Selective decontamination of the digestive tract in critically ill patients treated in intensive care units: a mixed-methods feasibility study (the SuDDICU study)

Sample

All UK ICUs delivering SDD, identified from a recent national SDD survey9 or known by the study investigators to deliver SDD, were deemed eligible for inclusion (15 ICUs). Two ICUs were purposively selected to represent different lengths of time since SDD adoption (one had adopted SDD < 5 years earlier and the other had adopted SDD > 5 years earlier) and different geographical locations (i.e. geographically dispersed ICUs to ensure different organisational profiles). Clinicians from different professions (i.e. intensive care consultants, clinical microbiologists, specialist clinical pharmacists and ICU nurses) have responsibility for the implementation and/or delivery of SDD. In each of the case study ICUs, all clinicians with potential involvement in the implementation and/or delivery of SDD were eligible for interview. From these, we recruited a purposive sample of clinicians from different professions. 33 Not all eligible clinicians were interviewed owing to lack of availability or time. Purposive sampling was appropriate for this small-scale exploratory qualitative study and the sample was not intended to be statistically representative. 33

Data collection

Direct observation offers the opportunity to record and analyse clinical behaviours and interactions as they occur in ‘real world’ contexts. 34 The use of direct observation as a data source allowed the process of SDD delivery to be ‘seen’ through the eyes of the researcher. Observations were conducted using an investigator-designed form to record all behaviours relating to ‘real time’ delivery of SDD. Additionally, the context (i.e. the physical environment where behaviours were performed), timing of procedures and physical presence of clinicians at time of delivery were recorded (see Appendix 1).

Semi-structured face-to-face clinician interviews were conducted in the study hospitals using a topic guide with prespecified prompts to ensure consistent coverage of key issues including behaviours relating to SDD implementation and SDD delivery as well as barriers and facilitators of described behaviours1 (see Appendix 1).

Finally, all written documentation relating to SDD implementation and delivery (e.g. SDD protocols, training documents) was provided by the participating ICUs for systematic analysis. This documentary information provided an unobtrusive and verifiable data source, which augmented and corroborated information from interviews and observations. The presence of relevant documents was established during the interviews. After the interviews, these documents were obtained.

Procedure

Data collection commenced with observation of SDD delivery performed by various ICU nurses to different patients at the bedside. A single researcher (SUD) visited the ICUs for 2 (case study 1) or 3 days (case study 2). Observation opportunities were identified by senior ICU nurses (i.e. they informed the researcher when and to whom SDD would be delivered). The number of patients eligible for, and receiving, SDD varied from day to day; therefore, it was not possible to prespecify the number of direct observations of SDD delivery that would occur during SUD’s visit. Semi-structured interviews were conducted in parallel with observations. Observed nurses were included in the interview sample to gain an in-depth understanding of observed behaviours. With participants’ permission, interviews were audio recorded, transcribed verbatim and anonymised. All observations and interviews were conducted by SUD. Written documentation from each ICU was examined (by SUD) following completion of all observations and interviews, in order to minimise researcher bias during these stages.

The written documentation was examined to identify clinical and behavioural components of SDD delivery. Clinical components were defined as the pharmaceutical regimens forming part of SDD including drug, dose, route, frequency and duration. Behavioural components were defined as any actions that were/would be directly observable. We recorded the behaviours involved in delivering clinical components and those not related specifically to drug administration.

Data management and analysis

Data from the three sources were analysed within case to describe the clinical and behavioural SDD components, and synthesised across case to identify emergent themes describing SDD implementation and delivery in context. The analytical process was guided by the study aims, which included identification of SDD clinical and behavioural components and exploration of SDD implementation and delivery.

The three data sources were analysed separately within ICUs and in reverse order to data collection. First, we systematically examined written documentation and extracted clinical and behavioural components of SDD delivery. Second, we performed content analysis35 of interview transcripts to identify additional behaviours involved in SDD delivery (i.e. those not specified in the documents). Third, direct observations provided contextual ‘real time’ data32 and identified new and corroborative evidence on SDD clinical and behavioural components (i.e. data triangulation from multiple sources). 32

To identify features of SDD implementation and delivery across ICUs, a thematic analysis of the interview data was conducted using a framework approach. 36 This approach involves five stages: (1) familiarisation with the raw data, (2) identification of emergent themes associated with SDD implementation and delivery (i.e. relating to the behaviours and clinician groups involved), (3) systematic coding/indexing of all data relevant to each theme within each transcript, (4) creating charts (Microsoft Word tables) that contain the coded data for each theme and distilled summaries of views and experiences, and (5) interpretation of the data (e.g. identifying associations between themes and providing explanations for the findings). 36 A single researcher (SUD) conducted the thematic analysis, a second researcher (EMD) independently coded randomly selected portions of the data set to identify clinical and behavioural components and three researchers (MEP, JJF, LR) provided critical comments on analyses drafts.

This study was classified as service evaluation by the Research Ethics Committee (10/MRE00/32) and, therefore, was deemed by them not to require ethical approval. All participants who were observed and interviewed were aware of the study purpose and provided verbal consent prior to data collection.

Selective decontamination of the digestive tract clinical and behavioural components

Protocols documenting the specific clinical behaviours required for drug preparation and administration in the two ICUs are detailed in Table 1, demonstrating the degree of clinical complexity and also the variation encountered in clinical aspects of SDD. The documents identified in the interviews and subsequently analysed listed nine different medications and a total of 13 different preparations as part of SDD in the two case studies (see Table 1). Several behaviours directly relevant for drug administration were identified in examined documentation.

Aside from clinical and behavioural components directly relevant to SDD delivery, documents from both cases revealed several additional delivery behaviours performed by multiple clinicians in various clinical and environmental contexts (Table 2). To complement understanding of behavioural components that are important in SDD, but not specifically mentioned in the examined documentation, Table 3 outlines additional delivery behaviours identified through interviews and observations. Behaviours outlined in Tables 2 and 3 were performed by various clinician groups (e.g. nurses, physicians, pharmacists) in a variety of clinical and environmental contexts (e.g. bedside, ICU nursing stations, pharmacy).

Participant interviews provided most data relating to behavioural components, 49 components were identified through interviews, 22 through documentation and 12 through observations. Each data source gave rise to unique behaviours not mentioned in other sources (28, seven and four unique behavioural components for interviews, documentation and observations, respectively), confirming the added value of analysing multiple information sources. The number of unique behavioural components was 29 (case study 1) and nine (case study 2). Twenty-six behavioural components were common across ICUs, being identified in at least one data source for each case.

Selective decontamination of the digestive tract implementation and delivery

Based on our analysis, SDD implementation and delivery was conceptualised as a complex procedure consisting of four overlapping processes, each involving specific behaviours: adoption, operationalisation, provision and surveillance. Adoption concerned the decision to introduce SDD; operationalisation referred to the processes required to introduce SDD into clinical practice; SDD provision included actions involved in delivery of the clinical components; and surveillance, mentioned in both case studies, provided the foundation for adoption, operationalisation and provision by checking that SDD was effective in preventing infection.

Adoption and operationalisation

For adoption, we identified that actions often occurred at the organisational and team level involving organisational and group processes as well as individual action. As the implementation process moved from adoption to operationalisation, more behaviours emerged that were performed by individual staff (see Tables 2 and 3). Although operationalisation was complete following SDD introduction, elements of operationalisation continued owing to clinician staff turnover (e.g. although SDD was a standard procedure within the ICUs, the low national baseline adoption meant that additional training for clinicians new to these ICUs and SDD delivery was required).

Provision of selective decontamination of the digestive tract

Three themes emerged from the interviews on SDD provision: complexity/difficulty, protocol adaptation in practice, and facilitators and barriers.

