 15.08 (SD = 17.98) in the 4LB group (Table 25). In the HH group a total of 662 compression treatments were applied/given in a total of 2627 nurse consultations, whereas in the 4LB group new compression treatments were applied/given in most of the 3453 recorded nurse consultations.

It is possible that, during ‘trial treatment’, participants received other treatments or procedures, including compression treatments, to the reference leg, for example because the allocated compression treatment was unavailable for a short period of time. A total of 226 participants received other compression treatments during their trial treatment period (151 in the HH group and 75 in the 4LB group); these were included in the cost analysis.

Consultations with health-care providers

Consultations with health-care providers were reported by participants in quarterly questionnaires and these participant data were used in the base-case analysis. Participants reported both ulcer-related and non-ulcer-related consultations; however, only ulcer-related consultations were used in the economic analysis as differences across groups were expected only for ulcer-related resource use.

Participant self-reported data suggested that the number of ulcer-related consultations with health-care professionals was lower in the HH group in all categories of consultations except for the hospital day admissions, which were marginally higher in the HH group. The number of non-ulcer-related visits was similar across the treatment groups (Table 27).

The proportion of consultations taking place in participant’s homes was similar between groups (29.3% for the HH group and 28.6% for the 4LB group). Doctor consultations accounted for a small proportion of the total number of consultations, and the number of such consultations was marginally lower in the HH group.

There were few hospital admissions (day cases and inpatient admissions). In total, 14 hospitalisations in the 4LB group and five in the HH group were reported by participants; therefore, as anticipated, there were small and potentially random differences in the number of stays that could result in large differences in costs and bias the results. Furthermore, evaluation of nurse-reported hospitalisation data revealed poor consistency with the participant-reported inpatient data (only 6/19 hospitalisations were also recorded in SAE forms by nurses and only one seemed to be related to leg ulcer based on the description of SAEs). We attribute these inconsistencies to the fact that participants may not easily be able to distinguish ulcer-related and unrelated hospital inpatient admissions. It was decided to exclude inpatient hospitalisation costs from the analysis.

Mean time to healing

On average, participants allocated to the HH group healed 1.4 days later than those allocated to the 4LB group. However, this difference was not statistically significant (95% bias corrected CI of the difference was from 36.8 days to –29.3 days) (Table 33).

Utility and quality-adjusted life-years

Quarterly utility scores (per participant by treatment group) were calculated using EQ-5D responses, and quarterly QALYs were computed using time-weighted averages of the utility scores measured at the beginning and end of each interval. Quarterly utility scores are presented in Table 34 and unadjusted average QALYs per group described in Table 35 (note: only 41% of the sample had complete cases, i.e. utility scores available at all time points). There is a trend for an increase in quality of life over follow-up time for participants allocated to HH; however, this trend is less clear in the 4LB group.

After adjustment for baseline utility scores and stratification covariates, and after accounting for the censored nature of data, individuals in the HH group had, on average, more QALYs than individuals in the 4LB group [annual difference in QALYs of 0.034 (95% CI –0.0006 to 0.0778): Table 36].

Ulcer healing

We found no evidence of a difference in time to ulcer healing between HH and the 4LB. The HR for healing was 0.99 (95% CI 0.79 to 1.25) meaning there is almost the same hazard (or ‘chance’) of healing in the HH group and the 4LB group. The CI indicates that HH may reduce the hazard (or ‘chance’) of healing by as much as 21% or increase it by as much as 25%.

This imprecision was observed despite this being one of the largest RCTs comparing compression treatments for venous leg ulcers ever conducted. 19 Although undertaking further large RCTs may reduce uncertainty around this treatment decision, it is important to consider the cost of investing in a further RCT in relation to the value collected data might have in improving decision-making. These issues will be investigated further in relation to high-compression treatments for venous leg ulceration in Part III of this report, where the amount of uncertainty around treatment effects will be investigated when all available data are synthesised and the value of further research considered.

The median time to healing in the 4LB group in VenUS IV was similar to that in VenUS I7 (99 days compared with 92 days, respectively) and both were pragmatic trials with wide inclusion criteria. However, the VenUS IV estimate was higher than the 70-day median time to healing reported for 4LB participants in the only other RCT comparing this treatment with HH. 32 However, this study excluded participants with clinical signs of infection so it may have evaluated a participant population with less-complex ulcers. Additionally, we note that Finlayson et al. 32 reported that the hazard of healing in the 4LB group was twice that of the HH group when their HH delivered a maximum of 35 mmHg compression at the ankle. Given trial findings from VenUS IV, one further interpretation may be that HH used to treat venous leg ulcers should aim to deliver 40 mmHg at the ankle in order to achieve similar time to healing as the 4LB.

As observed in VenUS I–III,7,13,61 as well as several other studies,88–90 we found that baseline ulcer area and duration were statistically significant predictors of time to healing (p ≤ 0.001). We also found that study centre is a significant predictor of ulcer healing, a finding that remained when adjusted for the number of participants per centre. As the application of compression bandaging requires skill, with the bandager relying almost entirely on technique for correct application, we postulated at the start of the study that a centre effect may be driven by differing bandaging skills between centres. This notion was supported by our tentative post hoc analysis, which suggested stronger evidence for a centre effect in the 4LB group than the HH group (which does not require the same skill level in application).

The decision to account for a potential centre effect in our sample size calculation was shown to be valid. We suggest that triallists planning future RCTs with a bandaging arm should consider the possibility of a centre effect and take this into account in sample size calculations and analyses.

Health-related quality of life

Changes in health-related quality of life over the duration of the trial were measured using the SF-12. At baseline the study population had a low quality of life in terms of physical health as observed in previous studies,7,13,61 whereas the MCS score was similar to population norms.

There was little change in the PCS and MSC scores over time in both the HH and the 4LB group of VenUS IV. The adjusted analysis showed slightly higher (all non-significant except at 3 months) scores in the HH group compared with the four-layer group, which agrees with the EQ-5D results reported in the cost-effectiveness analysis. There was a reduction in mean ulcer-related pain scores (over previous 24-hour period), during the trial but with no evidence of a difference between groups.

