A randomised placebo-controlled trial of oral and topical antibiotics for children with clinically infected eczema in the community: the ChildRen with Eczema, Antibiotic Management (CREAM) study

Article history

The research reported in this issue of the journal was funded by the HTA programme as project number 09/118/03. The contractual start date was in June 2012. The draft report began editorial review in July 2015 and was accepted for publication in November 2015. The authors have been wholly responsible for all data collection, analysis and interpretation, and for writing up their work. The HTA editors and publisher have tried to ensure the accuracy of the authors’ report and would like to thank the reviewers for their constructive comments on the draft document. However, they do not accept liability for damages or losses arising from material published in this report.

Declared competing interests of authors

Professor Christopher C Butler received fees for acting in an advisory capacity to Alere™ and is supporting a study on which he is the chief investigator with diagnostic devices in the form of an unconditional educational grant. He is also a National Institute for Health Research (NIHR) Efficacy and Mechanism Evaluation board member. Professor Kerenza Hood is a member of the NIHR Clinical Trials Unit standing committee. Dr Mandy Wootton has declared that she received a speakers honoraria from Nordic Pharma Ltd (who manufacture fosfomycin).

Copyright statement

© Queen’s Printer and Controller of HMSO 2016. This work was produced by Francis et al. under the terms of a commissioning contract issued by the Secretary of State for Health. This issue may be freely reproduced for the purposes of private research and study and extracts (or indeed, the full report) may be included in professional journals provided that suitable acknowledgement is made and the reproduction is not associated with any form of advertising. Applications for commercial reproduction should be addressed to: NIHR Journals Library, National Institute for Health Research, Evaluation, Trials and Studies Coordinating Centre, Alpha House, University of Southampton Science Park, Southampton SO16 7NS, UK.

Infected eczema

Eczema is a relapsing–remitting condition, yet there is considerable uncertainty about the cause of flares. 17–19 Eczema often results in skin that is dry, red, itchy, broken and sore, and can lead to a breakdown of the skin barrier. This makes the skin susceptible to trigger factors such as irritants and allergens, as well as microbial colonisation and infections. 20

Staphylococcus aureus has long been known to be more prevalent on the skin of patients with eczema. The organism can be isolated from up to 90% of patients with eczema21,22 compared with between 5% and 25% of healthy subjects. 21,23 Furthermore, more severe eczema is associated with higher densities of the organisms21,23,24 and more resistant strains. 25

There is evidence that a number of factors contribute to this propensity for colonisation and infection, including dysregulation of the adaptive immune response, reduced antimicrobial peptide levels, diminished recruitment of cells to the skin, toll-like receptor defects and epidermal barrier abnormalities. 26 The exact role of S. aureus in the maintenance or exacerbation of eczema is not clear. However, there is increasing evidence for the role of toxins with superantigenic properties (superantigens). 26

Use of antimicrobial treatments for eczema

Despite clear evidence for a relationship between eczema and the presence of S. aureus, there is a lack of clarity about what constitutes infection and when antibiotic treatments are likely to confer benefit. 27 A recently updated Cochrane review28 of antimicrobial interventions for people with eczema, which included 26 studies and 1229 participants, found that most studies were small and of poor quality, and that although there was evidence that interventions reduced the presence of S. aureus on the skin, none of the studies showed any meaningful clinical benefit from antibiotics or other antimicrobial interventions, for either clinically infected or non-infected eczema. The authors concluded that ‘Their continued use should be questioned in such situations, until better and longer-term studies show clear evidence of clinical benefit’. 28

This is important not only because of the need to identify effective treatments for children with eczema, but also to reduce the use of ineffective treatments currently being prescribed. Clinical experts estimate that children in the UK experience approximately 900,000 eczema flares a year, and that approximately 40% of them are treated with topical antibiotics. 29 Widespread use of antimicrobials is a key contributor to the development of antimicrobial resistance and exposes children to possible harms from adverse effects. Therefore, use of antibiotics is only justifiable where there is clear evidence of benefit. Topical antibiotics may be preferable to systemic treatment, as they maximise the effective doses at the site of infection while minimising the systemic effects. However, the prevalence of resistant strains of skin bacteria is steadily increasing and cases of allergy or skin sensitisation are not uncommon. 30,31 Fusidic acid resistance has been shown to be related to high levels of use, for example. 32 Therefore, once a decision is made to prescribe antibiotic treatment, it is unclear whether topical or oral antibiotics are most effective and which cause the least ‘collateral damage’ to the microbiome in terms of driving resistance.

Summary

Eczema has a negative impact on the quality of life of paediatric patients and their families. Oral and topical antibiotics are widely used to treat clinically infected eczema in primary care and yet there is insufficient evidence to be sure whether this established practice either helps or harms patients and, if antibiotics help, which route (topical or oral) does most good, causes least harm and is preferred by parents. Therefore, there is a clear need to identify whether or not oral and topical antibiotics confer meaningful benefit to children with clinically infected eczema. The ChildRen with Eczema, Antibiotic Management (CREAM) study aims to benefit patients with eczema (and their families), as well as helping to address the important issue of antibiotic use and resistance, by addressing this gap in the evidence.

Summary of study design

The CREAM study was a three-arm, double-blind, randomised placebo-controlled trial which aims to determine the clinical effectiveness and cost-effectiveness of the most commonly used oral and topical antibiotics, in addition to topical corticosteroids (TCSs), in the management of clinically suspected infected eczema in children compared with placebo treatment. The CREAM study was based in the community, primarily in primary care, but also included four dermatology clinics that saw patients who could self-refer. Participating clinicians were asked to identify children (aged < 8 years) with clinically infected eczema. Children with clinically severe infections or significant comorbid illnesses were excluded. Eligible, consented children were randomly assigned to one of three treatment groups: (1) oral antibiotic and placebo topical cream; (2) topical antibiotic cream and placebo oral treatment; or (3) placebo oral and placebo topical cream treatment. In addition, all participants received standard advice about eczema care and were prescribed TCS creams and emollients.

At the baseline visit, a trained research nurse assessed the extent and severity of eczema, and parents reported quality of life, impact of the eczema on the family and health status, using established assessment tools. The research nurse conducted these assessments again, and enquired about side effects and health-care resource use, at 2 and 4 weeks after baseline. Parents were asked to complete a diary for the first 4 weeks following enrolment to record daily symptom severity, use of medication and health-care consultations. Participants were followed up at 3 months by a postal or telephone questionnaire and through a review of their primary care medical record in order to identify health-care consultations, medication usage and any subsequent episodes of infected eczema. Skin, mouth and nose swabs were collected at the baseline, 2 weeks and 3 months to assess for the presence of S. aureus on the skin (from the most severe site) and the impact of treatments on bacterial resistance.

The main analysis compared subjective eczema severity at week 2 in each of the two active treatment groups (oral and topical antibiotics) with the double placebo group.

The schedule of events and participant flow for the trial is summarised in Figure 1.

FIGURE 1.

Study schema and participant flow diagram. CDLQI, Children’s Dermatology Life Quality Index; DFI, Dermatitis Family Impact; EASI, Eczema Area and Severity Index; IDQoL, Infant’s Dermatology Quality of Life instrument; POEM, Patient-Orientated Eczema Measure.

Trial objectives

Setting

The study was community based and originally aimed to recruit patients only through general practices. However, in order to address a low recruitment rate, the study was opened in four dermatology sites.

Site recruitment

The study was open to participant recruitment from 12 July 2013 until 28 November 2014. The study was conducted in regional centres in Wales (Cardiff), England (Bristol) and Scotland (Dundee). A principal investigator led each region.

General practices and dermatology clinics (providing first point-of-contact dermatology services) were considered sites for the purpose of the study. The Primary Care Research Network in England and Scotland, and the National Institute for Social Care and Health Care Research Clinical Research Centre Network in Wales supported site recruitment.

General practitioner (GP) practices were invited to take part in the study either by letter or via a Primary Care Research Network newsletter. Interested practices were contacted initially by e-mail to provide further information about the study and were followed up by telephone or face-to-face visit with the practice manager or lead research GP to discuss the study in more detail.