Complexity/difficulty

Reflecting the theme of complexity, one intensive care consultant and several nurses reported that SDD provision represented additional and time-consuming work that made it unpopular with staff. When examining the sequencing and flow of actions, we identified evidence of complexity – multiple clinicians were involved in managing various behaviours within multiple clinical and environmental contexts using a range of materials delivered in specific sequences in a continuing flow of action (see Box 1 for quotations: P, participant). However, most nurses and doctors refuted the idea that SDD was complex and time-consuming, stating that providing SDD was effortless (Box 1). Low complexity/difficulty of SDD for these staff was supported by observational data indicating that administration of clinical components took no longer than 5 minutes, often less, and was performed in a swift sequence of actions. However, it is important to note that these were highly practised actions and may require considerable skill development to achieve this high level of expertise.

Protocol adaptation in practice

Protocol adaptation in SDD delivery was noted in observational and interview data. Preparation of antibiotics/antifungals varied, suggesting some deviation from recommended practice. Further adaptation was evident in the provision of SDD oral components such as different ways of applying oral drug components and timing with other nursing interventions such as oral hygiene. Authorising SDD involved multiple staff and deviation from recommended practice was noted. Although documentation indicated that patients should be routinely commenced on SDD, this did not always occur, owing to more pressing clinical concerns. As a result, multiple layers of control to ensure protocol adherence were described (Box 2).

Facilitators and barriers

Facilitators and barriers to SDD delivery were evident across both cases (Box 3). One facilitating factor frequently reported was dovetailing of SDD with other established and routine procedures. Thus, intensive care consultants might include SDD delivery behaviours as part of the admission process, nurses might include SDD as part of oral hygiene or other activities and clinical microbiologists and pharmacists dovetailed SDD actions within ward rounds. Dovetailing was evident in multiple interviews and in documentary data on SDD provision for oral hygiene. Although barriers were commonly reported during interviews in response to specific prompts, these were often referred to as minor inconveniences, rather than significant obstacles to SDD delivery (see Box 3).

Infection surveillance

A fourth theme emerged from the documentary analysis. Surveillance was specified in the SDD protocol in one of the case study sites, but not in the other, in which it was part of the wider regimen to combat HAIs. Despite these differences, surveillance was integral to the provision of SDD and included the performance of multiple behaviours of various clinicians in several clinical and environmental contexts.

Answers to research questions

We found variation in SDD clinical components, in terms of the drug regimen, mode of drug delivery and specification of components (i.e. surveillance) between the two study sites. This may be appropriate to make the intervention simple and feasible to deliver within a local context. Various behaviours related directly to drug provision as well as other aspects of the SDD intervention (e.g. authorisation of SDD delivery) were performed by multiple clinicians in differing contexts. In terms of clinical components, topical antibiotics/antifungals and i.v. antibiotics were identified as SDD components in documents, but surveillance, general hygiene and general infection control regimen were not. Such inconsistency is also identified in the literature. 18 Both ICUs administered i.v. as well as topical/oral components. Overall, SDD implementation and delivery comprised the interrelated phases of SDD adoption, operationalisation, provision and surveillance.

Additional behaviours to those specified in documentation were identified and these behaviours are essential for SDD delivery. SDD involved a range of health-care professionals performing various behaviours in differing contexts. These findings emerged in the interview and observational evidence but were not always clearly specified in the documentation. Ensuring that these additional behaviours are specified in protocols, guidelines and the academic literature should lead to improvements in implementation, delivery and reproducibility of SDD. 24,25

Various behaviours were identified for SDD implementation, many at the organisational and team levels and others at the individual level. Several features of operationalisation involved an ongoing process (e.g. nurse training for SDD provision) as a result of staff turnover. SDD could thus be construed as a simple and easy intervention from the individual behavioural perspective that becomes increasingly complex when focusing on the flow of actions required at an organisational level for its delivery in practice. Consequently, some of the barriers and facilitators to SDD provision tended to centre on the environmental context and resource issues, rather than specific attitudinal (e.g. beliefs about SDD effectiveness) or skills barriers.

Strengths and limitations

The current study is the first to systematically identify and specify a full range of SDD components throughout the steps of SDD implementation and delivery. A limitation is the potential lack of generalisability owing to the use of two cases only. Additional clinical and behavioural components, as well as alternative methods of SDD implementation and delivery, may be evident if investigating SDD practice in a larger number of ICUs. However, the study was exploratory in nature with the goal of providing information-rich case studies that facilitate in-depth understanding of SDD in practice rather than a comprehensive picture of SDD across all UK ICUs. We recruited one microbiologist only, limiting the perspective from this profession. Finally, clinicians in ICUs that did not deliver SDD may have different views about barriers to SDD implementation. This was investigated in subsequent stages of the study.

Sample

A Delphi process gauges views from a panel of experts. 44 Ideally, potential Delphi participants would thus be experts in delivering SDD, or experts in terms of their knowledge of the SDD evidence base. Because of the low SDD uptake in UK ICUs at the time of this study, restricting the study sample to those with direct experience of SDD delivery or with a special interest in SDD research could systematically bias findings in favour of SDD adoption and delivery. Therefore, we decided to define ‘expertise’ more broadly to include the four stakeholder groups likely to exert decisional authority with regard to an ICU’s SDD policy or to how such a policy would be implemented in practice. Hence, the participants for the Delphi study were intensive care consultants, ICU pharmacists, clinical microbiologists with ICU responsibility and an ICU leaders group (including medical leads, nurse managers and educators working in NHS hospitals throughout the UK).

There is a broad range of estimates of suitable sizes for a Delphi panel, but smaller sizes (such as 10 for each stakeholder group) have been deemed appropriate where panel members have similar training. 45 Our minimum target sample size was thus set at 40 (10 in each stakeholder group). We also sought participant representation from across the four UK home nations.

Three clinical members of the research team (GB, APRW, RS) compiled lists of their clinician group and ranked them according to predetermined diversity factors (location, ICU size, current SDD practice and academic affiliation). A list of ICU nurse managers/educators was compiled by three members of the research team (GB, BHC, MEP) and ranked for the above diversity factors. Study invitations were issued to individuals according to rankings and in order of approvals made by the research and development (R&D) offices of each participating hospital trust. Sample diversity was tracked during recruitment. Additional participants from stakeholder lists (those working in NHS Trusts for which we had R&D approval) were specifically targeted to maximise variation. To preserve a minimum sample size of 10 in each stakeholder group by Delphi round 3, we oversampled by one to three participants in each of the four groups.

Data collection

At the start of each interview, to establish a shared understanding of SDD components, participants were first asked what they understood ‘selective decontamination of the digestive tract’ to mean. Irrespective of their initial definition of SDD, participants were, for the remainder of the interview, asked to consider SDD as application of antibiotics comprising all the following: (1) oral administration, to the mouth and throat, (2) gastric application via a nasogastric tube or similar and (3) a short course of i.v. antibiotics. We then asked whether or not SDD was delivered in the participant’s ICU. Responses to this second question determined which of two topic guides was used for the remainder of the interview (‘ICU currently delivering SDD’ or ‘ICU not currently delivering SDD’). Both topic guides are presented in Appendix 2 and included questions about:

• factors that might influence adoption of SDD, such as participants’ views about the likely positive and negative consequences of SDD and their knowledge of the evidence base

• participants’ views on the need for further research to settle questions around harms/benefits of SDD, what type of research [an effectiveness study or an implementation study (i.e. a study to evaluate strategies to increase uptake of SDD)] would be most informative and whether or not further effectiveness research on SDD was ethical

• whether participants would be willing to participate in an effectiveness study to evaluate SDD and/or an implementation study to assess strategies that aim to increase uptake.

Procedure

Potential participants from stakeholder groups were invited to take part by an investigator-signed e-mail invitation (GB, APRW or RS). Expressions of interest were followed up with a short telephone call or e-mail from the study co-ordinator to further describe the study, answer any questions and, if the participant agreed to take part, arrange a convenient time for a 30-minute telephone interview. After 1 week, non-responders were sent a reminder e-mail signed by the appropriate clinician researcher. No further contact with non-responders was attempted. Recruitment continued until target sample sizes were achieved for each stakeholder group.

At the start of each telephone interview, participants were reminded that the aim of the study was to identify their personal views and opinions on SDD (there were no right or wrong answers). Consent to audio record the interview was requested and received from all participants.