Although previous work has shown that the SF-12 is responsive to changes in the health-related quality of life of patients with venous leg ulcers upon ulcer healing,44 we note here that although 70% of trial participants had a reference ulcer heal during the study, there was no perceptible impact on mean PCS scores over time. It is important to acknowledge that it is people with a number of underlying physical comorbidities who are at increased risk of develop venous ulceration; these conditions include congestive heart failure, diabetes, peripheral vascular disease of the lower extremity and rheumatoid arthritis. 62

Adverse events

Although several adverse events were reported in VenUS IV, this had been anticipated given the older study population at risk of a number of comorbidities. 7,13,61 There was no statistically significant difference in the number of participants with one or more SAEs between trial groups, neither was there a difference in the total number of SAEs reported. Conversely, more participants in the HH group experienced one or more NSAEs compared with participants in the 4LB group (67.0% compared with 58.0%; p = 0.050); although again there was no evidence of a difference in the total number of NSAEs reported between each group (RR 1.12, 95% CI 0.95 to 1.32). Neither was there a difference in the total number of NSAEs recorded as being probably or definitely related to trial treatment (41.7% HH, 37.8% 4LB; p = 0.26).

Within the NSAEs that were probably or definitely related to trial treatment, the most common event across trial groups was bandage/hosiery failure (22%). It is not clear if these treatments were considered as failing from the perspective of the treating health professional or the participant themselves. A potential issue with the HH may have been difficulty in self-application, especially for those with grip and/or coordination problems, as the hosiery remain relatively tight and can require some dexterity to apply. Application aids are available but we did not record the level of their use in this trial.

Bandage- or hosiery-related pain and discomfort were also commonly reported as definitely or probably treatment-related NSAEs across trial groups (20%). Potential reasons for pain and/or discomfort with hosiery may have related to presence of friction via rubbing on bony prominences/discomfort caused by the hosiery top band and the lack of a padding layer as used in many bandaging systems. It is also possible that issues of discomfort could be related to the accuracy of measurement for hosiery wear. These issues would need to be investigated further.

Participant use of compression treatments

Although 61% of participants in the HH group remained on their allocated treatment, there was evidence that significantly more participants in this group changed from their trial treatment to a non-trial treatment compared with those allocated to the 4LB (where 72% remained on allocated treatment). A post hoc analysis showed that as well as being allocated to HH, a previous NSAE and age were also predictive of a shorter period on trial treatment. The most common specified reason given for the treatment change across groups was ‘compression uncomfortable’ and 71% of these reports were from the HH group. These data triangulate with the NSAE findings, further suggesting that, for some people, HH is not an optimal treatment, especially in terms of comfort.

The increased treatment change in the HH compared with the 4LB group was somewhat contradictory to what we had postulated before starting this RCT. Previously, reasons for patient non-compliance with compression bandages have been reported as pain, discomfort, application difficulty, discomfort with footwear and poor physical appearance of legs. 91 We suggested that HH might be more acceptable to venous leg ulcer patients because it is less bulky than bandages, as well as potentially easier to apply and less reliant on bandager skill. Indeed upon entry into the trial (but prior to randomisation) 50% of trial participants expressed a preference for HH over the 4LB (compared with 13% expressing a preference for four layer and 37% having no preference). In reality there may be people for whom HH is not an optimal treatment (e.g. in those who cannot apply it easily themselves).

Recurrence

In VenUS IV there was a statistically significant reduction in the chance of recurrence with HH compared with 4LB (HR 0.56, 95% CI 0.33 to 0.94; p = 0.026). This translates to a reduction in the chance of recurrence of ulceration for people allocated to HH of between 6% and 67%. This estimate was relatively imprecise as the sample size was limited to only those whose reference leg completely healed during the trial. Additionally, we acknowledge that this analysis was not of the population as randomised and thus must be interpreted with some caution. However, given that the same number of participants healed in these two groups, recurrence results would have been affected only if the characteristics of participants healing differed across the two groups, which is unlikely.

Although VenUS IV focused on HH as a treatment for healing ulcers, when planning the study we hypothesised that perhaps those who become used to wearing HH and healed were more likely to wear compression hosiery post healing (and perhaps more likely to stay in HH than those moving maintenance hosiery following treatment with a compression bandage) and thus have a reduced risk of recurrence. This hypothesis may potentially explain the results observed with those who found HH an agreeable (and effective) treatment and continuing to wear it. However, we did not collect data on use and adherence to maintenance compression treatments in this trial so could not investigate this hypothesis further. We also note that our protocol gave no guidance as to the provision of maintenance compression therapy and we assumed that standard procedures would be followed. We note there are currently limited data available on the proportion of people with a previous venous leg ulcer who are compliant with wearing compression hosiery as a maintenance treatment. A national audit conducted over 10 years ago reported that of patients with venous leg ulcers, on whom data were received, 88% had be given prophylactic compression hosiery (level of compression not reported). 92 However, the uptake, and use of these stockings was not assessed/reported.

Likewise, there are few data reporting how widely HH is currently used as a treatment for active venous leg ulceration. However, we know from VenUS IV that only 6.5% of trial participants were receiving compression hosiery as a treatment prior to entering the trial (compared with 49% receiving the 4LB). Although we acknowledge that these figures could be an underestimate (if people currently receiving HH did not enter the trial so as not to ‘risk’ randomisation to a non-hosiery treatment), if we accept the 6.5% figure as a guide to current HH use for the treatment of venous leg ulcers we can postulate that use is low. Yet, based on findings regarding recurrence here, even with the levels of treatment change recorded, HH may be the better treatment of choice for some patients in terms of reducing overall time spent with venous leg ulcers.

It is important to note that we do not suggest that there is no place for other compression therapies in the treatment of venous leg ulcers. By default those coming into this study would have been considered eligible for treatment with either of the trial compression therapies. There will be other patients, for example those with large, oedematous or awkwardly shaped legs, for whom experienced health professionals may select other compression therapies from the different options available.

Sample size

As far as we are aware this is the largest, individually randomised RCT that has been undertaken to evaluate leg ulcer interventions, although we fell just short of our recruitment target of 489. Even so, it is important to note that we did obtain the precision around the point estimate that was planned for in our original sample size calculation.

