Three wound-dressing strategies to reduce surgical site infection after abdominal surgery: the Bluebelle feasibility study and pilot RCT

Study design

Qualitative research methods, using semistructured interviews with HCPs and patients, were used to address objectives A1 and A2:

• understand the reasons underpinning the current choice and use of dressings (e.g. simple vs. complex) across elective abdominal, obstetric and paediatric surgery

• explore patients’ and HCPs’ attitudes towards a trial of dressing type, including the range of procedures and comparisons (e.g. simple, complex, no dressing) that will be deemed acceptable for inclusion in a trial, perspectives on random allocation to dressing type, and views on the important outcomes to measure in a trial of dressing use.

Scope: clinical specialties

The qualitative research integrated throughout the Bluebelle study (i.e. Phases A and B) focused on abdominal [upper and lower gastrointestinal (GI)] and obstetric (caesarean section) surgery, as these were surgical specialties specified in the description of the study population in the funding application. Interviews were also conducted in paediatric surgery as a means of comparing dressing practices with those in adult surgery. This was because informal preliminary enquiries and the literature14,16 had indicated that dressings are not routinely used in paediatric surgery, thus providing the potential to further the team’s understanding of clinical practice related to dressing use.

Setting

Interviews were conducted across six university and district NHS hospitals in the South West and West Midlands regions of England.

Sampling strategy

Data collection

Interviews were conducted by Leila Rooshenas, Christel M McMullan, Daisy Elliott and Jonathan M Mathers over a 7-month period (July 2014–January 2015), either via telephone or in person. Face-to-face interviews took place on NHS premises or in informants’ homes. The qualitative researchers obtained written consent to conduct and audio-record all interviews. Separate topic guides were used for HCPs and patients to ensure that broad topics were consistently covered across each set of interviews. The topic guides evolved as data collection proceeded, either through rewording of questions to elicit more detailed responses or through the addition of new topics based on emerging lines of enquiry from earlier interviews (see Results, Objectives A1 and A2: Understand practice and views in relation to dressings). The initial and final topic guides used are shown in Report Supplementary Material 1. All interviews were transcribed ‘smart verbatim’ [i.e. a full transcript of the recording with some amendments to spoken content to ease readability (e.g. removal of filler utterances such as ‘um’)].

Approach to analysis

Interview transcripts were analysed thematically using the constant comparison method adopted from grounded theory methodology. 23 This involved line-by-line coding of text, whereby descriptive words or phrases were attached to lines of transcript and arranged into themes. Themes were subsequently arranged into hierarchies (with overarching themes and subthemes). Data collection and analysis proceeded in tandem, with the coding frame evolving as new data were collected and analysed. Previously coded interviews were reread in the light of the evolving coding frame to ensure that this comprehensively captured data across the full set of interviews. HCP interviews and patient interviews were analysed separately (i.e. with distinct coding frames).

The analysis of Phase A interviews was conducted primarily by the researchers who led each interview. A subset of interviews (10%) were double-coded by two members of the qualitative team in the early stages of analysis. This helped to inform the earlier versions of the coding frame. The coding frame was regularly discussed among the team as it evolved during the study. Given the wider study aims, the team agreed that themes that had implications for the pilot RCT protocol would be prioritised and coded in greater detail.

Descriptive accounts summarising emerging themes and overall study findings were prepared for the SMG throughout the analytical process. One researcher took responsibility for collating and synthesising findings from all sites for the purpose of producing final report(s) of Phase A findings. These tasks involved consulting descriptive accounts from individual sites and referring to coding frames and raw data (transcripts/recordings) where needed. There was an attempt to search for ‘negative cases’ in relation to particular themes or theories; where present, these are fully reported in the findings. All final reports were scrutinised by all members of the qualitative team.

Presentation of findings

The key Phase A findings that had implications for the pilot RCT design are summarised in Chapter 4, Objectives A1 and A2: Understand practice and views in relation to dressings, and cover HCPs’ accounts of current wound-dressing practice, HCPs’ perspectives on the proposed pilot RCT and patients’ perceived acceptability of the proposed pilot RCT.

Survey design

A prospective multicentre study was undertaken. All hospitals located in the two trainee-led research collaborative networks were invited to participate by e-mail and personal communication. A surgical trainee-level principal investigator, responsible for local co-ordination of data collection and entry, was identified within each participating hospital. The study was registered with the clinical audit department in each hospital.

Survey population

Abdominal wounds created during elective or unplanned abdominal surgery, and closed wounds primarily, were surveyed during a 2-week interval in January 2015. We considered a wound to be a closed primary wound if the edges of the incised skin were apposed (using suture material, tissue adhesive or clips) at the end of the procedure. Vascular, gynaecological, urological and paediatric procedures were excluded. Cases were included only if trainees were present (and therefore able to collect the data prospectively).

Data collected

Trainees completed anonymised data collection forms at the end of each surgical procedure, recording information about skin closure and dressings. Dressings were categorised as complex (those with advanced practical and/or therapeutic properties, including amorphous material, silicone, hydrocolloid, foam, antimicrobials or negative pressure) or basic (those without therapeutic properties that are adherent around the perimeter or entire surface, with or without a pad to absorb exudate). ‘No dressing’ was documented when an already closed wound was left without a covering at the end of the operation. The use of tissue adhesive to cover an already closed wound (i.e. when it was used as a dressing rather than wound closure technique) was categorised separately.

Operative and patient-related risk factors that might influence dressing selection were recorded. Operative risk factors included the type of procedure performed and access (open, laparoscopic or laparoscopically assisted), whether or not a stoma was formed and the degree of wound contamination (clean, clean-contaminated, contaminated and dirty). 26,27 Procedures were classified as planned (elective) or unplanned (emergency). The following patient-related risk factors were recorded: age, sex, body mass index (BMI), diabetic status and American Society of Anesthesiologists (ASA) fitness grade.

The reason for dressing selection (by the surgeon responsible for closing the wound) was recorded in the following three categories: (1) personal preference, (2) selected for specific wound characteristics or (3) the dressing was simply handed to the surgeon at the end of the procedure, without discussion. Dressings could be selected for multiple reasons and space was provided for free-text answers. To supplement this, procurement officers from each hospital were contacted to obtain information about local policies for purchasing dressings.

Data analysis

Data were entered into a password-protected online database held on a server [developed and maintained by the Bristol Clinical Trials and Evaluation Unit (CTEU)] in one of the participating hospitals. Analyses summarised the frequency of different dressing types using descriptive statistics. Descriptive statistics were also used to examine whether patient characteristics or the type and urgency of surgery were associated with particular dressing strategies. Analyses were performed in Stata^® version 13 (StataCorp, College Station, TX, USA).

Study design

A mixed-methods study was conducted in three steps:

1. analysing existing tools and in-depth interviews with patients and HCPs to establish the content of the questionnaire

2. developing questionnaire items and designing the tool

3. pre-testing the questionnaire for content validity, acceptability and understanding using cognitive interviews with patients and HCPs.

Step 1: generation of questionnaire content

Step 2: designing the questionnaire

A comprehensive list of all possible domains identified from step 1 was considered for inclusion in the new questionnaire by the study team, surgeons and methodological experts in trial design and outcome data collection. Domains unsuitable for a patient-reported questionnaire were dropped and those retained were constructed into items for a questionnaire. Domains were ‘operationalised’ into items for a questionnaire. Items were designed to be clear and unambiguous, written using plain language targeted at a lay audience without technical or medical terms. At the end of the item, where possible, the medical term for the underlying domain that the item was intended to measure was included in parentheses. Response categories took the format of either a ‘yes/no’ option or an ordinal scale (initially ‘not at all’, ‘a little’, ‘moderately’, ‘quite a bit’, ‘very much’), depending on what was most appropriate for the individual item. Options for ‘do not know’ were available for some signs and symptoms and wound care interventions anticipated to be potentially unclear to patients. Instructions for questionnaire completion were written, and adapted for patient and professional use. The only other difference between the questionnaires for patients and professionals was use of the first- or third-person pronoun in the phrasing of the items.

Step 3: pre-testing the questionnaire

The provisional questionnaire was pre-tested in cognitive interviews.

Study design

We planned to develop a measure based on an existing framework for developing patient-reported outcome measures,34,35 incorporating guidance on eliciting health domain concepts using qualitative methods. 36,37 Step 1 aimed to produce a comprehensive list of potential issues relating to wound and dressing experience and practical management issues. Step 2 developed issues identified from step 1 into questionnaire items. Step 3 evaluated the measures for acceptability and relevance using cognitive interviews with patients and HCPs. Steps 1–3 were overseen by a working group (DE, JB, LR, RM and CMM) that was part of the wider SMG. The SMG met before and between each step to discuss progress and make decisions about how the measures should be adapted.

Step 1: generating relevant issues

Step 2: designing the questionnaire

The working group used the item-tracking matrix to agree on those issues that should be written into questionnaire items. Items featured words and phrases used by patients in the interviews to enhance content validity. 37,43

Step 3: pre-testing the questionnaire

Cognitive interviews, with a new sample of patients who had undergone surgery, were conducted. Cognitive interviews are used widely in questionnaire development36 and involve asking respondents to verbalise their thoughts while answering questions. This methodology enabled us to explore the acceptability of the measure and coverage of patients’ and HCPs’ concerns (in terms of language, accuracy and relevance, as well as layout).

Patients who had undergone abdominal general surgery or caesarean section at one of five hospitals in two cities in the UK were identified and approached by research nurses and surgical trainees. Contact details were passed on to members of the working group to follow up and arrange an interview. HCP participants were identified and approached directly by the working group. As in step 1, sampling was purposeful to achieve maximum variation in relation to clinical role, age, sex and geographic location (for HCPs) and age, sex, ethnicity, type of surgery, dressing type and location (for patients).

Interviews explored the acceptability of the measure and coverage of patients’ and HCPs’ concerns (in terms of language, accuracy and relevance, as well as layout). 37 During each interview, participants were asked to complete the measure by reading each item aloud and commenting on their understanding. Interviews were guided by a series of probes (e.g. ‘What does this item mean to you?’, ‘Are there other ways you would describe it?’). 44 Participants’ body language was also observed and prompted further discussion about specific items (such as the participant nodding in agreement or frowning). 36 A copy of the topic guide is available.

Operationalisation and modification of the measures was an iterative process. Findings from cognitive interviews and suggestions for amendments were regularly disseminated to the Bluebelle SMG. Each stage of feedback informed amendments to modify and reword items to improve understanding, which was repeated following efforts to revise questions and eliminate problems. 44 This process continued until no new issues were identified and no further refinements were believed to be necessary.

Feasibility of photographing wounds in theatre

We attempted to collect wound photographs in theatre at three of the participating hospitals. This required obtaining approvals from the hospitals, separately from the research ethics approval for doing this. Special safeguards were in place at NHS trust level in all three hospitals. We were ultimately successful in gaining approval to take photographs in theatre but the process was time-consuming and cumbersome. The challenges encountered are described under the results for this objective.

Feasibility of assessing the quality of wound closure

Poor wound closure may be a contributory factor to SSI development that has not previously been investigated. The reasons why poor wound closure may increase the risk of SSI include the presence of gaps in the wound allowing either exudate to leak out or contamination to enter, subcutaneous suture material visible on the skin acting as a nidus for infection, tethering or puckering of the skin edges that may invite bacterial colonisation and even tension and poor vascularisation increasing wound ischaemia. These issues have not been intensively studied but it is notable that cosmetic and plastic surgeons (wound experts) take great care to ensure that primary surgical wounds are closed well and heal well to optimise cosmesis and it is plausible that the quality of wound closure also affects the risk of SSI.

Therefore, we sought to investigate markers that constitute good-quality wound closure, in the context of such markers providing the basis for the development of a metric to measure the quality of wound closure and its relevance to the development of a SSI.

We were also interested in measuring the quality of wound closure in the Phase B pilot RCT because of concern about the feasibility of randomising after wound closure and the risk of performance bias. Randomising after wound closure was preferred to avoid knowledge of dressing allocation influencing how a surgeon closed the wound. For example, if a surgeon is aware that the patient has been allocated to ‘no dressing’, then the surgeon may take extra care to ensure that the wound edges are well approximated and that there is no puckering and tethering. Conversely, if the surgeon is aware that the patient has been allocated to a simple dressing, then the surgeon may take less care, knowing that the dressing will cover the wound.

Study design

We used a mixed-methods design to address this objective, comprising literature reviews, qualitative interviews with surgeons and observation of wound closure in the operating theatre. Mixed methods were selected to enable an in-depth exploration of a complex phenomenon (closure of surgical wounds) within its original setting (the operating theatre).

Literature review

We searched for literature pertaining to wound closure, with the aim of identifying theories about ‘what makes a good wound closure’, and how wound closure might influence wound healing and the development of SSI. Relevant articles were identified using snowballing techniques, and by reviewing the grey literature.

Interviews and observations

Decision question

Economic model

The total cost of dressing k, cost_k, is the sum of the cost of dressing k and the probability of a SSI (pSSI) using dressing k multiplied by the cost of a SSI. We also include a utility decrement associated with a SSI, so that the decision model is to identify the dressing k that maximises net benefit:

where NB is net benefit, SSIQALYloss is the QALY decrement resulting from a SSI and WTP is the willingness to pay per QALY threshold.

This model is shown as a decision tree in Appendix 1 (see Figure 13).