Data management and analysis

Recordings were transcribed verbatim, checked for accuracy and anonymised. The objective of the analysis was to develop questionnaire items for the second round of the Delphi study, the quantitative questionnaire round. Analysis of the transcripts proceeded through a number of stages. First, ‘specific beliefs’ were identified within the transcripts. A specific belief was defined as a statement for which the content may indicate a perceived influence on SDD adoption or delivery. Specific beliefs that expressed the same theme or were polar opposites of the same theme were grouped together and were considered as repeats of the same belief. Summary statements representing these beliefs were devised, and these became the basis of the round 2 questionnaire items. This analysis was performed using an iterative and parallel process with the SuDDICU Canada team (who had adopted an identical topic guide and sampling strategy). All summary statements identified in the analysis were discussed by an international working group of study investigators to identify appropriate wording for representing the beliefs in round 2 of the Delphi study. When it was justifiable from the interview data, identical wording of questionnaire items was agreed across nations. When specific beliefs emerged only in the UK study, these were included in the UK version of the round 2 materials.

The next stage in the analysis involved allocation of the specific beliefs to the prespecified TDF domains. This was carried out independently by two researchers (JJF and one research assistant) using the TDF as an analytic framework and content analysis methods previously employed by the research team in the context of intensive care. 46 When there was disagreement between coders, these were discussed with a third researcher (MEP). Agreement was achieved for 42 of the 46 UK items; the remaining four items, for which discrepancy in domain allocation persisted, were discussed with an expert group of 16 researchers who were familiar with the TDF (the Aberdeen Health Psychology Group) and the majority view was taken for domain allocation.

To address research question 4 we examined the Beliefs about Consequences domain in detail. A preliminary assessment was made of the perceived importance of each specific belief in this domain. In round 1, assessment of importance was based on evidence from cognitive psychology, which identifies that the ‘cognitive accessibility’ of a belief (i.e. the readiness with which it comes to mind) is an indicator of its importance to an individual. 47 In ‘social cognitive’ models of planned action,48,49 importance is assessed at the group level by identifying the ‘modal salience’ of the belief, i.e. how many times the same belief is elicited across the whole sample of study participants, relative to other beliefs. Hence, the frequency with which a belief was elicited in the interviews was taken as an indicator of its relative importance. It was recognised that this procedure is based on a range of assumptions and so this early indicator of importance was later compared with data from rounds 2 and 3 (reported in Chapter 4), in which participants were asked to report their ratings of the importance of each belief. Importance was also examined quantitatively in the national survey (see Chapter 5) so that these three methods for assessing the relative importance of consequences could be compared. This was done to ensure that robust evidence could be used to identify the most important outcome measures to include in a possible effectiveness trial of SDD.

Using similar assumptions, the importance of each theoretical domain was assessed, in round 1, by identifying the level of elaboration provided by participants in their responses to the interview questions. The relative importance of the domains would identify key barriers to implementation, which would provide an evidence base for the design of an intervention to increase uptake in a possible implementation trial of SDD. For each domain, we identified how many utterances were coded, how many specific beliefs were generated across the sample and the content of those beliefs. For example, if the belief was essentially that the issue was not important or not a problem, this content was taken into account.

The project was coordinated by the Health Services Research Unit, University of Aberdeen. Ethics approval for Delphi study was granted by NHS North of Scotland Ethics Service (10/S0801/69).

The project was subject to extreme delays and difficulties in obtaining research governance approvals for sites in England, Northern Ireland and Wales. These delays negatively impacted upon recruitment and hence may have compromised our attempt to sample for diversity. Despite repeated attempts to contact trusts, 15 NHS trusts had failed to issue a decision on research governance by the time data collection closed (6 months after submitting the documentation). Further details can be found in Appendix 3.

Participant characteristics

Ninety-four health professionals were invited to take part. Eight participants were excluded as not meeting eligibility criteria (had left the hospital or their profession) and three e-mails were returned undelivered. Forty-seven participants consented to participate and were interviewed (57% consent rate). Consent rates were highest among pharmacists (92%) and lowest among clinical microbiologists (39%), shown in Table 5.

The mean age of the 47 participants was 45.7 years and 31 (66%) were male. Participants were working in 25 different hospitals with ICU bed numbers ranging from 8 to 75. Five participants worked in ICUs that currently delivered SDD, another five had personal experience of SDD from previous positions and the remainder had no direct experience with SDD. Participants had a mean of 18.2 [standard deviation (SD) 7.2] years of ICU experience. Participants were recruited from all four home nations (30 participants from England, 12 from Scotland, two from Wales and one from Northern Ireland).

Importance analysis at the level of theoretical domains

Table 6 provides details of the number of times participants made comments that were coded into each theoretical domain and a brief description of the content of each domain, in so far as it indicates the perceived importance of the domain.

As shown in Table 6, most utterances were coded under the Beliefs about consequences domain and these utterances were coded into 18 specific beliefs. However, not all domains showed a direct relationship between the level of elaboration in interviews and the number of specific beliefs that were identified. For example, although Memory, attention and decision processes was the second most elaborated domain, the participants’ responses could be distilled into three specific beliefs relating to the decision-making processes required for SDD adoption (see Table 9). Furthermore, Motivation and goals produced the second highest number of specific beliefs but, in terms of the number of utterances, was elaborated less than the Memory, attention and decision processes or Knowledge domains.

The following sections provide further detail about the specific beliefs that emerged in each of the theoretical domains.

Beliefs about consequences

Participants discussed 18 Beliefs about the consequences of SDD, both positive and negative, and noted that SDD could produce positive benefits such as reduced morbidity, rates of HAIs including VAP and ICU length of stay. The most frequently mentioned concern about the negative consequences of SDD adoption was the potential for increased antibiotic resistance. Participants also discussed their concerns about the potential for increased rates of C. difficile infections.

Non-nurse participants expressed concern that SDD delivery would increase nursing workload, specifically relating to administration of SDD components, taking cultures for microbiology and dealing with patient side-effects related to SDD such as diarrhoea. However, nurse participants raised none of these concerns. The only nurse to mention the impact of SDD delivery on nursing workload reported that it was not a barrier to implementation.

People who are wittering on about increased nursing time . . ., you cannot really use that as a great reason not to do something.

P41

Participants also stressed the importance of weighing risks and benefits associated with SDD. Specifically, they questioned whether the potential reductions in mortality and VAP were enough to balance the potential risk of increased antibiotic resistance and associated consequences. We noted considerable variation in viewpoints in this domain.

Financial consequences also were discussed both as potential cost-savings (due to reducing VAP and length of stay) but also the potential for increased costs in terms of SDD drugs and additional human resources required to deliver them.

Memory, attention and decision processes

Three beliefs relating to decision-making processes were identified that would be required prior to adoption of SDD: the need to review and appraise current evidence (n = 35), the need for consensus among colleagues (n = 32) and the role of senior ICU doctors (consultant level) as key decision-makers (n = 13).

Knowledge

Participants varied in their self-assessed levels of knowledge about SDD components and in their views of their own and others’ knowledge of the evidence base. Importantly (but not surprisingly), knowledge of the evidence base was linked by some to their rating of the importance of SDD:

I have never quite got round to looking at it and I have to be honest that . . . when I got the original email to say . . . there’s going to be a questionnaire on this, I thought . . . this is something I have been meaning to look into for a while so . . . having looked at the evidence now, it is very important.

P05

Some participants were very knowledgeable:

There have been some very, very good trials recently . . . I think there is certainly plenty of evidence there that some of us should be looking at and I think the big problem is . . . not everybody has fully appraised the papers.

P05

Other participants identified a limited knowledge of existing evidence, which was potentially an important factor in interpreting the study findings. Knowledge data are reported in Chapters 4 and 5.

Motivation and goals

Six beliefs were coded into this domain. These beliefs included the perception that SDD was not currently a ‘front-and-centre issue’ and, therefore, not a topic of discussion in their units or among colleagues. Two intensive care consultants and one microbiologist reported that VAP was already adequately addressed by other interventions and, therefore, there was no motivation to pursue other options to reduce HAIs such as SDD. Other clinical priorities were reported to be more important, such as the adequate implementation of existing VAP bundle procedures.

