Clinical and cost-effectiveness of spironolactone in treating persistent facial acne in women: SAFA double-blinded RCT

Inclusion criteria

• Women aged 18 years or over.

• Facial acne vulgaris, symptoms present since at least 6 months.

• Acne of sufficient severity to warrant treatment with oral antibiotics, as judged by the trial clinician; women with an Investigator’s Global Assessment (IGA) ≥ 2 were eligible to participate in the trial.

• Women of childbearing potential at risk of pregnancy must be willing to use their usual hormonal or barrier method of contraception for the first 6 months of the trial [while taking the trial investigational medicinal product (IMP)] and for at least 4 weeks (approximately one menstrual cycle) afterwards.

• Willing to be randomised to either treatment.

• Willing and able to give informed consent.

• Sufficient English to carry out primary care Acne-QoL.

Exclusion criteria

• Acne grades 0–1 using IGA (i.e. clear or almost clear).

• Has ever taken spironolactone.

• Oral antibiotic treatment (lasting longer than 1 week) for acne within the past month.

• Oral isotretinoin treatment within the past 6 months.

• Started, stopped or changed long-term (lasting more than 2 weeks) hormonal contraception, co-cyprindiol or other hormonal treatment within the past 3 months.

• Planning to start, stop or change long-term (lasting more than 2 weeks) hormonal contraception, co-cyprindiol or other hormonal treatment within the next 3 months.

• Pregnant/breastfeeding.

• Intending to become pregnant in the next 6 months.

• Contraindicated to spironolactone:

  • currently taking potassium-sparing diuretic, angiotensin-converting enzyme inhibitor, angiotensin II receptor blocker or digoxin

  • hereditary problems of galactose intolerance, lactase deficiency or glucose–galactose malabsorption (as the spironolactone and placebo tablets contain lactose)

  • androgen-secreting adrenal or ovarian tumour

  • Cushing syndrome

  • congenital adrenal hyperplasia

  • estimated glomerular filtration rate below 60 ml/minute/1.73 m²

  • serum potassium level above the upper limit of reference range for the laboratory processing the sample.

Primary care

Secondary care

Trial staff at recruitment sites screened referral letters and opportunistically invited patients in outpatient clinics. Where there was an available database and patients had given permission to be contacted about research, potential patients were contacted by the local site trial team by e-mail or mail-out.

Community advertising

The trial was advertised through media local to recruiting trial sites and advertising in appropriate institutions, such as posters in universities, pharmacies and hospitals, directing potential patients to the trial-specific website.

Social media advertising

Baseline assessment

After initial telephone contact from site staff to carry out initial eligibility check, potential participants were invited to attend a baseline appointment (face-to-face) at their local trial site. Baseline appointment included:

• discussion of participant information sheet and written informed consent

• clinician assessed the eligibility

• pregnancy test

• blood test for renal function and potassium level

• height, weight, waist circumference, polycystic ovary syndrome (PCOS) characteristics, blood pressure

• acne medication and oral medication use

• Investigator’s Global Assessment (IGA) to assess acne severity

• the site team also took photos on an instant camera or the participant took selfies of their acne in order to compare progress at the follow-up visits

• questionnaires as outlined in schedule of observations (see Table 1).

Participants were offered a £20 voucher at baseline appointment to thank them for their time and a £10 voucher at subsequent appointments (weeks 6 and 12).

Follow-up

As shown in Figure 1, initial trial design included follow-up based on week-6 and week-12 visits which were initially intended to be held face-to-face in secondary care clinics, in order to discuss dose adjustment (see Interventions below), carry out IGA and reiterate contraceptive counselling. However, following redesign of trial delivery during the COVID-19 pandemic, sites were given the option of holding follow-up visits either remotely or face-to-face (if local trust guidance permitted). Additionally, (1) participants had the option to send digital photos of their acne for the clinician to assess during the consultation and grade the participant’s acne using IGA and (2) site teams had the option to post the trial medication directly to the participants. These modifications ensured that participants who had been recruited to the trial were not lost to follow-up and also allowed for flexibility during subsequent lockdowns.

Unblinded follow-up period (questionnaire)

Participants were unblinded at week 24 (if returning the week 24 questionnaire) or at week 28 (if the week 24 questionnaire was not returned) and then entered an unblinded follow-up period. Originally, all participants were to be followed up for 52 weeks and would be sent the final follow-up questionnaire 52 weeks after baseline. However, in order to deliver the trial on time, the unblinded follow-up period was truncated so that, once the last participant received the week 24 questionnaire, all remaining participants who had not already received the final follow-up questionnaire were sent it at this point.

Primary outcome

The primary outcome was comparison of mean Acne-QoL symptom subscale score between groups at 12 weeks, adjusted for baseline variables. Of the many different participant-reported outcome measures available, the Acne-QoL is the most extensively validated. 29,30 The Acne-QoL contains 19 questions with 7 response categories, each referring to the past week, organised into 4 domains (self-perception, role-social, role-emotional, acne symptoms).

Effectiveness of acne treatments is usually judged clinically at 8–12 weeks so primary outcome at 12 weeks was chosen. A survey carried out prior to the trial (reported in the published protocol paper1) suggested that people with persistent acne may not be willing to stay on a trial treatment of unknown benefit for longer than 12 weeks, which is why this time point was chosen, with blinded treatment continuing to 24 weeks to assess medium term outcomes and continuing to follow-up participants beyond this to investigate longer-term outcomes.

No changes were made to trial outcomes after the trial commenced.

Secondary outcomes

• Acne-QoL symptom subscale score at 6 weeks and at end of treatment (24 weeks). 29,30

• Acne-QoL other subscales (self-perception, role-emotional and role-social), and total score, at 6, 12 and 24 weeks.

• Participant self-assessed overall improvement at 6, 12 and 24 weeks recorded on a six-point Likert scale with photographs taken at the baseline visit to aid recall, as was carried out in the previous HTA-funded trial on acne. 31

• FDA IGA of acne at 6 and 12 weeks (5-point scale: 0 – Clear; 1 – Almost clear; 2 – Mild; 3 – Moderate; 4 – Severe). 32

• PGA at 6, 12 and 24 weeks (5-point scale same as IGA but written in plain language).

• Participant satisfaction with trial treatment (asked prior to revealing treatment allocation at 24 weeks).

• Health-related quality of life using EQ-5D-5L at 6, 12 and 24 weeks.

• Cost at 6, 12 and 24 weeks (participant report) and cost-effectiveness over 24 weeks.

Other outcomes

• Acne-QoL symptom subscale score at up to 52 weeks.