Model inputs

Surgical site infection risk with simple dressings (standard practice)

Surgical site infection risk is variable across surgery types, and Jenks et al. 50 and the PHE SSI surveillance programme51 report this risk by surgery type (Figure 1). The two different data sources are broadly comparable, although the risk of SSI following large bowel surgery was higher in PHE surveillance survey than in Jenks et al. 50 The PHE survey is optional and the data may not be representative of all hospitals, and it is not clear how the selective inclusion of data may have had an effect; hospitals may have wished to report good results, but managers may also have wanted to report poorer results to encourage improvements. The data in the Jenks et al. 50 study may be unrepresentative (i.e. expected to be better than average owing to awareness of SSI risk in participating hospitals and to the restriction to elective surgery only). PHE includes more hospitals and regions than the Jenks et al. 50 study (multiple hospitals in the University Hospitals Plymouth NHS Trust) and PHE covers a broader and more recent time period. We therefore prefer to use the SSI risk estimate from the PHE survey but, because PHE does not give information for the ‘multiple: intra-abdominal’ surgery type included in Jenks et al. ,50 we have used the Jenks et al. 50 data on the proportion of procedures and SSI risk for this category. The overall SSI risk under this assumption across all surgery types of interest was estimated to be 13.8% (95% CI 13.45% to 14.15%). We use this estimate in our base case but use the estimate from Jenks et al. 50 of 8.94% (95% CI 7.75% to 10.13%) in a sensitivity analysis.

FIGURE 1.

Surgical site infection risk by surgery type based on Jenks et al. 50 (circle, estimate; solid line, 95% CI), and PHE SSI surveillance survey 2014/1551 (square, estimate; dashed line, 95% CI).

The Jenks et al. 50 study is based on data from April 2010 to March 2012, whereas the PHE survey covers the period from April 2010 to March 2015 (between 50 and 70 hospitals per year). The estimated risk across all these surgery types is labelled ‘ALL’ for Jenks et al. ;50 this is the overall SSI risk, but for PHE, this is calculated on the basis of the above assumption for the surgery type multiple: intra-abdominal.

Odds ratios of surgical site infection comparing different dressing types

To estimate the odds ratios of SSI comparing different dressing types, we use results from a recent update of the Cochrane review of dressings for the prevention of SSI. 17 Appendix 1 (see Table 40) shows the included studies that report the SSI outcome, surgery type and interventions (I) with their classifications according to the Cochrane review classes (C) and according to the Bluebelle definition.

The intervention comparisons made in the included studies can be represented pictorially in a network plot, in which interventions are joined by lines if there is a RCT comparing those two interventions, and the thickness of the lines indicates the number of RCTs making that comparison. We display these plots for the surgery types representative of the Bluebelle population (Figure 2) and for all surgery types (see Appendix 1, Figure 14). Each figure shows network plots at three different classification levels of the interventions: Bluebelle project classification, intervention level and Cochrane review classification (see Appendix 1, Table 40). It is apparent that some level of grouping of the interventions is necessary because the networks are ‘unconnected’ when using the most detailed level of intervention definition; comparisons cannot be made between interventions when there is no connecting path.

FIGURE 2.

Network plots showing comparisons that have been made between interventions (for three different classification schemes) in RCTs included in the Cochrane update review17 where surgery type was representative of the Bluebelle population of interest. The term ‘exposed’ in the figure presents leaving the wound uncovered, described as ‘no dressing’ in the pilot Bluebelle trial. BWCD, basic wound contact dressing; PHMB, polyhexamethylene biguanide.

We considered two different NMA models to estimate the relative efficacy of wound dressings according to the Bluebelle classification: (1) a random-effects model with intervention effects defined by the Bluebelle classification and (2) a hierarchical model with three levels: studies nested within interventions defined using the Cochrane classification, which in turn are nested within the coarser Bluebelle classification. The fit of these two models was practically identical, and heterogeneity was not decreased by including the additional level of hierarchy. Therefore, we present only results from the simpler random-effects model using the Bluebelle classification. To put credible bounds on the estimates of intervention efficacy, we used informative priors on relative effects, chosen to give a 99% prior credible interval that the odds ratios lie between a factor of 3 either way (i.e. between 0.33 and 3, which represent extremes of effect). This corresponds to a 95% prior credible interval of 0.43 to 2.3. Any posterior intervals similar to this indicate comparisons where evidence is lacking, so that results are dominated by the prior.

Table 1 shows results from the NMA for each dressing type compared with simple dressings (Bluebelle definitions). Owing to the low SSI event rate and relatively small sample sizes in the included studies, all estimates have wide credible intervals, and all include 1 (no effect). The results for the ‘all surgery’ population are comparable to those for Bluebelle population surgery types. There is very little improvement in model fit by allowing for surgery type and heterogeneity increases. We therefore use the estimates from all surgery types because they are more precisely estimated.

TABLE 1 - Posterior mean odds ratios and 95% credible intervals for each dressing category relative to simple dressings

Grey font indicates estimates associated with considerable uncertainty (posterior approximately equal to the prior).

Surgery type	Odds ratio (95% credible interval)
	Exposed vs. simple	Glue vs. simple	Complex vs. simple
All surgery types
All wound types	0.979 (0.561 to 1.546)	1.049 (0.371 to 2.413)	0.858 (0.535 to 1.263)
Clean wounds	0.787 (0.403 to 1.388)	0.847 (0.278 to 2.001)	0.740 (0.397 to 1.277)
Mixed/unclear/contaminated	1.153 (0.608 to 1.92)	1.263 (0.392 to 3.029)	1.068 (0.534 to 1.865)
Only surgery types representative of Bluebelle population
All wound types	0.889 (0.465 to 1.497)	0.956 (0.315 to 2.214)	0.726 (0.329 to 1.382)
Clean wounds	0.805 (0.348 to 1.629)	0.860 (0.245 to 2.232)	0.753 (0.224 to 1.908)
Mixed/unclear/contaminated	1.064 (0.532 to 1.841)	1.172 (0.353 to 2.887)	0.902 (0.394 to 1.757)

Results are shown for all surgery types included in Dumville et al. 17 and also restricting to studies with surgery types representative of the Bluebelle population of interest. Results are also shown on the basis of whether or not there were clean wounds only. Some of the estimates are associated with considerable uncertainty (posterior approximately equal to the prior), reported in grey font.

Most of the evidence is for complex versus simple dressings and shows a trend for reduced SSI rates with complex dressings, although the credible interval crosses 1 (no effect). Exposed wounds, that is wounds left uncovered with no dressing, show similar effectiveness to simple dressings and a trend towards a reduced SSI rate when used on clean wound types, although credible intervals are wide and cross 1 (no effect). There is no direct evidence for glue versus simple wound dressings, and, although an indirect comparison can be formed, this estimate is extremely imprecise, so there is effectively no evidence for this comparison. Similarly, there is effectively no evidence for wounds that are non-clean.

In the economic model we use the results for ‘all wound types’ and ‘all surgery types’ in the base case, but note in interpreting the results that these results apply mainly to clean wounds. We use the results for ‘all wound types’ and ‘Bluebelle population surgery types’ in a sensitivity analysis.

The probabilities of each dressing being ranked first, second, third or fourth in terms of effectiveness in preventing SSIs are shown in Appendix 1 (see Figure 15). Complex dressings have the highest probability of being most effective for SSI outcomes, but this probability is < 0.5 for ‘all surgery’ and only just over 0.5 for the Bluebelle population surgery types, suggesting a high level of uncertainty. Appendix 1 (see Figure 16) also presents these probabilities cumulatively (i.e. the probability that each dressing is ranked in each position or better, so all dressings have a probability 1 of being fourth or better). The ideal curve would go straight to 1 and stay there, and we prefer dressings with curves above the others. Complex dressings are preferred according to the cumulative ranking curves, but the curves are all close to each other. Simple dressings perform least well. Table 2 gives the mean rank (where rank 1 is best and 4 is worst). Complex dressings have the best mean rank and simple dressings have the worst; however, these are very similar and 95% credible intervals span all ranks from 1 to 4 for each dressing type, reflecting the high degree of uncertainty.

TABLE 2 - The mean rank for SSI outcome and 95% credible intervals for each of the dressing types, reported separately for ‘all surgery types’ and ‘Bluebelle population surgery types’

Rank 1 is the best and rank 4 is the worst.

Dressing type	Mean rank (95% credible interval)
	All surgery types	Bluebelle population surgery types
Complex	1.93 (1 to 4)	1.71 (1 to 4)
Glue	2.52 (1 to 4)	2.53 (1 to 4)
Exposed	2.62 (1 to 4)	2.56 (1 to 4)
Simple	2.94 (1 to 4)	3.20 (1 to 4)

Surgical site infection cost

Jenks et al. 50 provide estimates of mean costs attributable to SSI, shown in Figure 3. Some of these estimates are very uncertain (very wide or incalculable CIs). SSI costs associated with gastric surgery are estimated to be much higher than those associated with other surgery types, although this estimate is based on only six patients. It is possible that gastric surgery included bariatric surgery (not our population of interest), which involves prosthetic ‘bands’ and may have led to higher costs. With the exception of gastric surgery, mean costs are quite similar across surgical sites. The estimated mean cost over all surgery types is £7179.79 with a 95% CI of £5225.01 to £9865.88. Assuming that these costs were correct for 2012, we inflate them to 2015/16 prices using an inflation factor of 1.097694841. 52 This gives a mean cost of £7881.22 with a 95% CI of £5724.25 to £10,850.95 in our model, described by a log-normal distribution with a mean of 8.972 and a standard error of 0.1631 on the log-scale.

FIGURE 3.

Mean cost (GBP) attributable to SSI by surgery type based on University Hospitals Plymouth NHS Trust data April 2010–March 2012 (Jenks et al. 50), assumed 2012 prices. CIs for the mean cost are obtained from reported confidence limits on the median cost, assuming that costs can be described by a log-normal distribution.

Study design

Phase B was a pragmatic feasibility and pilot parallel group RCT, using mixed methods. The trial used a factorial design, allocating participants to one of three dressing types and to disclosure of the dressing allocation before or after wound closure (Figure 4). If a participant had multiple closed primary wounds, all wounds were intended to be managed on the basis of the allocation.

FIGURE 4.

Trial schema for the Bluebelle Phase B external pilot trial, showing the double randomisation.

The time-of-disclosure factor was included because disclosing allocation after wound closure was preferred, to prevent surgeons closing the wound in different ways depending on the allocation, but was considered likely to be problematic by surgeons. This factor allowed the trial to test (1) the feasibility of randomising after wound closure and if photographs of the closed wound could be obtained in the operating theatre and subsequently assessed for the quality of wound closure and (2) the effect of timing of disclosure of dressing allocation on wound closure. This effect was potentially important as an indicator of the risk of performance bias if randomisation after wound closure proved impossible.

Study setting and population

The trial was set in secondary care (i.e. acute and maternity NHS hospitals). Patients aged ≥ 16 years undergoing primary elective or unplanned open or laparoscopic abdominal general surgery [including, but not limited to, gastrectomy for benign or malignant disease, cholecystectomy, anti-reflux procedures, hepatic resection, small or large bowel resection for benign or malignant conditions, abdominal wall hernia surgery (inguinal, femoral, incisional, epigastric and para-umbilical)] or elective or unplanned obstetric surgery (caesarean section) were eligible. Patients undergoing simultaneous abdominal and chest surgery were eligible but only those with abdominal wounds were allocated to one of the study interventions. At the time of recruitment, research nurses emphasised the need to attend a follow-up clinic 4–8 weeks after surgery and did not recruit patients who said that they would be unable to do so.

Patients with any of the following characteristics were ineligible:

• abdominal or other major surgery < 3 months before the index operation

• those for whom the surgeon intends to ‘close’ the wound with tissue adhesive (glue)

• any contraindication to one of the dressing allocations, including allergy to dressings

• undergoing surgery where no skin incision occurs

• lacking capacity to consent

• an inability to complete patient-reported outcome questionnaires

• detained in the prison service.

Reasons for ineligibility were recorded.

Eligible patients were provided with a participant information leaflet (PIL), either at a clinic visit before the operation or sent in advance of admission. They were given as long as possible to consider the study before being approached for consent (usually > 4 hours for elective surgery and usually > 1 hour for unplanned surgery). Consent was not requested if a patient asked for longer thinking time or appeared visibly distressed.

Participants were also asked to consent to four optional aspects of the trial, designed to explore feasibility:

1. willing to be interviewed about the acceptability of, and adherence to, the allocated dressing type (objective B3)

2. willing for a member of the research team to take a photograph of the wound(s) in the operating theatre

3. willing to take a photograph of their wound(s) themselves 4–8 weeks after the operation and to send it to the research team (described to participants as a ‘wound selfie’, a term that was readily understood; objective B7)

4. willing for a skin transfer to be applied after the operation to remind staff that the participant was in the study.

Surgery was carried out in accordance with local protocols. Apposition of wound edges and the method of closure of the skin were at the discretion of the surgeon and could include sutures, clips, wound closure strips or combinations of these wound closure methods.

Randomisation

The factorial design created six groups (see Figure 4). Both factors were randomised in blocks of varying size and stratified by hospital trust and specialty (abdominal surgery/obstetric surgery). The random allocation sequences were generated by computer in advance of starting to recruit. Local research team members accessed a participant’s allocation via the internet. The allocation was concealed until a participant’s eligibility and consent had been documented and information to identify the participant uniquely had been entered.