The main reason is that we are on a very steep improvement curve for our intensive care unit, in terms of trying to improve the outcome of our patients and just haven’t got to SDD yet. We haven’t reached the level of sophistication by which we can look at interventions like SDD. We are still working on simpler things like sedation holding and breathing trials, you know, much simpler things and I think SDD will be on our list, I think we just haven’t got to it yet.

P19

Finally, four participants (two with previous experience of delivering SDD and two with no experience) reported that SDD was considered ‘old news’ and no longer a relevant clinical topic.

Environmental context and resources

Potential barriers to SDD were discussed by participants with respect to the domain Environmental context and resources. Previous trials, conducted outside the UK, were perceived to have limited generalisability to the UK ICU context owing to different patient characteristics, ICU ecology or microbial flora and standards of care.

It would be very helpful . . . if there is good research from the UK, because we use antibiotics differently, our ecology is different, our patients tend to be slightly different to other European ITUs [intensive therapy units].

P25

Professional role and identity

Three beliefs were identified within the Professional role and identity domain. Specifically, three participants reported that SDD conflicted with their professional obligations, most notably in reducing the administration of prophylactic antibiotics.

It is kind of bred into us, hammered into us publicly that we should limit and target the use of antibiotics as much as is possible.

P44

Participants reported that there were conflicting opinions among clinical microbiologists and intensive care consultants (on antibiotic delivery used within SDD) that could influence SDD adoption.

Behavioural regulation

Three beliefs coded into the Behavioural regulation domain were mentioned by five participants. They discussed the influence of national guidelines and regulatory requirements on local policy with respect to SDD adoption. There was a perception that endorsement of SDD from an authoritative body (e.g. mandate by NICE) would lead to the adoption of SDD into a unit’s routine clinical practice.

. . . if, certain NICE or a group like that . . . came forward and said this is an absolute . . . necessity for treatment in the ICU then I think it would happen in our unit.

P15

Social influences

Two beliefs were coded within the domain Social influence. Participants suggested that adoption would require a clinical champion or SDD expert to put SDD on the agenda, educate others and drive SDD forward. Two participants (a clinical lead and a microbiologist) also reported that their practice was influenced by the practice of other ICUs; more specifically, they felt reassured that their position on not delivering SDD was in line with the position of other ICUs.

Well I guess as the years go by and the lack of pressure to institute it, in other words, you know, lack of a persuasive argument that what we are doing leaves us as an outlier, then it becomes less important.

P31

Emotion

Finally, the domain Emotion was discussed by one participant, who described the emotion associated with colleagues’ certainty about the SDD issue:

. . . it is very difficult to have a dispassionate discussion with people about it [SDD]. They have already made up their minds.

P45

In the context of intensive care, three domains were similar conceptually, and all were viewed as not being barriers to adoption. These domains were Skills, Nature of the behaviour and Beliefs about capabilities. Skills was discussed by 34 participants, and the vast majority (n = 31) felt that delivering SDD was within the existing competencies of ICU nurses and, therefore, skills were not a barrier to SDD adoption.

I don’t think in the ICU there would be any reason to say that the nursing staff shouldn’t be able to give the drugs because they are doing that already.

P39

All participants (n = 47) made comments coded under the domain Nature of the behaviour, with most participants reporting that delivering SDD would not be a dramatic shift from current practice. This domain was not considered to be a barrier to SDD adoption and included in round 2.

I don’t think it would be that difficult [in comparison to what already doing].

P14

Participants’ Beliefs about capabilities (to influence adoption) were not considered to be a barrier to SDD adoption. Those key professionals who were responsible for instigating changes in policy (medical leads, intensivists and clinical microbiologists) reported that they felt they could influence the decision on whether or not to adopt SDD.

I would say that if I personally made it my crusade to do it [adopt SDD], we would do it.

P19

Nurses and pharmacists were less certain of their ability to influence SDD adoption, but most noted that they could suggest changes in policy to colleagues.

I can definitely suggest changes and I would have to get agreement with the consultants before we could institute a major change, but I could certainly discuss instituting major changes.

P24

In summary, when the theoretical domains were used as a framework for analysis, it was possible to identify beliefs about the likely positive and negatives consequences on SDD and other factors that are likely to be barriers to SDD adoption. Nine of the 12 domains were potentially important in this context; Beliefs about consequences, Memory, attention and decision processes, Knowledge, Motivation and goals, Environmental context and resources, Professional role and identity, Behavioural regulation, Social influence and Emotion.

Participants were also asked their views on the existing SDD evidence base and about potential future research directions. For discussion of data relating to these topics, see Views about the existing evidence base.

Views about the existing evidence base

There was considerable variation in participants’ reported views of the existing evidence base for SDD. Six participants reported that the evidence base for SDD was sufficient and three specifically stated that further effectiveness research would not be helpful in determining the next steps for SDD.

I think they have done enough research on it to show it is a good thing . . . the research has been done and it’s time to implement it.

P05

In contrast, 26 participants were not convinced by the existing evidence.

[Would further research settle some of the issues surrounding SDD?] I think so. I mean the problem with ITU is that we know that whatever is published one day, in about 6 months time or a year, there could be something different that will contradict it. But it [the research] has to be pretty definitive before, you know, we rush into anything.

P27

Most felt that further research was needed to address clinical uncertainties (see Table 9), in particular beliefs about the consequences of routine delivery of SDD.

The 26 participants who discussed further research needs mentioned the limitations of SDD studies reported to date, including the length of follow-up (as this was not sufficient to assess antibiotic resistance in the longer term) and the need for studies conducted in the context of patient populations with a resistance profile similar to that of the UK. Given the variation in self-assessed knowledge of the evidence, we investigated associations between knowledge and views about further research. Views about further research that are based on poor knowledge of the current evidence would clearly have a different status from views based on a sophisticated level of knowledge. Such associations were more effectively explored in the context of the quantitative Delphi rounds and are reported in Chapter 4.

Views about further research

Despite variation in participants’ views of the existing evidence base, most participants (n = 32) reported that they would be willing to facilitate recruitment of patients to a randomised controlled trial (RCT) to evaluate the effectiveness of SDD. The importance of obtaining agreement from colleagues to participate in any further research was raised by nine of these participants. Reasons for willingness to participate in effectiveness research included:

• such a study would be important as it would help to resolve an unanswered question in critical care research

• a general interest in research

• a desire to improve patient care.

The case in favour of an effectiveness RCT is strengthened by the extent of elaboration of the Beliefs about consequences domain. Table 7 provides illustrative quotes that suggest an effectiveness RCT may be appropriate.

In contrast, five participants reported being unwilling to recruit patients to further effectiveness research as they perceived that enough research had been conducted (n = 3), they already delivered SDD and lacked equipoise (n = 3) and they wanted to implement SDD without further delay (n = 1). Contrasting views about the ethics of conducting further research were expressed: 27 participants reported that further research was ethical, 10 participants qualified this answer by stating that questions remained unanswered and five explicitly mentioned that equipoise existed. In contrast to this view, three participants reported that further SDD research would not necessarily be ethically sound because the question of effectiveness had already been answered in favour of delivering SDD.

When asked whether or not they would support an implementation study (i.e. a study to evaluate strategies to increase uptake of SDD), participants were hesitant. Seven participants felt that they would need more information about an implementation study before they could comment on participating in such a study and nine participants reported that there was insufficient evidence of effectiveness to proceed with such a study at this time. However, there was substantial evidence to suggest that the reasons for holding back on adoption of SDD were not strong. Table 8 presents quotes that suggest an implementation study may be appropriate or that propose specific implementation strategies.

In addition, there was considerable evidence that some participants were not aware of what is meant by ‘implementation study’. For example, when asked if they would be willing to participate in a study whose aim was to increase adoption of SDD based on current evidence, responses included:

It is not really a study is it? It is just a NICE guideline?

P19

I am not sure how that could be a study, it would be educational programme to raise awareness but I am not sure how in practice this would be a relevant study.