• Acne-QoL other subscales (self-perception, role-emotional and role-social), and total score, at up to 52 weeks.

• Participant self-assessed overall improvement at up to 52 weeks recorded on a six-point Likert scale.

• Participant’s Global Assessment at up to 52 weeks.

• ARs of special interest.

• Use of other oral acne treatment (e.g. antibiotics, isotretinoin) (participant report).

• Health-related quality of life using EQ-5D-5L up to 52 weeks.

• Cost up to 52 weeks (participant report).

• Participant experiences during the trial (qualitative interviews).

Original sample size calculation

Based on comparison of mean Acne-QoL scores between groups at 12 weeks, power 90%, alpha 0.05 and seeking a difference between groups of two points on the symptom subscale [standard deviation (SD) 5.8, effect size 0.35], we calculated that 346 participants would be needed. Allowing for 20% loss to follow-up gave a total target of 434 participants (217 per group). 30,33

Revised sample size

No assumption was made initially regarding anticipated correlation between baseline and 12 weeks Acne-QoL subscale (r = 0). Following preliminary data, an adjustment was made on the basis of correlation between baseline and 12-week Acne-QoL subscale (r = 0.293). A deflation factor of 1-ρ²,34 allowing a reduction in the required sample size (including allowing for 20% loss to follow-up) to 398 participants (199 per group). This revision was approved in consultation with the Trial Steering Committee (TSC), Data Monitoring Committee, REC and the Funder.

Subgroup analysis

Table 18 describes the subgroup analysis that was performed to compare results in certain pre-specified populations. Interaction terms were included in the model; however age (categorised as below 25 years and 25 years and over)36 and treatment were the only significant interaction terms. The coefficient and 95% CI for this interaction was: 4.2 (1.3 to 7.1). The difference in Acne-QoL symptom subscale score in the 25 years and above population was 2.42 (95% CI 1.00 to 3.84), which was statistically significant and reached the desired two-point mean difference between groups in favour of spironolactone. The population of those below 25 years was considerably smaller (44 participants) and therefore it is difficult to draw definitive conclusions, although the difference of –0.87 (95% CI –3.67 to 1.92), which does not favour the spironolactone group, suggests that spironolactone may be less effective in younger women.

Summary statistics are presented in Table 19 for the primary outcome pre and post COVID as well as for ethnicity. The mean Acne-QoL symptom subscale score at 12 weeks pre and post COVID was similar for both groups. Those who identified as non-white had on average a better score in the placebo group; however this population was small (n = 22), therefore it is difficult to draw meaningful conclusions.

Complier-average causal effect analysis

The results of the CACE analysis performed on self-reported treatment adherence at 24 weeks can be seen in Table 20. ‘Compliance’ was defined as participants reporting that they had taken either 50%, 80% or 100% of the prescribed medication over the 12- to 24-week period. Compliance was only considered over the 12- to 24-week period due to the collection of these data before this time point being poorly reported in part due to the COVID-19 pandemic, as pill count at 6 and 12 weeks was initially planned but often not possible. When considering 80% as the threshold for compliance, we observed 74% of women complied. The adjusted mean difference between groups was 5.13 (95% CI 3.17 to 7.08), suggesting a greater treatment effect amongst women who took 80% or more of the prescribed medication. For all thresholds of compliance (50%, 80% and 100%), the proportion of participants achieving ‘compliance’ were similar in both spironolactone and placebo groups, suggesting that spironolactone was well-tolerated.

Reported serious adverse events/serious adverse reactions and suspected unexpected serious adverse reactions

Table 25 details the SAEs reported. There were nine SAEs reported in total which were split evenly across the groups. There was one grade 2, six grade 3s, one grade 4 and one grade 5. None of the SAEs were related to treatment.

Pregnancies

Seven participants became pregnant while in the trial. The pregnancies were recorded at different sites and different contraception methods were used. Contraception was discussed with each participant at their baseline visit. Pregnant participants were withdrawn from the trial as soon as site staff became aware of the pregnancy (level 2 withdrawal). Pregnant participants were approached for consent for pregnancy follow-up. Three participants did not consent to the outcome of the pregnancy being followed up. Four participants consented to pregnancy outcome follow-up. All pregnancies were reported to Medicines and Healthcare products Regulatory Agency (MHRA) as part of the Development Safety Update Reports over the course of the trial following recommendation from the DMEC group, although the pregnancies did not meet the requirements to trigger an urgent safety measure.

Overview of economic analysis

The base (or reference) case within-trial economic analysis uses individual participant level data collected over 24 weeks from the SAFA trial to compare the cost-effectiveness of spironolactone used alongside routine topical treatment compared to no active treatment (placebo) alongside routine topical treatment for moderate to severe persistent adult female acne. This comparison was chosen because it reflected the clinical question being addressed and data being collected within the SAFA trial. There are potential questions with alternative comparisons that may be of more interest to clinicians. For instance, in practice women with persistent acne are likely to be given an active treatment on top of routine topical treatment. These women would most likely receive an oral antibiotic (lymecycline or doxycycline) as per NICE guidance12 suggesting a systemic antibiotic alongside topicals for more moderate to severe disease. We explore this question, which is more reflective of clinical practice, in sensitivity analysis but there are limitations to such an analysis given the data we have available.

The base case analysis undertakes a cost–utility analysis from an NHS and Personal Social Services (PSS) perspective, where outcome is defined in terms of incremental difference in the mean number of QALYs. A secondary analysis is undertaken using a cost-effectiveness analysis approach. This was chosen as the secondary analysis due to uncertainties about what an incremental cost per point change on the Acne-QoL symptom subscale means or how much a decision maker would value this.

Published guidelines for the economic evaluation of healthcare interventions were followed as appropriate. 37,38 NICE published updated guidance after analysis for this trial had started. 39–41 While much of the guidance remains unchanged, the preferred mapping function to be used in reference case analyses for the EQ-5D-5L was updated from that published by van Hout et al. 42 to that of Alava et al. 43,44 We have adhered to the earlier guidelines as the analysis started prior to the change in guidelines.

A copy of the health economic analysis plan that was written and reviewed prior to the trial database being locked can be found in the project documents.

Resource use

In keeping with the trial being conducted in the UK, where the NHS provides publicly funded health care, the analysis takes a health service perspective which is also in line with the NICE reference case. 45 Resource use was collected at baseline (for the 6 weeks prior to the baseline visit), week 6, week 12 and week 24 via case report forms and participant questionnaires. PSS resource use was not captured explicitly, as it was anticipated that these types of services would be unlikely to be incurred as a result of acne. Costs incurred by the participant and/or a family member or friend were also collected, informed by input with the patient and public involvement (PPI) members of the study team. This included personal out-of-pocket costs (including complementary therapists, non-prescribed medication, travel costs to healthcare appointments, parking costs at healthcare appointments, cosmetic and skin care products, and other costs), impact on employment (productivity costs) for the woman or a family member/friend, and support services outside of official services.