A member of the local research team (trainee or consultant surgeon/research midwife or nurse) accessed a secure password-protected computer system (within the study database) and entered the information needed to proceed to randomise the participant, usually at the beginning of the operation. Depending on the randomisation result, the dressing group allocation was disclosed immediately, or the user was advised to log into the system after the wound had been closed and to enter the time of wound closure, after which the allocation was disclosed.

Blinding

It was not possible to blind surgeons, participants or staff caring for participants to the dressing allocation. However, we planned to blind research staff assessing outcomes 4–8 weeks after randomisation. Methods to achieve blinding were piloted to test their feasibility and acceptability for the main trial. These included requiring the reference SSI assessment and the Wound Healing Questionnaire (WHQ) to be completed by HCPs who had not been involved in a participant’s care during the index admission. (The study also required these assessments to be carried out by different people, in order to validate the WHQ.) The success of blinding among assessors of SSI 4–8 weeks after randomisation (HCPs completing the reference SSI assessment or the WHQ) was assessed by asking them which study group they thought the participant was allocated to.

Trial interventions

The three dressing groups were simple dressing, glue-as-a-dressing and no dressing. The interventions for the first two groups were defined as described in Objective A6: use the literature and views of experts to define and categorise commonly used dressings into three pragmatic groups. Tissue adhesives are topical skin adhesives. In this trial, the glue-as-a-dressing intervention was defined as tissue adhesive applied in accordance with the manufacturers’ instructions and only to the surface of an already closed primary wound, acting as a dressing (i.e. not to close the wound, which would involve applying the tissue adhesive below skin level).

Bioclusive^® (Systagenix, Gatwick, UK), C-View^® (Aspen Medical, Ashby-de-la-Zouch, UK), Hydrofilm^® (Hartmann, Heywood, UK), Opsite^® (Smith & Nephew, Zaventem, Belgium), Mepore^® (Mölnlycke, Oldham, UK) and Tegaderm^® (3M, Bracknell, UK) are examples of commonly available simple dressings. 49 Hospital trusts could have stocked other types and were advised to use the dressing that represented usual care. Dermabond ProPen^® (Ethicon, Wokingham, UK), Epiglu^® (Meyer-Haake, Ober-Moerlen, Germany) and Histoacryl^® (Braun, Sheffield, UK) are examples of commonly available tissue adhesives that could be used. 49 In the no dressing group, no simple dressing or tissue adhesive was applied to the wound at the end of the operation.

The following aspects of wound care applied to all interventions:

• A participant’s wounds should be dressed on the basis of the participant’s treatment allocation throughout their hospital stay.

• When a participant had more than one wound (e.g. multiple port sites for laparoscopic surgery), all eligible wounds should be dressed on the basis of the randomised allocation.

• A wound might manifest slow discharge or ongoing seepage of fluid (‘ooze’) from the wound in the first 24 hours. We allowed a simple gauze swab to be applied to the area that was oozing in all groups without compromising adherence; nursing staff were instructed to tape the swab in place temporarily and not around its entire perimeter. Gauze swabs (filmated or not), non-woven fabric swabs (filmated or not), knitted viscose and paraffin gauze dressings are examples of swabs that could be used. If oozing continued, or if a SSI occurred (i.e. after the outcome of interest had been ascertained), the clinical team could apply any dressing or resuture the wound if necessary; when this represented a deviation from the allocated dressing, the action was documented.

• Cointerventions that might have influenced SSI rates (e.g. the use of prophylactic antibiotics and other aspects of pre-, peri- or postoperative care) were allowed at the discretion of the team and hospital looking after the participants. Their distribution by allocated group was monitored, because the differential implementation of cointerventions arising because the usual care team was not blinded to the allocated dressing could introduce bias. This information informed decisions about the need to standardise care when designing the main trial.

• To encourage adherence to the treatment allocation, colour-coded skin transfers showing the study logo and the dressing allocation were applied on the participant’s skin, near the surgical wound(s), as a reminder to HCPs looking after the patient (see Appendix 2, Figure 17). The skin transfer was not visible by the time of the 4- to 8-week assessment.

Study centres and surgeons

Four NHS trusts took part: University Hospitals Bristol NHS Foundation Trust (Bristol Royal Infirmary), North Bristol NHS Trust (Southmead Hospital), University Hospitals Birmingham NHS Foundation Trust (Queen Elizabeth Hospital) and Worcestershire Acute Hospitals NHS Trust. One recruited participants having either abdominal surgery or caesarean section. One trust joined in month 6 of recruitment. All general surgical teams carried out a wide range of operations. The participating obstetric unit carried out approximately 1800 caesarean sections each year. Because all operations were being carried out as part of the usual care of participants, there were no restrictions on operating personnel or ward care.

Primary feasibility outcome

The primary outcome was successful screening of a participant, and determination of the participant’s eligibility and consent to be randomised in the pilot trial. This information, together with the denominator describing the total number of participants approached, established whether or not recruitment into the main trial was possible.

Secondary feasibility outcomes

These comprised:

• adherence to disclosure of dressing category allocation at the designated time

• adherence to the allocated dressing type by the usual care team during the index hospital admission and, if applicable, the reason for non-adherence

• quality and completeness of the data for different outcomes anticipated to be measured in the main trial (see below; assessed at different times), including component assessments contributing to an overall judgement about the occurrence of a SSI

• adherence to the follow-up schedule

• documentation of cointerventions (e.g. details of hair removal, use of skin cleansing agents, type of wound closure methods, prescription of prophylactic antibiotics) and classification of surgery as ‘clean’, ‘contaminated’ or ‘dirty’ at the time of the operation

• completion of the reference SSI assessment at 4–8 weeks by a blinded observer

• completion of a WHQ57 at 4–8 weeks by a blinded observer and by the participant

• completion of a Wound Management Questionnaire (WMQ) (developed during Phase A) up to 4 days after randomisation by an observer or the participant (if discharged early, e.g. day case surgery)

• completion of a Wound Experience Questionnaire (WEQ) (developed during Phase A) up to 4 days after randomisation by the participant

• assessment of wounds from digital photographs at 4–8 weeks, if submitted.

Some of these secondary outcomes were developed in Phase A and piloted in Phase B, and detailed scoring methods were not necessarily available. This report focuses on the questionnaire response rates, the number of fully completed questionnaires and rates of missing items in this pilot RCT as part of the evaluation of the acceptability of the new questionnaires to patients. The data from the pilot trial were also used to validate some of the questionnaires.

Anticipated outcomes in a subsequent main trial

The following outcomes were expected to be assessed in a main trial. Therefore, we documented the collection of these outcomes carefully:

• occurrence of a SSI up to 4–8 weeks after randomisation (primary)

• wound management – WMQ

• patient-reported outcomes (WEQ) and generic health status, assessed by the EuroQol-5 Dimensions, five-level version (EQ-5D-5L)58

• wound complications arising up to 4–8 weeks after randomisation

• resource use up to 4–8 weeks, including length of postoperative hospital stay, rates of readmission and duration.

Trial procedures and data collection up to 4 days after randomisation

The schedule of data collection is described in the SPIRIT (Standard Protocol Items: Recommendations for Interventional Trials) diagram (Table 5). Surgery and closure of wounds took place as per usual practice. A member of theatre personnel, or a member of the research team, recorded (1) the time taken to close the wound, (2) the use of laparoscopic or open surgical methods and (3) the method of wound closure. A member of theatre personnel or the research team could also take one or more digital photographs of the wound(s). If a participant was randomised to disclosure of allocation after wound closure, the time of completing wound closure had to be entered into the database to obtain the allocation. A range of methods were used to promote adherence to the randomised dressing allocations in hospital, adapted to the circumstances of participating hospitals. These included simple labels on medical notes or placed by the bedside, as well as skin transfers (see Appendix 2, Figure 17).

Up to 4 days after surgery, after any early wound care that was required, a HCP completed the WMQ. This questionnaire captures information about aspects of the participant’s wound management. If discharged early (e.g. after day case surgery or the day after surgery), the research team gave the participant the WMQ to complete by day 4 and asked the participant to return the completed WMQ in a prepaid envelope.

The local research team gave each participant the WEQ to complete up to 4 days after surgery. For participants who were in hospital at this time, the WEQ was collected by the research team. If discharged early (e.g. after day case surgery or the day after surgery), the participant completed the WEQ at home by day 4 and returned it in a prepaid envelope.

Data collection up to 4–8 weeks after randomisation

Local research teams gave each participant a copy of the EQ-5D-5L and a prepaid envelope to take home and instructed the participant to complete and return it if the wound became infected or problematic. The reason for asking for the EQ-5D-5L to be completed if there was a problem with a wound was to try to document the peak impact of a wound problem on health status to inform a future health economic evaluation.

Several follow-up assessments were required 4–8 weeks after randomisation. First, the participant was asked to complete a copy of the EQ-5D-5L and the WHQ; these questionnaires were given to participants on discharge from hospital or posted shortly before they were due. Around this time, participants who had agreed to send a photograph might also have been contacted about this. A blinded health professional completed the WHQ; this could be done face to face, typically at the same clinic attendance but before the reference SSI assessment, or administered by telephone. The face-to-face reference SSI assessment was completed by a blinded health professional, who had to be different from the professional completing the WHQ. This assessment included eliciting information about potential wound-related complications. Whenever possible, the face-to-face clinic visit was arranged to coincide with a usual care clinic appointment. When this was not possible, travel expenses were offered to the participant. For women who had had a caesarean section, the WHQ questionnaire was administered by telephone and research midwives then carried out the reference SSI assessment during a home visit.

To promote retention, reminders were sent to participants who failed to return postal questionnaires promptly. Participants who failed to attend the face-to-face SSI assessment were offered further appointments.

Assessment and analysis of resource use

Trial duration

The trial was timetabled to take up to 11 months, randomising the target number of 330 participants in 9 months and then following up the last participants for 4–8 weeks.

Sample size

A target sample size of 920 eligible participants was set, allowing a recruitment rate of 36% (corresponding to the target number randomised of 330) to be estimated with a 95% CI of 32% to 39% and a recruitment rate of 60% (95% CI 56% to 64%). For the simple dressings group, we anticipated an adherence rate of 90%. Assuming a 36% recruitment rate and 110 randomised participants per group, a 90% adherence rate would be estimated with a 95% CI of 82% to 95%. We had no information on which to base any estimate of adherence in the no dressing and glue-as-a-dressing groups. However, if adherence were < 70% in either group, we stated in the protocol that we would conclude that randomisation to the group in question in the main trial was not feasible.

Statistical analyses

Summary descriptive statistics to inform plans for the main trial were compiled, including:

• the number of potentially eligible participants per month per centre

• the percentage of potential participants confirmed as eligible

• the percentage of participants consenting to the pilot RCT

• the percentage of randomised participants receiving the allocated treatment and completing outcome measurements at the 4- to 8-week assessment

• the rate of, and reasons for, non-adherence to allocation at both a wound and a participant level

• the mean number of wounds per participant

• the mean (or median, if skewed) time from wound closure to randomisation

• the mean (or median, if skewed) time to complete the randomisation process

• the completeness of data items and reasons for missing data

• the rate of unblinding of outcome assessors and reasons for unblinding.

Only the statisticians (CAR and RAH) had access to the data. The analysis population comprised all randomised participants. Results were described by centre and by specialty as well as overall. If the data allowed, we planned to carry out subgroup analyses of the secondary outcome measures relating to wound closure, estimating the interaction of timing of randomisation by dressing group.

The primary analysis took place when follow-up was complete for all recruited participants. During the pilot trial, we monitored recruitment and adherence periodically. The Study Steering Committee (SSC) reviewed safety data. In any interim reports, for example about withdrawals after randomisation, the data were presented by group, with uninformative labels used to keep the allocation masked.

Study design

We conducted a qualitative study, consisting of semistructured face-to-face or telephone interviews with Bluebelle participants (patients) and HCPs involved in delivering the pilot RCT.

Sampling and recruitment

Data collection

Interviews were conducted face to face or by telephone up to 30 days post randomisation, between February 2016 and August 2016. Interviews were conducted by Leila Rooshenas or Christel M McMullan and audio-recorded following receipt of written consent. Field notes were recorded to note any contextual information that could have influenced the interview (or have a bearing on analysis). Face-to-face interviews took place in participants’ homes or on NHS premises.

Analysis

Audio-recorded interviews were transcribed verbatim and analysed thematically using the constant comparison method, as described previously (see Objectives A1 and A2: understand practice and views in relation to dressings). Transcripts were coded line by line in NVivo version 10. HCP and patient interviews were analysed separately. Data collection and analysis proceeded iteratively, enabling the qualitative team to refine topic guides to explore new lines of inquiry that emerged from the ongoing analysis. The initial and final topic guides for HCP and patient interviews are shown in Report Supplementary Material 9.

Interviews were analysed by Leila Rooshenas, with Christel M McMullan analysing a subset in the early stages of analysis to enhance coding reliability. This helped to establish an early version of the coding frame, which subsequently evolved over time. There was an attempt to search for ‘negative cases’ in relation to particular themes from both patient and HCP interviews; where present, these are fully reported. Some of the key themes relating to wound healing and wound management from the patient interviews were presented in a matrix (as columns), set out against patient interviewees (rows) grouped on the basis of their allocated dressing strategy. This facilitated comparison of patients’ accounts from the three trial comparison groups.

Study design

Evaluation of the WHQ used data from a specifically designed cohort study (Phase A) and the pilot RCT.