P44

In summary, participants reported a range of views on the existing evidence base and directions for future research. In general, they were willing to participate in effectiveness studies (although some voiced uncertainty about the ethics of conducting further effectiveness research in SDD). There was less support for participating in implementation studies, but also less clarity in understanding what an implementation study involves.

Findings from round 1 used to develop round 2 materials

In collaboration with the SuDDICU research teams in Canada and Australia/New Zealand, the results from the round 1 interviews were used to develop a set of 47 domains-based statements for inclusion in the instrument used for the quantitative rounds of the Delphi study (round 2 and round 3). The finalised items are displayed in Table 9. There were eight positively worded items (i.e. in favour of SDD), 31 negatively worded items (i.e. opposed to SDD) and eight neutrally worded items. Further detail on the methods used to decide item wording is provided in Chapter 4.

Answers to research questions

Round 1 of the Delphi study provided preliminary answers to research questions 3, 4 and 5:

1. 3. Stakeholders held mixed views of the internal/external validity and adequacy of the existing evidence base for SDD. They were willing to participate in further research but the most appropriate kind of research was not clearly informed by these interviews.

2. 4. Stakeholders also held mixed views about the likely positive and negative consequences of implementing SDD in ICUs. The relative importance of these beliefs in influencing overall views about SDD was assessed using assumptions from cognitive psychology (but a range of further importance indices was examined in later stages of the study).

3. 5. Stakeholders had a lot to say about the likely barriers to implementing SDD in ICUs. These tended to focus on Motivation and goals (simply not high on the priority list) or on Behavioural regulation (a national guideline would change practice) but there was also considerable doubt about the evidence base.

These preliminary results were used to generate questionnaire items for further consideration in later rounds of the Delphi study (see Chapter 4). All the beliefs generated in round 1 of the Delphi study (including minority beliefs) were to be taken forward for consideration by the panel in rounds 2 and 3. 1 In view of the importance of considering the full range of views about SDD, we used two approaches. First, we based the round 1 interviews on the TDF of clinical behaviour, as this framework has been shown to generate more beliefs (including lesser discussed issues such as emotion). 52 Second, we conducted a data saturation analysis, using previously published methods, to provide evidence that saturation had been reached.

Evidence of data saturation

Data saturation principles did not drive our sampling strategy as sample size was based on recommendations for Delphi studies. 53 Therefore, data saturation analysis was conducted after completion of data collection and content analysis. Published guidance for determining data saturation54 recommends a two-step process involving analysis of consecutive interviews to (1) plot all ‘new’ beliefs (i.e. not counting repeats of beliefs previously mentioned) in an ‘initial analysis sample’54 of 10 cases with adequate diversity and (2) declare saturation when there are three consecutive interviews with no new beliefs emerging. This is referred to as the ’10 + 3 rule’,54 according to which the smallest sample size for data saturation would be 13. Saturation analysis of our data illustrating these two steps is displayed in Figure 2. After the 13th interview, the 10 + 3 rule had not been met. After interview 22 (represented by the dotted vertical line), there were three consecutive interviews without new beliefs emerging (at interview 25, represented by the solid vertical line), whereas recruitment continued to interview 47. This indicates that the sampling process was more than adequate to achieve data saturation in terms of beliefs about SDD. However, Figure 2 also shows that new beliefs were elicited in interviews 40 and 45. These beliefs were included with all other beliefs in the materials for round 2 (see Chapter 4).

FIGURE 2.

Evidence of data saturation and cumulative frequency of new beliefs about SDD for Delphi round 1 analysis.

Details of findings

The aim of the round 1 interviews, described in this chapter, was to identify – in four groups of key clinical stakeholders – the perceived positive and negative consequences of SDD implementation in UK ICUs, likely barriers to adoption, stakeholders’ views on the existing evidence base related to SDD, and potential willingness to facilitate recruitment into future studies of either SDD effectiveness or SDD implementation. Nine domains of the TDF55 were potentially important in characterising perceptions regarding SDD adoption and delivery. These included Beliefs about consequences (which was the focus of research question 4) and eight possible types of barriers to SDD adoption and delivery: Memory, Attention and decision processes, Knowledge, Motivation and goals, Environmental context and resources, Professional role and identity, Behavioural regulation, Social influence and Emotion. This is a coherent finding in the context of health care delivered in ICUs. First, knowledge of the research evidence is fundamental to the provision of evidence-based care. Second, such evidence focuses on the consequences (benefits and risks) of providing, or not providing, certain interventions. Third, clinical decision making is a core aspect of practice and covers both decisions about individual patients but also (importantly for SDD) decisions about local policy. Hence, these three domains encapsulate the kind of thinking that is required for delivering high-quality care. Fourth, motivation is important in the context of intensive care as there are many potentially important interventions to consider and clinicians need to consider the relative importance of SDD in comparison with other urgent priorities. Environmental context and resources are potentially important for two reasons: if the settings in which evidence has been developed have poor environmental fit with the UK, in terms of background infection rates, the evidence may not be seen as applicable. In addition, the financial costs in terms of drug supplies and staff time should be weighed against the benefits and other priorities in the intensive care setting. Some interesting findings relating to Professional role and identity highlighted that the training and responsibilities of the different clinical professions may have a bearing on SDD policy. Beliefs relating to Behavioural regulation drew on participants’ hypotheses about what might work to change their own behaviour. Such hypotheses are opinion based and do not represent a high level of evidence; indeed, there is evidence to suggest systematic bias in people’s views about what will change their behaviour. 56 However, such hypotheses are, in principle, testable and could provide a list of strategies to consider for implementation. There was a clear view that Social influence could be important, as intensive care involves a collaborative approach between disciplines and policies are influenced by ICU practice more broadly. Emotion was potentially a factor determining whether or not people were prepared to engage in dispassionate discussion about SDD.

Domains that appeared to be unimportant were Skills, Nature of the behaviours and Beliefs about capabilities. In other words, the interviews identified beliefs that lack of appropriate skills would not be a barrier to SDD delivery; the demands of the behaviour itself, while complex, would again be unproblematic in so far as they reflect the kinds of demands that intensive care staff routinely manage. Consistent with this, participants were confident that staff would be capable to manage a SDD regimen if it were adopted into the ICU. Notwithstanding their apparent unimportance, beliefs within these domains were taken forward into round 2 for further consideration, as specified in the study protocol.

Even when there was clear evidence of the importance of certain issues, participants placed themselves differently along the continuum from negative to positive. In terms of Beliefs about consequences, divergent opinions were expressed about the effect of SDD on patient outcomes and impact on clinician workload. The need for informed decisions, as well as collegial and leadership support to implement SDD, was acknowledged. However, lack of relevance to current perceived priorities in delivery of ICU care was also noted, as well as potential differences in the UK ICU ecological environment that may influence SDD effectiveness and potential adverse consequences, such as increased antimicrobial resistance. In general, participants perceived that acquisition of new skills and behaviours was not a major barrier to SDD adoption. This is in contrast to a previous survey that identified perceived difficulty associated with the behaviours needed to deliver SDD. 31

Divergent opinions also were noted in terms of the adequacy of the existing evidence about SDD effectiveness. In particular, respondents reported that existing evidence conducted in countries such as the Netherlands may not be applicable to the UK context, owing to differences in the microbial ecological profile. Another key area in which there was perceived to be insufficient evidence was in the long-term effect of SDD.

Most respondents indicated they would support further research on the effectiveness of SDD. Fewer positive views were expressed for a trial of SDD implementation but what was also noted was a lack of awareness among some participants about what implementation research involves and what it might achieve. There were many instances in which participants’ views would be consistent with criteria to support an implementation study (see Table 8). The findings of this first Delphi round were then used to inform the content of the next two Delphi rounds designed to refine and confirm participant views on barriers to SDD implementation and the need for further research.

Strengths and limitations

It should be noted that the sample was mixed in terms of direct experience of delivering SDD in practice. Hence, some of the opinion-based data presented in this chapter may reflect erroneous beliefs. For example, the concerns of non-nurse participants about nursing workload were not shared by the nurse participants and the belief that people have already made up their minds about SDD was not consistent with other participants’ reports about their own change of mind (reported above under Knowledge and Behavioural regulation). Hence, it would not be appropriate to make recommendations for future research based solely on these opinion-based data. Therefore, later stages of this study tested the robustness of these views but could not fully offset the problem of using data involving opinions.