Valuing costs

The cost of the intervention was estimated using data collected within the SAFA trial and costed using published unit costs for spironolactone in the prescription cost analysis (PCA). 46 In the base case analysis, the intervention is costed as per the schedule shown in Table 26, where the intervention is delivered in primary care. The delivery of the intervention in a primary care setting (i.e. GP practice) reflects how the intervention would be delivered in clinical practice. In the trial, the medication was either posted directly to participants’ homes or participants collected their medication at secondary care follow-up visits (if face-to-face). In clinical practice, patients would collect their medication via repeat prescription from their GP surgery or pharmacy. As a result, postage costs were not included in the economic evaluation.

Resource use relevant to the NHS perspective was valued using UK unit costs (Great British pounds) for the most current price year available at the start of the analysis (2021). Unit costs were identified from published sources and are clearly reported in the results section.

A mean cost (SD) per participant per intervention group was estimated for 24 weeks.

Outcomes

The primary economic outcome measure was incremental cost per QALYs over 24 weeks. The EQ-5D-5L is a generic preference-based measure of health-related quality of life instrument with five dimensions (mobility, pain/discomfort, usual activities, self-care, anxiety/depression) with five levels ranging from ‘no problems’ through to ‘unable to’ or ‘extreme problems’. 47,48 Responses were converted to utility scores using the EQ-5D-5L Crosswalk UK preference weights, as this was in line with recommendations at the point analysis started, where utility ranges from –0.594 to 1. 42 Utility values were used to estimate QALYs over 24 weeks, using both linear interpolation and area under the curve analysis with and without baseline adjustment. 48

Secondary analysis reports a cost-effectiveness analysis (CEA) where the incremental cost per unit change on the Acne-QoL symptom subscale score is estimated (i.e. summing items 15–19, see the Health Economics Analysis Plan for details). The Acne-QoL symptom subscale consists of five questions, each with answers on a scale of 0–6 (‘extensive, a whole lot, a lot, a moderate amount, some, very few, none’ for items 15–17 and ‘extremely, very much, quite a bit, a good bit, somewhat, a little bit, and not at all’ for items 18–19) with total scores ranging from 0 to 30 [Acne-Specific Quality of Life Questionnaire (Acne-QoL) Manual and Interpretation Guide]. 49 We undertook a cost-effectiveness analysis despite it not being clear what decision makers would be willing to pay for a unit change on the Acne-QoL symptom subscale because Klassen et al. 50 found that clinical measures were more responsive to change than the generic measures (shown by larger effect sizes) and concluded that it is desirable to combine generic preference-based measures with the use of disease-specific measures for acne.

Economic evaluation at 24 weeks

A full economic evaluation was only planned if spironolactone was found to be clinically effective. The economic base case analysis incorporates all randomised participants with complete cost and outcome data available. Given that the time horizon was 24 weeks, costs and benefits are not discounted. 45 The main base case analysis was a cost–utility analysis to estimate the incremental cost per QALY to enable comparison with the cost-effectiveness of other interventions. A cost-effectiveness threshold (ʎ) of £30,000 (£20,000) per QALY is used in line with NICE guidance. 45

Mean (SD) resource use and mean (SD) cost per participant is estimated per randomised group. The incremental cost (95% CI) between groups is estimated adjusted for randomisation stratification variables [centre, baseline severity (IGA < 3 vs. 3 or more)], baseline variables as in the primary analysis [including baseline Acne-QoL symptom subscale score and use of topical treatments (Y/N)] and baseline outcome variable (baseline EQ-5D) and unadjusted. Given that baseline characteristics and baseline outcome measurements are often predictive of total costs and QALYs, it is usual to give more weight to adjusted analyses than unadjusted analyses. 48,51 Mean (SD) utility and mean (SD) QALYs per participant per randomised group are presented along with the incremental QALYs (95% CI) between groups adjusted and unadjusted. The base case cost–utility analysis and secondary cost-effectiveness analysis are undertaken using a regression-based approach, seemingly unrelated regression equations [sureg command was used in Stata 17 (StataCorp LP, College Station, TX)], which allows covariate adjustment without requiring covariates for cost and effectiveness be the same. 51

To determine the level of sampling uncertainty surrounding the mean incremental cost-effectiveness ratios (ICERs), non-parametric bootstrapping was conducted to generate 10,000 estimates of incremental costs and benefits. From this, cost-effectiveness acceptability curves (CEACs) were generated to show the probability that the intervention is cost-effective at different values of willingness to pay. Stata MP version 17 was used to conduct the analysis.

A single subgroup analysis was undertaken for age (categorised as below 25 years and 25 years and over) as the clinical analysis found age to be a significant interaction term (see Subgroup analysis).

Sensitivity analysis

The following sensitivity analyses were undertaken to explore key uncertainties around important parameters in the economic evaluation:

Sensitivity analysis 1 (SA1): The base case economic evaluation over 24 weeks did not impute missing data, instead a complete case analysis (CCA) was undertaken. To evaluate the impact of missing data on the cost-effectiveness estimates, multiple imputation was employed, assuming that the data was missing at random (MAR). The mi impute chained [pmm, knn(10)] command was used in Stata 17, which imputes missing data using chained equations (MICE) and predictive mean matching (to ensure only plausible values are imputed). The model included the outcome measure (cost or utility), baseline value of the outcome (cost or utility), randomisation group, and all covariates {randomisation stratification variables [centre, baseline severity (IGA < 3 vs. 3 or more)] and baseline variables [including baseline Acne-QoL symptom subscale score, use of topical treatments (Y/N)]} included in the analysis model. To handle the missing cost and outcome data to assess the impact on the conclusions reached,52 a further sensitivity analysis (SA1a) was undertaken to test the assumption of MAR. A common concern in economic evaluations, particularly if data are missing to different degrees in different treatments groups, is that the probability of being missing may depend on the underlying unobserved values, for example the chances of completing a QoL questionnaire may depend on the patients (unobserved) QoL status: the lower their QoL, the less likely they are to complete a questionnaire. This is referred to as ‘missing not at random’ (MNAR). 53 Therefore, following review of the patterns of missingness, to test for the impact of possible MNAR on the results, we use a pattern-mixture model. 53 This model involves modifying the multiply imputed data to reflect possible departures from the MAR assumption, by multiplying the imputed values by a plausible value, and reanalysing the results. We tested the following scenarios:

• missing EQ-5D data were 5% lower than imputed in both groups of the study

• missing EQ-5D data were 10% lower than imputed in both groups of the study

• missing cost data were 5% higher than that imputed in both groups of the study

• missing cost data were 10% higher than that imputed in both groups of the study

• missing EQ-5D data were 5% higher and cost data 5% lower than imputed in both groups of the study

• missing EQ-5D data were 10% higher and cost data 10% lower than imputed in both groups of the study.