Working with the local research networks in all three regions provided the study team with the knowledge of which practices had experience of recruiting to studies within primary care and also highlighted practices to approach.

Each region recruited one or more study research nurses and a local site pharmacy. All sites, research nurses, site pharmacies and research centres received training on study procedures and protocols.

Participant selection

Children were eligible to join the study if they were consulting in a participating NHS site for their routine care, and met the following inclusion criteria and did not meet any of the exclusion criteria.

Participant recruitment

Participating clinicians were asked to identify suitable children during routine consultations. Some clinicians were aided in the identification of suitable children through the use of a software package called Trial Torrent (Tay Dynamic Ltd, Dundee, UK) (TT; see Trial Torrent recruitment software).

Trial interventions

The active medications being evaluated in this trial were well established and already widely used within their licensed indications. The active medications were not used outside their licensed indication during the course of this study.

Trial procedures

Data management and monitoring

Research governance

This study had clinical trials authorisation from the UK Competent Authority, MHRA reference 21323/0035/001-0001 and was reviewed as risk category type B. Ethical approval was granted from a NHS Research Ethics Committee. The initial approval was granted by the National Research Ethics Service South Wales Ethics Committee on 17 July 2012, reference # 12/WA/0180. NHS R&D approval was sought from the respective NHS relevant organisations in Wales, England and Scotland.

The trial was assigned European Union Drug Regulating Authorities Clinical Trials (EudraCT) number 2011-003591-37 and the International Standard Randomised Controlled Trial Number (ISRCTN) 96705420.

Outcome measures

Statistical considerations

Analysis

Owing to lower than expected recruitment rates the study was closed early, before the proposed reduced target sample size was reached. Because of this, the analysis focuses on estimating effect sizes and confidence intervals (CIs) rather than tests of significance, which would have been undertaken if the sample size had been achieved. Similarly, the cost-effectiveness analysis was replaced with a summary description of the NHS resource used and time off taken by the parent. This amended analysis plan was approved before database lock.

Patient and public involvement

In the CREAM study we were lucky enough to benefit from an extremely experienced and able patient and public involvement representative. Amanda Roberts has eczema herself and has had many years of experience caring for her children with eczema. She is also an experienced patient and public involvement representative, having sat on a number of bodies including the patient panel at the Centre for Evidence Based Dermatology. She has also set up a support group for parents of children with eczema (Nottingham Support Group for Carers of Children with Eczema) and through this has had direct contact with many carers of children with eczema. Amanda was a co-applicant and joined the study as a member of the Trial Management Group. The Trial Management Group also had members in professional roles that had personal experience of children with eczema.

Some of the key areas that Amanda made important contributions to included reviewing parent and child information sheets, providing feedback on the study protocol, logo and randomisation process, and providing guidance on strategies for successful recruitment. The views of a wider audience were canvassed through Twitter (Twitter, Inc., San Francisco, CA, USA) as appropriate, together with input from other Trial Management Group members with relevant experience from a patient or parent perspective. This proved invaluable, particularly in providing advice on how to best inform potential participants about the trial and recruitment strategies. Recruitment was considered throughout the conduct of the study, with discussions at Trial Management Group meetings and e-mail contact in the interim periods. The acceptability of the study design from a public and patient perspective was reflected in the positive feedback from participants and the follow-up rates achieved.

The study has also benefited from Amanda’s contribution during the analysis and dissemination of study results and will continue to contribute with dissemination activities.

Summary of changes to the study

The main changes to the protocol that occurred during the conduct of the study are summarised below.

The taking of photographs of the effected eczema area was deemed too great a burden on both the families and also technically difficult for research nurses to take good-quality photos during the baseline visit, therefore this was replaced with a features of infection CRF completed by research nurses.

Originally a week 1 visit by research nurses was included for collection of the study medication, this was deemed as an additional burden to families and a logistical challenge for the research nurse and this visit was removed from the protocol, and study medication was collected at the week 2 visit. A telephone call was made at week 1 to parents by research nurses to remind them to discontinue the study medication as the prescribed course had been completed and to continue completing the diary as instructed.

As recruitment was slow, the inclusion of dermatology clinics in addition to primary care sites was included. The original plan included a 12-month primary care notes review, this was changed to a notes review conducted at 3 months.

The original plan also included a secondary objective to assess the short- and long-term cost-effectiveness of treating suspected atopic eczema with oral or topical antibiotics, in terms of cost per unit reduction in subjective eczema severity. This objective was dropped when a decision was made to close the trial early. We have restricted our analyses to describing resource use and associated costs only.

Sites

A breakdown of the number of sites approached, expressing an interest, agreeing to participate and actively recruiting, for each of the four study centres are presented in Figure 2.

FIGURE 2.

Site recruitment flow diagram. EOI, expression of interest. a, Five or more recruits was the target for initiated sites. This was achieved by 12% of initiated sites (n = 11).

General practitioner practices and dermatology clinic sites were recruited between 9 July 2013 and 21 October 2014. A total of 147 sites (143 primary care and 4 dermatology clinics) expressed an interest in the study and 104 (71%) agreed to participate. Of the 95 sites that were initiated only 35% (32 GP sites and 1 dermatology clinic) actively recruited one or more participants into the study. Characteristics of recruiting and non-recruiting sites are given in Table 3.

An explanation of the problems experienced by sites that agreed to participate but did not progress any further, and for those that did not recruit any participants, is presented in Chapter 5.

Participants

Participants were recruited between 16 July 2013 and 28 November 2014. Of the 171 children referred to the study team, 113 (35 from Wales, 71 from England, 3 from Scotland and 4 from dermatology clinics) were randomised. The numbers referred and recruited, the reasons for not recruiting and the numbers followed up in each arm are given in the consort flow diagram (Figure 3). One participant failed to provide a complete set of baseline data (including baseline POEM score) and withdrew 4 days later. Three children were recorded as having a penicillin allergy and were therefore randomised to erythromycin or its placebo. However, none of these children were randomised to the active erythromycin arm (two children were randomised to a topical antibiotic and once child to the control), so no child actually received oral erythromycin. Eight other children withdrew (two with worsened eczema, one who was intolerant to study medication and five who were unable or unwilling to follow the study protocol). A further four children were lost to follow-up prior to the 2-week follow-up point, giving us a follow-up rate for our primary outcome of 89.4% (n = 101). We were able to follow up 98 (86.7%) and 74 (65.5%) participants at 4 weeks and 3 months, respectively, and were able to review the medical records (at 3 months) of 97 (85.8%) participants.

FIGURE 3.

Consolidated Standards of Reporting Trials flow diagram. a, Condition worsened; b, unwilling/unable to follow protocol; c, intolerant to study medication.

Primary outcomes

Mean baseline POEM scores were 13.42, 14.62 and 16.90 in the control (placebo/placebo), oral antibiotic (oral antibiotic/placebo topical) and topical antibiotic (topical antibiotic/oral placebo) groups, respectively (Table 7). These are all towards the upper end of the ‘moderately severe’ band of eczema severity. 39 By 2 weeks, the mean POEM score had reduced in all three groups (see Table 7). Patients with oral and topical antibiotics had non-significantly higher (worse severity) POEM scores at 2 weeks with mean differences from control of 1.52 (95% CI –1.35 to 4.40) and 1.49 (95% CI –1.55 to 4.53), respectively (see Table 7). The 95% CIs cross zero and so are compatible with harm, no effect or benefit. However, the lower bands of the CIs (–1.35 and –1.55) are less than the published MCID for POEM of 3.4. 50

Secondary outcomes

Subjective severity at 4 weeks and 3 months

The effects of oral and topical antibiotics on POEM scores at 4 weeks were –0.18 (95% CI –3.10 to 2.75) and 0.00 (95% CI –3.07 to 3.07), respectively (Figure 4 and see Table 7). These point estimates suggest no (or minimal) effect, and although the CIs are slightly wider than at 2 weeks (because of slightly lower numbers with follow-up data) they are still smaller than the MCID and therefore exclude a clinically meaningful benefit. The effect sizes for oral and topical antibiotics at 3 months [–0.21 (95% CI –3.12 to 2.70) and –1.13 (95% CI –4.32 to 2.06), respectively] are negative (consistent with small benefit) and, again, include the possibility of a harmful effect, no effect or a beneficial effect. The CI around the effect size for topical antibiotics is –4.32, which is greater than the MCID and therefore does not exclude the possibility of a clinically meaningful beneficial effect (see Table 7).