Study setting and population

Centres and participants in the pilot RCT are described elsewhere (see Objective B1: establish the numbers of potential participants at different hospitals who are considered likely to be eligible and who can be approached about the trial, and the proportions confirmed as eligible, recruited and randomised). Three NHS trusts participated in the cohort study. Participants with similar inclusion and exclusion criteria to those of participants in the pilot RCT were identified, approached and recruited by research nurses and surgical trainees.

Data collection

The collection of questionnaire data in the Phase B pilot RCT has been described elsewhere (see Objective B1: establish the numbers of potential participants at different hospitals who are considered likely to be eligible and who can be approached about the trial, and the proportions confirmed as eligible, recruited and randomised).

In the cohort study, participants were recruited either prior to or within a few days of surgery. Sociodemographic and operative details were collected at the time of recruitment. Data were collected on study data collection forms and recorded and managed using purpose-designed electronic data capture tools developed using Research Electronic Data Capture (REDCap) software. 68

Participants were asked to complete the EQ-5D-5L (baseline assessment). Approximately 3 weeks after surgery, participants were posted the WHQ (self-assessment). Included was a ‘debriefing’ questionnaire asking questions on the feasibility and practicality of completing the WHQ (e.g. time taken to complete the questionnaire and whether any items were confusing or difficult to answer) and the EQ-5D-5L (follow-up assessment). A stamped addressed envelope was provided to return the questionnaires.

A subset of participants (n = 50) within the Phase A cohort were invited to complete a second self-assessment to evaluate the test–retest reliability of the WHQ. Sampling was carried out by convenience during 1 month of the study. Participants were posted a second copy of the WHQ within a few days of receipt of the first questionnaire. This was done to provide data.

Approximately 30–35 days after surgery, staff attempted to contact participants and complete the WHQ (observer assessment) by telephone. This assessment took place face to face if the patient was still in hospital or had been readmitted. During the telephone assessment, a number of participants were invited to a face-to-face follow-up appointment in clinic. Convenience sampling methods were applied, inviting participants based on staff and clinic availability, and participant willingness and proximity to the clinic. At the face-to-face appointment, a clinical member of the research team (blinded to the wound self-assessment and observer assessment) made a SSI diagnosis using the CDC criteria (reference SSI assessment).

Statistical analyses

Combined cohort and RCT data were used to examine the adherence to and acceptability of the WHQ (response rates and missing data), inter-rater patient and observer agreement in responses and ability of the WHQ to discriminate between SSI/no SSI [receiver operating characteristic (ROC) curve analyses]. Phase A cohort data alone were used to examine the feasibility and practicability of completing the WHQ (debriefing questions) and intrapatient test–retest reliability. Possible scale structures of the WHQ were initially explored in Phase A cohort data and subsequently tested (as a method of internal validation) with data from the Phase B RCT.

Photographing wounds after wound closure

We encountered major information governance challenges when seeking permission to take photographs of wounds in theatre after wound closure (see Objective A7: investigate the feasibility of photographing wounds in theatre and assessing the quality of wound closure). Consequently, we did not attempt to implement photographs after wound closure in theatre during Phase B.

Photographing wounds after discharge

If the occurrence of a SSI can be assessed from a photograph after a dressing has been removed, the assessment of the photograph could be blinded. For this reason, the trial tested the feasibility of participants submitting ‘wound selfies’ securely to the trial database.

There were challenges in developing the information technology to allow participants to submit a photograph confidentially, ensuring that the submitted photograph was linked with the rest of a participant’s data and stored securely. A system was successfully put in place that relied on sending a secure web link in an e-mail but we achieved this only towards the end of the recruitment period. The text of the e-mail is in Report Supplementary Material 11. We implemented the system during the last 4 weeks of recruitment to pilot participants’ willingness to take and submit a photograph.

Consent for this part of the study was obtained at recruitment. This pilot was limited to a subgroup of participants who had consented to take and submit a photograph and who offered an e-mail address for correspondence. A letter providing instructions on taking a photograph of the wound, along with a disposable ruler, was posted in advance of the due date for the completing the 4- to 8-week WHQ questionnaire. The success of the pilot was assessed by the proportion of participants in the subgroup who successfully submitted a photograph. The submitted photographs were not assessed for quality of the photograph or wound healing.

Meeting 1

The aim of the first meeting was to explore PPI members’ thoughts about and reactions to the protocol and materials for the Bluebelle pilot RCT (Phase B). Key objectives were to elicit views about three main topics: (1) the presentation, content and delivery of the PIL for the pilot RCT, (2) the design and delivery of the WEQ and (3) strategies for improving adherence to trial allocation in the pilot RCT.

All invitees had participated in qualitative interviews in Phase A of the feasibility study. The meeting took place at the School of Social and Community Medicine, University of Bristol. Two members of the research team co-ordinated this meeting, and a third member of the research team attended part of the meeting to discuss the WEQ.

Meeting 2

The aim of the second PPI meeting was to update the group on the progress of the Bluebelle study and discuss thoughts and ideas for a possible large-scale RCT, with a focus on findings from the feasibility study. Key objectives were to discuss and elicit views about two main topics: (1) ideas for encouraging participants to stay in a study and be followed up by the study team (retention) and (2) methods for patients or their carers to take and send digital photographs of their wounds to the study team (to allow remote, and potentially blinded, wound assessment).

Phase A

Phase A was registered with Current Controlled Trials as ISRCTN06792113 (registration assigned on 20 March 2014). Research ethics approval was granted by the NHS Health Research Authority (HRA) National Research Ethics Service (NRES) Committee London – Camden & Kings Cross (reference number 14/LO/0640). All participants gave written informed consent. One substantial amendment was approved (18 November 2014, implementing v3.0), allowing patients having unplanned abdominal surgery or an unplanned caesarean to be recruited for the SSI interviews.

Phase B

The pilot trial was registered with Current Controlled Trials as ISRCTN49328913 (registration assigned on 20 October 2015). Research ethics approval was granted by the South West–Frenchay REC (reference number 15/SW/0008) in February 2015 and subsequently by the HRA NRES (24 August 2016). All participants gave written informed consent. Two substantial amendments were approved. The first (8 December 2015, implementing v3.0) added patients having unplanned surgery to the population, substituted glue as-a-dressing for complex dressing as one of the groups, and included the use of the WMQ and the WEQ. The second (6 September 2016, implementing v5.0) reduced the time required between giving the PIL and seeking consent, allowed observer-completed WHQs to be completed over the telephone, clarified that an exclusion criterion applied to contraindications to dressing allocation, not just to dressings, and recruited a fourth NHS trust (Worcester Acute Hospitals NHS Trust). This ethics approval covered all participating sites.

Centres and patients

In total, 25 hospitals within the SPARCS and WMRC networks were approached, and 20 (80%) participated. Data were included from 727 patients (1794 wounds), of whom 193 (27%) underwent upper GI surgery (Table 6). The number of wounds per patient varied from 1 to 7. The proportions of patients with each number of wounds was as follows: one (n = 299, 41%), two (n = 51, 7%), three (n = 155, 21%), four (n = 190, 26%), five (n = 25, 4%) and more than five (n = 7, 1%). Complete data sets were submitted for 675 (93%) patients. There was one missing data item for 36 (5%) patients and 16 (2%) patients had more than one missing item.

Wounds and dressings

Sutures were most commonly used to achieve skin closure (n = 1531/1770, 87%), with clips (n = 158, 9%) and wound closure strips (n = 48, 3%) less commonly used.

Of the 1794 wounds, dressing type was recorded for 1769, with 1706 out of 1769 (96%) covered and 63 out of 1769 (4%) not covered by a dressing. The majority of dressings were classified as simple (n/N = 1203/1769, 68%); just 18 out of 1769 (1%) were classified as complex. Tissue adhesive was applied over closed skin to 485 out of 1769 (27%) of wounds.

Phase 1: generating questionnaire content

Phase 2: designing the questionnaire

Two domains (microbiology and prolonged hospital stay) were excluded as they were considered inappropriate for patient self-report after hospital discharge as well as not being specific to all SSIs. Retained domains were constructed (‘operationalised’) into items for the first version of the questionnaire. Initially, eight medical terms were included in items in parentheses after plain language descriptions.

Phase 3: pre-testing the questionnaire

A total of 42 cognitive interviews were conducted (with 28 patients and 14 HCPs) (Table 11). Interviews lasted a median time of 27 minutes (range 13–52 minutes). The mean time between patients’ surgery and Interview was 46 days.

During pre-testing, interviews highlighted that the initial plain language description for some items required modification to ensure a more accurate interpretation of the intended sign/symptom or intervention. 72 Throughout the cyclical pre-testing phase of interviews and revisions, a total of eight versions of the WHQ were tested. Modifications included changes to the wording and structure of items, layout, instructions and response categories.

The final version of the WHQ after pre-testing included 16 core items (Table 12). Items 1–8 related to patient-reported signs or symptoms that are potentially indicative of SSI. Items 9–16 related to wound care management and clinical interventions for treating SSI. Two core items assessing SSI signs and symptoms (items 3 and 4) had further subitems, to be completed if the response to the core item was anything other than ‘not at all’. Other core items (numbered 9, 10 and 12) had subitems that were included specifically to collect health economic data for the pilot trial.

Medical terms were included in parentheses at the end of nine items (six core items and three subitems) in the final version of the questionnaire after pre-testing.

Response categories for signs and symptom items were changed from a five-point option to a four-point option, removing the middle ‘moderately’ option. ‘Very much’ was change to ‘a lot’.

Written instructions informed responders to complete the questionnaire in relation to just one wound only: either the main wound or another wound if there had been any concerns about how it had been healing. Instructions asked responders to answer items based on what had happened since leaving hospital after having surgery.

A full report on the development of the WHQ has been published. 72

Phase 1: generating questionnaire content

Phase 2: designing the questionnaire

A provisional measure was designed based on the findings from Phase 1. Nine key categories were included: (1) wound comfort, (2) exudate and its impact, (3) allergic reactions to the dressing, (4) dressing removal, (5) dressings to protect the wound, (6) impact on daily activities, (7) ease of movement, (8) anxiety about the wound and (9) satisfaction with dressing. Issues relating to the appearance of the wound were not included as they were relevant only to longer-term outcomes of wound healing (not within first days after surgery). In addition, because most patients reported having an adhesive dressing, many had not seen their wound within this time frame. The first version of the measure included 16 items and was provisionally called the Practical WMQ.

Phase 3: pre-testing the questionnaire

Cognitive interviews (n = 40) were conducted between July 2015 and March 2016. All interviews were conducted face to face. These consisted of interviews with 25 patients who were in hospital and had undergone abdominal general surgery and 15 HCPs involved in surgical wound care. Demographic characteristics for participants who had cognitive interviews are shown in Table 13.

Interviews highlighted issues with content in the initial measure. For example, items regarding the colour of the wound exudate were removed. Questions were rephrased to focus on the experience of having a dressing rather than general recovery after surgery [i.e. ‘Have you been able to perform everyday tasks? (i.e. showering/bathing)’ was changed to ‘Has your dressing prevented you from showering/washing?’]. In addition, because four patients commented that a question on the smell of their wound was missing from the measure, an item was added to capture this.

The measure had intended to be administered 2 days after surgery, although feedback suggested that this needed to be completed up to day 4 as the patient may be disorientated from surgery in the first few days. However, because there were clear differences in recovery between caesarean section and abdominal surgery patients, a time frame of within 4 days of surgery was set, and the measures recorded the date of surgery and the date completed to determine context of responses.

Feedback from patients suggested that it was difficult to respond to questions about exudate, because a HCP cared for their wound while they were in hospital. If their dressing had been changed, they were also uncertain about the reason why (i.e. if it was simply as part of standard practice or for other reasons). Therefore, the study team decided to separate the measure into two. The first measure related to the practical aspects of wound management and the second related to the patient’s experience of the wound/dressing and the psychological aspects (anxiety, satisfaction, etc.). The two measures were named the WMQ and the WEQ.

Seven versions of measures were modified throughout the pre-testing phase. Pre-testing continued until no new issues were identified and no further refinements were believed to be necessary. The final version of the WMQ contains four items, whereas the WEQ contains 10 items (see Appendix 3). Overall, the final versions of the measures were well received. In addition, 96% of participants stated that each measure took < 5 minutes to complete. Further details of the development of these outcome measures have been published elsewhere. 101

Feasibility of photographing wounds in theatre

We attempted to obtain permission to take photographs in theatre in three trusts. Although we had already obtained research ethics approval to take photographs in theatre as a substudy, all three trusts required additional approvals.

At one trust, anyone taking a photograph had to:

1. read the trust’s policies for taking photographs of patients

2. obtain ‘level 3’ consent for photography from the patient, separately from study consent for the photograph substudy

3. register as a Camera User

4. register with the trust’s Information Management and Technology (IM&T) department to be able to upload a digital photograph to the study website.

Level 3 consent required participants to give:

consent for these photographs/video recordings to be used for publication in clinical journals and textbooks etc. including the Internet. (Personal details will not accompany the images. The general public as well as medical professionals can view the images). Please Note images consented for publication can only be removed from our system and cannot be called back once published in the public domain.