Sensitivity analysis 2 (SA2): The cost–utility analysis was repeated, but the intervention was costed as per the SAFA trial protocol (i.e. intervention was accessed via secondary care) in order to provide a range on the possible cost-effectiveness estimate from the base case analysis (i.e. where intervention accessed via primary care). This was a more conservative analysis as secondary care delivery of the intervention is more costly.

Sensitivity analysis 3 (SA3): The cost–utility analysis was repeated assuming the NICE guidance12 on acne was followed, meaning all women in the placebo group were assumed to have taken oral antibiotics (lymecycline or doxycycline, one tablet daily for 24 weeks), in addition to topical treatment. As this patient population had persistent acne of sufficient severity to warrant treatment with oral antibiotics, this is a reasonable assumption. In this analysis we assumed that the health-related quality of life (measured using the EQ-5D-5L) remained unchanged from that which was collected in the SAFA trial. In reality we did not know what the health-related quality of life of these women would be. However, assuming that these women took oral antibiotics (and the antibiotics were effective), the incremental QALYs would be less than that observed in the SAFA trial for women in the placebo group. A threshold analysis was performed to ascertain what level of QALYs would switch the intervention between cost-effective and cost-ineffective.

Sensitivity analysis 4 (SA4): The cost–utility analysis was repeated taking a wider perspective by including the participants’ and family/friends’ out-of-pocket costs, productivity costs and non-official services costs.

Sensitivity analysis 5 (SA5): We performed a descriptive analysis of data collected after participants were unblinded to treatment (at week 24), from week 25 and up to week 52. The mean costs (SD) and mean (SD) QALYs by randomisation group are reported together with incremental costs and QALYs for up to week 52. Due to the number of missing data up to week 52, costs and QALYs were not combined.

This economic evaluation is reported in line with the Consolidated Health Economic Evaluation Reporting Standards (CHEERS) guidance. 41 We did not, however, examine the distributional effects, that is how costs and outcome were distributed across different individuals, as this item was not on the previous version of the CHEERS checklist at the time the HEAP was written.

Resource use and costs

Unit costs, together with their source and any assumptions made, are presented in Table 26. Mean baseline resource use per participant for each randomised group is summarised in Table 27 using the available case data. The level of resource use in each group was very similar prior to randomisation.

Intervention resource use and costs

If implemented in clinical practice, spironolactone would most likely be used alongside a topical treatment and would be obtained by the patient through their GP (i.e. primary care). Figure 4 shows a flow chart of resource use for the intervention group (spironolactone) as if it was being delivered in primary care. The majority of women (182/184, 99%) increased to two tablets of spironolactone at week 6. In the 24-week period where spironolactone was used and data were available, the mean number of doses per participant was 293 (SD 9.47). The standard of care approach (expected resource use if implemented widely in primary care) used in the base case economic evaluation, gave rise to a mean cost of £122.87 (SD £13.04) per participant (including medication, GP visits and blood tests) (refer Table 28 for breakdown).

FIGURE 4.

Intervention resource use as per proposed standard of care (base case). a, Assumes all patients escalated to two 50 mg tablets spironolactone at 6-week visit. Based on the data from the trial, this was the case for 182/184 (99%) of available patients in the spironolactone group at 6 weeks. b, As those with relevant comorbidities or on treatments with increased risk were not included in the trial, it is not possible to estimate the proportion of such patients that might receive spironolactone in reality and that would need blood test monitoring.

Other health resource use and costs

Mean resource use for the spironolactone and placebo groups are shown in Table 28. Table 29 reports the disaggregated mean discounted costs per participant for both groups using available case data. When intervention use was combined with other health resource use, the mean difference per participant was £126.35 (95% CI, £112.88 to £139.82) for women receiving spironolactone compared to women receiving placebo in the base case (see Table 29). The difference in total costs between both groups largely reflects the cost of the intervention; other NHS costs were not significantly different between groups (£3.98 higher per participant, on average, in the intervention group, 95% CI, –£9.30 to £17.26).

Participant out-of-pocket and productivity costs

Personal out-of-pocket expenses costs were £69.41 (SD £113.05) per participant for the spironolactone group compared to £82.57 (SD 148.60) for the placebo group, giving a mean difference of –£13.15 (95% CI –£45.23 to £18.92) per participant. Out of the six categories of personal costs explicitly asked about, cosmetics skincare products formed the largest cost category, with a mean cost of £56.77 (SD 82.90) per participant in the intervention group and £60.05 (SD 78.00) per participant in the control group, followed by complementary therapy, with a mean cost of £9.96 (SD 57.54) per participant in the intervention group and £17.37 (SD 106.11) per participant in the control group. The five main categories included under out-of-pocket expenses, as informed by input with the study team’s PPI members, captured nearly all out-of-pocket costs; the ‘other’ category included to capture anything that may have been missed was rarely used.

Four participants reported using other support services, including an online acne support forum or social media groups for PCOS/Acne, but did not report any costs for using these.

In terms of productivity costs, response rates to these questions varied between 98% at baseline, 70% at week 24 and < 7% at up to week 52. Very few women, in paid employment, reported an impact on their employment as a result of their acne. The number of women reporting an impact on their paid employment was in single figures. It was not always possible to attach a cost on the basis of the information provided.

In terms of whether the participants had a family member or friend who had taken time off paid work to accompany them to healthcare appointments related to their acne, six women reported this at baseline, three at 6 weeks, four at 12 weeks, one at 24 weeks and none at 52 weeks.

Base case cost–utility analysis

Table 34 presents the results of the cost–utility analysis in terms of incremental costs and QALYs, together with an estimate of the ICER and separately the CEAC (see Table 34). In the CCA, the incremental cost for the spironolactone group (n = 118) compared to the no active intervention (placebo) group (n = 101) was £125.36 (95% CI, £111.13 to £139.58) [unadjusted this was £125.53 (95% CI £112.15 to £138.91)]. The adjusted incremental QALYs for the spironolactone group compared with the no active intervention (placebo) group was 0.0019 (95% –0.0096 to 0.0133) (unadjusted was 0.0036, 95% CI –0.0117 to 0.0189). The ICER was £67,191 (unadjusted £34,770) per QALY, which given a threshold value of £30,000 (£20,000),45 means the intervention appears cost-ineffective in both the adjusted and unadjusted analyses at baseline without taking account of missing data. At a willingness to pay of £30,000 per QALY there was a 35% (unadjusted 47%) chance of spironolactone being cost-effective in this population for women with persistent moderate to severe acne.