FIGURE 4.

Baseline, week 2, week 4 and 3-month POEM scores for each treatment group.

Daily symptom scores

Cronbach’s alpha for the eight symptom scores in the daily diary was 0.985, indicating good reliability between these symptoms and suggesting that they could be combined into a single symptom score. The factor analysis also suggested that these symptoms could be used to form a single scale (see Appendix 3).

Table 8 shows the correlation between total symptom scores (sum of the scores of each of eight symptoms) and the POEM score at baseline, 2 weeks and 4 weeks. The high correlations demonstrate the validity of using the total symptom score to measure eczema severity. Figure 5 shows a plot of the mean daily total symptom score for each group.

TABLE 8 - Total symptom score correlation

**Significance at 0.01 level.

Time point	Pearson’s correlation	Spearman’s correlation
Baseline	0.57**	0.53**
2 weeks	0.61**	0.65**
4 weeks	0.72**	0.73**

FIGURE 5.

Daily total symptom score for each treatment group over 4 weeks.

The mean (95% CI) of the area under the curve by treatment groups, which measures the accumulated daily symptom scores during the first 4 weeks of treatment, is shown in Table 9, and demonstrates that symptom recovery was similar in all three groups.

TABLE 9 - Area under the curve by treatment group for daily total symptom score

Control		Oral antibiotic		Topical antibiotic		Overall
n	Mean (95% CI)	n	Mean (95% CI)	n	Mean (95% CI)	n	Mean (95% CI)
32	225.14 (–135.66 to 585.95)	33	215.55 (–90.56 to 521.66)	29	225.03 (–16.38 to 466.45)	94	221.74 (–83.65 to 521.73)

Sensitivity analyses, process measures and post-hoc analyses

Introduction

NHS resources are always finite and often scarce, whereas demand on these resources can seem to be unlimited and is certainly far in excess of the budget allocation in the UK. Health economists have developed instruments to help identify the optimal decision around the distribution of these finite resources – quality-adjusted life-years (QALYs) is one such measure. 54 Cost–utility analysis is the economic evaluation of alternative treatment options assessed in terms of costs per QALY and, as such, the QALY provides a common metric to compare different treatments for the same condition and also across conditions. In the UK, the National Institute for Health and Care Excellence (NICE) recommends the use of QALYs as a measure of health benefits and the use of the generic preference-based health-related quality of life (HRQoL) measure European Quality of Life-5 Dimensions (EQ-5D)55 to determine health status. 56

National Institute for Health and Care Excellence guidelines recognise that there might be population subgroups (i.e. children) for which using EQ-5D to derive QALYs might not be appropriate ‘when EQ-5D is not appropriate for the condition or effects of treatment, the valuation methods should be fully described and comparable to those used for EQ-5D’ (i.e. preference-based HRQoL measure). 56 In relation to paediatric cost–utility analyses, 2013 NICE guidelines56 state that ‘when necessary, consideration should be given to alternative standardised and validated preference-based measures of health-related quality of life that have been designed specifically for use in children’. Nonetheless, the guidelines do not advocate the use of any specific instrument, leaving practitioners with no definite guidance.

In 2014 Adlard et al. 57 carried out a systematic review to assess within the UK how paediatric cost–utility analyses have evolved over time and the extent to which practice adhered to NICE guidelines. The authors identified 43 papers. Most used adult measures and none used children-specific instruments such as Child Health Utility Instrument – 9 Dimensions. 58 More recently, Thorrington and Eames59 carried out a systematic review to assess the methods used by researchers and health economists to estimate health utilities for children and adolescents. They concluded that researchers were left to use instruments not specifically designed for children because specific instruments are still being developed and validated. In relation to very young children and toddlers, Griebsch et al. 60 concluded from their systematic review that there was a lack of health-state classification instruments for use in children and infants who are younger than 5 years, and there is a need to understand the role of proxies for measuring and valuing child health.

The aim of this work is to use the children recruited into the CREAM study to validate the condition-specific preference-based HRQoL measure for children with eczema developed by Steven et al. ,44 known as the ADQoL, and from which QALYs can be generated. The ADQoL includes four dimensions and each dimension is covered by two options: (a) activities (‘Your child cannot join in some activities with other children’, ‘Your child is not limited in activities with other children’); (b) mood (‘Your child is very moody’, ‘Your child is not very moody’); (c) being settled (‘Your child cannot be comforted’, ‘Your child is quite settled’); and (d) sleep pattern (‘Your child sleeps badly most nights’, ‘Generally, your child sleeps well’) (see Appendix 2).

The ADQoL (a preference-based index) uses the parent as proxy to assess the quality of life of the child and there is no age limit to the child. In the CREAM study the majority of the children and toddlers are 5 years old or younger. This cohort, therefore, produced a unique sample for this validation exercise. Testing the construct validity of the instrument constitutes an important step towards the wider implementation of this preference-based outcome measure. To the best of our knowledge this instrument has never been applied to a trial population.

Methods

Results

Table 25 describes the available data collected at baseline and follow-up time points at 2 weeks, 4 weeks and 3 months.

TABLE 25 - Answers provided and missing data for each question

Questionnaire Item	Baseline, n (%)	2 weeks, n (%)	4 weeks, n (%)	3 months, n (%)
Joining in activities with other children
Your child is not limited in joining in activities with other children	87 (77.0)	85 (75.2)	91 (80.5)	65 (57.5)
Your child cannot join in some activities with other children	23 (20.4)	15 (13.3)	7 (6.2)	7 (6.2)
Missing	3 (2.7)	13 (11.5)	15 (13.3)	41 (36.3)
Your child is very moody/your child is not very moody
Your child is not very moody	76 (67.3)	87 (77.0)	79 (69.9)	61 (54.0)
Your child is very moody	34 (30.1)	13 (11.5)	18 (15.9)	10 (8.8)
Missing	3 (2.7)	13 (11.5)	16 (14.2)	42 (37.2)
Your child cannot be comforted/your child is quite settled
Your child is quite settled	91 (80.5)	93 (82.3)	87 (77.0)	66 (58.4)
Your child cannot be comforted	19 (16.8)	7 (6.2)	11 (9.7)	5 (4.4)
Missing	3 (2.7)	13 (11.5)	15 (13.3)	42 (37.2)
Your child sleeps badly most nights/generally, your child sleeps very well
Generally, your child sleeps very well	57 (50.4)	73 (64.6)	74 (65.5)	56 (49.6)
Your child sleeps badly most nights	53 (46.9)	27 (23.9)	24 (21.2)	16 (14.2)
Missing	3 (2.7)	13 (11.5)	15 (13.3)	41 (36.3)

The children with missing data were not dissimilar from the children with available data in terms of age but carers of girls were less likely to answer the questionnaire (Table 26).

TABLE 26 - Missing data according to child’s sex

Missing data	Female, n (%)	Male, n (%)
Baseline	3 (5.8)	0 (0.0)
2 weeks	11 (21.2)	2 (3.3)
4 weeks	13 (25.0)	3 (4.9)
3 months	28 (53.8)	14 (23.0)

The developers of the instrument did not design a set of rules for imputing missing data and as a result it was decided that we would not impute any missing data for the purpose of this validation exercise.

The box plot graph (Figure 6) shows the level of dispersion of the data. Most observations are close to the maximum value possible for the health state with a series of observations tailing towards the minimum possible value for the health state (i.e. negative skewed distribution).

FIGURE 6.

Box plot of ADQoL at each time point.

Table 27 lists the key summary statistics of the ADQoL scores at each data collection point, minimum and maximum reflects the possible range of values for the measure.