Box 5 presents the key findings that inform the need for, and acceptability of, a clinical effectiveness RCT or an implementation RCT. There were some initial indications that these participants consider the evidence base sufficiently uncertain to warrant further effectiveness research. However, there was also evidence that implementation would be appropriate. Four ideas to support implementation, based on four of the theoretical domains, emerged in the analysis. First, in the domain Beliefs about consequences, some participants felt that there was sufficient evidence in favour of SDD to warrant adoption now. Second, in the Motivation and goals domain, there was a view that SDD was simply not high enough on the unit’s priority list to be considered at this time. Third, in the Social influence domain, it was clear that the personal influence of key individuals (rather than problems with the evidence) could be sufficient to trigger adoption. Fourth, in the part of the interview that probed potential Behavioural regulation strategies, a recurring response was that SDD would be adopted (apparently quite straightforwardly) if adoption was mandated by regulatory bodies.

Sample

All participants interviewed in round 1 were invited to participate in round 2. Only those who responded at round 2 were invited to participate in round 3.

Materials

An instrument was developed in questionnaire format. Item content was based on the findings from the first round of the Delphi study (see Table 9). The number of items included in round 2 relating to the Behavioural regulation domain was limited to ensure the resulting questionnaire did not suggest a bias towards adopting SDD. For example, beliefs reported in the Delphi interviews included hypothetical views about what would change practice and made the hypothetical assumption that SDD adoption is desirable.

No specific item relating to the emotion domain was taken into round 2 because participants were able to represent the extent of their emotional reaction in the questionnaire response scale (i.e. strongly against, neutral or strongly in favour). All other beliefs identified in the analysis were converted into questionnaire items. The content of these statements was decided using two principles:

1. The wording was chosen to reflect the language used in the interview data when possible.

2. The principles of questionnaire design and correct grammar were also observed (i.e. to assure brevity and maximise clarity and to avoid double negatives in the context of a ‘disagree to agree’ response format).

In round 2, participants were asked to consider 47 items based on the TDF20 and 10 questions relating to their views about further SDD research. For 46 of the 47 domains-based items, there were two questions: (1) ‘to what extent do you agree or disagree?’ (on a nine-point Likert scale, with 1 indicating strongly disagree and 9 indicating strongly agree) and (2) ‘how important is this issue in your overall opinion about the delivery of SDD to critically ill patients?’ (on a nine-point Likert scale, with 1 indicating not at all important and 9 indicating very important). 57 Importance ratings were not requested for the item relating to overall opinion on SDD (‘I am opposed to SDD’) because the importance of this item is reflected in the response format that assessed strength of agreement/disagreement.

The questionnaire was piloted with five clinicians who were not part of the Delphi study sample (one from the UK, one from Canada and three from Australia). The pilot process was also used to inform the wording of the participant information sheet. In round 3, participants were provided with feedback about the overall group responses to each question (in the form of a frequency histogram, Figure 3), they were given a reminder of their own previous response and were then asked to rate to the item again. Both rounds 2 and 3 were delivered online.

FIGURE 3.

Screenshot of a round 3 item.

Procedure

E-mails were sent to all 47 participants from round 1 with individual links to the electronic questionnaire. Responses were monitored and reminders sent to non-responders after a 2-week interval. After a further week, remaining non-responders were sent a second reminder (by telephone when possible, or by e-mail if unreachable by telephone).

Data management and analysis

Two participants responded late to round 2 (after feedback material had been produced). They were nevertheless included in the analysis reported in this chapter.

Ethics approval for the Delphi study was granted by NHS North of Scotland Ethics Service (10/S0801/69).

Participant characteristics

Forty-four participants completed round 2 (94%) and 42 completed round 3 (95% of round 2 participants, 89% of total sample). The breakdown of participation by stakeholder group is given in Table 10. The table shows that the recruitment target of 10 participants (see Chapter 3, Methods) per professional group was met at the end of round 3. The participants who had experience of using SDD were spread across all professional groups (one, one, five and two of the intensivists, pharmacists, microbiologists and ICU leads, respectively).

Participants’ perception of their own knowledge was measured with the item ‘I know the SDD evidence base well enough to have an informed opinion of its use’. As shown in Figure 4, the participants generally tended to rate themselves as knowledgeable about SDD (modal score of 7 on a nine-point scale). However, 12 participants rated themselves below the mid-point of the scale. It was expected that these participants would be more likely to alter their responses in round 3 after viewing the round 2 feedback. This prediction was tested on the round 3 data, reported below (see Figure 7). Further participant characteristics are reported in Chapter 3.

FIGURE 4.

Round 3 responses to knowledge item ‘I know the SDD evidence base well enough to have an informed opinion of its use’.

The frequency distributions of all items at round 3 are presented in Appendix 4. The following sections report the remaining results of the quantitative Delphi rounds in four parts. First, we report the stability of responses between round 2 and round 3, addressing both individual-level and group-level stability. Second, we provide summary data relating to beliefs about the consequences of delivering SDD, addressing the question ’what is the perceived importance of these beliefs in influencing overall views about SDD?’ Third, we describe the range and level of participants’ agreement with each questionnaire item in terms of levels of consensus among the Delphi panel. Finally, we describe data relating to participants’ views about further SDD research.

Stability of ratings from round 2 to round 3

The Delphi literature provides a range of guidance about assessing stability. 58,59 Our innovative approach was to assess stability at the within-individual level (individual change scores) and at the within-group level (changes in group means) as follows. Individual-level change scores were calculated, such that a score of zero, signifying identical responses in round 2 and round 3, demonstrated high stability. As shown in Table 11, the percentage of individuals whose round 3 response differed by ≤ 1 scale point ranged from 100% (for the item ‘SDD reduces VAP’) to 76% (for the item ‘There are conflicting opinions on antibiotic use among clinical microbiologists/ID physicians and ICU physicians’). Histograms of these two items are given in Figure 5. There was thus high individual-level stability of responses from round 2 to round 3, with > 75% of change scores in the range −1 to 1 for all items.

FIGURE 5.

Histograms of change scores (to indicate individual-level stability) from round 2 to round 3 for the items with (a) the most and (b) the least stability.

Group-level stability for each item was assessed by computing the change in arithmetic mean from round 2 to round 3. On the nine-point scale, a change of one scale point in the mean agreement level for any item was considered a potentially important change in views. Mean change scores are displayed in the last column of Table 11 (items presented in order of stability). These showed a high level of stability of views about SDD, with mean differences ranging from 0 to 0.52. In summary, at both the individual and group levels, the data appeared to be highly stable; therefore, the Delphi study did not proceed to a fourth round.

Group-level stability was also compared across stakeholder groups. Analysis of variance on change scores by group showed that there was a greater level of change from round 2 to round 3 in the clinical lead/nurse managers group than in the other stakeholder groups (Figure 6).

FIGURE 6.

Stability of responses from round 2 to round 3 for each stakeholder group. Note: the clinical lead group includes ICU clinical leads and ICU nurse managers or educators.

Participants who reported a low level of perceived knowledge (i.e. scoring below the mid-point for the item ‘I know the SDD evidence base well enough to have an informed opinion of its use’ at round 2) showed greater change in their responses from round 2 to round 3 (mean change 0.23, SD 0.35) than participants with a higher level of perceived knowledge (mean change 0.08, SD 0.18), as displayed in Figure 7.

FIGURE 7.

Self-reported knowledge of SDD evidence base and stability of responses from round 2 to round 3.

Importance ratings at round 3: beliefs about the consequences of selective decontamination of the digestive tract

The importance of each questionnaire item was assessed by calculating the mean rating of importance given by participants to the question ‘How important is this issue in your overall opinion about the delivery of SDD to critically ill patients?’. Table 12 presents the mean ratings of importance for each item assessing beliefs about consequences, presented in order of importance.

Importance ratings at round 3: domains level

Table 13 presents a range of indices that may indicate the relative importance of barriers to implementation for the other 11 theoretical domains on which this study was based.