The CEACs (Figure 5) of the adjusted and unadjusted base case analysis show that the probability of spironolactone being cost-effective only approaches 50% as the threshold value approaches £120,000 (adjusted), demonstrating a high degree of uncertainty associated with the decision under these conditions.

FIGURE 5.

Cost-effectiveness acceptability curve, CCA, adjusted and unadjusted QALYs.

Secondary cost-effectiveness analysis

Refer to Table 35 for a summary of results. The adjusted incremental difference in cost per point change on the Acne-QoL symptom subscale for the spironolactone group (n = 119) compared to the non-active treatment (placebo) group (n = 102) was £38.21 (unadjusted £35.91) based on a CCA. The adjusted mean point change on the Acne-QoL symptom subscale was higher in the spironolactone group suggesting their acne-related quality of life improved more than for those in the placebo group. The spironolactone group was more expensive and more effective than the placebo group. It is impossible to say whether the ICER reported is deemed cost-effective or not as there is uncertainty about how much a decision maker would value a point change on the Acne-QoL symptom subscale.

Missing data analysis (SA1)

Using multiple imputation, under the assumption of MAR, to evaluate the impact of missing data on the cost–utility results, it can be seen that the ICER is less than in the complete case, adjusted analysis (£27,879 compared with £67,191 per QALY in the adjusted base case). The probability of spironolactone being cost-effective at £30,000 per QALY was estimated as 53%.

Table 37 reports the proportion of missing values (%) for key variables – it can be seen that there was differential rates of attrition for cost and outcome variables, with greater missing data in the placebo group, compared to the spironolactone group, at 12- and 24-week follow-up for costs and outcome variables. There also seems to be a pattern of increasing missingness with time, suggesting some patients are being lost to follow-up.

Logistic regression was used to explore any association between total costs over the treatment period, QALY score at 24 weeks, and Acne-QoL subscale score change at 24 weeks with randomisation group, randomisation stratification variables [site and baseline severity (IGA < 3 vs. ≥ 3)], baseline variables (Acne-QoL symptom subscale score and use of topical treatments) and value of baseline outcome (EQ-5D).

The fact that some baseline variables, including treatment allocation, predict missingness confirms that data are not MCAR and support the methods for SA1 under a MAR assumption as more plausible (Table 38). However, MNAR is a possibility given that missingness is associated with treatment allocation: patients with worse outcomes may be dropping out at higher rates in the control arm if placebo plus usual treatment is not effective. As it is not possible to ascertain the extent to which this may have occurred, this supports the need for SA1a to explore the impact MNAR may have on outcomes.

The results of SA1a are presented in Table 39 and demonstrate that changing costs to assume that patients with missing data may have had higher resource use, did not have a large impact on results. Changing QoL data to assume that patients with missing data had lower QoL scores, did have an impact on results, reducing the ICER and increasing the probability that spironolactone is cost-effective compared with placebo. Indeed, assuming a 10% reduction in QoL scores pushed the ICER below the £20,000 threshold.

Intervention costed as per protocol with delivery of spironolactone in secondary care (SA2)

In a sensitivity analysis, costing the intervention as it was delivered in the trial (i.e. in secondary care, see Figure 6 for detail), the mean cost of the spironolactone intervention (medication, dermatology nurse visits, blood tests and pregnancy test) was £264.94 (SD £36.14). The ICER was estimated as £141,955 per QALY and a probability of 12% of spironolactone being cost-effective at a £30,000 threshold. The increased ICER reflects the cost of delivering the intervention through regular appointments with a dermatology nurse (trial protocol), rather than in primary care, as would be the expected standard of care if delivered in practice.

FIGURE 6.

Intervention resource use as per trial protocol (i.e. accessed via secondary care) (SA2).

Cost-effectiveness of spironolactone compared to treatment with oral antibiotics in line with NICE acne guidance (National Institute for Health and Care Excellence, 2021) (SA3a: complete case and SA3b: multiple imputation)

There were 13 participants who reported taking an oral antibiotic (lymecycline/oxytetracycline/doxycycline) between baseline and week 24, 9 participants in the placebo group and 4 participants in the spironolactone group. In a sensitivity analysis undertaking a CCA, assuming all women in the placebo group took an oral antibiotic for the 24-week treatment period in line with NICE guidance,12 an ICER of £9169 per QALY was estimated with a probability of 59% of spironolactone being cost-effective at a £30,000 per QALY threshold. Repeating this but using multiple imputation to deal with missing data, the ICER was £2683 per QALY with a probability of 82% that spironolactone would be cost-effective at a £30,000 threshold (see Table 34). This sensitivity analysis assumed that the incremental QALYs observed in the SAFA trial would still be valid, if oral antibiotics are effective, we would expect the incremental QALYs to be less than that observed in the trial. A threshold analysis was performed which found that incremental QALYs would have to decrease to 0.00057 in order to switch the ICER from cost-effective to cost-ineffective at a £30,000 threshold.

Wider case (SA4a: complete case and SA4b: multiple imputation)

Sensitivity analysis was undertaken to take into account costs outside the NHS, including personal out-of-pocket expenses and lost patient and carer productivity as a result of participant’s acne. This wider perspective CCA only had complete cost and QALY data for 97 (48%) women in the spironolactone group and 85 (40%) of women in the control group. The CCA found spironolactone to be dominated by the control group, that is the spironolactone group had higher costs and lower QALYs than those in the control group. This is likely to reflect the small numbers in this analysis. Due to the level of missing data when combining responses from multiple questionnaires, multiple imputation was also undertaken in this sensitivity analysis. Handling the missing data in this way resulted in an estimated ICER of £30,249 per QALY with a probability of 50% of spironolactone being cost-effective at a £30,000 threshold (see Table 34). Changes to reported productivity were found to be minimal, with only nine such changes to paid employment reported for participants themselves and six reported for a family member or friend taking time off work to accompany them to appointments for their acne.

Costs and outcomes over 52 weeks

As described in the methods, data were also collected beyond the treatment period (24 weeks) for up to 52 weeks. Response rates were, however, very low at this time point, with 58% of participants missing EQ-5D data and 93% missing resource use data (see Table 37). As such, it is difficult to draw conclusions from these results, but as presented in Table 29, incremental QALYs over 52 weeks was 0.0644 (95% CI 0.0093 to 0.1194) and incremental cost (NHS perspective) (Table 40) over the same period was £95.44 (95% CI 8.29 to 182.70).

