A randomised controlled trial to compare the safety, effectiveness and cost effectiveness of doxycycline (200 mg/day) with oral prednisolone (0.5 mg/kg/day) for initial treatment of bullous pemphigoid: the Bullous Pemphigoid Steroids and Tetracyclines (BLISTER) Trial

What is pemphigoid?

Bullous pemphigoid (BP) is the most common form of a group of autoimmune pemphigoid diseases characterised by autoantibodies directed at skin adhesion proteins of the epidermal–dermal junction. 1 BP often starts with intense itching followed by an urticated and/or eczematous-looking rash and tense blisters on erythematous or normal-looking skin (Figure 1). The blisters may develop several months after the appearance of the initial symptoms and may be quite large and filled with blood. They often break down to form raw skin erosions, which can become infected.

FIGURE 1.

Large tense blisters and erosions in a man with BP (by kind permission of Professor Enno Schmidt with written consent from the patient).

The pain and severe itching associated with BP can greatly impair quality of life. BP typically affects people aged around 80 years and the incidence, which appears to be on the increase, is estimated to be between 6 and 43 cases per million per year. 2,3 The cause of BP is unknown but it is associated with neurological conditions such as cerebrovascular disease, dementia, Parkinson’s disease, epilepsy and multiple sclerosis4,5 and the chronic use of several drugs including spironolactone (Aldactone^®, Pharmacia Ltd), neuroleptics, certain diuretics and phenothiazines in some patients. 4,6–9 Neurological disease has been suggested as a predisposing and prognostic factor. 10

Increased mortality

Mortality rates are increased for people with BP. In one study the risk of death was more than six times that of the general population. 11 Mortality rates for people with BP of between 11% and 41% have been suggested to be related to advanced age and comorbid medical conditions rather than the disease itself. 12,13 It is likely that the increase in mortality rate is owing, at least in part, to side effects of oral prednisolone, which is commonly used to treat BP. 14–17

Topical treatment

Some widely reported studies have shown that superpotent topical corticosteroids applied to the whole skin surface are effective and safer than oral corticosteroids. 18,19 Practical guidance on how this can be carried out is provided elsewhere. 15 However, the application of topical steroids to the whole body for weeks or possibly months is not a practical option in many elderly outpatients with limited support. Therefore, there still remains a need for a convenient oral treatment that is both effective and safe.

Rationale for testing tetracyclines further

Treatment of BP with antibiotics possessing anti-inflammatory actions, such as those from the tetracycline group, is widely used in clinical practice. In a national survey of 326 UK dermatologists conducted in 2013, around 80% stated that they had used doxycycline (Efracea^®, Galderma), minocycline (Aknemin^®, Almirall Hermal GmbH) or lymecycline (Tetralysal^®, Galderma). 20 However, a Cochrane systematic review21 found only one small poorly reported clinical trial22 and concluded that further evidence was needed before the effectiveness of tetracyclines could be established. No further trials of tetracyclines were identified in a subsequent systematic review23 nor in a search of the Cochrane Central Register of Controlled Trials using ‘pemphigoid’ as a search term (28 May 2015). The limited data available and practical experience to date suggest that it is very unlikely that tetracyclines would be more effective than long-term use of oral corticosteroids, yet they are likely to be safer in this elderly population given the known adverse effects of oral corticosteroids, including diabetes mellitus, serious infections and osteoporosis, leading to fractures and death. Because some patients with BP might not respond to tetracyclines at all, it was clearly important to test how they might be used in clinical practice by permitting additional application of potent topical corticosteroids to affected areas or a switch to oral corticosteroids if symptoms and blister control were inadequate. Similarly, in the UK, those initiated on oral steroids are typically given topical corticosteroids for localised application to blisters to help with initial control and doses of oral corticosteroids are typically adjusted over several months, according to blister control or side effects. We therefore sought to investigate whether or not a strategy of initial treatment of BP with anti-inflammatory tetracyclines (200 mg/day of doxycycline) is effective enough to produce an acceptable degree of blister control compared with initial treatment with oral corticosteroids (0.5 mg/kg/day of prednisolone) in a non-inferiority comparison and whether or not tetracyclines confer a long-term advantage in terms of safety over oral corticosteroids in a superiority comparison.

Rescue medication

Up to 30 g/week of topical corticosteroids in the potent class [preferably mometasone furoate (Elocon^®, Merck Sharp & Dohme Ltd)] was permitted throughout the study, except between weeks 3 and 6. The cessation of topical corticosteroids from week 3 to week 6 provided a washout period to minimise the potential effect of systemic absorption of topical corticosteroids on the primary effectiveness outcome at 6 weeks. To minimise potential systemic effects, topical corticosteroids were applied only to blisters and erosions. Participants were permitted to apply a moisturiser to blisters and erosions at any time.

Choice of tetracycline

Doxycycline was chosen for this trial because (1) it is associated with a lower incidence of gastrointestinal side effects than other tetracyclines and (2) the alternative option of oxytetracycline would have required participants to swallow approximately eight large tablets a day.

Although the only published randomised controlled trial investigating tetracycline antibiotics for the treatment of BP used a combination therapy of tetracycline plus nicotinamide,21 a single therapy of doxycycline was chosen for this trial to allow the effects of the tetracycline to be clearly defined.

Primary outcomes

The primary outcomes were the absolute difference between the two treatment arms in the:

• Non-inferiority comparison. The proportion of participants classed as a treatment success (three or fewer significant blisters present on examination) at 6 weeks. A significant blister was defined as an intact fluid-filled blister at least 5 mm in diameter or a ruptured blister with a flexible (not dry) roof over a moist base. Mucosal blisters were excluded from the count.

• Superiority comparison. The proportion of participants with grade 3 (severe), 4 (life-threatening) and 5 (death) adverse events that were possibly, probably or definitely related to the treatment in the 52 weeks following randomisation. A modified version of the Common Terminology Criteria for Adverse Events v3.0 was used [see http://ctep.cancer.gov/protocolDevelopment/electronic_applications/docs/ctcaev3.pdf (accessed 12 November 2015)] and the relatedness of grade 5 (fatal) adverse events was judged by an independent adjudicator. Although only grade 3 and above related adverse events were captured in the primary outcome, less medically important related adverse events (such as weight gain and skin fragility) can bother patients when taking corticosteroids and so were included in a secondary outcome.

For the primary outcome, treatment success was defined as three or fewer significant blisters, regardless of whether or not treatment had been modified because of a poor response (either by changing the dose or by changing the treatment) during the first 6 weeks. However, for all secondary and tertiary end points, participants were classed as a treatment success at each visit only if (1) they had three or fewer significant blisters present on examination and (2) their treatment had not been altered because of a poor response prior to that visit.