TABLE 27 - Description of the ADQoL scores at each time point

IQR, interquartile range.

ADQoL scores	n	Minimum	Maximum	Mean	SD	Median	IQR
Baseline	110	0.36	0.84	0.7268	0.11	0.7440	0.19
2 weeks	100	0.36	0.84	0.7805	0.09	0.8410	0.10
4 weeks	97	0.49	0.84	0.7848	0.08	0.8410	0.10
3 months	71	0.36	0.84	0.7889	0.10	0.8410	0.10

Change of preference-based Atopic Dermatitis Quality of Life Index scores over time

Figure 7 shows the change of the preference-based scores at each time point and, as expected, the health scores suggest that the biggest improvement to the child’s health takes place between baseline and 2 weeks. This is reflected in the sensitiveness of the preference-based outcome measured by the SRM (Table 32).

FIGURE 7.

Preference-based ADQoL scores change over time.

TABLE 32 - Standardised response mean at each time window

Time window	Mean difference	SD	SRM
Baseline to 2 weeks	0.0525	0.1083	0.485
2 weeks to 4 weeks	–0.0022	0.0848	0.026
4 weeks to 3 months	0.0062	0.0939	0.066

The SRM suggests a low to moderate response for the first change and very low sensitiveness to the changes after 2 weeks. Although given the lack of evidence of any ‘real’ change on other measures after 2 weeks it is unlikely that there is any change occurring for the SRM to detect.

Discussion

Preference-based outcome measures are key to helping policy-makers support the implementation of the health-care interventions that are most valued by the population. In the UK the EQ-5D is the generic preference-based HRQoL measure accepted by NICE; however, this generic measure might not be sensitive enough to capture the effect on quality of life for specific conditions (e.g. eczema) and/or specific patient groups (e.g. very young children). Steven et al. 44 developed a condition-specific preference-based HRQoL measure for children with atopic dermatitis, which is capable of generating QALYs and, as far as we are aware, has not been previously validated alongside a randomised controlled trial. The aim of this study was to validate this specific preference-based HRQoL (ADQoL preference-based index) on the patients recruited in the CREAM study.

Non-response to questionnaires and individual items are often used as rough indicators of the ease of completion of the questionnaire. In the CREAM study only one participant did not fully complete the 3-month questionnaire and the cohort of the non-responders to the questionnaire overlapped almost completely with the participants who did not complete the other questionnaires. This would suggest that the likelihood of answering the questionnaire was not linked to the questionnaire itself. Furthermore, only 9 (12%) participants reported that the questionnaire was somewhat difficult or very difficult to complete. However, although most participants found the questionnaire reasonably easy to complete, 17 participants (15%) thought that the questionnaire was not fully describing the effect of the condition on their child’s life. Not surprisingly, those who found that the questionnaire was not fully capturing their child’s quality of life tended to find it more difficult to answer. Reasons reported for this included the young age of the child, the absence of important health dimensions and the lack of enough levels attached to each health dimension.

Correlations with health outcome measures used in the study were significant and in the right direction and of moderate strength. The instrument showed good discriminate validity at 2 weeks and similarly the sensitivity to change was moderate for the change between baseline and 2 weeks.

The instrument was designed to allow parents of older children to directly ask the questions to the child and for younger children to act as a proxy. As the CREAM study mainly recruited children younger than 4 years, most questionnaires were completed by the parent acting as proxy. This did not allow separation of the performance of the instrument between these two groups.

Conclusions

If preference-based outcome measures are to be used to inform policy decisions it is paramount that they are validated and reflect the true values that the population attaches to the health states of a patient and that any change in the health status is captured. Our evaluation of the construct validity and performance characteristics of this preference-based HRQoL measure are encouraging; however, some of the respondents felt the questionnaire was not relevant to them because the child was too young, whereas others commented that each dimension could have benefited from additional response options to accommodate health status in between those currently presented in the questionnaire.

Manufacture and supply of trial treatments

Significant problems were encountered in sourcing the pharmaceutical products required for the study. Extensive negotiations with a number of pharmaceutical manufacturers and suppliers were required to secure the supply of the required oral and topical active antibiotics, and matched oral and topical placebo products, with sufficient assurances in terms of liability protection and provision of necessary good manufacturing practice documentation (summary product characteristics, IMP dossier, stability data). This ultimately required contracting with a pharmaceutical service provider for the supply of the matched placebo antibiotic suspensions (manufactured by a third party) and a separate pharmaceutical manufacturing unit for the supply of the active antibiotic suspensions, the active and placebo antibiotic creams, and the supply and packaging of the trial packs. Unfortunately this solution, including the protracted contracting process, the process for matching taste and appearance of the oral placebo product and required pharmaceutical stability testing, resulted in significant time delays. The additional costs for contracting with a number of pharmaceutical providers, along with changes in the market value of the active antibiotics and relatively short shelf life of the pharmaceuticals, resulted in an increase in the cost of pharmaceuticals that exceeded the original funded costs.

Contracts and research and development approvals

Initial delays in contracting processes had a knock-on effect that prolonged the set-up phase of the study and ultimately significantly delayed the start of site and participant recruitment. This impacted on other processes, such as NHS approvals as GP and pharmacy agreements need to be reviewed by NHS R&D officers prior to issuing local approvals.

The delays in obtaining NHS approvals meant that recruitment of sites and pharmacies could not progress, and therefore the initiation and training required before the study could open to recruitment of participants was delayed. The study involved 91 primary care sites and four dermatology clinic sites across England, Wales and Scotland, which required NHS approvals through three systems for co-ordinating permissions (Central Permissions Unit, National Institute of Social Care and Health Research – Permissions Coordinating Unit and NHS Research Scotland respectively). Delays were encountered due to multiple submissions of documentation and local checks being carried out at various stages and not necessarily in parallel to the global checks initially undertaken by the National Institute of Social Care and Health Research – Permissions Coordinating Unit. Responding to queries from multiple individual R&D offices, often using a variety of different processes, rather than a single centralised body or uniform system, was a significant burden on staff time during this period.

Delays in obtaining local R&D approvals were compounded by problems in obtaining required documentation from, and arranging visits with, sites. In some cases sites reported that the delays had had an effect on motivation. The impact of time taken to gain statutory approvals for research sites has previously been reported. 64,65 Despite the large number of sites recruited the majority did not recruit any participants.

The study team also encountered some difficulties in obtaining excess treatment costs in England. There are differences in the assessment and handling of excess treatment costs between England and Wales, and R&D offices in different regions in England did not have a uniform approach to the assessment or handling of excess treatment costs. This caused delay in getting approvals for excess treatment costs, and resulted in delays in recovering costs for the active treatments.

Recruitment of sites

A number of challenges were experienced in the recruitment of primary care sites. A key decision in the design of the trial had been to have research nurses visit potential participants at home, in order to minimise the burden for GP practices. However, as a result, we were limited to practices that were within travelling distance of research nurse support. This was a major challenge, both in terms of recruiting our original cohort of practices and attempts to expand once it was clear that recruitment was below target. Another key factor was distance from pharmacy sites. Our initial plan had been for GP practices to store and dispense IMP; however, given the need for reconstitution and temperature monitoring the study sponsor (Cardiff University) assessed the risk as being too great, and therefore we contracted a hospital pharmacy site in each region to store and dispense IMP. However, this led to further challenges for the study team as the research nurses’ travel time was increased by having to attend a hospital pharmacy before visiting a potential participant and, again, it made it more challenging to try expanding the study to other regions. Alternatives to dispensing by hospital pharmacy departments were considered at various times during implementation of the study, but were deemed unfeasible due to the limitations for controlled temperature storage, reconstitution of IMP and the short shelf life of IMP once reconstituted (7 days) that required same day delivery to study participants. There is a growing role for the involvement of community pharmacies in research (Royal Pharmaceutical Research Ready Accreditation Scheme – URL: www.rpharms.com/research-ready/benefits-of-research-ready.asp). However, there were no participating pharmacies that had appropriate good clinical practice and blinded IMP monitoring or accountability processes to undertake the study in the required geographical locations. Therefore, hospital clinical trials pharmacies, who generally experience high volumes of workload and long waiting times for dispensing, were required to dispense study medication.