Main findings

Our economic evaluation provides a range of estimates for the cost-effectiveness of spironolactone compared to no active treatment (both groups could use routine topical treatment). In the base case analysis, where the comparator is that used in the trial (no active treatment) and the delivery of the intervention is costed as delivered via primary care, we found no evidence of cost-effectiveness for spironolactone in the adjusted analysis. However, this is not particularly meaningful as women are unlikely to be given no treatment in clinical practice, so a more appropriate comparator would be another oral treatment, such as oral antibiotics.

In adjusted analysis using multiple imputation (MI), spironolactone was estimated to be cost-effective at the £30,000 per QALY threshold. This divergence in conclusion between the complete case and MI analysis demonstrates the impact of missing data (41.95% missing for costs and 27.32% for QALYs at 24 weeks). The cost-effectiveness of spironolactone is potentially understated by the decision to use the CCA as the base case. Furthermore, exploration of the MAR assumption for this MI analysis, given the greater proportions of missingness in the placebo arm, suggests that this MI analysis may under-estimate the cost-effectiveness of spironolactone. The large difference in the estimated ICER for the adjusted and unadjusted complete case analyses reflects the non-significant differential mean utility scores between study groups at baseline.

The base case within trial economic evaluation is limited to the question and data explored in the trial; for this study the comparator was not that which would be considered to represent current practice. 12 Rather, given the dearth of evidence of benefit for spironolactone for acne vulgaris in adult women,25,54 the comparator was a placebo with routine topical treatment in order to show any benefit. To test the impact of this on the findings of the economic study, we present a sensitivity analysis where the placebo group was assumed to take oral antibiotics in addition to topical treatment in line with current NICE guidelines for moderate acne. 12 In this analysis we assumed that antibiotics plus routine topical treatment would convey no more benefit (as measured by the EQ-5D-5L) than placebo plus routine topical treatment. We believe this to be reasonable given that the available evidence from a systematic review demonstrates little to no benefit of combined oral antibiotics with topical treatments compared to topical treatment combinations alone. 55 Furthermore, placebos are known to convey some effectiveness through the placebo effect. 56,57 This analysis found that spironolactone is highly cost-effective with an ICER of £9169 per QALY in a CCA and £2683 per QALY using multiple imputation. Though there are limitations of this analysis, notably we cannot be certain as to the impact on utility of taking an oral antibiotic on top of topical treatment, as this was not the comparator within the trial. To explore this, we undertook a threshold analysis to estimate how much incremental QALYs would have to change in order for the conclusion to move from cost-effective to cost-ineffective. We found that as long as the incremental difference in QALYs was more than 0.00057 QALYs when the willingness-to-pay threshold is £30,000 per QALY in the CCA, then spironolactone would remain cost-effective compared to oral antibiotic use. We await further evidence to determine whether this is a likely scenario.

Strengths and limitations

The strength of this economic evaluation is that it is able to provide reliable estimates of cost-effectiveness based on individual participant level data collected at little marginal cost alongside a RCT. This is, however, also a potential limitation in that within trial health, economic evaluations are constrained by the question and data collected in the RCT. This is particularly so for placebo-controlled trials. The base case set out to answer the question of whether spironolactone is cost-effective compared to no active treatment (both groups could use routine topical medications) to align with the clinical question funded. In line with recommendations, the costs related to the placebo tablets were excluded as these were a research cost. However, it should be acknowledged that placebos may themselves result in some effectiveness through a placebo effect that could potentially lower cost and that were such an effect present it would not be possible to observe the size of this effect given the study design or account for it in the economic analysis. In practice this analysis may also not be the most informative for clinicians who would be unlikely to send women away with no active treatment if they consulted with acne persisting beyond 6 months. We therefore performed a sensitivity analysis assuming all women in the placebo group received an oral antibiotic for 24 weeks. Limitations of this analysis were acknowledged above, and a further potential limitation is that this approach assumes that the downstream resources used, if receiving oral antibiotics, will be the same as in the placebo arm. In reality, patients may use fewer resources if their acne is well-managed and/or use more resources in order to manage AEs. However, despite these limitations and while the results should be interpreted with caution, the analysis serves to provide a lower range estimate for the cost-effectiveness of spironolactone that better reflects accepted standard of care based upon current NICE guidelines. 12 Other studies of spironolactone are underway in France and the USA,58,59 which compare spironolactone directly to oral antibiotic treatment, and therefore the economic results of these studies will add to the evidence base in this regard in due course. It should also be acknowledged that this study was not able, given the time frame, to capture the potential costs and benefits of changes to antimicrobial resistance that may result from less prescribing of antibiotics for acne if the use of spironolactone resulted in such. Since acne is one of the most common dermatological conditions and since the prescription of antibiotics is one of the most common systemic treatments prescribed for acne worldwide, the impact of increased use of spironolactone on antimicrobial resistance is likely to be significant. Further research based on this sensitivity analysis comparing spironolactone to oral antibiotics could be undertaken to conduct a model-based cost–utility analysis using the results from the aforementioned ongoing trials comparing spironolactone to doxycycline, when their results are published. 58,59

At the design stage there was discussion about whether the EQ-5D-5L was an appropriate instrument to use in a study of acne. The EQ-5D was included to enable a cost–utility analysis to be performed. Such an analysis enables cost-effectiveness to be compared across a range of health conditions and interventions and is most useful to decision makers who have to prioritise health care. The limited published evidence available at that time supported the use of the EQ-5D for acne, although this evidence was based on the EQ-5D-3L, whereas the 5L version increases the potential for the instrument to pick up smaller changes in health-related quality of life. Yang et al. 60 systematically searched for evidence on the reliability, validity and responsiveness of three generic preference-based measures [EQ-5D, short form questionnaire-6 dimensions (SF-6D) and Health Utilities Index (HUI)] in skin conditions. They only found evidence, albeit a limited amount, to support the use of the EQ-5D in skin diseases; they found no studies looking at measurement properties for the SF-6D or HUI in skin disease. For acne there was one paper by Klassen et al. 50 who found that problems on the EQ-5D-3L domains were substantially higher in the acne sample than in an age-truncated population sample (aged 20–39 years) particularly on the pain/discomfort (42.1% in the acne sample vs. 17.7% in an age-truncated population sample) and anxiety/depression domains (52.8% vs. 12.5% respectively). It also found the EQ-5D-3L to be responsive to change, with moderate effect sizes at 4 and 12 months (–0.44 and –0.53 respectively). The authors argued that the use of the EQ-5D would ensure the benefits of acne treatments could be compared to those of other treatments (Klassen et al.,50 p. 232).