Secondary outcomes

The secondary outcomes were the absolute difference between the two treatment arms in the following:

• non-inferiority comparisons:

  • proportion of participants classed as a treatment success (three or fewer significant blisters present on examination and no treatment modification) at 6 weeks

  • proportion of participants classed as a treatment success at 13 and 52 weeks

  • proportion of participants who had a further episode of BP during the study

• superiority comparisons:

  • proportion of participants reporting adverse events of any grade that were possibly, probably or definitely related to BP medication in the 52 weeks following randomisation

  • quality of life [European Quality of Life-5 Dimension (EQ-5D) and Dermatology Life Quality Index (DLQI) questionnaires at 6, 13, 26, 39 and 52 weeks]

  • cost-effectiveness over 12 months from a NHS perspective

• combined comparison: proportion of participants classed as a treatment success at 6 weeks and who were alive at 52 weeks.

Tertiary outcomes

The tertiary outcomes were the absolute difference between the two treatment arms in the following:

• non-inferiority comparisons:

  • proportion of participants completely blister free at 6 weeks

  • proportion of participants classed as a treatment success at 3 weeks (to compare the speed of onset of action)

• superiority comparisons:

  • mortality over the 52-week follow-up period

  • amount of potent and superpotent topical corticosteroids used during the 52 weeks following randomisation.

Retention of participants

To help retain participants in the study, in addition to being able to speak to the investigator, participants were able to telephone the trial manager if they wished to discuss any aspect of the study. For medical queries, participants were directed to a medical member of staff. In addition, the trial administrator made telephone calls to participants to support them throughout the duration of the study. Participants were sent birthday and Christmas cards while they were participating in the study.

Withdrawal of participants

Patients whose immunofluorescence test results were not available until after randomisation and which were subsequently both negative, indicating that they did not have BP, were withdrawn from the trial, replaced and not included in the analysis. This approach reflects normal practice: if there is a clinical picture of BP, treatment is commenced and this is changed later if the laboratory tests are subsequently negative.

Participants who withdrew from study treatment were followed up for the remainder of the year unless they had withdrawn their consent.

Informed consent

Patients presenting with suspected BP were assessed by the recruiting investigator as per normal clinical practice. If a diagnosis of BP was suspected, the patient was given details of the study verbally. If the patient was interested in taking part he or she was given time to read the full participant information leaflet and the investigator answered any questions. A consent form was signed before any study procedures were carried out.

If the disease severity was such that immediate oral treatment was required or the patient did not wish to delay the start of treatment, he or she was able to give consent and be randomised to the study during the first visit to the dermatologist. Otherwise, the patient was given a second appointment for consent and randomisation.

Subgroup analyses

For some patients the blister count at 6 weeks may have been performed by an investigator who knew what treatment the patient was on. To determine whether or not this introduced bias into the results of the 6-week effectiveness outcome, an interaction test was performed to compare the treatment effects in patients who were and patients who were not assessed by an investigator who knew the treatment allocation. This interaction analysis was performed on both the primary and the secondary definition of treatment success at 6 weeks.

Subgroup analyses of treatment success (using both definitions) at 6 weeks by the three categories of baseline disease severity (mild, moderate and severe) were also performed. A global test for a treatment interaction was used to determine whether or not treatment effects were different in the three categories of disease severity.

The EuroQoL tool

The EuroQol tool is a two-page questionnaire consisting of the EQ-5D descriptive system and the EuroQol visual analogue scale (EQ VAS). 32 The tool is a standardised measure of current health status developed by the EuroQol group for clinical and economic studies. The EQ-5D three-level version (EQ-5D-3L) was used, consisting of five questions addressing the health dimensions of mobility, self-care, usual activities, pain/discomfort and anxiety/depression. Each dimension is assessed at three levels: no problems, some problems and extreme problems. EQ-5D and EQ VAS data were collected using patient-completed questionnaires at baseline and 6, 13, 26, 39 and 52 weeks. Scores were converted to a single health-related index ranging from 0 (death) to 1 (perfect health), with negative scores possible for some health states. Patients who died during the study were subsequently scored 0 at later scheduled follow-up visits.

Resource use

Resource use assessments were carried out at 3, 6, 13, 26, 39 and 52 weeks during mandatory clinical visits and were augmented by telephone calls. Recall was assisted by the use of patient diaries. Patients’ use of study and non-study drugs was recorded and costed using weighted average prices determined from Prescription Cost Analysis (PCA) data. 33 Health service contacts were recorded by asking patients to recall general practitioner (GP) clinic and home visits, practice and district nurse visits, outpatient visits and inpatient stays. Health-care resource use was costed using published national reference costs. 34–36 Patient-level resource costs were estimated as the sum of resources used weighted by their national reference costs.

Economic analysis

The economic analysis followed intention-to-treat (ITT) principles and a prospectively agreed analysis plan (see Appendix 2). No discounting was applied to economic data reflecting the follow-up period of 1 year. Costs were estimated in UK pounds sterling using patient resource use and 2013 reference costs. 35 The analysis took an English NHS perspective, reporting generic (EQ-5D, EQ VAS)32 and disease-specific (DLQI)37 health outcomes. Repeated scores over time were used to construct area under the curve (AUC) estimates for each patient, using the trapezoidal method. Missing values at individual follow-up points were managed using two scenarios: multiple imputation (the base-case analysis38) and analysis of complete cases (in which patients with any missing data were excluded).

Patient estimates of costs and quality-adjusted life-years (QALYs) at 1 year were used to derive an estimate of the cost-effectiveness of doxycycline-initiated therapy compared with prednisolone-initiated therapy for patients with BP. Estimates using imputed missing data provided the base-case analysis and estimates using complete data provided supportive sensitivity analysis. Analysis and modelling were undertaken in Stata 13 (StataCorp, College Station, TX, USA). The base-case analysis included the imputed within-trial incremental cost/QALYs gained, adjusted for trial covariates (age, sex, baseline blister severity and baseline Karnofsky score). QALY estimates were also adjusted for baseline EQ-5D score.

Primary effectiveness outcome

The primary effectiveness outcome was a non-inferiority comparison of the proportions of patients achieving treatment success at 6 weeks, defined as three or fewer significant blisters. It was anticipated that doxycycline would be less effective than prednisolone but that this would be accompanied by an improvement in the safety profile. As is best practice for non-inferiority comparisons, both mITT and PP analyses were performed.