Blinded outcome assessment

We were unable to conduct blinded outcome assessment of wounds in person for VenUS IV as the provision of a community-based blinded assessor across all trial centres would have been too resource intensive. Additionally, as described previously, the actual logistics of performing blinded outcome assessment would be challenging. 61 Thus, blinded outcome assessment was conducted centrally with two blinded assessors examining digital photographs taken monthly from baseline until healing, and then weekly from healing for 4 weeks. This frequency of data collection was implemented in this study for the first time and was based on our previous experience of the blinded outcome assessment processes used in VenUS II61 and III. 13

The use of sequential images from the healing period was found to work extremely well, as it allowed blinded outcome assessors to give a healed date subsequent to that given by the treating nurse if required. In VenUS IV we also implemented an electronic system for uploading pictures that helped to ensure that we received most photographs and had limited missing data.

A recent study examined data from 21 RCTs (4391 participants) that had conducted both blinded and unblinded assessment of their binary, subjective outcomes. The study reported that, on average, non-blinded outcome assessment exaggerated the odds ratio by 36% when compared with blinded outcome assessment. 93 However, this meta-epidemiological study included data from our previous study, VenUS II,61 which, like VenUS IV, found little difference in blinded and unblinded outcome assessments of healing. In VenUS IV, the unblinded assessment of ulcer healing resulted in slightly more ulcers recorded as healed, and the corresponding HR was slightly lower than the blinded assessment (HR 0.91, 95% CI 0.73 to 1.12). Thus there was still no statistically significant treatment effect although there was greater precision around the point estimate. Although giving similar results, blinded outcome assessment remains an important strength of our venous ulcer trials.

Attrition

Attrition was not an issue in analysis of the trial primary outcome (with 453/454 patients contributing at least some follow-up in the survival analysis and 6.5% of 2LB and 6.7% of four-layer hosiery patients being classified as lost to follow-up during the study), however, as with VenUS trials I–III;7,13,61 we observed a marked reduction in participant response rates for postal questionnaire containing SF-12, EQ-5D and resource-use data. This drop in response occurred despite using strategies that aimed to maximise response with the use of a £5 incentive, as well as reminders to return questionnaire. As in VenUS II,61 analysis suggested that participants were less likely to return questionnaires if they had not healed, meaning that we may have overestimated health-related quality-of-life estimates, although there was little change in these estimates during the trial. In non-healing participants there was some evidence of lower return rates in the HH group; it is unclear whether this might led to an overestimate of health-related quality of life in this group.

Initial classification

Data were obtained from 40 RCTs (Table 42; for further details see Appendix 11), but three were excluded as they did not report suitable endpoints for the MTC analysis (e.g. reduction in wound area reported rather than time to healing or proportion of participants healed). Of the remaining 37 RCTs, many evaluated well-established compression treatments that were recognised as aiming to deliver either high (≥ 40 mmHg) or non-high (< 40 mmHg) compression. However, a number evaluated ad hoc treatments, for example hybrids of compression components, with a compression level that we could not easily classify. Thus, to formally establish whether evaluated treatments aimed to deliver high compression (or not) an internationally recognised expert (Hugo Partsch) in compression systems classified each compression treatment, based on its component composition (Box 2) [see Table 42, ‘(2)’]. Subsequently, the expert classified each treatment as high or not high compression, based on the number of components, the initial pressure (reported or estimated by expert depending on detail report) and stiffness of the final treatment (estimated by expert). Among all available 82 treatment groups, 45 treatment groups were classified as high compression and 37 as non-high compression [see Table 42, ‘(I)’].

Further grouping

Subsequently, where possible, treatments were further grouped into key categories consistent with current practice. For example treatments defined as ‘Beec’ and ‘Beea’ were grouped into the 4LB group. Similarly, ‘Bii’, ‘Biic’ and ‘Bc’ were all grouped into a SSB group; ‘Bzc’ and ‘Bza’ into a zinc paste group (referred to as ‘paste’); ‘HH’ grouped as ‘HH (two-layer: aiming to deliver 40 mmHg compression at the ankle)’ and ‘Bic’ grouped as the 2LB system (subcompression wadding and cohesive bandage). The remaining, generally ad hoc, systems were not grouped further. This classification process was checked with members of the TMG, cross-checked with the source review and judged as valid. After this further grouping process the evidence evaluated 24 unique treatments, 13 of which were classed as high compression (see Table 42, ‘3’).

Statistical model for the data

For each included RCT, the most complete ulcer healing outcome data available were used in the MTCs: for two studies time to ulcer healing and time to censoring were available as IPD (VenUS I7 and VenUS IV); for the remaining RCTs, outcome data were reported as the number of healed ulcers by study group [aggregate data (AD)]. We maximally drew on all available data by statistically synthesising the AD with the IPD, extending the methodology described in Smith et al. 129 and Saramago et al. 130

Briefly, data were synthesised by assuming that the time to healing underlying both AD and the IPD was described by the same parametric distribution. For the IPD, where time to healing (under censoring) was observed for each RCT participant – these data could directly inform the distribution of the time to healing (the likelihood). For the AD, by assuming a binomial likelihood, the number of participants healed was used to inform the probability of participants being healed. In turn, the probability of participants being healed was related (algebraically) to the common distribution of time to healing taking into account the duration of follow-up in each study (analogous to the synthesis models defined by Soares et al. 131 This approach allowed all ulcer healing data (proportion of ulcers healed and time to ulcer healing) to be defined as time to ulcer healing thus the measure of effectiveness used to report MTC findings was the HR.

Had we only considered aggregate information (reducing the information on studies with available IPD to AD and synthesising this in the standard way with the remaining aggregate information), the synthesis would have been constrained to use of the exponential distribution, which imposes a constant hazard of healing over time (the hazard of healing at any instant is constant). The use of IPD here allowed more complex time-to-event distributions, namely the Weibull and Gompertz distributions, to be investigated. Both these distributions are flexible when compared with the exponential, in that they can reflect an increasing, decreasing or constant hazard of healing over time. 63 In addition to Weibull and Gompertz, we recognise that there is other time-to-event distributions such as the log-normal or the generalised gamma, which could potentially have provided a better fit to the time-to-healing data. However, given that these distributions do not allow the probability of healing over time to be expressed in a closed form (key to the approach proposed here for the joint synthesis of IPD and AD), we opted to use the Weibull distribution, as implementation issues prevented us from using the Gompertz distribution. Further work is required in this area to embrace these other modelling frameworks.