In this study, which uses the EQ-5D-5L as opposed to the 3L version, the conclusion reached about cost-effectiveness was sensitive to the estimates of QALYs generated using it. This may reflect a number of factors. Firstly, at baseline there was a non-significant difference in utility between study groups with higher mean utility in the intervention group. This meant that the complete case adjusted analysis, which takes into account baseline utility, reached a different conclusion to the unadjusted CCA, not taking account of baseline variables. Secondly, there were differential rates of missing data on the EQ-5D-5L between study groups, with lower completion rates in the control group (65%) than the intervention group (81%) at 24 weeks. When multiple imputation in an adjusted analysis is used, the incremental CCA, unlike the complete case analysis, fell below the £30,000 NICE threshold, suggesting that spironolactone alongside topical treatment is borderline cost-effective compared to no active treatment (placebo) alongside topical treatment. Finally, at baseline, 46% in the intervention group and 43% in the control group were in perfect health (the health state 11111) according to the EQ-5D-5L. For these participants, the EQ-5D-5L had no potential to measure any change in health-related quality of life from the treatments. It is unclear to what extent this differs from the proportion we would expect to be in the perfect health state in the general population as there do not appear to be published population norms for the EQ-5D-5L in the UK (Kind et al., 1998 and EuroQOL website). 61–63 Comparison can be made to other studies of different conditions, for instance in a recent study of the skin condition vitiligo, 55% of participants were in the perfect health state on the EQ-5D-5L at baseline suggesting that acne may have more impact on health-related quality of life than vitiligo as captured on the EQ-5D-5L. Like Klassen et al.,50 we find that women with moderate to severe acne report most problems on the pain/discomfort and anxiety/depression dimensions of the EQ-5D. Further research using the EQ-5D data generated in this study alongside that elicited in other studies of acne would be useful in order to inform future studies about the validity and responsiveness of using the instrument for acne, while acknowledging that this may vary depending on the severity of acne amongst the study group.

The health economics analysis plan specified that a CCA would be undertaken in the base case; this was to be consistent with the approach undertaken in the statistical analysis for the clinical primary outcome. However, such an approach is less reliable where there is missing economic data and where this missingness is at differential rates between study groups at follow-up (attrition bias). With the benefit of hindsight, primary concern ought to have been around the level of missing economic data, which is known to often be greater than that for clinical outcomes. However, both complete case and multiple imputation analyses are reported as planned so that the impact of missing data on the results can be clearly seen. Taking account of missing data in the analysis improves the incremental cost per QALY estimated and thus the cost-effectiveness of spironolactone.

Finally, given the timing of the study, with recruitment and follow-up occurring during the COVID-19 pandemic, it is probable that patient resource utilisation was affected by this. For example, patients may have had fewer opportunities to access healthcare professionals, which could have reduced the downstream resource use, thus potentially reducing the incremental cost in favour of no active treatment. This could have resulted in an under-estimation of the cost-effectiveness of spironolactone and the generalisability of these results should be considered in this context.

Patient-reported outcome measures in acne

Due to the pragmatic trial design, we wished to include a PROM for acne as the primary outcome as this is in line with what people with acne say is important to them, that is overall appearance, rather than counting the number of spots. 64 Furthermore, previous trials have concluded that lesion counts are time-consuming, show wide inter-assessor variation and give little additional information to global assessments.

However, although the Acne Core Outcomes Research Network65 has developed consensus around ‘what to measure’ in acne trials there is, as yet, no strong consensus on which specific measurement tools to use. At the time of designing the trial, there were few validated measures available, and we chose the Acne-QoL symptoms subscale as primary outcome on the basis of a mixed methods evaluation of existing scales. 66 More recent evidence suggests that, while evidence on measurement properties is lacking for all acne PROMs, newer measures (COMPAQ and Acne-Q) may have greater content validity. 67

Ongoing randomised controlled trials of spironolactone for acne

There are currently two further RCTs of spironolactone recruiting: spironolactone (100 mg daily) versus doxycycline (100 mg daily) in a non-inferiority trial in the USA59 and spironolactone (150 mg daily) versus doxycycline (100 mg daily) superiority trial in France. 58 It is hoped that the outcomes of these may provide opportunities for individual patient meta-analysis and greater power to examine effects within subgroups, particularly age, body mass index (BMI) and ethnicity.

Spironolactone dose

In the SAFA trial, women commenced on 50 mg daily of spironolactone, increased to 100 mg daily at or after 6 weeks if tolerated. We found that over 95% of women did increase their dose to 100 mg. It could be speculated that our finding of greater effect at 24 weeks than 12 weeks might not have been observed if participants had commenced on 100 mg, and it may be that starting at the higher dose and then down-titrating only if side effects develop may lead to more rapid improvement in acne symptoms. This treatment regimen was chosen on the basis of a survey of clinicians carried out at the design stage and reported in our protocol paper. 1

The ongoing US and French trials both chose a higher starting dose of spironolactone (100 mg and 150 mg respectively) and it will be interesting to compare their data on tolerability, adherence and adverse effects in these trials. With more data it may also be possible to explore whether people with PCOS or higher BMI benefit more from higher doses, as there is literature to suggest that this may be the case. 25

Side effects of spironolactone in this patient group

The most common side effect found in observational studies was menstrual irregularities (15–25%), although this was thought to be less common at lower doses (100 mg/day or less). 25 It is notable that reports of menstrual irregularities were very similar across both groups (39% in the spironolactone group and 38% in the placebo group combined across all time points). Menstrual irregularities have been reported as a very frequent side effect amongst women taking spironolactone at higher doses (particularly 200 mg daily),25 but it is reassuring that, in comparison with placebo, the proportion experiencing this side effect does not appear to be high amongst women taking 100 mg daily spironolactone.

Other side effects, also dose-related, reported in previous studies include breast tenderness, dizziness, nausea, headache, polyuria and fatigue. 25 It is a useful addition to the literature to be able to report the frequency of these adverse effects in a placebo-controlled trial. Whereas SAFA participants in both groups commonly reported several of these symptoms, in general, differences between groups were not substantial and the only statistically significant difference was headaches, experienced by 20% in the spironolactone group and 12% in the placebo group. Clearly the analyses of differences in reporting are not powered to detect small differences and, for instance, the difference in dizziness/vertigo/lightheadedness between groups suggests this may be more common for spironolactone than placebo, whereas there are some instances, such as fatigue, that were actually slightly more common in the placebo group.