Per-protocol analysis

Patients were excluded from the PP analysis for deviations from the protocol that could significantly affect the outcome. Patients may appear more than once under the various reasons for exclusion if they met more than one of the criteria. The results of the PP analysis were very similar to the results of the ITT analysis, with 92.3% of patients in the prednisolone group achieving success compared with 74.4% in the doxycycline group, an adjusted difference of 18.7% (90% CI 9.8% to 27.6%) (Table 8), which falls well within the 37% upper bound for the 90% CI.

TABLE 8 - Proportions of participants who achieved treatment success at 6 weeks: PP analysis

Estimates are from a regression model adjusted for baseline severity of BP and Karnofsky score; however, age was omitted from the model because the model failed to converge when age was added as an adjustment factor.

Note

For the study treatment doxycycline to be considered non-inferior to the control treatment, the upper bound of the 90% CI should fall below 37%. This analysis includes all patients who had a blister count at week 6, and excludes those participants without a blister count at week 6, who were lost to follow-up or who died before week 6. Treatment success = 0–3 significant blisters at week 6; treatment failure = ≥ 4 significant blisters at week 6. (See Appendix 3, Table 58 for reasons for exclusion.)

Outcome	Number (%) of patients
	Doxycycline	Prednisolone, n (%)
Success	58 (74.4)	84 (92.3)
Failure	20 (25.6)	7 (7.7)
Total	78	91
Difference in proportions (prednisolone – doxycycline) (90% CI) (%)	Adjusted:a 18.7 (9.8 to 27.6)
	Unadjusted: 17.9 (8.6 to 27.3)

Both the mITT and PP analyses are represented graphically in Figure 5.

FIGURE 5.

Proportions of participants who achieved treatment success at 6 weeks: mITT and PP analyses.

Subgroup analyses

Two subgroup analyses were performed to test for treatment interactions. There were no significant interactions in either the mITT or the PP populations between effectiveness at 6 weeks and baseline disease severity (mild, moderate and severe) (Tables 9 and 10 respectively) or between effectiveness at 6 weeks and the blinding status of the investigator who carried out the week 6 blister count (Tables 11 and 12, respectively).

Primary safety outcome

The primary safety outcome was the proportion of patients who, during the 52 weeks following randomisation, experienced at least one adverse event that was judged to be either grade 3 (severe), grade 4 (life-threatening) or grade 5 (death) and possibly, probably or definitely related to study treatment. The mITT population for this primary safety outcome included all those with data from at least one scheduled visit, regardless of any changes to treatment. An adjusted regression model in which preceding visits were set to zero was used to impute missing data to determine the sensitivity of the observed results to the missing data.

The risk of experiencing a treatment-related severe, life-threatening or fatal adverse event for patients started on doxycycline was 18.2%; this compared with 36.3% for those starting on prednisolone (Table 13). This represents a difference of 19.0% (95% CI 7.9% to 30.1%) after adjusting for baseline severity of BP. Similar results were obtained from a regression model in which missing data had been imputed: 22.5% of patients in the doxycycline group experienced a treatment-related severe, life-threatening or fatal adverse event compared with 40.0% in the prednisolone group, a difference of 18.4% (95% CI 6.0% to 30.8%) after adjusting for baseline severity of BP (Table 14).

A higher number of participants in the prednisolone group than in the doxycycline group had a maximum grade of adverse event of grade 3 (severe) and grade 5 (death) (Table 15). The pattern was the same for the total number of adverse events of each grade (Table 16).

Secondary effectiveness outcome: weeks 6, 13 and 52

For secondary effectiveness outcomes, the definition of treatment success was different from that used for the primary effectiveness outcome. A participant was required to have three or fewer significant blisters present on examination to be considered a treatment success but, unlike the primary effectiveness analysis, patients who had had their treatment modified because of a poor response (change of medication or dose of randomised medication increased) prior to the visit did not qualify as a success in the secondary analyses. This difference is because the primary effectiveness outcome assessed the strategy of starting on doxycycline or starting on prednisolone, whereas the secondary outcomes assessed the effectiveness of the treatments used as longer-term monotherapy. All secondary effectiveness outcomes are non-inferiority analyses.

The proportion of patients classed as a treatment success at 6 weeks in the mITT analysis, according to the definition of success described above, was 85.4% in the prednisolone group and 53.6% in the doxycycline group (Table 17). This represents a difference of 31.8% (90% CI 22.5% to 41.2%) in favour of prednisolone after adjusting for baseline severity of BP and age. The PP analysis showed similar results, with a difference of 34.4% (90% CI 23.7% to 45.1%) in favour of prednisolone after adjusting for baseline severity of BP and age (Table 18). The upper CIs for both analyses fall just outside the prespecified upper bound of 37%. Subgroup analyses of the effectiveness outcome data at week 6 using the secondary outcome definition failed to show any evidence of interaction between disease severity and treatment effect (Tables 19 and 20) in either the mITT or the PP population. There was no interaction between blinding status and treatment effect at 6 weeks using the secondary outcome definition in the mITT analysis (Table 21) but there was some evidence of an interaction effect when a PP analysis was done (Table 22).

At 13 weeks, in the mITT analysis, 75.3% of patients in the prednisolone group were considered a treatment success compared with 58.6% in the doxycycline group, a difference of 17.5% (90% CI 6.8% to 28.2%) in favour of prednisolone after adjusting for baseline severity of BP and age (Table 23). Again, similar results were noted for the PP analysis (a difference of 17.3% in favour of prednisolone, 90% CI 4.9% to 29.7%) (Table 24).

The longer-term assessment of effectiveness (the proportion classed as a success at 52 weeks) shows that, in the adjusted mITT analysis, 41.0% of patients started on doxycycline compared with 51.1% of those started on prednisolone achieved treatment success, a difference of 10.0% (90% CI –2.3% to 22.2%) in favour of prednisolone (Table 25). Similar results were seen for the PP analysis, with a difference of 7.1% (90% CI –7.1% to 21.3%) in favour of prednisolone (Table 26).

Secondary outcome: relapse rates

Another measure of the long-term effectiveness of the two treatments was the proportion of participants who had a further episode of BP during their participation in the study after previously being classed as a treatment success. A participant was classed as having a relapse if he or she had a further episode of BP, defined as more than three significant blisters, or a change or escalation of treatment because of worsening of disease during participation in the study after previously being classed as a treatment success (either three or fewer significant blisters present on prior examination or previously classed as a treatment success on the treatment log). After adjusting for baseline severity, age and Karnofsky score, a similar number of relapses occurred in the doxycycline group and the prednisolone group in the mITT population (2.1% more in the prednisolone group, 90% CI –8.3% to 12.5%) (Table 27). A larger difference was noted in the PP analysis (11.0% more relapses in the prednisolone group, 90% CI –1.2% to 23.2%) (Table 28).