Reproduced with permission from University Hospitals Bristol NHS Foundation Trust102

Lower levels of consent only applied to ‘treatment, diagnosis and medical records’ or ‘the teaching of healthcare professionals’. 102

The other trusts required that similar approvals and processes be followed. An additional requirement for one trust (not the sponsor) was to retain an electronic copy of the digital photograph at the site because they regarded the photographs as the property of the trust. After a lengthy process, the trust agreed that the information governance and our IM&T processes (storage of the photographs on the sponsor’s NHS IM&T servers, with access on request) were satisfactory.

These challenges had not been anticipated and it was time-consuming to address them. Given the short duration of the pilot RCT (only 9 months), we decided not to try to implement wound photography in theatre as part of data collection for the trial. Nevertheless, at one trust we piloted how to obtain a good digital photograph of abdominal wound(s) after wound closure, which resulted in draft instructions for wound photography. Photographs obtained during piloting were used to inform our investigation of the feasibility of assessing the quality of wound closure.

Feasibility of assessing the quality of wound closure

Summary of findings

Information from literature reviews, observations in the operating theatre and surgeon interviews led to the development of content for a tool to assess the quality of primary wound closure. The tool comprises metrics of wound closure (which are possible to visualise in a closed wound) and mediators, which comprise factors influencing wound healing.

Cost-effectiveness analysis

Table 17 shows the results of the cost-effectiveness analysis. All dressing types give a negative expected net-benefit, indicating that they are associated with costly outcomes (i.e. SSI). Costs should be minimised, and so the intervention with the highest net benefit (smallest costs) is preferred. Based on the evidence available currently, complex dressings have the lowest mean health-related cost (£1256/patient), followed by exposed dressing (£1411/patient). Complex dressings are the optimal dressing based on current information when they are included in the decision options; otherwise, exposed dressings are the optimal dressing.

There is a high level of uncertainty in the optimal dressing type. Glue has a probability of being most cost-effective (almost as high as complex) and glue has the highest probability of being cost-effective when complex dressings are excluded (even though exposed dressings have the highest expected net benefit). However, this is an artefact of the extreme levels of uncertainty in the effectiveness of glue; it also has the highest probability of being the least cost-effective intervention. This uncertainty can also be seen in the cost-effective plane (Figure 5), with each dressing type demonstrating uncertainty whether it is more costly and less effective than simple dressings, or vice versa.

FIGURE 5.

Cost-effectiveness planes showing the joint distribution of incremental costs and effects for each dressing type compared with simple dressings. Base-case model. (a) Exposed vs. simple; (b) glue vs. simple; and (c) complex vs. simple.

Expected value of partial perfect information

The EVPPI (Table 18) can be interpreted as the amount a decision-maker would be willing pay to eliminate all uncertainty in a subset of the model inputs. It can be seen that there is no value in eliminating uncertainty in the cost of a SSI or in the SSI risk on the reference intervention (simple dressings). There is, however, considerable value in eliminating uncertainty in the relative effectiveness of the different interventions, with an EVPPI of £171 per patient when complex dressing is included as a decision option. This corresponds to a population EVPPI of £1760M, which is substantial. The increase in EVPPI is higher when complex dressing is excluded as a decision option (£222 per patient wound, £2295M per population over 5 years) owing to the greater uncertainty about the most effective dressing.

The EVPPI analysis therefore suggests that there is, potentially, value in conducting new research to obtain better estimates of the relative effectiveness of the different types of dressings. To explore the value of particular trial designs, we compute EVSI.

Value of Bluebelle randomised controlled trial: expected value of sample information

Scenario B: decision options – exposed (E), simple (S), glue (G)

Below we present results for EVSI when complex dressings are excluded, so that exposed, simple and glue are the only decision options for Bluebelle population surgery types. As for scenario A, in scenario B (decision options: E, S and G) even a small study (50 patients randomised) has value, although the benefits of a trial increase steeply as sample size increases, until the EVSI levels off for total sample size of more than approximately 3000 patients. The EVSI is higher for scenario B (decision options: E, S and G) (Figure 6; see also Appendix 4, Table 43) than for scenario A (decision options: E, S, G and A; see Appendix 4, Table 42 and Figure 18), owing to the increased uncertainty in the optimal dressing.

FIGURE 6.

Scenario B: decision options – exposed, simple, glue. EVSI per patient for balanced three-group designs plotted against total sample size, at a willingness to pay per QALY of £20,000.

Figure 7 shows the per-patient-wound EVSI for balanced designs with two or three groups. There is greatest value in a three-group trial of exposed versus simple versus glue, but the added value of including an exposed group to a two-group trial of simple versus glue is apparent only for sample sizes of greater than around 750. Of the two-group trials, there is greatest benefit in a simple versus glue trial, followed by exposed versus glue, and least value in a two-group trial of exposed versus glue. In general, there is greater value in trials that include a glue group, reflecting the lack of existing evidence on this dressing type.

FIGURE 7.

Scenario B: decision options – exposed, simple, glue. EVSI per patient for balanced designs plotted against total sample size, at a willingness to pay per QALY of £20,000. Results are shown for designs with different numbers of groups and different included interventions. E, exposed, S, simple, G, glue. EvSvG, three-group trial of exposed vs. simple vs. glue.

On the basis of EVSI, there is little to choose between a 1 : 1 : 1 or a 2 : 2 : 1 (as proposed for Bluebelle) allocation ratio for a three-group trial of exposed versus simple versus glue, given the same total sample size (see Appendix 4, Figure 21).

Sensitivity analyses

Tables 19 and 20 show results from sensitivity analyses for scenario A:

A – using the NMA results from Bluebelle population surgery types only

B – using Jenks et al. 50 alone to estimate SSI risk on reference (simple) dressing

C – assuming SSIQALYloss = 0.06

D – assuming SSIQALYloss = 0.

Using a lower SSI risk on the reference (simple) dressing (from Jenks et al. 50) reduces the overall value of the expected net benefit and, consequently, reduces the VoI. However, it does not change the probability that each dressing is the most cost-effective, nor the optimal dressing (complex). Similarly, reducing the value of QALYs lost as a result of a SSI reduces the overall value of expected net benefit and VoI but does not change the optimal decision or uncertainty in that decision.

Cost-effectiveness results depend on the evidence used to inform the NMA. If the relative effectiveness of the different dressing types is based on studies with Bluebelle population surgery types only, then the probability that a dressing type is most cost-effective increases, although the optimal decision is unchanged and, hence, our uncertainty in the decision is reduced, as is reflected in the lower estimate of VoI.

Incorporating the Bluebelle Phase B study results

We reran the NMA (‘all surgery types’ and ‘all wound types’) including the Bluebelle Phase B study results. Table 21 presents the results from an analysis using the ITT denominator and the PP denominator, alongside the results from the analysis omitting Bluebelle (taken from Table 22). The results from the ITT and PP analyses are very similar, suggesting that participants who do not complete the full assessments are not associated with the treatment allocation. The results are very similar to those obtained without the inclusion of Bluebelle Phase B. However, uncertainty in the estimates has been reduced by the inclusion of Bluebelle Phase B. Because the results are so similar, we present the cost-effectiveness and VoI results only for the ITT analysis (as reported by other studies included in the NMA).

Table 22 shows the cost-effectiveness results with and without the Bluebelle Phase B results. Expected net benefit is still highest for complex dressings and is very similar between exposed, simple and glue (although glue now has the highest expected net benefit of these three decision options). There is still a high degree of uncertainty as to which is the most cost-effective option, especially if complex dressing is removed from the decision options. Table 23 shows the EVPPI for subsets of parameters. Adding the Bluebelle Phase B results leads to a reduction in EVPPI, but the results still suggest that the relative effects are the key areas of decision uncertainty and that there is likely to be value in conducting a trial to reduce uncertainty in these parameters.

Appendix 4 (see Figures 22 and 23) shows per-patient EVSI for a range of different, balanced study designs (depending on the intervention groups included and overall sample size). Figure 22 shows results when complex dressings are included as a decision option (scenario A) and Figure 23 shows results when complex dressings are excluded as a decision option (scenario B). We see that, although EVSI is reduced when including the Bluebelle Phase B study results, the overall patterns and conclusions are similar to those seen when the Bluebelle Phase B study results were omitted from the analysis (see Figure 7 and Appendix 4, Figure 19). There is highest benefit in studies that include glue-as-a-dressing group (reflecting the high level of uncertainty as regards the effectiveness of this intervention). The added benefit of including an exposed group is lower when the Bluebelle Phase B results are included in the NMA. This is because the results obtained for exposed and simple dressings in Bluebelle Phase B are similar, reducing our uncertainty in this comparison. If complex dressings are considered an option, then there is value in including this in a new trial.

Multiplying results to give population EVSI gives very large figures, suggesting that a trial comparing dressing types is likely to be a good use of resources. The benefits increase with sample size, but at a decreasing rate, with a high proportion of the benefits accrued after 3000 patients are randomised.

Discussion

Limitations

The RCT evidence on which the cost-effectiveness and VoI modelling was based was limited and at potential risk of bias (all studies except one17 were rated as being at unclear or high risk of bias). Limitations in the existing evidence base, however, support the generation of more robust evidence via a future well-designed methodologically rigorous trial.

Under the Bluebelle definition, an exposed wound is one to which a gauze may be applied as needed to soak up exudate. Some of the included RCTs classified as ‘simple’ have applied a simple gauze, and one could argue that these could have been classified as ‘exposed’ under the Bluebelle definition. It is assumed that in a RCT the dressing would be always applied, rather than ‘as needed’, and for this reason the simple classification has been retained in these cases. Without more information than is reported in the including studies, this assumption is a potential a limitation of our analyses.

We pooled evidence from all wound types (clean, mixed and contaminated), owing to limited evidence. Although 9 of the 21 studies did not restrict to clean wounds, these studies randomised relatively small numbers of patients, and the results for those study populations were very uncertain. The results that we used in our cost-effectiveness analysis were driven primarily by the studies that restricted analysis to clean wounds only. Generalising the results to non-clean wounds should be done very cautiously, and this remains a question for further research.

To estimate SSI rate for the reference intervention (simple dressings), we were required to make some assumptions:

1. The proportion of surgery types missing in PHE51 is the same proportion of all surgery types seen in Jenks et al. 50

2. SSI rate in surgery types missing in PHE51 is the same as that seen in Jenks et al. 50

3. The split between surgery types seen in PHE51 (and where missing in Jenks et al. 50) is representative of the population of the whole of England and Wales.

The VoI analysis found that there was no value in reducing uncertainty in the SSI rate on the reference dressing, suggesting that our decision is unlikely to be sensitive to these assumptions. However, the overall value of a new study and the optimal sample size may depend on the SSI rate on the reference treatment. This is because we would expect more events in a population with a higher SSI rate, and, therefore, more precise results can be obtained for smaller numbers of patients randomised.

We have presented EVSI, which measures the benefits of a new study of a given design. But these figures should be considered together with the costs of such a trial. This can be done formally using the expected net benefit of sampling, which is the difference between the population-level EVSI and the cost of the study for a given study design. We prefer study designs that give a larger expected net benefit of sampling. However, given our results, it is clear that the population-level EVSI will be substantially higher than the cost of a trial for the range of sample sizes that we explored, suggesting that such a trial is likely to represent an efficient use of resources.

Numbers analysed

The analysis population consisted of 388 participants (i.e. the 394 randomised participants, excluding three participants who withdrew and were unhappy for their data to be used, two participants who were allocated to disclosure of dressing allocation after wound closure and whose randomisation in theatre was not completed, and one participant whose surgery was cancelled).

Participant demographics and operative characteristics

The demographic characteristics and past medical histories of randomised participants are shown in Table 24, and intraoperative characteristics are shown in Table 25, by allocated group. (The same information by centre is described in Report Supplementary Material 15.)

Primary feasibility outcome

With respect to the primary feasibility outcome, a total of 862 patients were approached and had their eligibility determined; 163 were found to be ineligible (see Figure 8). Of the remaining 699 eligible patients, 284 declined to take part (41%), 21 consented but were not randomised (3%) and 394 (56%) were randomised. (The same information by centre is described in Report Supplementary Material 15.) Table 26 shows the primary feasibility outcomes by centre.

Sample characteristics

Interviews were conducted with a total of 55 participants during the pilot RCT (professionals, n = 18; patients, n = 37). None of the individuals who agreed to be contacted refused participation or withdrew. Three out of the 18 professional interviewees took part in a group interview (from site 3). Table 27 shows the breakdown of professionals interviewed by site and by their principal role.

Of the 37 patient interviewees, five had been allocated to receive a ‘simple dressing’, 18 had been allocated to receive ‘no dressing’ and 14 had been allocated to receive ‘glue-as-a-dressing’. Each of these three groups comprised a mix of patients who had undergone general or obstetric surgery and laparoscopic or open procedures (Table 28). Seven patients took part in two interviews (one in the early stages of post-surgical recovery and one in the later stages).

Except for two, all patients had been admitted as elective cases. The two ‘unplanned’ cases were both abdominal surgery patients recruited from site 1.

Presentation of findings

Findings have been presented in two parts, each mapping on to one of the objectives for the integrated qualitative study. Part 1 focuses on issues of adherence, drawing on HCP and patient interviews; and part 2 presents HCPs’ and patients’ perceived acceptability of the study processes and trial comparison groups. Findings have been supported by illustrative quotations, some of which have been edited for ease of comprehension (without altering the meaning). The following identifiers have been used for quotations: I, interviewer; S, surgeon; SR, surgical registrar; N, nurse; RN, research nurse; TS, theatre staff; and P, patient.