Delays in recruitment of sites, due to the problems described above as well as other practice demands (e.g. Quality and Outcomes Framework, holiday seasons, etc.) meant that we did not have a full complement of practices (we had 60 out of a target of 90) when the study opened to recruitment in July 2013. Further practices were recruited during the study, with a total of 90 practices opened to recruitment. The number of practices open at any given time fluctuated in response to a temporary lack of research nurse support in the South East England network, and the decision to close the network in Scotland. Overall, the number of practices open to recruitment was not clearly associated with participant recruitment rates.

After 6 months of participant recruitment the number of children recruited was significantly below target and therefore a number of strategies were introduced to try and improve recruitment.

Actions taken to improve recruitment at site level

Incidence of infected eczema

Feedback from participating sites was that the key challenge to recruitment was that eligible patients were not consulting very frequently. Prior to study funding, an assessment was conducted of recruitment potential by analysing data from 70 general practices in Scotland. The incidence of eczema in children aged < 5 years (the age range of children set in the commissioning brief) during a 1-year period was calculated. Then an estimate of the incidence of ‘infected eczema’ was obtained by calculating the proportion of these patients with eczema who had received a prescription for an antibiotic typically used to treat skin infections (flucloxacillin or a topical antibiotic). Using these data it was estimated that an average-sized practice (6000 patient list size) would see one or two patients per month with infected eczema. It was therefore estimated that the study would be able to meet recruitment targets in a 12-month period with 90 GP practices. This was supported by a reported 19% increase in primary care consultations for skin conditions in children and a 64% increase in prescription of flucloxacillin (used for S. aureus infections) between 1997 and 2006. 66

In response to below-anticipated recruitment and the feedback received from participating practices, we conducted some further assessments of the incidence of children presenting with infected eczema in primary care. Practices were asked to conduct a review of their medical notes using eczema-specific Read Codes, and the reported incidence was lower than anticipated.

Infected eczema represents a common and important problem across the UK, but a relatively low incidence makes it difficult for practices to identify eligible children. General practices are busy places with many service delivery demands that vary from day to day and month to month. Practices may well have had every intention of approaching all eligible children, but if they were only presenting once every month or two then it would have been harder to stay alert to the study. In addition, many practices tried to filter potential research participants to one or two clinicians who were involved in the study. With a low-frequency condition this was more difficult and patients were more likely to be missed.

We found considerable practice-level variation in coding for eczema. Practices were asked to send letters to children on their list with a Read Code for eczema in their record (using a standardised set of Read Codes), and we found significant variation in the number of letters sent out, despite relatively similar list sizes. There were also differences between practices in terms of the age of the patient populations that they served. A number of practices had a smaller than average proportion of children registered with the practice and served a local population that was relatively elderly. Some GPs reported that children with a history of recurrent infected eczema may receive antibiotics on repeat (or repeat acute) prescription so would rarely present at the surgery to be recruited and were likely to be excluded due to recent use of antimicrobial agents if they were seen in the practice. As GPs did not reliably record details of potentially eligible children screened/reviewed in their practices, there are no data regarding children who consulted their GP for infected eczema and were not referred because they were ineligible at the time of consultation. Some GPs reported they were more reluctant to prescribe antibiotics in recent years because of their concerns about antimicrobial resistance and general public awareness of antibiotic stewardship campaigns. GPs also reported that the incidence of infected eczema might be lower than previously reported due to improved eczema care such as prescription and use of emollients.

Practices with GPs with a special interest in dermatology were more likely to agree to participate in the study, but were more likely to provide an enhanced level of dermatology care overall. It is possible this may have resulted in better treatment and adherence, and therefore patients at these practices may have experienced fewer eczema flares, and/or less use of antibiotics than patients from other practices with lower levels of dermatology expertise.

Defining eligibility

There is a well-described relationship between the isolation and culture of S. aureus on the skin and the presence of eczema. The organism can be isolated from over 90% of atopic eczema lesions. 21,67–69 The presence of bacteria, particularly staphylococcus expressing exotoxins, has been shown to be associated with more severe eczema,70–72 and higher concentrations of S. aureus are associated with increasing clinical severity. 73 Therefore, it might seem reasonable to assume that (if not all) flares of eczema are probably infected. However, there is no clear, widely accepted definition for infected eczema, and guidelines often refer to more overt features such as pain, oozing, crusting, asymmetry, extensive disease, fever and malaise. 74 The presence of S. aureus on eczematous lesions does not prove a causal relationship, or that treating S. aureus will improve eczema – hence the need for the study.

We took the pragmatic approach of including children with eczema (as defined by U.K. Working Party33) who were suspected of having infected eczema by their treating clinician. This could include where:

• the eczema is failing to respond to standard treatment with emollients and/or mild–moderate TCSs

• there is a flare in the severity or extent of the eczema

• there is weeping or crusting.

However, despite lack of evidence about which children will benefit from antibiotic treatment, some clinicians reported during qualitative interviews, written responses to questionnaires and informal discussion that they felt uncomfortable with entering all children with ‘suspected infection’ into the trial. A number of the clinicians reported that it was unethical to include children with more pronounced signs suggestive of infection into a trial where they may not receive any antimicrobial treatment, and others reporting unease at including children with a flare but no specific signs of infection into a trial where they may receive systemic antibiotics.

Discussions with GPs participating in the trial attempted to tackle these concerns about equipoise by highlighting the current lack of evidence to guide treatment. The definition of an eczema ‘flare’ is, in itself, subject to extensive variation and lacking in validity. 17

Nevertheless, for some clinicians there were a limited number of patients with eczema flares who they would consider entering into the study, which limited potential recruitment.

Attrition between identification and recruitment

Of patients referred to the study by participating clinicians, 33% (58/171) were not recruited. Of these, 31% (18/58) were identified by the research nurse as not meeting the study inclusion criteria, and 21% (12/58) decided that they did not wish to participate, sometimes because their child’s condition had improved by the time they were seen by the research nurse, but more frequently because parents did not have a complete understanding of the requirements of the study when they consented to the referral being made. One-quarter of those who did not wish to take part cited antibiotic or placebo use.

Another 48% (28/58) were not recruited because of difficulties in arranging a baseline visit within the study parameters. This included cases where parents were either not contactable or were unavailable for a baseline visit (22%, 13/58). Reasons stated for unavailability of the parent included work commitments and the child being in school or childcare during the period for a baseline visit. Reasons for unavailability for the research nurse to arrange the visit within the study parameters were conflict between study appointments, or the time required for dispensing and collection of IMP prior to the visit (14%, 8/58). Twelve per cent (7/58) were not recruited because of a lack of time or resources, but the exact reasons were not adequately documented. In response to the challenges experienced in trying to recruit within tight time frames, a decision was taken by the study team (and agreed by the funder and Trial Steering Committee) to increase the maximum time allowed between referral and baseline assessment from 48 (usually 24 hours) to 72 hours. Although the research nurses endeavoured to contact parents at the earliest opportunity, it was left to GP discretion whether or not it was considered safe/appropriate to refer a child on a Friday, knowing they would not be seen by a research nurse over the weekend. Difficulties remained if the child was of school age, and well enough to attend school, due to the length of time needed for a baseline visit (up to 2 hours) and the need to assess the child’s skin and take swabs during the visit.

In retrospect, the decision to use research nurses to recruit participants in the days following identification in general practice may not have been the most effective strategy. Although it helped to address the issue of general practice clinical staff not having time or flexibility to be able to conduct an unplanned 2-hour baseline assessment, it resulted in loss between identification and recruitment, deterred some parents from participating and led to difficulties in expanding the study. It would not have been possible to conduct the baseline appointment within most GP surgeries, given the time required to complete all the assessments. A scaled down appointment might have been possible and, if combined with local GP storage of trial treatments, dispensed to parents at the practice, this may have enabled us to roll out the study on a much wider scale.