Secondary outcome: combined outcome of effectiveness at 6 weeks and safety over 52 weeks

To provide an overall measure of success, a combined analysis of effectiveness and safety is presented in a superiority analysis. In the prednisolone group, 74.8% of patients were classed as a treatment success at 6 weeks and were alive at 52 weeks whereas in the doxycycline group 50.0% of patients were classed as a treatment success at 6 weeks and were alive at 52 weeks (Table 29). This represents a difference of 25.0% (95% CI 13.1% to 37.0%) in favour of prednisolone after adjusting for baseline severity and age.

Secondary safety outcome: all adverse events

This secondary outcome includes all adverse events that are possibly, probably or definitely related to the trial medication, regardless of severity. All analyses were carried out on a superiority basis. The maximum grade of related adverse events experienced by participants during the trial is shown in Table 30. Patients in the prednisolone group were significantly more likely to experience an adverse event that was related to the study medication than those in the doxycycline group (95.7% vs. 86.2%, 95% CI 1.8% to 17.2%; p = 0.016, unadjusted because of non-convergence in the model) (Table 31). The total numbers of related adverse events by grade (raw data) are shown in Table 32.

Secondary outcome: quality of life

Quality of life was assessed using the generic EQ-5D and the skin-specific DLQI. For a specific visit, the EQ-5D tabulations and analyses were conducted on all patients in the ITT population who had all five scores for the five questions for that visit (mobility, self-care, usual activity, pain/discomfort and anxiety/depression). Only patients who had at least one visit (not including baseline) for which data were available were included in the analysis. The EQ-5D score (social preference score) was obtained, with a value assigned to each combination of scores from the individual five questions (see Chapter 4, Data sources). For the DLQI, the tabulations and analyses were conducted for a specific visit on all patients in the ITT population who had all 10 scores for the 10 questions for that visit. Only patients who had at least one visit (not including baseline) for which data were available were included in the analysis. The DLQI score was calculated as the sum of each of the scores for the 10 questions asked at each visit.

There was a median change in EQ-5D score from baseline to week 52 of +0.090 in the doxycycline group and +0.071 in the prednisolone group (Table 33). When this was adjusted for baseline EQ-5D score, baseline severity, age and Karnofsky score, the difference was not significant (0.045, 95% CI –0.015 to 0.106; p = 0.143) (Table 34). Patients in the two groups had a similar improvement in DLQI score, with a median improvement from baseline to week 52 of 9 points in the doxycycline group and 10 points in the prednisolone group (Table 35). When adjusted for baseline DLQI score, baseline severity, age and Karnofsky score, there was a significant difference of –1.8 (95% CI –2.58 to –1.01) in favour of doxycycline (Table 36).

Tertiary effectiveness outcome: proportion of participants who achieved treatment success at 3 weeks (non-inferiority)

As a measure of the speed of onset of action, the proportions of participants classed as a treatment success (three or fewer significant blisters) were assessed after only 3 weeks of treatment. In the mITT population, a higher proportion of those started on prednisolone were classed as a treatment success at 3 weeks than those started on doxycycline, with a difference of 23.4% (90% CI 14.4% to 32.5%) after adjusting for baseline severity, age and Karnofsky score (Table 40), although this remains within the prespecified bounds of non-inferiority. A similar difference was seen in the PP analysis (Table 41).

Tertiary safety outcome: mortality over the 52-week follow-up period (superiority)

An assessment of all deaths, regardless of any relatedness to the study medication, was performed. Those who started on prednisolone were more likely to die during the year of follow-up than those started on doxycycline (83.5% alive at 1 year compared with 89.4%, respectively) (Table 42). The hazard ratio for reduction in death in favour of doxycycline, adjusted for baseline severity, age and Karnofsky score, was 0.61 (95% CI 0.30 to 1.24; p = 0.173) (Table 43). Time to death is shown in the Kaplan–Meier plot in Figure 6.

FIGURE 6.

Kaplan–Meier survival plot showing time to death.

Tertiary effectiveness outcome: use of potent and superpotent topical corticosteroids during the 52-week follow-up period (superiority)

Use of topical steroids (potent or superpotent) was permitted during the first 3 weeks of treatment for symptomatic relief (no more than 30 g per week to localised lesions only) and also after 6 weeks, as might occur in normal practice in the UK. Although use of topical steroids was discouraged from the end of week 3 to the week 6 effectiveness assessment, some participants did use topical corticosteroids for local relief of the affected area. Data on quantities of topical corticosteroids used were not collected accurately over the 1-year study period and so Table 44 presents topical corticosteroid use during different study periods. As anticipated, topical corticosteroid use for symptomatic relief was greater at all time points for those initiated on doxycycline treatment.

Post-hoc analysis: primary effectiveness outcome repeated at week 52

The primary effectiveness outcome at 6 weeks assessed the strategy of starting on doxycycline or prednisolone by including all participants regardless of any treatment modification. It was decided at the Trial Steering Committee meeting to conduct an additional post-hoc effectiveness analysis using the same definition of the population included to assess the strategy of starting on doxycycline or prednisolone over the whole 52-week period. There was no significant difference between the two groups when assessed over the full 52 weeks in either the mITT analysis (Table 45) or the PP analysis (Table 46).

Post-hoc analysis: action (treatment changes) taken in response to relapse

In total, 76 patients were identified as having relapsed after previously being classified as a treatment success, either because they had three or fewer significant blisters present or because they had ceased trial medication because of treatment success (Table 47). For those started on prednisolone, if they experienced a relapse, patients usually either received an increased dose of prednisolone (30/39, 76.9%) or recommenced prednisolone if treatment had ceased completely (8/39, 20.5%). For those who started on doxycycline, the most common course of action on relapse after having previously achieved disease control was to change to prednisolone (13/37, 35.1%) or increase the dose of prednisolone (16/37, 43.2%).

Post-hoc analysis: time to switching treatment

For those patients who switched treatments, the time between starting the study and switching treatment was assessed in a post-hoc analysis. This included all of the patients in the mITT population who contributed to the secondary efficacy analysis at week 6. The rate of switching from doxycycline was greatest during the first 6 weeks; after 13 weeks there was very little switching (Figure 7). There was a lower rate of switching for those starting on prednisolone and the rate was more constant over the whole 52 weeks.