Table 13 shows that across different measures of importance, a number of other theoretical domains (in addition to Beliefs about consequences) included potentially important barriers to the implementation of SDD. The most elaborated domain in round 1, and the domain rated as most important in round 3, was Memory, attention and decision processes. However, as shown above, some less well-elaborated domains at round 1, for example, Behavioural regulation, were rated within the most important at round 3.

Agreement ratings at round 3: consensus of opinions about selective decontamination of the digestive tract

The Delphi literature operationalises consensus in various ways. In this study, we were interested not only in the proportion of participants who agreed with each item, but also in the proportion of participants who were uncertain about their agreement with the items. Hence, levels of consensus for the questions ‘To what extent do you agree or disagree …?, were assessed by noting the highest percentage of participants whose scores fell within any three-point band on the nine-point scale. 53 Table 14 displays a summary of the item content, grouped by the mid-point of the most populous three-point band and by consensus level. As shown in the table, there was high consensus (> 90% of sample within a three-point band) around strong agreement (mid-point of 8) for a subset of items. There was also consensus around uncertainty (mid-points around 4, 5 and 6) for different kinds of items.

Beliefs about consequences items (n = 18) made up 38% of the domains-based items, but a striking feature of Table 14 is that all but one of these items had either a low level of consensus (e.g. 50–75% consensus for SDD increases C. difficile) and/or consensus around the middle of the scale (e.g. SDD reduces VAP, concerns about specific antimicrobials, cost-effectiveness, length of stay, mortality benefit). In general, the Beliefs about consequences items that were rated as more important (see Table 12) had lower agreement scores (e.g. mortality benefit was important, nursing workload was less important).

Another interesting finding is that no-one reported strong opposition to SDD although scores ranged from 1 to 7 (where 1 means not opposed to SDD and 9 means opposed to SDD).

To examine potential differences between professional groups, we closely inspected four key beliefs about the consequences of SDD (‘There is no mortality benefit’, ‘The risks outweigh the benefits’, ‘SDD reduces VAP’ and ‘SDD increases antibiotic resistance’). We selected these beliefs to illustrate patterns of group responses because they were rated as important and had a range of levels of consensus, but the modal values were towards the centre of the response scale (reflecting uncertainty). Responses of the mean scores for the four professional groups showed that the most negative views were reported by pharmacists and clinical leads for ‘There is no mortality benefit’, by microbiologists for ‘The risks outweigh the benefits’, by pharmacists for ‘SDD reduces VAP’ and by intensivists for ‘SDD increases antibiotic resistance’. In other words, there was no consistent pattern of unfavourable views based on professional group. Furthermore, based on these items, the responses of participants who were current SDD users were only slightly more positive (in the order of 0.5 of a scale point) than the responses of participants who were not current SDD users. Owing to small group sizes, it was not appropriate to test the significance of these differences.

Non-normal distributions

Where consensus was ≤ 60% of responses occurring within a three-point band, the statistical description no longer portrayed the nature of the distribution of responses. For over half of these items (n = 9), there was evidence suggesting bimodality as shown in Figure 8.

FIGURE 8.

Non-normal distributions for items with consensus at ≤ 60% of responses occurring within a three-point band. A notable pattern was that in seven of these nine items, at least 20% of the participants registered a score of 5, reflecting uncertainty. (a) ‘SDD is straightforward to deliver’; (b) ‘I am reassured that our position on SDD adoption is in line with other hospitals’; (c) ‘I would be more likely to participate in a RCT if mortality is the end point’; (d) ‘I am opposed to the i.v. component of SDD’; (e) ‘SDD is outdated  ’; (f) ‘The SDD evidence base is not generalisable to my patient population’; (g) ‘I am opposed to SDD’; (h) ‘Prophylactic antibiotic use in SDD is at odds with my professional training’; and (i) ‘My concerns about antibiotic resistance limit my willingness to participate in future RCTs of SDD’.

Answers to research questions

1. The validity and adequacy of the existing SDD evidence base was perceived to be limited. There was consensus around moderate agreement (7 on the nine-point scale) with the item ‘Research to date has not addressed concerns of antibiotic resistance and SDD’ and consensus around slight agreement (6 on a nine-point scale) with the items ‘The SDD evidence base is not generalisable to my patient population’ and ‘The SDD evidence base is not generalisable to my country’. There was consensus around the mid-point of the scale (reflecting uncertainty or equipoise) for the item ‘SDD evidence from countries with different resistance profiles’. Taken together, these findings suggest that the Delphi panel was not persuaded of either the adequacy or the external validity of the evidence base.

2. A striking aspect of the results was the high level of consensus around uncertainty (or equipoise) for many of the items from the domain Beliefs about consequences. For the nine items rated as most important in driving participants’ overall opinion about SDD (i.e. the top 50% of items from Table 12), the majority of agreement responses occurred within a three-point range with a mid-point of 4, 5 or 6. The only exception to this was the item ‘SDD would increase ICU C. difficile infections’, for which the majority of responses occurred within a three-point range with a mid-point of 7. This pattern of findings was similar across the four professional groups and across users and non-users of SDD, and thus suggests a need for further evidence about the consequences of SDD.

3. In contrast to the Beliefs about the consequences of SDD, the theoretical domains relating to the likely barriers to implementing SDD in ICUs showed a different pattern of responses. Greater than 90% consensus around the three-point range 7 to 9 (reflecting strong agreement) was evident for items from the two domains that had the highest importance ratings (> 7). These items focused on perceived barriers to the decision processes that would be required in order to adopt SDD (‘Requires agreement among colleagues and agreement about which patients would receive SDD’) and aspects of motivation (‘VAP and HAI are already being addressed using other strategies’). The other important barrier (i.e. other domain with a mean importance rating > 7) was Knowledge. This is discussed in Strengths and limitations. For items with levels of consensus below 60% (within a three-point range), true consensus was not met as there was evidence of bimodality.

4. Specific trial design and feasibility issues were also explored. The majority of participants would support their ICU’s involvement in a nationwide RCT to evaluate the effectiveness of SDD and this finding was assessed further in the national survey (see Chapter 5). There was a high level of consensus (> 90%) in favour of participating in such a trial if the control arm received VAP bundles and if a cost–benefit analysis was included. There was also consensus (> 85%) that such a trial should include pretrial, during-trial and post-trial monitoring of antibiotic resistance. Trial design features and their associated challenges were also discussed in the interviews with international clinical triallists and are reported in Chapter 6.

Strengths and limitations

This study benefited from the inclusion of four key clinical groups that could have an influence on SDD policy in ICUs or on the way in which SDD is delivered in practice. There was a high retention rate (85%) of participants across the three Delphi rounds. There was evidence of high stability of responses between rounds 2 and 3, although less knowledgeable participants shifted significantly more than the rest of the sample after viewing the group-level feedback from round 2. This stability of opinion suggested that it was appropriate to interpret the round 3 results and that proceeding to a fourth round was unlikely to produce new findings. As for round 1 of the Delphi study, rounds 2 and 3 gathered opinion-based data that have clear limitations. One belief that emerged in the Delphi study, and for which there was low consensus, reflected the theoretical domain Professional role and identity and was worded in rounds 2 and 3 as ‘There are conflicting opinions on antibiotic use among Medical Microbiologists/ID Physicians and ICU physicians’. The national survey provided the opportunity to test the accuracy of this belief systematically as we sampled these two professions (intensive care consultants and clinical microbiologists) for the survey stage of the study (reported in Chapter 5).

Self-rated knowledge of the field was generally high, although a third of the participants rated their knowledge of the evidence base as uncertain or low. This variation in knowledge of the evidence base is a potential limitation of the study as it makes uncertainty difficult to interpret. There has been longstanding debate in the literature regarding the meaning of the ‘neutral’ response in Likert-type response scales (in this case, the score of 5 in a 1–9 scale). 60 When investigating views that may be based on knowledge of evidence, interpretation is rendered even more uncertain. Box 6 presents three different meanings of a neutral score in this context.