Part 1: adherence

Part 2: acceptability of dressing strategies

Summary

This integrated qualitative study indicated that there had been no major adherence or acceptability issues in the pilot RCT. The variable awareness and uptake of skin transfers to promote adherence limited the possibilities for fully exploring stakeholders’ perspectives on these adherence aids; however, those who had used the transfers were generally positive about their application and utility, and those who had not encountered them envisaged that they could be helpful. Relying on written documentation to flag patients’ participation in Bluebelle was thought to be unreliable, as this was not necessarily consistently reviewed at the start of shifts. Notably, patients’ understanding and awareness of the Bluebelle study and their allocated dressing strategy appeared to be key in promoting adherence.

In general, there was no indication that the Bluebelle study processes had any major impact on HCPs’ usual practice. Wound management for patients with ‘no dressing’ and ‘glue-as-a dressing’ was not perceived to be any different from that for patients who received simple dressings (i.e. routine care). The only exceptions to this were accounts of being more careful with cleaning undressed wounds, and the possibility of more incidental observation of wounds that were visible (i.e. the ‘no dressing’ and ‘glue-as-a-dressing’ groups).

Interviews with patients and HCPs were typically characterised by an absence of notable issues or concerns about any aspect of the Bluebelle study or its impact on routine care. Although a few patients who received ‘no dressing’ experienced issues with exudate, they did not attribute this to their allocated dressing strategy, believing that these issues would have still arisen had they received a dressing after surgery. Although some patients and HCPs talked about specific advantages of ‘no dressing’ or ‘glue-as-a-dressing’, it was the absence of issues (rather than notable positive experiences) that underpinned the sense that a future RCT would be acceptable to key stakeholders.

On a logistical note, research nurses from all sites indicated that a future trial’s success would be contingent on upfront training to optimise staff engagement and co-ordination across clinical teams. Although they had successfully delivered the pilot RCT, most felt the current model of having research nurses lead most components of the trial would not be feasible on a larger scale.

Adherence to allocation

Adherence to the timing of disclosure of the dressing allocation and to the dressing allocation itself by group is shown in Table 29. (The same information by centre is described in Report Supplementary Material 15.) Overall, adherence was good, with 99% adherence to the timing of disclosure when allocated to disclosure before wound closure and 86% when allocated to disclosure after wound closure. Adherence to the allocated dressing was > 97% for the initial dressing and > 86% for patients requiring one or more wounds to be redressed. Redressing wounds in the glue-as-a-dressing group was rare but, when required, glue was not used to redress the wound.

Co-interventions

Co-interventions by group are shown in Table 30. (The same information is described by centre in Report Supplementary Material 15.) Most wounds were closed with sutures and approximately three-quarters of participants were prescribed prophylactic antibiotics. There is no indication that these co-interventions were used differentially by group.

Adherence to the follow-up protocol

Participation in follow-up is described by group in Appendix 5 (see Table 44). (The same information is described by centre in Report Supplementary Material 15.) The WMQ and WEQ at 4 days were completed for > 90% of participants (355/385) and completion rates were similar in the three groups. Completion of the EQ-5D-5L was excellent at recruitment (385/388, > 99%) but fell to 269 out of 382 (70%) at 15 days and 242 out of 377 (64%) at 4–8 weeks. Again, response rates were similar across the three treatment groups. The response rate for the participant-completed WHQ was similar, at 254 out of 378 (67%). Face-to-face assessments were higher: 303 out of 377 (80%) had a face-to-face SSI reference assessment and 281 out of 378 (74%) had an observer-completed WHQ assessment. The participation rate in the reference assessment was highest for patients allocated to no dressing (107/128, 84%) and lowest for those who had a simple dressing (97/127, 76%) but this trend was not observed for the observer-completed WHQ assessment.

The SSI reference assessor was unblinded in 58 out of 302 (19%) assessments, most often for participants who had their wounds dressed with glue (31/100, 31%) and least often for those who had a simple dressing group (11/96, 11.5%). The overall unblinding percentage for observer-completed WHQ assessments was 22% (28/177) and was similar across the groups.

Completeness of data for outcomes anticipated to be measured in the main trial

Completeness of key data items by group is shown in Appendix 5 (see Table 45). (The same information is described by centre in Report Supplementary Material 15.) The reference SSI assessment, when done, was well completed. Wound complication data were completed for 326 out of 388 (84%) participants during the postoperative hospital stay and for 315 out of 378 (83%) participants at 4–8 weeks, with similar completion rates for the three wound-dressing groups.

Resource use data were well completed (> 97%), with the exception of data relating to the level of care that participants received during their postoperative stay, particularly the time of admission to a particular level of care, which was complete for approximately three-quarters of participants.

Health economics

The numbers of available observations for the five cost categories are described in Appendix 6 (see Table 48). The percentage of participants who contributed data for each of these categories ranged from 81% (for primary care appointments and use of medicines) to 98% (for dressings taking place during the follow-up period) of the total number of participants. The CCA is reported below and the ACA in Appendix 6 (see Tables 47–49 and Figure 26).

Complete-case analysis

Complete-case analysis comprises patients for whom all data are available. Effectively, this involves excluding patients for whom data on one or more cost categories are missing.

Overall, the mean per-patient cost was higher in the ‘no dressing’ group (approximately £147) than in the ‘simple’ and ‘glue-as-a-dressing’ groups. Participants allocated to simple dressing were associated with a mean per-patient cost of about £146, whereas those who were randomised to receive ‘glue-as-a-dressing’ presented the lowest mean cost (£118) (Table 32).

TABLE 32 - Total cost per allocated intervention (CCA)

SE, standard error.

Estimated using bias corrected and accelerated bootstrapping.

Allocated intervention	Cost (£)
	Mean	SEa	95% CI
No dressing	147.13	68.05	13.76 to 280.50
Simple dressing	145.69	32.87	81.26 to 210.11
Glue	118.19	16.58	85.70 to 150.68

The cost associated with each resource use category by treatment group is detailed in Table 33 and depicted in Figure 9. In the ‘no-dressing’ group, the greatest share of the total per-patient cost is attributed to hospital visits after discharge, followed closely by the cost of new medicines. In the ‘simple dressing’ group, the main contributor to the total cost is the cost of new medicines, followed by the cost of in-hospital dressings and redressings, and the cost of hospital appointments after the initial discharge. In the ‘glue-as-a-dressing’ group, the largest contribution to the total cost is the cost of dressings and redressings in the hospital, followed by the costs associated with primary care appointments and new medicines.

TABLE 33 - Total cost by allocated dressing (CCA)

Cost category	Number of observations	Percentage of all patients (n = 388)	Total cost per patient (£)	Total cost by allocated intervention (£)
				No dressing	n	Simple dressing	n	Glue-as-a dressing	n
Cost of dressings and redressings in hospital	268	69	33.54	6.86	96	42.72	84	53.89	88
Cost of hospital appointments after initial discharge	268	69	34.36	59.99	96	23.59	84	16.66	88
Cost of primary care appointments after initial discharge	268	69	16.54	14.99	96	15.29	84	19.45	88
Cost of redressings after initial discharge incurred by NHS	268	69	8.21	4.66	96	11.17	84	9.25	88
Cost of redressings after initial discharge incurred by patients	268	69	0.38	0.04	96	0.59	84	0.56	88
Cost of complications arising while in hospital	268	69	1.15	3.22	96	0.00	84	0.00	88
Cost of new medicines	268	69	42.99	57.38	96	52.33	84	18.39	88
Total cost (CCA)	268		137.18	147.13	96	145.69	84	118.19	88

FIGURE 9.

Total cost by cost category and allocated intervention (CCA).

Differences in mean per-patient cost between trial groups are presented in Table 34. Although the mean cost differences range from approximately £20 (for no dressing vs. simple dressing) to > £50 (for no dressing vs. glue-as-a-dressing), none of these differences reached statistical significance.

TABLE 34 - Differences in costs between allocated interventions (CCA)

Estimated using bias corrected and accelerated bootstrapping.

Allocated intervention	Cost difference (£)	p-value	95% CI (£)a
No dressing vs. simple dressing	1.45	0.985	–146.87 to 149.76
No dressing vs. glue	28.94	0.673	–105.56 to 163.44
Simple dressing vs. glue	27.49	0.459	–45.30 to 100.28

Discussion

The findings of this pilot study showed that it is feasible to obtain detailed information on relevant resource use categories, including the type of dressings used, the frequency of dressing changes and use of health-care resources in response to wound-related problems. The proportion of available data ranged from slightly over 80% to 98% for all categories of resource use; a future full trial could use imputation methods to ameliorate problems of missing data.

The fact that much of the information for the cost analysis was sourced from CRFs meant that participants were called to provide information only in cases in which this was not available through routine sources, which avoids placing a great burden on participants. 103 This, and the short time period within which this information was collected, helped to alleviate the issue of recall bias, that is, bias arising from the fact that respondents are asked to recall all instances of health-care use based on memory. 104

The analysis showed that some additional information would be useful to collect in a definitive trial. In particular, more detailed information is needed around the place where dressing changes were performed during the follow-up period (e.g. at a patient’s home, during a follow-up appointment), as well as detailed information on the person who carried out (or assisted with performing) the dressing change in the hospital (e.g. grade of hospital-based nurse) or at a patient’s home (e.g. participant, family member, community nurse). It would also have been useful to have details on additional care provision for non-serious SSIs. More details around patients’ travels to hospitals would allow a better appreciation of the costs associated with such trips.

Translating the use of resources into costs showed that ‘glue-as-a-dressing’ was associated with the lowest cost, followed by ‘simple dressing’ and ‘no dressing’, in this small-scale feasibility study. Cost drivers varied across dressing options, although it is evident that key cost drivers were hospital appointments, dressings and redressings, use of new medicines and primary care appointments. As expected, the cost of dressings and redressings made a sizeable contribution to the total cost associated with ‘simple dressing’ and ‘glue-as-a-dressing’, although, unsurprisingly, it did not have a notable influence in the total cost for the ‘no-dressing’ group. In general, instances of care that are particularly costly were rare.

Overall, the collection of economic data for this trial proved feasible and the major cost categories that should be focused on in a future trial have been established.

Data sets

In total, data from 792 and 791 participants from the cohort study and pilot RCT contributed to the WHQ self-assessment and observer-assessment validation analyses, respectively (Figure 10).

FIGURE 10.

Phase A and Phase B participant data contributing to validation of the WHQ.

Analyses

Participant and observer agreement (inter-rater reliability)

Data from a self-assessment and an observer assessment were available for 470 out of 791 (59.4%) participants. The median time between self-assessments and observer assessments was 8 days (IQR 2–16 days).

Agreement between self-assessments and observer assessments for signs and symptoms (items 1–8) was generally high, although patients showed a trend to report signs and symptoms slightly more severely than observers. This is demonstrated in Figure 11.

FIGURE 11.

Bar charts comparing responses to the participant self-assessment and the observer assessment, for participants with data from both assessments (n = 470). (a) Item 4b: Did the deeper tissue separate?; (b) Item 5: Has the area around the wound become swollen?; (c) Item 6: Has the wound been smelly?; (d) Item 7: Has the wound been painful to touch?; (e) Item 8: Have you had, or felt like you have had, a raised temperature or fever (fever ≥ 38 °C)?; (f) Item 9: Have you sought advice because of a problem with your wound?; (g) Item 10: Has anything been put on the skin to cover the wound? (dressing); (h) Item 11: Have you been back into hospital for treatment of a problem with your wound?; (i) Item 12: Have you been given antibiotics for a problem with your wound?; (j) Item 13: Have the edges of your wound been deliberately separated by a doctor or nurse?; (k) Item 14: Has your wound been scraped or cut to remove any unwanted tissue (debridement of wound)?; (l) Item 15: Has your wound been drained (drainage of pus/abscess)?; and (m) Item 16: Have you had an operation under general anaesthetic for treatment of a problem with your wound?

Although cases were rare, examination of the raw data showed that there was slight discrepancy in agreement between self-assessments and observer assessments for some of the intervention items, with patient reports of having had that intervention not being supported by the observer assessment (see Report Supplementary Material 16). For example, six patients responded that their wound had been scraped or cut to remove unwanted tissue (item 14). Observer data, however, agreed with only two of these cases. Furthermore, observer assessments implied that another two patients had had debridement of their wound, which was not indicated by the patients themselves.

Sensitivity and specificity of the Wound Healing Questionnaire for surgical site infection/no surgical site infection discrimination

A simple summated scoring system without weighted scores was used, supported by findings from the scaling analyses, which meant that the WHQ total score had a possible minimum and maximum range of 0–44. Actual scores in the data from self-assessments ranged between 0 and 30 (median 3; IQR 1–5).

A contingency table of raw data showing the number of patients scoring all levels of WHQ total score against the reference assessment SSI diagnosis is provided in Report Supplementary Material 19. Sensitivity and specificity for selected WHQ ‘cut-off’ scores are shown. Analyses showed that the sensitivity and specificity of the WHQ patient self-assessment for discriminating SSI/no SSI was high, with an area under the ROC curve of 0.9056 (95% CI 0.8271 to 0.9841) (Figure 12).

FIGURE 12.

Receiver operating characteristic curve for overall WHQ score (after item reduction) for discriminating SSI diagnosis compared with reference assessment (CDC SSI classification). Area under ROC curve = 0.9056.

Revised Wound Healing Questionnaire

The findings from the validation study were considered and a revised version of items for the WHQ was produced (see Appendix 7).