The parametric survival modelling of the IPD studies was implemented using regression analysis that allowed for baseline covariate adjustment. All relative treatment effects were presented as HRs with the 4LB arbitrarily chosen as reference treatment in all cases. As several of the treatment comparisons in the MTC network were populated by a single RCT (exceptions are SSB vs. 4LB and 2LB vs. 4LB), a fixed-effects modelling approach was used. In additional analysis (not reported here), we also considered the use of a random-effects model. We found no significant gains in quality of fit, and thus adopted the more parsimonious fixed-effects approach throughout. This is consistent with previous published work synthesising evidence on the comparison between SSB and 4LB,24 which also chose to undertake a fixed-effects approach because there was no significant evidence of between-study heterogeneity. The baseline covariates considered were participant mobility, duration and area of ulcer and centre, all available in VenUS I7 and VenUS IV. The effects of these covariates on the hazard of healing were assumed equal in both IPD studies (both studies were used to estimate the covariate effects). The effect of centre was described using a common frailty effect across the IPD trials.

Besides adjusting for baseline covariates, we tested the inclusion of interaction terms between alternative treatments and baseline ulcer area and duration simultaneously in the RCTs with IPD and aggregate outcome data – as described by Cooper et al. 41 and Saramago et al. 130 In considering treatment–effect interactions, the use of IPD allows individual patient covariate information to be modelled and, consequently, avoids potential ecological fallacy. 132 Interaction terms were not found to be important in explaining the hazard of healing over time in this case study and were excluded from the final model.

All analyses were conducted from a Bayesian perspective, using WinBUGS software version 1.4.3 (MRC Biostatistics Unit, Cambridge, UK. URL: www.mrcbsu.cam.ac.uk/bugs/WinBUGS/contents.shtml). Alternative model specifications were implemented to assess the inclusion of baseline covariates and treatment–interaction terms. Goodness of fit was assessed using the deviance information criterion (DIC). 133 The DIC is a measure that balances fit and complexity, allowing parsimony to be considered in model choice. Alternative model specifications were used (see Appendix 12) and model fit was assessed using DIC. Results for the best fitting model only are presented along with 95% credible intervals (Crls), the Bayesian equivalent of CIs.

The treatment with the lowest HR estimated in the MTC (with respect to the reference compression system, the 4LB) is expected to have the highest effectiveness in healing venous leg ulcers. However, it is important to be fully aware of uncertainty around these estimates. Thus, in addition to presenting CrIs, for the selected model in the base-case analysis we explored uncertainty regarding treatment choice, presenting this as the probability that each compression system was the ‘best’ treatment in terms of being the most effective (when compared with all other evaluated treatments).

Quality assessment and consistency of evidence

As with other methods of evidence synthesis, the quality of the data included in the model must be reflected in conclusions made. Although there is no recognised system to undertake such quality assessment for MTC we have previously published a modified Grading of Recommendations Assessment, Development and Evaluation (GRADE) approach (we called this iGRADE) to allow us to assess and communicate the quality of MTC-derived evidence. 134 The iGRADE approach uses the five GRADE categories that allow the quality of evidence to be decreased, with focus of some categories altered to be relevant for MTC assessment (see Appendix 12 for further information). We conducted a cautious application of iGRADE (see Appendix 13) to the key estimates of this base-case analysis MTC, in which estimates could be graded as very low-quality evidence, low evidence, moderate evidence and high-quality evidence.

One of the dimensions assessed in the iGRADE is of evidence inconsistency, i.e. discrepancy between the direct and the indirect evidence in existing evidence loops. 135 We formally assessed for inconsistencies using the back calculation method suggested by Bucher et al. 136 Briefly, within an evidence loop, direct and indirect estimates of a pairwise treatment effect were compared against a null hypothesis that there would be no difference between them.

Inclusion of non-high-compression treatments in the mixed-treatment comparison

Despite findings from non-high-compression treatments being less relevant here, it is possible that the inclusion of data from RCTs evaluating these treatments could impact on inferences on the high-compression treatments of interest. For this reason a network of evidence with non-high-compression treatments was also considered as a secondary ‘scenario’ analysis (studies numbered 3, 7–9, 24, 27, 31, 33, 36, 38, as shown in the network diagram in Figure 11b). In addition to the nine high-compression treatments, the network consisted of four non-high-compression treatments (Bzee, H, Bi and Be), adding 10 further RCTs and 22 additional treatment groups to the network (where two of the additional RCTs were three armed).

This analysis tested the robustness of the evidence synthesis findings for the base-case analysis by assessing how any supplementary indirect evidence had an impact in the estimation of treatment effects for high-compression treatments.

The effectiveness of the two-layer bandage

Because the quality of RCT data driving the relative effectiveness of the 2LB to the 4LB was so low, we conducted a post hoc sensitivity analysis assuming that there was no difference in the effectiveness of these treatments.

Presentation of mixed-treatment comparison results in terms of probability of healing over time

Mixed-treatment comparison results were initially presented using the HR. However, the hazard (and its ratio) can sometimes be a difficult measure to interpret, as it reflects the instantaneous risk of healing, which is a somewhat abstract (but mathematically useful) measure of the likelihood of healing. In order to aid the interpretation of the MTC, the hazard of healing findings was mathematically transformed to give the probability of participants healing over time for each of the high-compression systems of interest. The data were presented in this way to aid the interpretation of findings and the implications of these to clinical practice.

Uncertainty over the probability of healing with four-layer bandage (reference treatment)

As the 4LB group was a reference treatment in the MTC analyses, its estimates were not shown as uncertain. For completeness, we also presented uncertainty over the probability of healing in the 4LB group (taken from the MTC model as previously described).

Determinants of ulcer healing

Finally, for general interest to the field, and to aid analysis in Part III regarding the cost-effectiveness of subgroups of venous ulcer patients, we explored the implication adjustments made in the MTC for participant’s baseline characteristics (ulcer area, ulcer duration at baseline and mobility), which investigate whether these have an impact on healing. This analysis indicates which participants characteristics are expected to be associated with a higher, or lower, probability of healing over time. For the characteristics deemed relevant (those that showed statistically significant effects) we illustrated how the probability of healing over time was affected, using inferences obtained in the MTC (note that these parameters were informed by the IPD trials, VenUS I7 and VenUS IV).