FIGURE 7.

Kaplan–Meier survival plot showing time to change of treatment. Note: A change in treatment refers to a change from the allocated treatment (either doxycycline or prednisolone) to the other treatment. This graph includes the 213 patients who formed the mITT population for the primary efficacy outcome.

Post-hoc analysis: sensitivity of the primary effectiveness outcome at 6 weeks to topical corticosteroid use

A slightly different definition of a PP violator was used from that defined in the statistical analysis plan so that any use of a prohibited potent or superpotent topical corticosteroid between weeks 3 and 6 resulted in exclusion from the PP analysis of blister control at 6 weeks. The results of this post-hoc analysis are shown in Table 48. This shows that the non-inferiority status of doxycycline was robust to excluding those who used additional topical corticosteroids in the 3 weeks prior to the 6-week blister count assessment, at which the presence of three or fewer significant blisters was considered a treatment success.

Post-hoc analysis: safety data at 52 weeks according to baseline disease severity

To assist with interpreting the health economic analysis, an additional post-hoc analysis was performed for the primary safety data according to baseline disease severity (Table 49).

Similar post-hoc subgroup analyses for imputed data showed similar results: the differences in the percentages experiencing a related adverse event (prednisolone – doxycycline) were 25.2% (95% CI 4.0% to 46.3%) for mild baseline severity, 28.6% (95% CI 9.9% to 47.3%) for moderate baseline severity and –6.1% (95% CI –32.0% to 19.7%) for severe baseline severity.

Data sources

Resource use and quality-of-life measurements were collected from patients by health-care professionals at mandatory clinics at baseline and 6, 13, 26, 39 and 52 weeks. Generic health-related quality of life was assessed using the EuroQol questionnaire (containing the EQ-5D-3L and EQ VAS). The DLQI was included as a disease-specific quality-of-life measure.

The EQ-5D scores were converted to health status scores using the value set (time trade-off recommended by the EuroQol group). 40,41 Using the trapezoidal rule, the AUC of health status scores was calculated, providing patient-level QALY estimates for the cost-effectiveness analysis. 42 Similarly, EQ VAS and DLQI scores were integrated discretely over time. As AUC estimates were predicted to correlate with baseline scores (and thus potential baseline imbalances), AUC estimates were adjusted for baseline scores. 43 Sensitivity analyses explored adjusted and unadjusted estimation including a range of patient covariates: age, sex, baseline blister severity and baseline Karnofsky score as well as baseline outcome measure score.

Resource capture included use of study and non-study drugs as well as health service contacts: GP clinic and home visits, practice and district nurse visits and outpatient visits and inpatient stays. Patients were asked to recall use of resources since their last clinic visit and were given a diary at the beginning of the study to aid recall. Drugs recorded related to the management of BP or its sequelae. For health service contacts patients were asked to identify all patient contacts as well as those related directly to the management of BP or its sequelae. Consequently, ‘attributable costs’ are used in the base-case analysis and a sensitivity analysis included ‘all costs’.

Unit costs of resources were estimated or drawn from national reference sources for 2013 and are shown in Table 50. Study and non-study drugs were prescribed at varying doses. Using national PCA data,33 average costs per unit weight of therapeutic were determined and applied to patient drug use records. The list of drugs used included a number of occasionally prescribed drugs; pragmatically, drugs that were prescribed on more than five occasions during the course of the trial were costed. Costs for inpatient stays (in days) and outpatient visits were estimated using Hospital Episodes Statistics (HES)36 and NHS reference costs. 35 National HES data were explored for inpatient episodes with a primary diagnosis of International Classification of Diseases L12.0 Bullous Pemphigoid. The 10 most common Healthcare Resource Group (HRG) HRG4+ codes associated with that diagnosis were included, accounting for 96.2% of admissions. Per diem costs for each HRG4+ code were estimated from NHS reference costs and a volume-weighted average cost per admission for BP was estimated allowing for mean stay and cost per day. GP clinic and home visits, and practice and district nurse visits were costed using unit costs provided by the Personal Social Services Research Unit (PSSRU). 34 Patient costs were estimated as the sum of resources used weighted by their reference costs. The base-case analysis included only BP-related resource use.

Multiple imputation

In any form of analysis, missing data can cause serious problems. The problem is greater than a simple loss of statistical power as missing data may be not a random event but related to treatment and outcome. Examples include getting better (and no longer needing care) or dying, although QALY estimates allow for and include patients who die during follow-up. Quality-of-life measures typically feature repeated assessments over time, which are used to construct AUC estimates. This approach is both a strength and a weakness. With a complete-case analysis the weakness is evident: more observations may mean more incomplete assessments and loss of one follow-up assessment means the non-inclusion of that patient. However, repeated observations may provide (partial) assessments on more patients, allowing their experience to be reliably imputed. The base-case analysis used multiple imputation, conducted according to good practice guidance. 44,45 A range of alternative model specifications provided sensitivity analyses.

Multiple imputation provides a method of replacing each missing value with a predicted value, potentially permitting analysis of the entire trial sample. The process begins with reporting levels of missingness and the missing at random (MAR) assumption is then explored in the data. The MAR assumption requires that the probability of data being unobserved is dependent on the observed values but independent of unobserved values. Multiple imputation provides unbiased estimates of treatment effect if data are MAR. Additionally, complete-case analysis provides unbiased estimates only if data are missing completely at random, that is, the probability of data being unobserved is independent of both observed and unobserved values. The imputation model included all variables used within analysis models to preserve correlation structures.

A regression model was used to generate multiple imputed data sets (or ‘draws’), in which missing values were predicted drawing on predictive covariates. These included age, sex, baseline blister severity and baseline Karnofsky score as predictors only and outcome measures (at each time point) and costs as predictors and imputed variables. Each draw provided a complete data set that reflected the distributions and correlations between variables. Predictive mean matching was used to enhance the plausibility and robustness of imputed values, as normality could not be assumed. The model used fully conditional Markov chain Monte Carlo methods (multiple imputation by chained equations), which are appropriate when missing and correlated data occur in more than one variable. Each draw was analysed independently and the estimates obtained were pooled to generate mean and variance estimates of costs and QALYs using Rubin’s rule, a method that captures within and between variances for imputed samples. 46 To minimise the information loss of finite imputation sampling, 50 draws were taken. 46 This resulted in a loss of efficiency relative to infinite sampling of < 0.5% in all imputed values. The distribution of imputed and observed values was compared to establish that imputation did not introduce bias into subsequent estimation.