Revisions were made as follows:

1. Subitems that were specific to collecting health economic data for the purpose of the pilot RCT (10a–d, 11a–e and 12a and b) were removed.

2. Subitem 4a (‘Did the skin separate?’) was removed owing to the overlap with core item 4, ‘Have the edges of any part of the wound separated/gaped open on their own accord?’ (spontaneous dehiscence).

3. Core item 3 (‘Was any part of the wound leaking fluid?’) and the associated subitems 3a–c were combined and restructured to form three separate items to minimise missing data and avoid overlap in responses.

4. Items were renumbered to accommodate these changes.

5. Any ‘do not know’ response categories were dropped.

Further testing and validation of the WHQ is now needed in the light of these minor revisions.

Digital photographs after wound closure in theatre

As described above (see Objective A7: investigate the feasibility of photographing wounds in theatre and assessing the quality of wound closure), we did not attempt to implement obtaining digital wound photographs after wound closure in theatres for logistical reasons.

Digital photographs by participants after discharge

During the last month of recruitment, we piloted the submission of wound photographs taken by participants after discharge from hospital in three of the four participating trusts. Those participants who had consented to this substudy and who had e-mail addresses were sent instructions by post and then were e-mailed a secure web link, which invited them to upload a photograph securely to the study database; those who did not submit a photograph were subsequently sent a reminder. The flow of participants during this pilot is shown in Appendix 8 (see Figure 27). Of the 26 participants who consented to the substudy and provided an e-mail address, nine (35%) submitted a photograph.

Meeting 1: informing the design of the pilot randomised controlled trial (Phase B)

Three key issues were discussed.

Meeting 2: informing plans for a possible randomised controlled trial

Two key issues were discussed.

Impact of patient and public involvement

Members said that they had enjoyed being involved in a PPI group and attending the meetings. Written feedback on the PPI meeting was rated as excellent (4/7 response) or very good (3/7 responses). Comments from members included that PPI was regarded as very important and that it had been valuable to meet others and share experiences.

One PPI member read and commented on the lay summary for this HTA report. All members wanted to be kept informed about the study, for example in terms of whether or not a large trial will go ahead and any findings and results.

Key issues and suggestions will be taken forward and incorporated into plans for the design and conduct of a future large RCT and will inform the development of PILs and interview topic guides for a study to test methods for own-wound photography.

Phase A

The study comprised multiple elements that investigated the feasibility of a main trial. These were found to be valuable to inform the pilot RCT design. However, the complexity of the work, combined with the requirement to design and set up the pilot RCT within the time frame, was challenging.

The qualitative research showed that additional work was needed. This was achieved by increasing collaborations and gaining an extension to Phase A. Collaborations with the trainee research collaboratives, the Medical Research Council (MRC) ConDuCT-II Hub (NJW for objective A7 and LR and DE for objective A4), and with the Cochrane wounds group (JCD) were established. The funded extension allowed the additional work to be completed to inform the pilot RCT.

The process of applying for funding for a commissioned topic is the same, regardless of whether the topic is requesting a feasibility, pilot or substantive study. The Bluebelle study demonstrates the difficulty of defining the scope of initial feasibility research in advance of doing the research. The research expanded to address additional uncertainties that were identified from the planned feasibility work. These included a survey of current practice, a VoI analysis, literature reviews and the development of secondary outcome measures.

The terminology used in the commissioning brief illustrates this issue. The qualitative case studies carried out to address objectives A1 and A2 found that the concept of a ‘complex’ dressing for a primary closed surgical wound was not recognised by staff. Informants reported examples of using products not traditionally marketed as dressings (e.g. tissue glue) to cover primary wounds. This led to the survey of dressings use in the NHS (objective A3). A second example relates to our original objective to develop a patient-centred measure of practical wound management (A4). The case studies and literature showed that two new measures, one of wound management and one of patient experience, were needed rather than a combined measure (as conceived by the Cochrane systematic review). 33 Therefore, more research was required than had been anticipated in Phase A.

On reflection, feasibility research shares some features with an adaptive trial design. 105,106 When feasibility research is commissioned, it is likely that there will be considerable uncertainty about the precise nature of the research. If the research is going to address fully both the uncertainties identified in a commissioned brief and the emerging uncertainties, both the researchers and the funders may need to be flexible in their responses to new and emerging findings. The Bluebelle study demonstrates this flexibility, to the extent that the HTA programme agreed contract extensions and additional funding. However, the ability to draw on wider collaborations (above) was vital in ensuring that the study generated the maximum value.

The need for flexibility when undertaking feasibility research potentially creates risks; researchers may seek extensions, justifying these as demonstrating the required flexibility to adapt to emerging findings, when the objective of the extended research is not sufficiently aligned with the original commissioned topic. Arguably, this is no different from any request for a contract variation to extend a study. However, in a ‘standard’ adaptive trial, a proposal to add or drop a group is similar to prioritising any effectiveness research question, so it is likely to be familiar to panels and commissioners. In the case of feasibility research, it would be important to ensure that the emerging findings and request for extensions truly did align with the NHS agreed research priorities.

Phase B

It was planned that recruitment to the pilot RCT would start 3 months after completion of Phase A. REC approval for the pilot RCT was applied for in advance of completion of Phase A. It was envisaged that approval for the main elements of the RCT could be obtained and an amendment to update the approval for the latest findings from Phase A could subsequently be submitted.

• The approach with respect to the REC was successful but research to address some Phase A objectives extended up to and beyond the start of Phase B.

• Given that the primary objective of Phase B was to test whether or not recruitment would be satisfactory, starting to recruit to the pilot trial was prioritised over the need to have all of the trial management features in place. For example, local research teams were able to consent and randomise participants but CRFs for later stages of follow-up had not been finalised and the database for data entry was not ready. Not having all of the trial management features in place at the outset made trial conduct inefficient and frustrating for all involved.

• Other consequences were that processes for following up participants were largely manual, which may have influenced the questionnaire response rates and led to a backlog of data entry.

• It became apparent as the study progressed that the CRFs contained redundant items. We had intended to pilot the CRFs and optimise them during the study (as is customary in a pilot). This was not possible with the resources available; once recruitment started, all of the resources were invested in this.

• Submission of the substantial amendment coincided with the introduction of the new NRES/HRA systems. This led to a 3-month delay in implementing a revised and improved consent process for patients having emergency surgery and obtaining approval to open a fifth centre.

Phase A

The main Phase A findings were as follows.

It was found that the terminology for dressing types used in the HTA commissioning brief was not recognised by health professionals and did not appear to be relevant to the management of primary wounds (objectives A1 and A2). Dressings were referred to by their trade names. The selection of dressing type for use on primary abdominal wounds was reported to be dependent on hospital policy rather than clinician preference and did not vary between emergency and elective settings. Complex dressings as envisaged in the commissioning brief (e.g. antibiotic-impregnated dressing) were rarely used, and primary wounds were generally covered. Some interviewees described the use of tissue adhesive as a dressing on a closed wound, which was one reason for surveying current practice.

The survey (objective A3) found that simple dressings were used most frequently on primary abdominal wounds and that tissue adhesive was also used as a dressing (27%) despite a lack of evidence for this practice. Patients and staff reported that it was acceptable to leave a closed wound uncovered in the context of a RCT, acknowledging that there were uncertainties around the role of dressings and SSI outcomes. They postulated there might be practical and psychological advantages and disadvantages around dressing use, indicating the importance of a future RCT.

It was planned to develop and test a questionnaire to ascertain SSI after discharge to be completed at 30 days, potentially supplemented by a wound photograph taken by patients themselves (objective A4). The intention was to devise a way to reduce the research costs of collecting outcome data in a main trial (i.e. avoiding the need for a participant to have a SSI assessment by a trained researcher/clinician, whether in hospital or at home), while at the same time developing a validated instrument that could be used for SSI surveillance. The questionnaire, for completion by a patient or an observer, was successfully developed. After refinement through cognitive interviews, this WHQ included 16 items; the WHQ was acceptable and quick to complete, and was completed with few missing data. Two new measures of wound management and patient experience (WMQ and WEQ) were developed (objective A5), based on the literature and interviews to identify domains.

Working definitions of ‘simple’ and ‘complex’ dressings were established and the parameters of ‘no dressing’ were defined (objective A6). It was found to be feasible to photograph wounds in theatre but challenging to obtain the necessary approvals to do so (over and above the research approvals). Surgeons’ perceptions of what constituted ‘good quality’ wound closure were investigated, as were methods to assess these (objective A7). It was shown that a definitive trial, comparing three groups (simple vs. complex dressing vs. no dressing) or only two groups (simple vs. complex dressing), would be valuable to the NHS (objective A8). All of these findings were used to design the pilot RCT (objective A9) carried out in Phase B.

Phase B

The main Phase B findings were as follows.

Participants were successfully recruited and randomised, exceeding the target sample size within the scheduled recruitment period of 9 months, and 59% of approached eligible patients consented to take part (objective B1). The consent rate might be higher in a future main RCT because we experienced constraints on recruitment in the pilot RCT (see Strengths and limitations of the study). The qualitative research supported the conclusion that randomisation to the different dressing strategies was acceptable to patients and professionals (objective B2).

There were similarly consistent qualitative and quantitative findings about adherence to the allocated dressing strategies (objectives B3 and B4); it was found that adherence was not problematic and that the adherence aids that were used (notably the skin transfers) were acceptable. Adherence to the allocated dressing was > 97% for the initial dressing and > 86% for patients who needed to have a wound redressed. Adherence to the timing of disclosure of allocation was 99% when disclosure was before wound closure and 86% when disclosure was after wound closure.

Adherence to the follow-up protocol (objectives B4 and B5) varied depending on when outcomes were collected. More than 90% of participants completed the WMQ and WEQ at 4 days; the EQ-5D-5L was well completed at recruitment, but less so at 15 days (70%) and 4–8 weeks (64%). The response rate for the participant-completed WHQ at 4–8 weeks was 67%. The reference assessment was carried out for 80% of participants, with blinding of this assessment being compromised in 19% of participants.

Considering resource use as a secondary outcome, there were no obstacles to collecting the data required (available for > 80% of participants in all categories identified and > 95% of participants in relation to dressings and redressings, both before and after discharge). The costs analysis showed that some additional information would be useful, including information about dressing changes after discharge (by whom and where), additional care provided for non-serious SSIs, and how patients who need additional wound care travel to hospital. Cost drivers varied across dressing options, although it is evident that key cost drivers were hospital appointments, dressings and redressings, use of new medicines, and primary care appointments. In general, instances of care that were particularly costly were rare and, therefore, did not have a major prominent effect on the mean cost per patient.

The WHQ was implemented in a cohort study in Phase A and in the pilot RCT, and data contributed to validation of the questionnaire (objective B6). The response rate at 4–8 weeks (67%) would probably be considered insufficient for the primary outcome of a main trial. However, the low response rate may be due, at least in part, to the limitations of the pilot RCT (see Strengths and limitations of the study), which could be addressed. With respect to a main trial, the authors recommend that researchers implement measures to enhance the response rate in an internal pilot phase and specify an adequate response rate as a progression criterion.

The WHQ was found to have good psychometric attributes: when the questionnaire was completed, few data were missing, and responses were distributed as expected and were consistent with the frequencies of observed wound complications and SSI. Agreement between participant and observer, and within participants, was good, although, compared with observers, patients tended to report signs and symptoms as being slightly more severe. All items in the WHQ fitted a single plausible scale structure; factor analyses of Phase A data supported a single factor model, which was replicated using data from the pilot RCT. Cronbach’s alpha coefficients for a single scale were between 0.8 and 0.9 in both data sets, whether the WHQ was patient completed or observer completed. The new measure demonstrated good sensitivity and specificity compared with the CDC clinical assessment.

The short duration of the pilot trial meant that we were able to test the taking and submitting of wound photographs (objective B7) in only a small subgroup of individuals. We showed that it was possible to implement secure uploading of photographs, linking the photographs to other trial data, but only one-third of participants who consented to this substudy uploaded a photograph. The impact of involving patients and the public in the design of Phase B (objective B8) is covered below (see Patient and public involvement and engagement). Our proposals for a main trial (objective B9) are discussed in Future research recommendations.

Strengths

The collaborations within the Bluebelle group made it feasible to conduct multiple preliminary studies to inform a pilot RCT. This was a major strength because the multidisciplinary inputs were critical to the pilot design. Without the MRC ConDuCT-II Hub and close working with the surgical trainee research collaboratives this may have not been feasible because of the range of skills required. These collaborations enabled us to complete and publish within the extension period for Phase A, and to recruit into the pilot trial in Phase B on time and target. Participants engaged extremely well with the whole study.

We included a new centre, a district general hospital that was keen to participate, towards the end of the pilot RCT (month 7 of 9, delayed by the HRA moratorium). We included this centre to test the feasibility of delivering the study in a less experienced setting; it was the first time that the general surgical team had participated in a National Institute for Health Research (NIHR) portfolio study. The hospital recruited successfully for 3 months, confirming the acceptability of the design to patients and staff.

There were some teams that had not used tissue adhesive as a dressing prior to participation in this study and they requested training. This was provided by a principal investigator from one trust, who visited the team to demonstrate the intervention. Application of tissue adhesive on a closed wound is a straightforward process and one training session was sufficient to build the team’s confidence. It was also necessary to reassure the teams about the safety of leaving wounds exposed. None had routinely left wounds exposed without a dressing prior to the pilot study. We consider the successful recruitment and adherence rates as an important strength of the pilot trial, given the unfamiliarity of some local surgical teams with the interventions.