Pregnancies

Spironolactone is cautioned against in pregnancy, due to a theoretical risk of feminising the male fetus in the third trimester, although this has not been observed in humans. 68 Spironolactone is likely to be less teratogenic than oral tetracyclines, commonly used for acne in young women, where there is concern about dental discolouration and possible effect on fetal bone growth in the second or third trimester,68 so it seems appropriate that in usual practice, spironolactone would be treated the same as these agents.

The SAFA trial took a pragmatic approach to pregnancies, again in terms of seeking to reflect real-life practice. SAFA participants were advised to use contraception at the baseline visit and at subsequent visits, but pregnancy tests were only carried out at baseline and not at every visit. It was felt that a baseline test would be necessary in the context of a RCT, although this is unlikely to be usual practice. Seven pregnancies were observed in the trial, which would suggest the importance of regular counselling about contraception in this age group.

Renal function monitoring

There is increasing consensus that, while baseline check of renal function and potassium levels is likely to be advisable prior to commencing spironolactone, ongoing monitoring is unlikely to be necessary for most young women. 13 The largest study on this topic to date found that 13 (0.7%) of 1802 women commencing spironolactone were found to have raised potassium, but that subsequent testing showed this to be either erroneous or transient or quickly self-resolved. 69 A study using routinely collected US data found that, of 618 women starting spironolactone, 145 had no blood tests either at baseline or for ongoing monitoring and concluded that raised potassium rarely occurs in otherwise healthy women aged 45 years or less. 70

While UK guidelines make no mention of the use of spironolactone for acne, US guidelines13 advise that:

Serum potassium testing is therefore not required, but should be considered in older patients and in patients who are also taking angiotensin-converting enzyme inhibitors, angiotensin receptor blockers, nonsteroidal anti-inflammatory drugs, and digoxin.

Gaps in prior evidence

Prior to carrying out this research we had previously carried out systematic reviews, both of the use of spironolactone for acne25,26 and on the views and experiences of people with acne. 71 The systematic review of spironolactone for acne showed that, unfortunately, spironolactone is not a suitable treatment for men due to related gynaecomastia (breast swelling).

The systematic review of the views and experiences of people with acne showed that under-represented groups include younger people, people from ethnic minority backgrounds and men. Unfortunately, men cannot use spironolactone, as above, so were not included in this trial. The HTA call sought a proposal on persistent acne, and we therefore chose to limit the trial to women aged 18 or over, which is also clinically where more consensus for the use of spironolactone previously converged. Regarding ethnicity, we aimed to address this by including sites covering geographical areas with high proportions of populations from ethnic minority backgrounds.

The INCLUDE Ethnicity Framework72 will be helpful in designing future trials and seeking to maximise inclusion and applicability to underserved groups, although was not available at the start of designing this trial. It is unfortunate that ethnicity was not included initially in study documents although during the progress of the trial, efforts from the trial staff and site staff to collect this data for participants who had already been recruited mean that it has been possible to report this, though with some missing data.

Members of research team

The research team was made up predominantly of women, including the two Co-Chief Investigators. There was no great ethnic diversity within the research team, although at least five members were of European background and did not have English as their first language.

Both the TSC and Data Monitoring Committee were chaired by women, and the DMEC chair was of SE Asian background.

Participant representation

Of 356 participants where ethnicity data were available, 92.1% were white and 7.9% were from non-white background. The reason for the predominantly white trial participant population is unknown but may have related to the ability of participants to travel to recruiting centres during the COVID-19 pandemic. The three recruiting centres with the highest number of participants (accounting for 59.3% of recruited participants) were located in Portsmouth, Bristol and Harrogate.

Public involvement

Public contributors were relatively representative of the target population, as both the two public contributors on the TMG and the public contributor on the TSC had had acne for many years and had been frustrated by treatments on offer. They represented differing geographic areas (NE, SE, SW) but no non-white ethnicity.

How public contributors were involved prior to the randomised controlled trial

A James Lind Alliance Priority Setting Partnership, funded by the National Institute for Health and Care Research (NIHR), identified the need to establish the best way to manage acne in women who may or may not have underlying hormonal abnormalities. 25

We gained feedback on key questions relating to research design from a virtual acne-specific patient panel, convened through ‘People in Research’ (www.peopleinresearch.org), as well as a patient survey carried out with the support of the UK Dermatology Clinical Trials Network (UK DCTN). Findings suggested that participants would find it difficult to abstain from using topical treatments and that asking participants to take a placebo for 1 year would also be a barrier to recruitment. This strongly influenced design decisions around use of topical treatments in the trial and choice of primary outcome at 12 weeks with unblinding at 24 weeks.

Two public contributors (IS and KaT) with experience of acne were co-applicants on the grant and influenced design decision, for instance highlighting that the originally planned upper age limit of 50 years was arbitrary, which was then abandoned, and contributing to the choice of primary and secondary outcomes.

How public contributors were involved in delivering the randomised controlled trial

Two public contributors (IS and KaT) with experience of acne attended all TMG meetings to ensure that decisions around trial design were informed by their perspective, trial procedures were feasible for participants and trial materials were readable and included all the relevant information that participants would want. Public contributors influenced the trial design and delivery, for instance by advocating the use of social media advertising to improve recruitment, designing and later redesigning recruitment materials. Due to the COVID pandemic, face-to-face meetings were limited. Indeed, the ‘results reveal’ meeting, planned as face-to-face in London, became a largely hybrid meeting due to so many attendees having COVID at that time.

Although there are not specific charities to liaise with in the field of acne, public contributors will also be involved in disseminating the trial and have suggested routes to dissemination.

Guidance for reporting involvement of patients and the public 2 short form

Clinical questions

• What is the comparative effectiveness of oral antibiotics compared with spironolactone for the treatment of acne?

• What is the optimal dosing of spironolactone in acne, particularly for women of higher BMI?

• Which women with acne are most likely to benefit from spironolactone, for instance which age groups, ethnicities and acne types?

Mechanistic questions

• Further research into the mechanism of action of spironolactone for acne could inform which patient groups and clinical subtypes of acne may respond better and which treatments could most usefully be co-prescribed with spironolactone, or help develop new treatments.

Health economics research recommendations

• Further research based on this sensitivity analysis comparing spironolactone to oral antibiotics could be undertaken to conduct a model-based cost–utility analysis using the results from the aforementioned ongoing trials comparing spironolactone to doxycycline, when their results are published.

• Further research using the EQ-5D data generated in this study alongside that elicited in other studies of acne would be useful in order to inform future studies about the validity and responsiveness of using the instrument for acne, while acknowledging that this may vary depending on the severity of acne amongst the study group.