Incremental analysis

The goal of economic analysis is to inform decision-makers about the incremental costs and benefits of change; when expressed together in the cost/QALY metric this facilitates comparison with other technology choices. The analysis reports costs, QALYs and cost per QALY comparing doxycycline-initiated therapy with prednisolone-initiated therapy for patients with BP.

Bivariate regression using seemingly unrelated regression equations was used to model incremental changes in costs and QALYs. This method respects the correlation of costs and outcomes within the data and allows adjustment for a set of covariates that can be explored. 47 Baseline quality-of-life scores were included within all QALY estimation models to allow for potential baseline imbalances. 38 The incremental cost-effectiveness ratio (ICER) was estimated as the difference in mean total costs between treatments divided by the difference in mean total QALYs. Value for money is determined by comparing the ICER with a threshold value, typically the National Institute for Health and Care Excellence threshold of £20,000–30,000 per QALY for UK studies. This represents the willingness to pay for an additional QALY and lower values than the threshold could be considered cost-effective for use in the NHS. The net monetary benefit (NMB) of changing treatment was also reported as a recalculation of the ICER at a range of thresholds of willingness to pay for an additional QALY. The NMB succinctly describes the resource gain (or loss) when investing in a new treatment when resources can be used elsewhere at the same threshold. 48

Uncertainty

A range of decisions are made in the construction of cost-effectiveness models, for example which costs to include, whether or not to adjust for covariates and whether or not to impute missing values. These options are presented and the most plausible assumptions form the base-case analysis; assumptions are explored using a range of supportive sensitivity analyses. The analyses carried out in this study are described in Table 51.

The imputation data set provides the most plausible base-case analyses, but a complete-case analysis and unadjusted analysis provide useful sensitivity analyses. No method of analysis can provide protection from bias if the assumption that data are MAR does not hold (adequately). Joint distributions of costs and outcomes were generated using the (non-parametric) bootstrap method, with replicates used to populate the cost-effectiveness plane and generate cost-effectiveness acceptability curves. The cost-effectiveness acceptability curve compares the likelihood that treatments are cost-effective as the willingness-to-pay threshold varies. 49 Bootstrapping jointly resamples costs and outcomes from the original data with replacement (maintaining the sample correlation structure) to create a new bootstrap sample from which a change in costs and QALYs is estimated. Using bias-corrected non-parametric bootstrapping, 5000 bootstraps were taken per model evaluated. Means estimates are reported with 95% CIs.

Completeness of data

Of the 253 patients included in the primary analysis of effectiveness, 220 patients (87%) had at least one EQ-5D record during follow-up and were included in the economic analysis (Table 52). Subsequent discussion of the completeness of the data focuses on the 220 included patients as most relevant when considering the extent of imputation. In total, 164 patients (75%) had complete EQ-5D assessments for all time periods. Patients who died were subsequently scored zero on visits that followed for both cost and EQ-5D score and are included as observed data. There was a pattern of decreasing completeness as follow-up proceeded. Cost data were complete for 191 patients (87%). It was not possible to explore completeness of health-care costs by follow-up visit as patients could use diaries to complete missing data at a later follow-up visit and the study drug report covered the entire follow-up period. When considering both utilities and resource use, complete information was available for 143 patients (65%). Completeness of data was similar when comparing treatment arms (see Table 52).

Thirty-four patients died during the trial period, 14 (10.6%) in the doxycycline arm and 20 (16.5%) in the prednisolone arm. Missing values were imputed to provide a base-case analysis including all 220 patients.

Complete-case estimates

To describe the observed data, findings in this section are reported unadjusted for patient-level covariates. Models in this section (Tables 53–55) are adjusted as shown in Table 51.

Mean EQ-5D scores by treatment group are reported in Table 53. There was a small non-significant difference in health status at baseline (lower in doxycycline-initiated patients). Over the 1-year follow-up period there were no significant differences in QALYs when comparing treatments.

Resource use (in natural units) by treatment group is reported in Table 54. Predictably, study drug use for prednisolone and doxycycline was significantly higher in patients allocated to each treatment. However, there was significantly subsequent crossover to the alternative study drug in patients with poor outcomes. Of patients starting on prednisolone, 12.6% subsequently received at least one prescription of doxycycline; of patients starting on doxycycline, 57.8% subsequently received at least one prescription of prednisolone. Patients commonly used four topical corticosteroids; the overall use of any topical steroid during the trial was very similar for prednisolone (76.8%) and doxycycline (72.5%) patients. However, the level of use (and steroid potency) was notably lower in the prednisolone group (average prescribed amount of all topical steroids among users: prednisolone 121 g vs. doxycycline 277 g). Similar proportions of patients received the immunosuppressant azathioprine (prednisolone 4.2% vs. doxycycline 8.8%) although again the quantity of prescribed use was far lower in the prednisolone group (average prescribed amount among users: prednisolone 2.9 g vs. doxycycline 12.4 g).

Although resource-use comparisons (attributed to BP) were generally not statistically significant, there was a pattern of greater care received by doxycycline patients, consistent with a lower level of clinical effectiveness reflected (by design) in the primary outcome.

Patterns of resource use were costed using national reference values (see Table 50) and are reported in Table 55. Although costs for patients starting on doxycycline treatment appeared greater over 1 year, the increase was not statistically significant.

Cost-effectiveness analysis

The joint distribution of incremental cost and outcome for the base-case analysis is shown graphically in Figure 8. Patients allocated initially to doxycycline experienced a slightly lower average quality of life (–0.024 QALYs, 95% CI –0.088 to 0.041) while tending to incur higher average health costs (£959, 95% CI –£24 to £1941), although neither finding was statistically significant (Table 56). These findings are consistent with the results of the clinical trial, which by design demonstrated a compromise between reduced effectiveness and increased safety for doxycycline. However, there remains the suggestion that doxycycline-initiated care may result in greater medium-term care costs, as shown by the 95% CI in Figure 8.

FIGURE 8.

Cost-effectiveness plane: base-case analysis [cost per QALY (£), 2013].

The likelihood of being cost-effective at different thresholds is shown in Figure 9.

FIGURE 9.

Cost-effectiveness acceptability curve: base-case analysis – doxycycline-initiated therapy.

Using a willingness-to-pay criterion of < £20,000 per QALY gained, there is a 4.6% probability that doxycycline-initiated treatment is cost-effective; conversely, there is a 95.4% chance that prednisolone-initiated therapy is cost-effective. [Note that these (one-sided) model probabilities should not be compared with inferential findings from (two-sided) statistical tests reported in Chapter 3.]