Probability of treatments being best

An important feature of the Bayesian methodology used here is the ability to assess the probability that each relevant treatment is the best with respect to the primary endpoint (i.e. time to ulcer healing). When making recommendations on which treatment to use, this measure reflects the impact of uncertainty on the relative treatment effects. In the base-case analysis model (including VenUS IV data) the treatment associated with greater probability of healing was the 2LB (71.9%) (Table 45), reflecting the fairly high relative effect point estimate and the wide uncertainty around this from available evidence (see Table 44).

Quality assessment and consistency of evidence

Interpretation of the MTC evidence must consider also its quality. The overall quality of the evidence for each treatment comparison was assessed using the iGRADE approach. Table 46 shows the result of this assessment together with the effect estimates from comparison of the five main ulcer treatments.

Although the MTC model finds that 2LB was associated with a higher hazard of healing than the other treatments, quality assessment shows that these findings were driven by direct, but low-quality, evidence encompassing two studies [see Table 42, study numbers 34122 and 39123 with unclear/high risk of bias (one study was particularly small, with 31 participants)]. Inferences on the 4LB compared with SSB were made using data from multiple RCTs; however, the results are driven by a high-quality study for which IPD were available7 and the quality classification for this estimate is high. Except for the two treatment comparisons informed by a high-quality study available in IPD format, estimates were considered to be of medium to very low quality.

In the network of evidence there was one closed loop through which both direct and indirect data informed relative treatment effect estimates. Thus, the existence of inconsistency was explored. There was no evidence of statistically significant discrepancies between the direct and the indirect data. However, we note, given the fairly high uncertainty in the evidence base and the obtained estimates, only large differences in direct and indirect data within the loop would have returned a statistically significant result.

Overall, the quality of the evidence limits the confidence that we have in some MTC findings. Although estimates for the 2LB suggest that it has the highest probability of being the most effective, these findings are supported by low- or very low-quality evidence. Estimates for the SSB and HH indicate that they have a fairly low probability of being the best; however, these estimates are informed by high-quality evidence.

Inclusion of non-high-compression treatments in the mixed-treatment comparison

The results of the sensitivity analysis considering evidence on both high-compression and non-high-compression treatments [see Table 45, (3)] show that extending the network had a very limited effect, with the relative effect estimates of the main five high-compression treatments remaining similar (see Table 45, (3) vs. (1)].

Sensitivity analysis regarding the effectiveness of two-layer bandage

Because the quality of RCT data driving the relative effectiveness of the 2LB to the 4LB was so low we conducted a post hoc analysis, assuming that there was no difference in the effectiveness of these treatments. In this scenario the high-compression treatment with highest estimated probability of being best was HH, with approximately 67% probability of being best (the 4LB and the 2LB, together, would have 5.5% probability of being best).

Presentation of mixed-treatment comparison results in terms of probability of healing over time

For ease of interpretation we also present MTC findings as the probability of healing over time (Figure 12). In Figure 12a, point estimates for the probability of healing are plotted showing that, at any time point, the 2LB is expected to be the most effective high-compression treatment and paste is the least effective one. The expected probability of healing for HH is equivalent to that of the 4LB.

FIGURE 12.

Reflecting uncertainty over relative treatment effects in the probability of healing over time for the five main high-compression ulcer treatments. (a) The expected probabilities of healing (point estimates) across time (25 months); (b) to (e) compare the expected values for 4LB with the healing probability (point estimates and uncertainty) of each of the other four high-compression treatments. Estimates reflect the average participant in the trial data from VenUS IV (mean ulcer area at baseline of 9.4 cm² and ulcer duration at baseline of 11.5 months).

Parts (b) to (e) of Figure 12 illustrate the uncertainty around the probability of healing using 95% credibility bands (the equivalent to confidence bands) represented by shaded areas. Thus, for a given time point and with 95% confidence, current RCT evidence indicates that the probability of a cohort of participants healing lies somewhere in the shaded area. As the 4LB group was a reference treatment in the MTC analyses, its estimates were not shown as uncertain here. Parts (b) and (c) of Figure 12 show that uncertainty regarding the relative probability of healing with the SSB and HH is less marked than with other treatments – the confidence bands are narrower in parts (b) and (c) than in parts (d) and (e). The expected probability of healing with the 4LB is on the upper bound of the shaded area for SSB (reflecting the upper bound of the CrI for the HR of approximately ‘1’). For paste and the 2LB, uncertainty in their HRs translates into a large uncertainty in the probability of healing over time – the full line in parts (d) and (e) is inside the shaded areas, indicating that no evidence of a statistically significant difference from 4LB. In both these cases, estimated uncertainty is particularly large at between 5 and 20 months after treatment initiation.

Uncertainty over the probability of healing with four-layer bandage (reference treatment)

All findings to this point have been based on the relative treatment effects in relation to the 4LB, the common comparator. For this reason, 4LB estimates shown in the plots of Figure 12 were not uncertain. However, the synthesis models also evaluated uncertainty over the probability of healing in the 4LB group, controlling for centre effect and participants’ baseline characteristics. The estimated probability of healing associated with the 4LB over time (specifically for a participant with area and duration of ulcer equal to the mean values observed in VenUS IV) is represented in Figure 13 and uncertainty in the estimates is shown in the shaded area.

FIGURE 13.

Probability of healing across time (36 months) for the 4LB considering baseline heterogeneity. (For an average participant with a mean ulcer area at baseline of 9.4 cm² and ulcer duration at baseline of 11.5 months.)