We used skin transfers close to the wound(s) to provide an immediate reminder of the dressing allocation. Although there were some concerns from nursing staff about the skin transfers, the qualitative research and PPI group confirmed that these were acceptable to patients and were a useful method to promote adherence. We recommend that this simple aid to adherence be employed in a main trial.

The outstanding research required to inform a main trial is described below (see Future research recommendations).

Limitations

The multiple activities required for the study, largely carried out in parallel with one another, were challenging to co-ordinate and caused some limitations. Research nurses and the trainee collaboratives were involved with recruitment, which was initially found complex to organise. Centres were required to set up additional processes to streamline communication between the teams. Trainees often identified patients and gave out information sheets. The nurses were then required to recruit and follow them up. A main trial would require additional administrative support and formal processes to co-ordinate the efforts of the different teams and trainees and track the large numbers of people involved.

A participant in the Phase B qualitative interviews described the Bluebelle study as having ‘many moving parts’. This sense of too many things going on at once, which were not always co-ordinated as well as we had hoped, was reflected in qualitative interviews with staff. Research nurses from all centres emphasised that the success of a future main trial would depend on staff engagement and co-ordination. The pilot RCT would not have succeeded without commitment from the research nurses, who went beyond the call of duty (the time of local research nurses was not costed in the budget, except in the case of a research midwife); the research nurses felt that the model of having research nurses lead most components of the pilot trial at a centre level would not be feasible on a larger scale.

A separate cohort study was added to Phase A to inform the validation of the WHQ, providing a larger sample for characterisation of some attributes of the WHQ and the opportunity to explore the scale structure in one sample and replicate it in the second. However, the Phase A cohort study captured fewer operative data than the pilot RCT and only a relatively small proportion of participants had a reference SSI assessment. The Phase A cohort study was resourced mainly through wider collaborations established for the project but, nevertheless, may not have represented the best use of these resources.

The pilot RCT could have recruited more quickly had it not been for the HRA moratorium on approving amendments, which delayed implementation of a discretionary time window between giving the PIL and requesting consent. This limited recruitment of patients having emergency operations and delayed the opening of the fifth centre.

The numeric description of adherence to timing of disclosure of dressing allocations disguises the fact that some centres devised ‘workarounds’ to make disclosure after wound closure more practicable – avoiding, in practice, the need for teams to log into the database twice. The short time taken to close a wound (< 5 minutes), and allowing for small discrepancies between theatre clocks (the time of wound closure had to be entered into the database manually in order to disclose the dressing allocation) and the server clock automatically recording time in the database (when the wound closure time was entered), makes these adherence percentages somewhat uncertain, but not to an extent that would undermine our conclusion about the feasibility of disclosing the dressing allocation after wound closure. Randomisation in theatre did work but was challenging for some surgical teams/hospitals, hence the need for workarounds. These included telephoning a research nurse (outside theatre) to ask him or her to enter the wound closure time and disclose the dressing allocation.

The logistics of obtaining the patient-completed WHQ, an observer-completed WHQ and an independent (blinded) reference SSI assessment at 4–8 weeks were challenging, because this required two members of staff. Some centres were not able to do this. Others co-ordinated the work with research nurses, academic surgeons and members of trainee research collaboratives. The low proportion of participants who had a reference SSI assessment might have introduced attrition bias (i.e. owing to participants being more or less willing to have a reference assessment depending on how the wound had healed). We cannot rule out this limitation but note that the percentages of participants who had a SSI were similar in each group, and that these percentages are consistent with published SSI percentages in similar clinical settings. It is important to remember that potential bias with respect to the SSI percentages was less important here because of the feasibility and pilot objectives. A main trial would be collecting data for a single measure of SSI at 4–8 weeks, not three measures.

Although the study recruited well, the response rates for the patient-completed outcomes at 4–8 weeks were 64% (EQ-5D-5L) to 67% (WHQ), which are considered inadequate for the primary outcome in a main trial. Owing to the manual processes for follow-up, we did not have up-to-date reports on response rates. We believe that it is possible to improve the response rate substantially and recommend that researchers be required to demonstrate a high response rate in an internal pilot phase of a main trial. Bluebelle hospitals achieve response rates of > 85% for periodic SSI surveillance in usual care for PHE.

We were unsuccessful in implementing the capture of digital photographs in theatre after wound closure, which prevented us from attempting to quantify the quality of wound closure and assessing potential performance bias from knowledge of the dressing allocation in advance of wound closure. We faced several challenges in capturing and transmitting photographs of wounds at the end of the procedure. Although we had obtained ethics approval and had consented patients to take photographs and use them, there was resistance from theatre staff. Each trust required us to apply additional local governance processes for digital data capture and transfer. These procedures in each of the trusts were burdensome, and it was not possible to implement them in all centres. With more time, we think that some of these obstacles could be surmounted. Jane M Blazeby has recently appointed a research photographer through the MRC ConDuCT-II Hub and her NIHR Senior Investigator award to investigate these issues.

The challenges in participants taking photographs of their own wounds and submitting them securely were different. We devised a way to send a single-use weblink to participants by e-mail that allowed a participant to upload a photograph securely and directly into the study database. We implemented this system during the last month of recruitment, sending instructions by post about how to take a photograph before e-mailing the weblink. We showed that the method worked, but only one-third of participants who agreed to do so uploaded a photograph (with one reminder). We did not have time to investigate further whether participants understood the instructions or experienced a problem with taking a photograph or following the weblink. We expect that this can be improved, and Rhiannon C Macefield is continuing this work as part of her Doctor of Philosophy project.

In the pilot RCT, data collection forms were completed on paper and entered into the online study database by the local research teams. This conventional method of data capture provides a clear audit trail but many of the surgeons involved had worked in other trials for which data had been directly entered online. We recommend that direct data entry be used in a main trial.

Further considerations in relation to a main trial

This study has shown that a main trial is feasible (see The health technology being assessed: dressings for primary surgical wounds) and the VoI analysis has established that the evidence generated by a main trial would be valuable to the NHS (see Objective A8: analyse the value of information to the NHS that would be provided by a definitive trial). We have outlined the main research questions for a three-group (as in the pilot RCT) or a two-group trial (simple dressing vs. glue-as-a-dressing) and proposed sample sizes in Objective B9: design a large, definitive RCT based on information from the pilot trial and from integrated and interactive meetings with nurses/midwives, surgeons, methodologists and patient partners. In addition to the choice between a three-group or two-group trial, some other aspects have not been resolved, namely ensuring an adequate response rate for the primary outcome, establishing a threshold for defining SSI when using the WHQ and choosing the time of randomisation.

Our preference is for a three-group trial, despite the daunting size of the proposed trial. The reason for this preference is to do with the potential interaction between the allocated dressing strategy and the quality of wound closure (which also affects the choice of time of randomisation). A key decision for a commissioner of a main trial is whether the question of interest is specifically about the effect of different dressing strategies or about the combined effect of different dressing strategies and their – potential – effect on the quality of wound closure. For example, one might hypothesise that glue-as-a-dressing reduces the risk of SSI because surgeons close the wound more carefully; this effect would not be evident if any effect of dressing strategy on wound closure was excluded by randomising after wound closure. We were unsuccessful in the pilot trial in answering the question of whether or not an interaction exists because we were unable to implement the capture of wound photographs in theatre on a routine basis. Nevertheless, we believe that the two-group trial comparing a simple dressing and glue-as-a-dressing would also be very important to the NHS because, if glue-as-a-dressing does not have benefits, the NHS needs to try to stop the spread of this practice as soon as possible given the relatively high cost of tissue adhesive.

We propose that the primary outcome should be a combination of information about SSI collected at discharge (as in this study) and SSI ascertained by the patient-reported WHQ. Using this primary outcome would require (1) a much better response rate for the WHQ than we achieved in the pilot RCT and (2) a threshold score for the WHQ to define SSI after discharge. We do not believe that a conventional ‘reference’ SSI assessment would be practicable to use as the primary outcome in a main trial because of the high cost, and the validation achieved in the Bluebelle study demonstrates that the WHQ identifies a high proportion of SSIs.

As described above, we believe that the response rate achieved in the pilot RCT could be greatly improved, primarily by implementing existing evidence-based methods107,108 and by using information technology solutions such as text messaging. We recommend that, if the HTA programme decides to commission a main trial, applicants should be required to include an internal pilot, progressing to a main trial only if the WHQ response rate at 4–8 weeks can be shown to exceed 90%.

Choosing an optimum WHQ threshold score to define SSI after discharge is not straightforward. The point on the ROC that lies further from the leading diagonal minimises total misclassifications but assumes that the clinical and economic consequences of false-negative and false-positive misclassifications are the same. This is unlikely to be the case: the consequences of an undiagnosed SSI are likely to outweigh those of treating a suspected SSI that, in fact, was not a SSI. We intend to research an optimum threshold using quantitative information collected in the study and consensus methods with the Bluebelle co-investigators. It is also important to explore whether or not subthreshold signs of SSI collected at discharge, combined with information from the WHQ, might improve diagnostic accuracy. The ASEPSIS instrument combined information collected in hospital and after discharge in an additive manner,32,109 whereas we have assumed that SSI should be defined by a Boolean ‘or’ operator (SSI ascertained in hospital or SSI ascertained after discharge, applying a WHQ threshold score). We propose to continue to research the WHQ to address these issues. We have incorporated the WHQ in the HTA-funded CIPHER cohort study (reference number 14/166/01), which aims to recruit 4000 participants having abdominal surgery. The data for the WHQ collected in CIPHER will complement those already collected.

Team members could not agree on the best time to disclose dressing allocation in a main trial, primarily because of differing opinions about the question of interest (see above), but also for logistical reasons. Although the surgical teams had worked hard to achieve good adherence to the timing of wound disclosure in the pilot RCT, surgeon members of the team felt that disclosing dressing allocation after wound closure would be very difficult to deliver consistently in a main trial, given its proposed magnitude. The triallists, however, were confident that, without the factorial aspect of the pilot RCT, it would be possible to disclose allocation after wound closure in a convenient manner. Both groups agreed that it would be very difficult to achieve a factorial design in a main trial and, therefore, the timing of disclosure of allocation needs to be decided by clearer specification by the HTA prioritisation panel of the research question of interest.

With respect to the timing of a future main trial, we recommend that the HTA programme waits for 12 months until it becomes clearer (1) how well the NHS can support the large wound trials that have recently been funded and that are starting to recruit (e.g. see www.birmingham.ac.uk/research/activity/mds/trials/BiSTC/trials/ROSSINI-2.aspx; accessed 25 April 2018) and (2) whether or not the large investment required for a main trial (on top of the existing research investment on SSI) fits with other NIHR priorities.

Research about wound photographs taken by patients

We remain attracted by the idea of patients taking and submitting a photograph of their wound(s) at 4–8 weeks, primarily because such a photograph could be assessed without knowledge of the way the wound was initially dressed (or, indeed, treated in other ways). We have shown that this is possible and we suspect that, by working with patients of different ages and aptitudes for smartphones, both the response rate and the quality of photographs could be enhanced.

Research about facilitating the capture of wound photographs for research in theatre

We were frustrated by the data protection challenges that we experienced in trying to implement wound photography in theatre, despite having ethics approval and consent to do so. All trusts in which we tried to obtain permission for wound photography in theatre insisted on obtaining additional permissions and imposed restrictions on storing digital images; the specific permissions required varied by trust. Photography and video imaging in theatre may create many research opportunities that extend beyond wound-related research (e.g. defining interventions, monitoring intervention fidelity and assessing outcomes). We believe that it is important to facilitate the collection of images by establishing a policy that would apply to all hospitals.

Assess the quality of wound closure, including from photographs

We made considerable progress in characterising aspects of wound appearance that could constitute a metric of quality of wound closure. Such a metric is needed to answer the question of whether or not there is an interaction between dressing strategy and quality of wound closure. Quality of wound closure could also be useful as a future outcome in surgical research (e.g. in assessing cosmesis). A further aspect of this research is to compare measurement of the metric ‘in vivo’ with measurement from digital images.

Carry out further validation of the Wound Management Questionnaire and Wound Experience Questionnaire

Similarly, we made considerable progress in developing patient- and practitioner-reported outcomes of wound experience and wound management to fill the gap for such tools identified in the Cochrane review of wound dressings. 17 We showed that these instruments are acceptable to patients and practitioners and are well completed. However, we recommend that both measures are validated so that these tools can be deployed in future trials.

Research whether patients can recall wound problems in hospital at 4–8 weeks after discharge

One of the challenges in designing the WHQ was to develop a tool that was appropriate for procedures with varying lengths of stay after surgery or carried out as day-cases. This was important so that the WHQ could be used for patients having a wide range of operations and because hospital stays after surgery continue to shorten. For patients having day-case surgery, the concept of SSI at discharge was irrelevant and SSI was ascertained only from the WHQ at 4–8 weeks. This raises the question about whether it is necessary to collect data about SSI signs and symptoms at discharge in patients who do stay in hospital (i.e. could the timeframe for the WHQ be extended to include time in hospital?). We recommend that this question be researched because, if the answer is yes, then data collection for trials of interventions to prevent SSI could be reduced. Recall of signs and symptoms of SSI in hospital may vary according to the nature of the surgical procedure undertaken so it is likely to be important to include a wide range of operations with varying average hospital stay in such research.