Clinicians and research nurses reported that parent preference either for antibiotic use or against antibiotic use was an important factor in parents’ decisions about participation. Four parents indicated that they did not want their child to participate because they did not want their child to receive antibiotics and one parent indicated that they did not want their child to participate because they wanted antibiotic treatment and did not want to risk receiving placebo. It has been reported that there has been a shift from parents’ expectation and demand for antibiotics when their child is unwell to concerns regarding antibiotic resistance related to unnecessary prescribing. 75 This may be a result of awareness campaigns about the links between antimicrobial use and antimicrobial resistance. Future research to evaluate parental treatment preferences in terms of the perceived efficacy of antibiotics, adherence to standard eczema care and impact on quality of life may be warranted.

The study protocol required that all children referred into the study to be prescribed a mild and/or moderate potency TCS for application to affected areas on the face and body respectively, as part of standard eczema care. This was to be used concurrently with trial medication for 1 week and continued for a second week after the trial medication had stopped. A number of referrals were received where prescriptions for concurrent moderate corticosteroid medications were absent. In most cases the prescriber was contacted and a prescription for moderately potent corticosteroids obtained. Some GPs reported hesitation in prescribing moderately potent corticosteroids for infants; however, it is not known if this had any impact on recruitment.

Amendment of inclusion and exclusion criteria

In addition to supporting clinicians to maintain an awareness and interest in the study, and to continue to refer potentially eligible children, amendments were made to the inclusion and exclusion criteria in order to try and increase recruitment potential while maintaining the validity of the research question.

The eligible age range was increased from 0–5 years (as set in the commissioning brief) to 0–8 years. This was done prior to opening to recruitment and with full consent from the study funders.

The inclusion criteria were simplified in November 2013 by removing renal or hepatic impairment as an excluding factor, allowing more potent topical steroids to have been used up until 2 days prior to recruitment and the previous use of antibiotics changed from any ‘use of oral antibiotics’ to only ‘use of antibiotics for a skin infection’ as an exclusion criterion.

Recruitment barriers at site

Recruitment and retention in trials in the demanding world of primary care is extremely challenging. 76 Unique barriers to recruitment were encountered during this study.

Enabling factors

Strategies employed to try and improve recruitment

A number of strategies were used to try and improve recruitment. Regular communication with practices and principal investigators was essential. Communication strategies included monthly CREAM study newsletters, and frequent practice visits by members of the research team, with a particular aim of raising awareness among the reception staff. Practices that took a co-ordinated approach to actively identifying potential participants were generally more successful at recruiting than those with a single GP participating without support from the wider practice team.

As part of the study set-up, GP practices sent information about the study to those identified on a search of patient records as aged < 8 years with a history of consultation for eczema. Where practices had not recruited a participant for a long period of time, practices were encouraged to rescreen and resend out the information to parents of children with eczema, as there may have been new children identified since the original mailshot and it would also raise awareness about the study where this had lapsed since the previous letter. This resulted in a small number of additional potentially eligible children being identified.

When contacting practices to chase prescription faxes or to verify patient contact information, it became apparent that reception staff in some practices were not familiar with the study. A laminated prompt for reception staff to position next to their booking-in screens was created and distributed to practices by research nurses when visiting for notes reviews. The simple visual aid was intended to raise awareness and encourage everyone to get involved in the study. The prompt sheet encouraged reception staff to allocate potentially eligible patients to doctors or nurses who were involved in the study.

We consider that the most effective strategy was researcher visits to the GP practices. Members of the research team contacted and visited GP practices where there had been lower levels of eligible children reported or low levels of recruitment activity identified. It was often reported at this contact that few eligible children had been seen during recent consultations. However, this contact, while time and resource intensive, raised the profile of the study in the practice and identified staff turnover at some sites, such as practice managers. This provided opportunities to inform new practice personnel about the study and provide study-specific training as required, and some visits were followed by a subsequent increase in recruitment.

Conclusion

The CREAM study experienced delays in set-up, followed by below anticipated recruitment rates. Set-up delays were mainly due to problems with IMP procurement and site recruitment and approvals, and the main factors contributing to slow recruitment appear to be a low rate of infected eczema, problems in defining infection and clinicians and parents having equipoise, problems integrating the study into busy general practice and loss of potential participants before they could be seen by a study nurse. Many strategies were implemented to try and simplify the study and improve recruitment rates; however, the design of the study did not allow for easy expansion to other sites, and the low numbers made it very difficult for practices to remain alert and ‘catch’ potential participants when they did present.

Summary of main results

The CREAM study randomised 113 children and over 90% had one or more ‘classical’ signs or symptoms of infection, and approximately 70% had S. aureus cultured from a skin swab. Treatment with oral and topical antibiotics was associated with non-significantly higher (worse severity) POEM scores at 2 weeks than placebo, after controlling for baseline severity. The CIs included zero, but the lower bands were substantially less than the published MCID, suggesting that these interventions are either associated with no effect or harmful effect, while ruling out a beneficial effect. POEM scores at 4 weeks and 3 months, and objective eczema severity (EASI), family impact (DFI), quality of life (IDQoL and CDLQI) and daily symptom scores all showed small effect sizes with 95% CIs that included zero. Adherence to oral antibiotic (or placebo) was moderate, with overall average adherence at 61.3% of treatment doses and adherence in the oral antibiotic group at 70.4%. Adherence to topical antibiotic was better, with 81.8% of applications being applied overall (80.8% in the topical antibiotic group). Sensitivity analyses, including adjusting for compliance and missing data and a subgroup analysis with those who had S. aureus cultured from skin swabs, were all consistent with the main findings. Flucloxacillin and erythromycin non-susceptibility in isolated pathogens was low at baseline and follow-up. However, 27% of S. aureus skin isolates were resistant to fusidic acid at baseline, and at 2 weeks this increased to 31% overall and 73% in the topical antibiotic group. By 3 months non-susceptibility rates had reverted to 15% overall and 25% in the topical antibiotic group. However, the numbers were small and the differences not statistically significant.

Strengths and limitations

The CREAM study is the largest trial to evaluate the effectiveness of oral and topical antibiotic treatment for clinically infected eczema in children, and the only trial to be conducted in primary care, where most people with eczema are treated. We used pragmatic inclusion criteria, based around clinical suspicion of infection, and interventions that are commonly used in routine clinical practice (flucloxacillin and fusidic acid are the most common oral and topical antibiotics prescribed for skin infections in the UK and are commonly used around the world). Randomisation was conducted independently, we used matched placebos and there was no indication of breaches in allocation concealment. All participants were prescribed TCSs and we measured use of all medications, including antibiotics, TCSs and other medications for eczema. In addition, we measured quality of life, impact on family, symptoms and skin pathogens. We found no evidence of differential use of medications, including TCSs, and a CACE analysis that adjusted for compliance did not alter our findings. Blinded outcome assessors used well-validated instruments to assess subjective and objective eczema severity at baseline and at follow-up. We achieved high rates of follow-up. Our primary analysis was by ITT. Analysis of longer-term follow-up and analyses controlling for missing data were all consistent with our main findings of no additional clinically important benefit from either topical or oral antibiotics.

The main weakness is that we failed to reach our recruitment target, and this limits the power of the study, particularly for longer-term follow-up and for subgroup analyses. Nevertheless, the 95% CIs around our primary outcome, difference in POEM score at 2 weeks, exclude a clinically meaningful benefit from either intervention. Other secondary findings are largely consistent with the estimate of our primary end-point analysis.

Almost a third of the children from whom we cultured S. aureus in the topical antibiotic group (but only 16% of the whole group) had organisms that were resistant to fusidic acid, and therefore this could be postulated as a reason for the lack of effectiveness of antibiotic treatments. However, resistance to flucloxacillin was only identified in one child in the oral antibiotic group (3.3% of those with S. aureus and 2.8% of all children) and yet there was a similar lack of effect in this group. Although we found no differential rate of TCSs between groups, we have few data about how use of TCSs changed following inclusion into the study or whether or not TCS use in this population that had relatively intensive education differs from that found in normal clinical practice. Nevertheless, the underlying message that antibiotics do not confer benefit in addition to optimal TCS treatment remains.