Determinants of ulcer healing

For completeness it is important to understand how patient characteristics might impact on venous ulcer healing and the models here provide an ideal opportunity to investigate this given the available baseline and healing data for VenUS I7 and IV. The base-case analysis model, including trial data from VenUS IV, suggests that no matter the compression treatment used, log baseline ulcer size and log ulcer duration are important predictors of ulcer healing time (HR of approximately 0.71 and 0.92, respectively, CrIs do not include 1 – Table 44), with increased time to ulcer healing as ulcer area and duration increase. Based on VenUS I7 and VenUS IV data, we found that an individual with a venous leg ulcer of one month’s duration (all remaining characteristics assumed at their average value, i.e. ulcer duration at baseline of 11.5 months approximately 36% possibility of having limited mobility and 0.1% of being immobile), who starts using the 4LB, is expected to heal within 12 months (with a probability of 0.92). If, instead, a patient has an ulcer of approximately 5cm² and starts treatment with the 4LB, the probability that this ulcer heals within six months is 0.55. For all sizes or durations of the baseline ulcer considered, patients that started treatment with the 4LB are expected to have healed with probability of 0.88 or higher at 36 months. Figure 14 shows additional information on the impact of these two factors on the effect of treatment with the 4LB. Participant mobility was also seen to be a relevant factor to explain time to ulcer healing as participants with ‘difficulty in walking’ were associated with slower healing than those that ‘walk freely’ (HR of 0.72 in Table 44).

FIGURE 14.

Probability of healing across time (36 months) for the 4LB considering (a) different durations of ulcer at baseline and (b) different ulcer sizes at baseline. All remaining characteristics assumed at their average value.

Ulcer recurrence

The main goal of treating venous leg ulcers is ulcer healing. However, many (possibly most) ulcers recur post healing. Although the frequency and speed of recurrence may not be determined by initial ulcer treatment, quantifying recurrence is important, as the event may affect the level of health after treatment: i.e. health-related quality of life after first healing should consider the possibility of a recurrence, otherwise QALY gains of treatment may be overestimated. In decision models, recurrence was represented using a transition backwards from the health state ‘healed’ to ‘unhealed’ (see Figure 15a). 7 Alternatively, recurrence was assumed to be a distinct state (see Figure 15b), allowing for healing or death to occur at different rates after recurrence.

Adverse events/complications

Complications and severe adverse events related to treatments are often relevant for economic analyses owing to the relative high cost of negative events compared with other categories of resource use. One of the models reviewed,142 which evaluated a surgical intervention, set up an additional health state representing the discontinuation of treatment after occurrence of such events. In this way, the model allowed for distinct costs and healing (and/or death) rates in patients with severe adverse events/complications. Figure 15c presents an adapted diagrammatic representation of how a model considering both recurrence and adverse events could be specified.

Compliance

Compliance with venous leg ulcer treatments may vary. Compliance is here used to describe the degree to which a patient uses the treatment as prescribed, with variations mainly concern duration of treatment. Poor compliance with treatment may reduce or annul the effectiveness of a treatment. However, delivery of treatments to non-compliant patients may still incur costs. The model reported by Korn et al. 143 examines the impact of compliance with venous leg ulcer treatments. Here patients entering the model are classed as compliant or non-compliant and undergo distinct transition rates to reflect this. Following this design, the model could be expressed by the diagram in Figure 15d.

Health-related quality of life/utility

In total, 72 citations were initially identified and full papers were obtained for 33 of these after which a further 30 studies were excluded (see Appendix 15). The main characteristics of the three included studies44,48,150 are described in Table 50. The study by Walters et al. 48 reported the health-related quality-of-life results from a RCT that investigated the clinical effectiveness of community compared with home-based service in England. All trial participants received the 4LB and were asked to report their EQ-5D index score at 12 weeks and 12 months. The study also reported the proportion of healed and unhealed patients at 12 months. However, the analysis methods utilised was unclear from the RCT report and utility results across the 12-month assessment appeared inconsistent. We noted a bigger decrease in utility score at 3 months for healed participants compared with unhealed participants (suggesting decreased health-related quality of life), whereas overall a positive average utility gain per month up to 12 months was estimated for healed patients (Table 51). We did not contact the authors for further details. The two more recent studies were National Institute for Health Research-funded RCTs, one comparing the clinical effectiveness and cost-effectiveness of the 4LB with the SSB (VenUS I),7 and the other comparing the effects of silver-donating with non-silver low-adherent dressings in the treatment of venous leg ulcers (VULCAN trial). 150 All participants in the VULCAN trial150 wore multilayer compression bandage or hosiery over their dressing. Both studies reported EQ-5D scores separately for healed and unhealed ulcers, and could be used to establish an effect of ulcer healing on health-related quality of life (see Table 51). These results were used to inform the decision model, for which an increase in health-related quality-of-life score for a patient whose ulcer healed was estimated to be 0.00251 in the VULCAN trial150 and 0.11 in VenUS I7 (adjusted analysis).

Costs and resource use

In total, 72 citations were identified. After screening, 25 full papers were obtained. Following further screening, 16 studies were excluded; these were studies conducted outside of the UK and those studies that did not include information relevant for our model (see Appendix 16). Thus two studies informed our model for this parameter: VenUS I7 and VULCAN150 (details of these RCTs have been described elsewhere).

Based on the trial-level cost-effective analysis of VenUS IV (see Part I) nurse consultations for ‘unhealed’ participants was known to be a key cost driver; consequently, data relating to this cost component was of particular interest. However, of the two studies identified, only VenUS I7 assessed the number of visits for specific compression treatments, reporting a mean of 5.1 consultations per month in the 4LB arm and 6.0 in the SSB arm. No similar evidence was available for the paste and 2LB.

Ulcer recurrence

In total, 182 citations were identified. Of these, 147 were excluded and full papers were obtained for 35 studies. Upon further screening 29 studies were subsequently excluded for a variety of reasons (see Appendix 17): 17 studies were either not primary research or did not report recurrence results, eight studies were conducted outside of the UK, four studies evaluated surgical treatments (data from these were thus considered to potentially misrepresent recurrence rates in patients using compression), two studies assessed prevention strategies not used in current clinical practice, and four studies did not report whether participants were routinely offered the use of prevention compression and were considered too old to be able to provide data that was contemporary to VenUS IV regarding routine use of prevention treatments. It was for this last reason that VenUS I7 was excluded.