The NMB associated with doxycycline-initiated treatment was found to be negative and diminished with willingness to pay (Figure 10). Using a willingness-to-pay criterion of < £20,000 per QALY gained, the NMB associated with doxycycline-initiated therapy was negative but failed to reach statistical significance (–£1432, 95% CI –£3094 to £230). Although statistically imprecise, analysis of the NMB suggests that resources displaced in the NHS by doxycycline-initiated therapy would be greater than the value of benefit gained.

FIGURE 10.

Relationship between net monetary benefit and willingness to pay for doxycycline-initiated treatment.

Sensitivity and subgroup analyses

Comparing mean costs and QALY estimates using different modelling assumptions supports the base-case finding (see Table 56). The qualitative similarity of NMB estimates comparing imputed and complete-case analysis, covariate adjustment and range of costs included supports the validity of the imputation process and assumptions.

There was no interaction between treatment effect and trial stratifying variables, except in the case of blister severity. Patients recruited with severe blisters demonstrated a different cost and outcome pattern from patients with mild or moderate blisters, as shown in Table 56 and Figures 9 and 11. For patients presenting with mild or moderate blisters, differences in costs and QALYs are very small and thus the costs and outcomes can be thought to be similar. For patients presenting with severe blisters, doxycycline-initiated treatment presents greater costs (£2558, 95% CI –£82 to £5198) and poorer quality of life (–0.090 QALYs, 95% CI –0.222 to 0.042), which together make this strategy appear a poor investment (NMB –£4361, 95% CI –£8283 to –£439) using a willingness-to-pay criterion of < £20,000 per QALY gained.

FIGURE 11.

Cost-effectiveness plane: subgroup analysis [cost per QALY (£), 2013].

Other quality-of-life measures

Additionally, EQ VAS and DLQI scores were used to derive AUC scores over the course of the 1-year follow-up period (Table 57).

The EQ VAS is scored from 1 to 100; equivalent QALY scores are obtained by dividing by 100, although the EQ VAS is not recommended for QALY estimation within trials as values are self-rated rather than societal. As with the EQ-5D estimates, there was no significant difference between treatments. DLQI estimates were significantly higher (denoting a poor outcome for doxycycline-initiated therapy). Being a disease-specific quality-of-life measure, the DLQI is potentially more sensitive to change than a generic measure; nonetheless, the changes correspond to an average of 1 point on a 30-point scale and are of uncertain clinical importance.

Strengths

This was a large, multicentre, multinational trial of 253 evaluable participants that had the power to detect clinically important differences. The trial had good external validity as patients were recruited from a large number of hospitals all over the UK and Germany, covering those with all types of disease severity, and from a range of hospital settings and socioeconomic areas. Additionally, patients were not excluded on the grounds of comorbidities and any patient with active BP was eligible. With the exception of incapacity to consent, patients were excluded only on safety grounds. Typically, BP patients are elderly and present with several comorbidities and so inclusion of such a population as in this study was an important issue.

This trial was designed to reflect clinical practice as far as possible, making the results useful in the real-life setting of a dermatology clinic. We assessed the degree to which the study was pragmatic using the pragmatic–explanatory continuum indicator summary tool25 and concluded that the trial was more pragmatic than explanatory.

There is no evidence of selection bias as the two groups were well matched for all baseline characteristics. Allocation of treatment was concealed by blinding the investigator until the 6-week blister count had been carried out (primary effectiveness outcome). Treatment allocation was revealed only once the blister count was entered into the trial database, thereby minimising the potential for performance and information bias. The investigator was unblinded for the remaining assessments for that participant, for example the primary safety outcome (adverse events over the full 52 weeks), as the investigator was often also the treating physician for this trial and it was not feasible to have a second (blinded) outcome assessor to collect adverse event data at every recruiting site. A relatively objective measure (adverse events graded according to the Common Terminology Criteria for Adverse Events) was chosen to minimise bias for the primary safety outcome measure.

As the investigator was blinded for the first 6 weeks of treatment (until the primary effectiveness outcome had been assessed), it was difficult for local investigators to comment on the relatedness of adverse events. Therefore, an independent adjudicator judged the relatedness of any adverse events to trial medication so that the blinding of the investigator for the primary effectiveness outcome could be maintained. In addition, all severe, life-threatening and fatal adverse events were independently assessed for relatedness to the trial medication to ensure that relatedness was being properly attributed.

The primary effectiveness results were similar in both the ITT population and the PP population. This is particularly important in this trial because of the non-inferiority comparisons, in which dependence on the ITT population alone risks falsely concluding a lack of difference between groups when there are significant amounts of missing data. 50 Similarly, findings from the cost-effectiveness analysis were robust to a range of sensitivity analyses, providing confidence in the modelling assumptions.

To minimise selective reporting bias, the trial was registered before recruitment started, the protocol published and the statistical analysis plan made freely available on the trial website [see https://ctsu.nottingham.ac.uk/ts0614/summary.asp (accessed 12 November 2015)]. All planned outcomes and analyses are reported here and any post-hoc analyses have been clearly indicated as such.

Limitations

This trial was a comparison between initiating treatment with doxycycline and initiating treatment with prednisolone rather than an explanatory trial evaluating the pharmacological effects of these treatments. It is very likely therefore that some of the treatment response in patients who switched from initial doxycycline treatment to prednisolone can be attributed to prednisolone. Despite treatment switches, there were still fewer medically serious side effects in those starting on doxycycline than in those starting on prednisolone over the course of 1 year. The strategy of initiating treatment with doxycycline and thus reducing the overall dose of corticosteroids is a safer treatment approach than starting on prednisolone over the course of 1 year.

To reflect clinical practice in which topical corticosteroids are often used early on in addition to systemic treatments,14 the use of potent topical corticosteroids applied only to blisters was permitted for the first 3 weeks. A washout period from week 3 to week 6 allowed a clearer assessment of the effectiveness of doxycycline at 6 weeks. It is unlikely therefore that topical corticosteroid use only in the first 3 weeks of treatment was responsible for the blister control observed at 6 weeks, given such a substantial 3-week washout period that followed. 51 More participants in the doxycycline group had to use potent topical corticosteroids during the washout period, although the non-inferiority findings for the primary effectiveness outcomes were robust to excluding such participants in a sensitivity analysis. Localised use of topical corticosteroids was permitted after week 6 when needed to reflect clinical practice and so assessments at all other time points could be affected to some degree by such usage.