Generalisability

Although we used broad inclusion criteria and, in terms of infection severity, only excluded children with ‘severe infection’ or who, ‘required immediate admission to hospital’, our research staff received anecdotal reports from clinicians that for some patients with ‘clear infection’, either they or the child’s parent felt unhappy about including them in a trial where they risked receiving no antibiotic treatment. Most clinicians did not accurately record recruitment logs and therefore it is difficult to quantify the number of children falling into this category. Nevertheless, all children included in the study were deemed to have suspected infected eczema and over 90% had ‘classical’ signs of infection (weeping, crusting, pustules or painful skin). Significant improvement in outcome measures (symptom scores, POEM and EASI) during the follow-up period in all three groups indicates that this was a population experiencing a flare in their eczema. We identified S. aureus from the skin of 70% of participants, which is lower than the proportion reported in some studies of patients with eczema and without a clinical diagnosis of infection. 21 However, these studies are primarily based in secondary care settings, used more complicated procedures for collecting samples (e.g. adhesion of Film Stamp Check) and immediately processed samples, while our study relied on swabs being taken in patients’ homes by research nurses and sent in the post to the laboratory. Therefore, we may have failed to identify some pathogens because of the sampling procedure and transportation processes. The proportion of children in whom S. aureus was cultured from the skin decreased from baseline to follow-up in all three groups, which also suggests resolution regardless of antibiotic treatment.

Nevertheless, there was a strong feeling among clinicians and parents that there were some clinical presentations that should be treated with antibiotics, or at least a belief that the risks of no treatment were so great that including them in a trial where they might be randomised to no antibiotic treatment would be unethical. Therefore, there were children who had ‘clear evidence of infection’ who were not included in the trial, and it was difficult to clearly define this boundary. We have described our population as clearly as possible, and cannot draw conclusions about the likely effects in children with more extensive and severe disease than those included in the trial.

The participants in our study all received a visit from a nurse who had training in eczema care and a booklet with standardised advice about eczema care. As such, we have compared ‘good advice about eczema care’ with ‘good advice about eczema care’ plus antibiotics, and our results may not be generalisable to a comparison with standard eczema care in primary care. Nevertheless, all patients should be receiving good advice about eczema care, and our results suggest that, in the population studied, antibiotic treatment does not confer meaningful benefit over this standard level of care.

More than two-thirds of our participants were recruited by one centre (Bristol) and only one in three research sites (practices or dermatology clinics) that were initiated actually recruited any participants, and this may have affected generalisability. However, we have no reason to believe that patients in Scotland or Wales are likely to respond differently from those in the Bristol area, and the characteristics of recruiting and non-recruiting sites were similar.

Interpretation

The CIs for our primary effect sizes suggest that both oral and topical antibiotics may be associated with no effect or a small harmful effect, but are not associated with clinically important benefit. We conclude this as the lower bands of the 95% CIs around our estimates of the POEM scores are less than the published MCID of 3.4. This MCID was derived in a secondary care population with higher baseline POEM scores, and may not be relevant to a primary care setting like this. A MCID for primary care has not formally been established, but a recent primary care trial, in which children had lower baseline POEM scores than in the CREAM trial, assumed a MCID of 2,80 and a recent secondary care trial has assumed a MCID of 3. 81 However, even if we adopt the most conservative MCID of 2, our results still suggest no clinically important benefit at 2 weeks from either oral or topical antibiotics for clinically suspected infected flares of eczema in children. The 95% CIs for the effect sizes at 4 weeks and 3 months do include the possibility of a POEM score that is lower (beneficial) in both intervention arms by more than 3, and the estimate of effect of oral antibiotic at 3 months on the POEM score had a 95% CI that went down to –4.3. However, the point estimates for these effects were close to zero and it is likely that the lower bands included potential beneficial effect only because of the smaller sample sizes at these points. It seems unlikely that antibiotics would have no beneficial (and potentially harmful) effects during the week of treatment but then be beneficial later on.

Although, overall, resistance to antibiotics in S. aureus isolated from eczema lesions was low and had mostly returned to baseline by 3 months, we did find an increased risk of carriage of fusidic acid-resistant S. aureus from treatment with topical fusidic acid. Effect on resistance requires further assessment in larger studies.

The results of our study, the largest and most rigorous study of antibiotics for clinically infected eczema in children, are largely in keeping with previously published research in this area. Only one previous study has assessed the effects of antibiotics in children (not combined with adults) with clinically infected eczema. 82 Weinberg et al. 82 showed that oral cefadroxil (a first-generation cephalosporin) was associated with no statistically significant difference in eczema activity (measured using the Hanifin/Rajka activity scale) at follow-up when compared with placebo in a trial of 33 children (aged 6 months to 12 years) with ‘S. aureus superinfected atopic dermatitis’. Both groups showed improved eczema severity scores at follow-up, and there were similar within-group improvement but lack of between-group differences for eczema signs and symptoms (pruritus, erythema, peeling, lichenification, induration and ulceration). The only between-group differences were the proportion with ‘positive’ culture results at follow-up [4/13 (30.8%) in the cefadroxil group and 14/17 (82.4%) in the placebo group; p < 0.0001] and the proportion with, ‘clinically apparent superinfection’ at follow-up [0/13 in the cefadroxil group and 9/17 (52.9%) in the placebo group]. However, this trial is at high risk of bias because of a lack of data about blinding, no clearly defined primary outcome, small sample size and no ITT analysis (three patients in the cefadroxil group were withdrawn because of side effects, non-compliance and presence of a resistant organism). The only other trial of antibiotics for clinically infected eczema that included children (as well as adults and with no subgroup analysis for just the children) was a three-arm randomised controlled trial comparing topical betamethasone combined with gentamycin, topical betamethasone alone, or topical gentamycin alone. 83 In this trial, 83 patients were enrolled and 79 were included in the analysis. No indication is given of the number of children included or the average age of participants. Investigator-graded severity improved in all three groups, but with significantly better final (day 22) total score in the betamethasone combined with gentamycin group when compared with either agent on its own. The study was reported to be ‘double-blind’, but no information was given about allocation concealment or matching of treatments. Given the subjective nature of the outcomes measured, and no pre-specified primary outcome, the risk of bias is also high.

Three studies have evaluated the effect of oral antibiotics in children without clinically infected eczema84,85 or in a mixed group that included both infected and uninfected and only excluded severe infection. 86 A parallel group randomised controlled trial84 and a cross-over study85 evaluated flucloxacillin and cefuroxime, respectively, and both used placebo as a comparator, and the final study used a before-and-after design to evaluate cloxacillin or erythromycin (depending on the sensitivity of the organisms identified) and made within-group comparisons over time. 86 The only one of these studies to report clinical improvement associated with the use of antibiotic was Boguniewicz et al. ’s non-blinded before-and-after study. 85 Neither of the randomised controlled trials found a significant clinical benefit associated with antibiotic treatments. Indeed, the trial of flucloxacillin found worse clinical outcomes (pooled efficacy scores) in the intervention group, which is similar to findings from our study. 84

Low levels of adherence to flucloxacillin could be a reason for the lack of effectiveness that we found (only 42.4% of patients in the oral antibiotic arm adhered to 80% or more of the recommended doses). Flucloxacillin is well known to have an unpleasant bitter taste which is likely to affect adherence. 87 However, adherence in clinical practice is likely to be as bad or even worse (without nurse monitoring and a diary to complete) and therefore our results reflect what is likely to be seen in clinical practice. Furthermore, adherence to topical antibiotic was much better (79.3% of those in the topical antibiotic group adhered to 80% or more of the recommended applications), but resulted in a similar lack of clinically important benefit.

Our evaluation of the construct validity and performance characteristics of the ADQoL are encouraging and suggest that it may be a good instrument for measuring QALYs in children with eczema. However, 15% of parents thought it was not reflecting the effect of the illness on their child, and it may be that it is not appropriate for younger children or that more items need to be added.

We found that a smaller proportion of the parents of girls completed the questions about quality of life than did the parents of boys. This may be a chance finding or may represent a genuine difference between the way that parents of girls and boys respond, although it is difficult to think of a rational explanation for such a difference.