Details of the six included studies6,150,151–154 all relevant to the UK, are shown in Table 52 (all were RCTs). Three RCTs examined the effectiveness of surgical approaches (i.e. venous surgery and grafting) in preventing ulcer recurrence. Surgery is not commonly used in the UK as a treatment to heal venous ulcers thus we extracted data from only the control arms here (where participants received compression treatments). The proportion of participants with a recurred ulcer ranged from 30% at 6 months, 28–36% at 1 year and 56% at 4 years. The RCTs by Brooks et al. 152 and by Nelson et al. 153 investigated the effect of using hosiery after healing in preventing recurrence. Brooks et al. 151 suggest that a nurse education programme significantly improved ulcer recurrence rates but there was no association with the time spent in maintenance hosiery. Nelson et al. 153 found no evidence of a difference in recurrence rates between class 2 and class 3 compression hosiery. The final included study was the VULCAN RCT,150 in which all RCT participants were advised to wear hosiery after healing to prevent recurrence.

Although our decision model explicitly accounts for the possibility of a recurrence, it does not differentiate between primary ulcers and recurrent ulcers in terms of differential times to healing or impact on death rates. First, this was because we found no evidence of such an effect (the studies reviewed did not investigate whether risk of ulcer recurrence depended on the treatment received for the primary ulcer). Second, the population of interest within the current evaluation included people with venous leg ulceration that could be primary or recurrent.

Regarding the rate of recurrence, studies were heterogeneous and results varied significantly across studies. VULCAN,150 the most recent study, reported the lowest recurrence rate. We postulated that developments in the management of patients at risk of venous leg ulceration may have led to a reduction in recurrence rates in more recent years – thus, in addition to trial data from VenUS IV, we used only evidence from VULCAN149 to inform the decision model for recurrence data. Despite the evidence suggesting that the likelihood of recurrence may decrease as time from healing increases (see Barwell et al. ),151 the data available in the literature alone was insufficient to allow this issue to be considered explicitly.

Mortality

A total of 414 citations were identified (see Appendix 18). After initial screening, six studies were obtained in full;155–160 however, we were unable to retrieve one paper. 155 Of the remaining five studies,156–160 three were conducted in Sweden,154–159 one in Austria160 and one in the UK. 159 Subsequently, we focused on the UK study. 159 This was a RCT that evaluated the effect of a national community intervention programme on healing rates of chronic leg ulcer in Scotland.

In this cluster RCT, 3949 participants were registered and followed up using quarterly surveys over a 30-month period. Overall, there were 489 deaths and 65 amputations, corresponding to a quarterly rate of death of 3.7%. The study found no evidence of a difference between the two treatment groups – participants subject to a programme of nurse training that was specific for leg ulcer care (intervention), which included workshops, lecture material, and a range of dressing and bandaging material, among other items compared with no training (control).

Specification of the patient population

In Part II, we have seen that the probability of healing is determined by ulcer duration and area. It is intuitive that the probability of dying is determined by the age of the cohort. It is thus important to specify the key characteristics of our model population using values representative of venous leg ulcer patients. For the base case VenUS IV information was used, whereas the scenario analysis (without VenUS IV scenario) reflected participant characteristics observed in VenUS I. 7 Table 53 shows detailed information on the values assumed.

Transitions from unhealed to healed

Effectiveness evidence was synthesised in a MTC as described in Part II. The Weibull MTC model was used to define the transition probabilities to healing. The base case included IPD trial data from VenUS IV and the scenario analyses excluded it. The MTC results were incorporated into the decision model using the posterior samples or predictive distributions when a random effect across centres had been considered [extracted from the Convergence Diagnostic and Output Analysis WinBUGS output (CODA)].

Specifically, the probability of healing for patients starting treatment with the 4LB was specified according to the patient’s ‘baseline’ mobility status, ulcer area and duration. The mean values (section 1 of Table 53) and the regression coefficients estimated in the MTC (β_{log_ulcer_area}, β_{log_ulcer_duration}, β_{mobility_difficult}, β_immobile, μ and β_{new_centre}) were linearly combined on the log scale to generate λ (representing the scale of the ulcer healing time distribution), which was used, in turn, to calculate the transition probability to healing according to the following formula:

where tpheal4LB(tu) is the transition probability for 4LB for each cycle (where the cycle length was 1 month, and thus the times at each cycle u were represented by t_u) and the (ancillary) γ parameter defines the shape of the distribution. The relative effects of the remaining treatments compared with the 4LB were then applied to this baseline value as follows:

where HRⁱ is the HR of treatment i in relation to the 4LB. Figure 17 illustrates the cumulative probability of ulcer healing with the 4LB in the model, for the base case (with trial data VenUS IV, reproducing Figure 15) and for the scenario analysis (without VenUS IV). Uncertainty in the estimates is shown in the shaded areas. As expected, for the majority of the time points considered, uncertainty is higher when excluding VenUS IV evidence than when including it.

FIGURE 17.

Probability of healing with the 4LB for the base-case and scenario analyses.

Transitions from healed to unhealed (ulcer recurrence)

For the base-case analysis, trial data from VenUS IV alone was used to inform this parameter, where approximately 19% of participants recurred during follow-up (65 out of 343). Additionally, differences were found between treatment arms with respect to recurrence (approximately 15.0% of participants allocated to the HH group had a recurrence event compared with 23.5% in participants allocated to the 4LB group). Given the availability of IPD, the probability of recurrence parameter was evaluated using time-to-event models. This analysis was parametric and we evaluated the following alternative distributions of time to event: exponential, Weibull, Gompertz, log-logistic, log-normal and gamma. The selection of the distribution was based on values of the Akaike information criterion (AIC),166 considering goodness of fit and penalising model complexity (Table 54). All regression models adjusted for baseline participant mobility (categorical variable) and baseline reference ulcer duration (months). Baseline ulcer area was removed from the model based on the lack of statistical significance.

The Gompertz regression was selected based on lower AIC value, which indicates a better fit (see Table 54). Figure 18 shows the cumulative probability of ulcer recurrence for the 4LB arm of VenUS IV, based on the selected model. This shows that, on average, 12 months after healing the probability of observing an ulcer recurrence was estimated to be approximately 27.0% in participants allocated to 4LB compared with 17.7% for participants allocated to HH. Again, uncertainty in the estimates is shown in the shaded areas.

FIGURE 18.

Probability of recurrence for the 4LB and hosiery from VenUS IV (base-case analysis).