The prespecified non-inferiority margin for effectiveness was relatively wide (with an upper CI of 37%), resulting in an effectiveness estimate for doxycycline that was both non-inferior and inferior compared with the prednisolone strategy, a paradox that was anticipated beforehand. 52 When surveyed prior to the trial starting (see Appendix 1), there was clearly a willingness among UK dermatologists to accept a considerable reduction in short-term effectiveness of doxycycline for a treatment that did not result in such severe long-term side effects.

There was a relatively high dropout rate in this trial, reflecting the multiple morbidities and frailty of a mainly elderly study population, with the potential for differential reporting bias. A low dropout rate of no more than 5% was predicted for the week 6 visit given the morbidity of the disease, but the dropout rate was higher than anticipated (although it should be noted that 2.8% dropped out as a result of death). There was a gradual loss over time resulting in a higher dropout rate than the 20% at 52 weeks allowed for in the sample size calculation.

It is unlikely that the success of both treatments at 6 weeks is a result of regression to the mean, reflecting the natural progression of the disease, because BP is a progressive disease. 53,54 Additionally, there is evidence that very low doses of prednisolone (0.3 mg/kg/day) are not effective,55 suggesting that there is little evidence to support a placebo effect. Even if one postulates a placebo/natural resolution response of around 20% at 6 weeks, the study response rates for effectiveness at 6 weeks (74.4% and 92.3% for doxycycline and prednisolone, respectively) are much higher and are likely to represent a true therapeutic response. Including a third placebo arm would have been unethical and would have probably failed to recruit.

As we collected data only on adverse events that were deemed to be possibly, probably or definitely related to the study treatment, the side effects of which are well known, it is possible that some degree of attribution bias could have occurred, for example sepsis occurring in a patient taking oral prednisolone might have been more likely to have been attributed to the drug rather than being recorded as a natural event. We think that this is unlikely because it was expected that the rate of severe adverse effects would be higher in the prednisolone group, based on knowledge from other trials,18 and because the number of deaths (regardless of attribution) was higher in the prednisolone group. Mild and moderate treatment-related adverse events for both drugs were fairly similar between groups, with grade 3 and above adverse events more common in the prednisolone group (see Table 32), arguing against a differential bias towards attributing more events to prednisolone. Determining the true attribution of adverse events to drugs is difficult in clinical practice and all unclear grade 3, 4 and 5 events in this trial were reviewed by a senior independent dermatologist who had access to the medical backgrounds of study participants and the chronology of events. In the absence of data on all adverse events, which was deemed unnecessary by the regulatory authorities for two such study drugs with well-established adverse effect profiles, some degree of attribution bias could have occurred in this study, but in a way that reflects how such attribution would be judged in everyday clinical practice, as per the pragmatic design of the study.

The generalisability of these results may be reduced by not including patients with dementia. Such patients were not included, as ethics committees insist that people with dementia, the majority of whom are unable to give informed consent, should not be recruited into trials if a trial can be delivered without involving such patients. However, in a disease that affects mainly the elderly, a large part of the patient population will be affected by dementia. It is unclear, though, whether or not having dementia would be likely to affect the treatment comparisons described in this study, as issues such as adherence are likely to have been similar in both groups. It is also worth noting that large numbers of potentially eligible new cases of BP for our study ended up not being eligible because their GP had already started them on oral prednisolone.

Although relapse rates were assessed, the results should be interpreted with some caution. To make it practical for clinicians in busy clinics to be able to include patients in this trial, they had to decide only whether or not the patient had < 4 significant blisters at each blister count (other than baseline), rather than undertake the time-consuming process of carrying out a full blister count. This means that it is possible that a patient who was considered a treatment success because of the presence of only two significant blisters could then ‘relapse’ by having just four at the next visit. Other patients may go from being almost blister free to having ≥ 30 blisters, which might be considered a more ‘true’ relapse. However, our relapse definition does give some indication of relapse rates and the direction is reflective of the other effectiveness outcomes in the trial.

All economic analyses in modelling the bivariate distribution of costs and QALYs involve assumptions. For example, judgements are made about the base-case model, the estimation method, adjustment for covariates, attribution of resource use and unit costs applied, as well as the quality-of-life measure used and societal weighting applied. As levels of missing data increase, complete-case analyses become progressively less satisfactory, whereas multiple imputation inevitably requires strong assumptions about data being MAR, which are only partially testable. Careful consideration of modelling issues and use of sensitivity analyses, exploring assumptions, provide some indication of the robustness of findings.

Patient and public involvement was not as strong as it could have been in this study. We failed to identify patients with BP or their carers who were willing to join our study team, despite repeated requests across study sites. Most patients were elderly and infirm and not able to travel to team meetings and no patient support group exists for BP that we could tap into. Perhaps we did not try hard enough. We did manage to find a patient and a carer who kindly joined the independent Trial Steering Group, one of whom (Penny Standen) was able to advise the team on the interpretation of the study. She also helped to write the plain English summary once the study was completed. Further qualitative work with patients to explore their views on the trade-off between short-term blister control and long-term safety gains would be worthwhile.

Independent members

• Professor Jonathan Barker (chairperson), St John’s Institute of Dermatology, Guy’s Hospital, London, UK.

• Professor Pascal Joly (clinical expert), Hôpital Charles Nicolle, Rouen, France.

• Dr Jonathan Leonard (clinical expert), St Mary’s Hospital, London, UK.

• Ms Helena Haywood (dermatology nurse), Amersham Hospital, Amersham, UK.

Patient representatives

• Penny Standen.

• Brian Lockwood.

Non-independent members

• Professor Hywel Williams (chief investigator), Centre of Evidence Based Dermatology, University of Nottingham, Nottingham, UK.

• Professor Fenella Wojnarowska (lead clinician), Nuffield Department of Clinical Medicine, University of Oxford, Oxford, UK.

• Dr Gudula Kirtschig (clinical expert and co-ordinator for German sites), Centre of Evidence Based Dermatology, University of Nottingham, Nottingham, UK.

• Professor Andrew Nunn (senior trial statistician), MRC Clinical Trials Unit at University College London, London, UK.

• Daniel Bratton, Sunita Rehal, Tom Godec (trial statisticians), MRC Clinical Trials Unit at University College London, London, UK.

• Dr Karen Harman (principal investigator representative), University Hospitals Leicester, Dermatology Department, Leicester Royal Infirmary, Leicester, UK.

• Dr Phillip Hampton (principal investigator representative), Royal Victoria Infirmary, Newcastle, UK.

• Dr Joanne Chalmers (research fellow), Centre of Evidence Based Dermatology, University of Nottingham, Nottingham, UK.

The current trial manager was also a non-independent member of the Trial Steering Committee.