Conclusions

Contributions of authors

Nick A Francis (GP and Reader in Primary Care and Public Health) was the lead applicant and co-chief investigator for the CREAM study. He had overall responsibility for the overall study design and implementation, and the writing of the background, methods, results and discussion sections of the report, with final approval of the report submission.

Matthew J Ridd (GP and Consultant Senior Lecturer in Primary Health Care) was a co-investigator for the study and contributed to the overall study design, study implementation and approved the final version of the report.

Emma Thomas-Jones (Research Fellow) was the lead for study management, contributed to the study design and implementation, and to the writing of the methods and results sections of the report. She also was responsible for co-ordinating the compilation, formatting, proofreading and final approval of the report.

Victoria Shepherd (Research Associate/Research Nurse) contributed to the study management, study design and study implementation, and to the writing of the challenges section of the report. She contributed to the proofreading of the report and final approval of the report.

Christopher C Butler (GP, Professor of Primary Care Medicine) was a co-investigator for the study and contributed to the overall study design, study implementation, interpretation of findings and approved the final version of the report.

Kerenza Hood (Professor in Statistics, Director of Centre for Trials Research, Cardiff University) was a co-investigator and contributed to the overall study design, study implementation, supervision of the statistical and economic analyses, and approved the final version of the report.

Chao Huang (Statistician) contributed to the study design, conducted the statistical data analysis, and to writing the methods and results sections, and final approval of the report.

Katy Addison (Research Assistant) contributed to the design of CRFs and to the data management of the study, methods and results sections, and to proofreading the report.

Mirella Longo (Health Economist) conducted the statistical analysis and reporting of the validation exercise of the preference-based ADQoL and of patient NHS resource use, and final approval of the report.

Charis Marwick (Senior Clinical Academic Fellow and Consultant in Infectious Diseases) was a co-investigator and contributed to conduct of the study, critical review of drafts and approval of the final version of the report.

Mandy Wootton (Lead Scientist) was responsible for the integrity of the microbiology data and contributed to the overall study design, data analysis and writing of the methods and results section of the report, with respect to microbiology.

Robin Howe (Consultant Microbiologist and Head of the Wales Public Health Laboratory Reference Laboratory) was a co-investigator, lead microbiologist and contributed to the overall study design and final approval of the report.

Amanda Roberts (Public and Patient Representative) contributed to the study design, study implementation in terms of public and patient involvement, and review of the report.

Mohammed Inaam-ul Haq (Trial Manager) contributed to the study design, study implementation and review of the report.

Vishnu Madhok (GP) contributed to the study design, study implementation and review of the report.

Frank Sullivan (GP and Professor of Primary Care) was a co-chief investigator for the study and contributed to the overall study design, study implementation and approved the final version of the report.

Other members of the study team

We would like to thank and acknowledge the input from Professor David Cohen (Health Economist) and Manju Kalavala (Dermatologist), who contributed to the study design.

In addition, we would also like to thank Dr Eleri Owen-Jones (Trial Manager), Dr Natasha Kalebic (Data Manager), Mr Miquel Cossio (Database Developer) and Ms Jane Woodhead (Data Manager) for their contribution to the study implementation and management.

We would also like to thank the following research nurses for their contribution to the conduct of the study: Nicola Ball, Anna Gilbertson, Rebecca Elliott-Jones, Lyn Liddiard, Pauline Jones, Linda Phillips, Melissa Van Der Bijl, Karen Jones, Mandy Cook, Mat Williams, Hayley Prout, Susan Macfarlane, Debbie Rice, Fiona Treanor and Fiona Gammie.

Finally, we would like to thank Jennifer Richards and Leanne Davies (Specialist Antimicrobial Chemotherapy Unit Laboratory, Cardiff), who contributed to the acquisition of microbiology data.

Independent members of the Study Steering Committee and Data Monitoring Committee

We would like to thank our Trial Steering Committee for their continued support. Dr Anthony Harden (Trial Steering Committee Chairperson, University of Oxford), Dr Kim Thomas (Professor of Applied Dermatology Research and Co-Director of the Centre of Evidence Based Dermatology, University of Nottingham) and Therese Gladys Hingco (Patient and Public Representative), and also our Independent Data Monitoring Committee, Dr Sonia Saxena (Independent Data Monitoring Committee Chairperson, Clinical Senior Lecturer and GP, Imperial College London), Dr Miriam Santer (Clinical Research Fellow and GP, University of Southampton) and Mark Mullee (Senior Lecturer in Medical Statistics, University of Southampton).

Administrative team

We would like to thank all the administrative team across the centres, namely Kerry Fuery (Research Administrator, Cardiff University), Sian Dawes (Research Administrator, Cardiff University), Julia Carver (Study Administrator, University of Bristol) and Janice Broomhall (Project Manager, University of Dundee).

Research networks

We also wish to thank the providers of nursing/clinical studies officer support in all three centres. These include staff at the National Institute for Social Care and Health Research, Clinical Research Centre in Wales [all three regions (North Wales, South West Wales and South East Wales)]. In Bristol the Primary Care Research Network in the South West, and in Scotland the Scottish Primary Care Research Network.

Recruitment sites

We would like to thank all our recruiting sites across the three regions.

In Wales:

• Ely Bridge Surgery

• Practice of Health Surgery

• Highlight Park Medical Practice

• Avicenna Medical Centre

• Cardiff Road Surgery

• Stanwell Surgery

• Pen Y Cae Surgery

• Meddygfa Gelligaer Surgery

• Clifton Surgery

• Llandaff North Medical Centre

• Plas Menai Surgery

• Clarence House Surgery.

In Scotland:

• North Glen Medical Practice

• Linburn Medical Practice

• Ravenswood Surgery.

In West England:

• Bradgate Surgery

• Malago Surgery

• Air Balloon Surgery

• Clevedon Riverside Group

• Wellspring Surgery

• Backwell & Nailsea Medical Centre

• Frome Valley Medical Centre

• Portishead Medical Centre

• Chew Medical Practice

• Southville Surgery

• Oldfield Surgery

• Yorkleigh Surgery

• Portland Practice

• St Chad’s Surgery

• Courtside Surgery

• Gloucester Road Medical Centre

• Kingswood Health Centre

• Yeovil District Hospital.

And finally we would also like to thank the children and families who participated in the study, without whom this study would not have been possible.

Data sharing statement

All available data can be obtained from the corresponding author.

Disclaimers

This report presents independent research funded by the National Institute for Health Research (NIHR). The views and opinions expressed by authors in this publication are those of the authors and do not necessarily reflect those of the NHS, the NIHR, NETSCC, the HTA programme or the Department of Health. If there are verbatim quotations included in this publication the views and opinions expressed by the interviewees are those of the interviewees and do not necessarily reflect those of the authors, those of the NHS, the NIHR, NETSCC, the HTA programme or the Department of Health.

Evaluation of the Trial Torrent software

General practitioner practices that were willing to use TT were randomly assigned to receive TT prompts or not. Randomisation occurred at the time of practice enrolment and was minimised by list size (< 6000, > 6000) and whether or not the practice had been research active in the past 12 months.

The plan was to compare the mean recruitment rate (recruited patients per month recruiting per whole-time equivalent participating clinician) for GP practices in each arm (of the TT evaluation) at 3 and 6 months (from the start of patient recruitment). In addition, at 3 months from the start of recruitment, to compare the time to first patient recruited in the two groups. A significant improvement in either the mean recruitment rate or time to first patient recruited at 3 months or in mean recruitment rate at 6 months would prompt the evaluation to stop and TT to be rolled out to the rest of the practices. As the analysis of the mean recruitment rate at 3 and 6 months were interim analyses, the significance level for those tests needed to be adjusted if we were to retain an overall significance level of 5%. The Pocock method was used implying a significance level of 0.022 and was used for all three analyses. With 90 practices we would have a power of 84% for detecting a hazard ratio of 0.5 with an estimated SD of 0.5 and at the 2.2% significance level.