The distribution chosen to represent time to recurrence – the Gompertz distribution – assumes that the hazard of recurrence depends on time since healing. This time dependency was built into the decision model structure using tunnel states to relax the Markov assumption. Tunnel states are a modelling ‘trick’, for which one state is split into as many states as the number of cycles for which we wish to consider the time dependency. In this case, the initial healed state was transformed into 13 states: ‘healed 1’, ‘healed 2’, etc. Patients were allowed to remain in states for only one cycle (1 month), after which they either recurred (transit to the unhealed state) or died. In this way, the time patients spent healed was tracked and the probability of recurrence depended on this duration. If, after 12 months, patients had not recurred or died then they transitioned into a ‘healed’ state for which the probability of ulcer recurrence was assumed null. The transition probability for recurrence, tpⁱ_recurr, was defined using the parameter estimates in Table 54 using the following relation:

where λ is a scale parameter (a linear combination, on the log scale, of the average population characteristics and estimated coefficients: β₀, β_{mobility_difficult}, β_immobile), γ the shape parameter of the Gompertz distribution and ‘HR’ the HR for recurrence of treatment i relative to the 4LB (note that only HH was assumed to affect recurrence, thus all remaining compression treatments for which models were run were assigned a HR of ‘1’).

In the scenario analysis, evidence from the VULCAN trial150 was the source of recurrence data (and trial data from VenUS IV were not included). The VULCAN trial150 observed that 13% of participants recurred within 12 months (24 from a sample of 185 participants healed within the study). This information was used in the model by defining a transition probability for recurrence using:

where p_recurr represents the probability of recurrence observed in the study (0.13). In this scenario analysis; recurrence was not assumed to depend on treatments used for ulcer healing as the evidence from the review conducted did not indicate that such an effect might exist.

Transition to death

Most ulcer patients are elderly and they generally have more comorbidities than the average population. 7,13,61 Although the presence of an ulcer may not directly affect patients’ mortality, the characteristics of this population may imply a higher mortality rate that the general population. Transitions probabilities to the absorbing state (death) were informed by official population statistics165 and adjusted using the mortality ratio of a population of patients with ulcers (calculated from Brown et al.,159 taken to be 2.36 times higher than of the general population). Mortality was not assumed to depend on the treatments evaluated. The same information was used in both the base-case and scenario analyses.

Costs, resource use and health-related quality of life/utilities

Table 55 identifies the evidence used for costs and resource use and health-related quality of life for both the base-case and scenario analyses. Two main cost components were defined:

1. Costs dependent on health state (healed and unhealed) The model considered only ulcer-related costs, so once a patient had healed then no costs were incurred in the model. When patients had an ulcer, i.e. were in the unhealed health state, clinician and hospital consultations were considered (assumed to be not treatment dependent) (see Table 55).

2. Costs dependent on treatment Treatment costs were a function of the average number of nurse consultations per month for each patient, the cost of each of those nurse contacts and the cost of the treatment applied in each visit (see Table 5), i.e. mean number of nurse consultations × (cost of nurse visit + cost of applied treatment).

The mean number of nurse consultations while unhealed was derived from VenUS IV (see Table 28) in the base case and VenUS I11 in the scenario analysis. Given the scarcity of information on resource use for other compression treatments, in the ‘without VenUS IV’ scenario, the number of nurse consultations while on HH, paste and the 2LB was assumed to be the same as the 4LB. The number of nurse consultations for the SSB group was assumed to differ (based on data from VenUS I). In the base case, the number of nurse consultations while on the paste and the 2LB were again assumed to be the same as the 4LB owing to absence of information. Consultations for HH were informed by VenUS IV (see Table 28) and consultations for SSB were again informed by VenUS I.

The ‘without VenUS IV’ scenario considered that treatments of interest were provided to patients throughout the unhealed period. Evidence from VenUS IV, however, established that the same treatments are not used throughout an ulcer episode. As this is relevant for costs, it was considered in the base-case analysis by costing the treatment of interest for a period of time (Tdur_treat) after which ‘standard treatments’ were costed. These standard treatments were costed as an average of the compression treatments reported in VenUS IV after the allocated treatment was discontinued. In VenUS IV, duration of treatment was found to differ between the 4LB and HH. No evidence was available with respect to the duration of treatment for SSB, paste and 2LB. Thus duration of these treatments was assumed to be the same as the 4LB (see Table 55).

The health-related quality-of-life/utilities scores of healed patients were assumed to vary with time according to UK population utility score norms (per year of age) [Centre for Health Economics (CHE) report, 1999]. 167 These values were adjusted by baseline utility values from VULCAN150 for the ‘without VenUS IV’ scenario and for the base case, to reflect the fact that a patient population with venous leg ulcers is expected to have worse health than the general population.

Patients with unhealed ulcers were assigned a decrement in utility. In the scenario without VenUS IV, adjusted estimates at 12 weeks from VULCAN trial150,164 and VenUS I7 – 0.00251 and 0.11, respectively – were pooled. To pool, a weighted average of the study estimates was used, with the weights being the inverse of the variance for each of the study estimates (i.e. the larger the variance, the smaller the weight, and vice versa). The pooled value obtained for this scenario was 0.08 (SE = 0.024, Table 56). For the base case, VenUS IV was used in addition to the other two studies; the pooled statistic assumed a value of 0.082 (SE = 0.021), a similar value to that of the scenario analysis.

Specification of distributions for probabilistic analysis

To perform probabilistic analysis input parameters were represented as a probability distribution. The type of distribution was dictated by the nature, characteristics and method of estimation used to derive the evidence for each parameter (Table 57). Once the type of distribution was defined, the mean statistics and SE were used to define the distribution parameters using the method of moments. For the ‘without VenUS IV’ scenario analysis, the probability of ulcer recurrence was modelled through a standard beta distribution parameterised by the number of individuals that recurred given those healed at 3 months (see Table 57). Consultations with health-care professionals were represented by a normal distribution, parameterised using evidence in Table 55. Adjusted utility decrements for unhealed ulcer patients were represented by a standard beta distribution. Adjusted population utility scores per year of age and decrements due to having an unhealed ulcer were also modelled using normal distributions.