This point is important because it would be unethical to randomise patients to an effective treatment in a new clinical trial of SDD if neutral scores on key items were interpreted as equipoise or uncertainty when actually they could reflect ignorance. To check this possibility, we inspected cross-tabulations between knowledge scores and scores on all the items relating to Beliefs about the consequences of SDD. If neutral scores on the consequences items were over-represented by participants whose self-assessed knowledge of the SDD evidence base was at or below the mid-point, this would suggest that a modal score of 5 largely reflected ignorance. These cross-tabulations showed that the mid-point on the Beliefs about consequences items was endorsed by a large proportion of participants whose self-assessed knowledge of the evidence base was high. The one exception was that participants with higher knowledge scores tended to agree more strongly that ‘SDD reduces VAP’. Overall, this suggests that endorsement of the mid-point signified options 1 or 2 from Box 6: uncertainty arising from evenly balanced evidence or gaps in the evidence base.

Further methodological strengths

As noted in Chapter 3, Delphi methodology is often associated with achieving consensus whereas, in this study, we aimed to assess consensus, identifying the most important views and barriers to implementing SDD and, in particular, assessing the views of the four stakeholder groups about the current evidence base for SDD. This objective (rather than any imperative to achieve consensus) informed the methodological approach in this study.

Indices of group stability may mask individual instability if different subgroups change their views in opposite directions. In this study there was evidence of stability at both the individual and group levels. To our knowledge, this is the first study to assess stability of views at both these levels. The observed stability enhances our confidence that the identified views will be relevant over time, unless the profile of evidence or the knowledge of the evidence changes.

It is also rare in Delphi studies to identify consensus around uncertainty, where this exists, and to contrast this with consensus around agreement. Given the overarching objective of the study (to identify the most appropriate SDD research agenda), consensus around uncertainty was of great importance. With regard to this, the findings were critically appraised to identify whether uncertainty was likely to have arisen from inadequacies of the evidence base or inadequacies in individuals’ knowledge of the evidence. For the first time in a Delphi study in this field, we explored, in detail, the distributions of responses beyond a simple cut-off criterion for consensus by identifying different levels of consensus (> 90%, 75–90%, 50–74%). Below 60% consensus levels, there was evidence of bimodality. In other words, instead of sample statistics describing the views of one group (spread around a modal score), the distributions of scores suggested that there were two or more subgroups each with their own (different) modal score. Importantly, the distribution of responses for the item ‘I am opposed to SDD’ was spread around the two modes 1 (i.e. not opposed) and 5 (i.e. uncertain). It is possible that this distribution reflected subgroups representing professional groupings with different views of SDD, or groups of participants with either in-depth knowledge or poorer knowledge of the SDD evidence base. The sample size for the Delphi study was not sufficient to address this question. Factors associated with opposition to, or support of, SDD were explored further with the larger sample in the national survey, reported in Chapter 5.

A further strength of this study was its grounding in a theoretical framework that enabled us to distinguish between factors related to the clinical evidence (i.e. Beliefs about consequences of SDD) and potential barriers related to more practical issues to do with professional roles, resources and the management of change. Box 7 presents the key findings from the Delphi study that inform the need for, and acceptability of, a clinical effectiveness RCT or an implementation RCT.

Sample

Materials

The development and validation of the questionnaire consisted of two steps:

1. Item generation. We included items relating to the issues identified in the Delphi study as important. This included overall views about SDD, intention (willingness) to participate in a randomised trial and the factors likely to influence these intentions (e.g. beliefs about the consequences of SDD, views about the ethics of further SDD effectiveness research). The questionnaire commenced with preliminary questions to check participant eligibility: (1) ‘Have you answered this questionnaire before?’ and (2) ‘Do you have clinical involvement in the care of patients in intensive care?’. The questionnaire was required to be relatively brief to maximise response rates and comprised 23 items. The first question asked about current SDD practice in the participants’ units. Further items assessed participants’ knowledge (item 2) and views about SDD (items 3 and 4). Beliefs about consequences was by far the most elaborated domain in the Delphi study; therefore, the seven items with the highest importance ratings from this domain at round 3 of the Delphi were included (items 5–11). Items 12–17 were included to address research question 7. Response options followed guidance for questionnaire design (seven options for a well-educated sample from 1, strongly disagree, to 7, strongly agree). 61 Finally, six items (items 18–23) were included to assess demographics characteristics in order to describe the sample and assess the representativeness of the responder group. Participants were also provided with the space to include free-format comments at the mid- and end point of the questionnaire. The comments data were used to check the content validity (coverage) of the questionnaire as it was assumed that participants would comment on any important issues that had been omitted from the questionnaire. The final version of the questionnaire is presented in Appendix 5.

2. Pretesting. The draft questionnaire (and cover letter) was pilot tested to assess wording, acceptability and length, using personal interviews with four clinical collaborators (these individuals were not from the sampling frame so that data from all intensive care consultants and microbiologists could be used in the analysis). Each question was evaluated by the research team in the light of the pilot test findings and reworded if necessary to enhance meaning and acceptability. 62

Procedure

An e-mail invitation, containing a link to the online questionnaire, was sent to all members of the ICS on 1 December 2011 by the society, on behalf of the study team. Two reminder e-mails were sent to all ICS members 3 weeks after the first posting and 3 weeks after the Christmas and New Year holiday period. Personalised e-mails were additionally sent to consultant members of the ICNARC case mix programme on 14 February 2012 by the Director of ICNARC (KMR; one of the SuDDICU study investigators) requesting their support for the survey and sending an electronic copy of the questionnaire and a link to the online questionnaire.

E-mail invitations were sent to the clinical microbiologists by the HIS on 1 December 2011. The second society (BSAC) sent the invitation to its members on 10 February 2012. We relied on the professional societies to send reminders to their members if they deemed this appropriate based on other communications being sent at the same time.

The questionnaire instrument was programmed so that clicking on the submit button resulted in upload of data into an electronic database. Questionnaires were anonymous and so it was not possible to target reminders to non-responders. The national survey was advertised at a national conference63 in order to generate publicity and interest from key stakeholder groups and maximise response rates.

Data management and analysis

Analysis included simple descriptive statistical testing and statistical prediction techniques. First, responses to each question were summarised using frequency distributions. Second, multiple regression techniques (including linear regression and, when appropriate, logistic regression) were used to identify the predictors (i.e. theoretical constructs and characteristics of responders) with intention to participate in the SDD research. The differences between intensive care consultants and microbiologists were explored.

The comments provided by participants were sorted into three categories: new information (reported in the results section), elaboration of the reasons for giving certain responses and comments about the procedure and quality of the questionnaire.

Ethics approval for the national survey was granted by the Research Ethics Board of the College of Life Sciences and Medicine, University of Aberdeen (CERB/2011/8/633).

Response rates

Invitations were sent to 2908 ICS members (including 1685 intensive care consultants), 528 members of HIS (excluding 101 invitations that were not delivered), 866 BSAC members and 500 consultants registered with ICNARC. As mentioned previously, there is likely to be considerable overlap across each of these membership lists. A recruitment flow chart is displayed in Figure 9.

FIGURE 9.

National survey recruitment flow chart.

Responses were received from 468 UK clinicians and the breakdown by professional group is shown in Table 15. Clinicians were likely to be a member of more than one professional society; therefore, only an estimate of the response rate could be generated. Furthermore, it was unclear how many e-mail addresses would have been out of date or how many e-mail messages would have reached spam folders rather than inboxes of eligible clinicians. We provide a worst-case response rate in Table 15.

Participant characteristics

Four hundred and nineteen participants were consultants in intensive care medicine and 49 were consultant clinical microbiologists. The sample as a whole reported a high level of experience in clinical practice (mean 19 years, SD 8 years). Participants also provided descriptive details about the hospital(s) they worked in, shown in Table 16. For comparative purposes, data from the ICNARC database and UK ICUs are presented alongside the participant characteristics of this sample. As shown in Table 16, the SuDDICU sample had greater representation from larger units and from units affiliated with universities.

Participants were asked to report their hospital’s current policy with respect to SDD delivery. Six options were offered:

1. not delivering SDD and have not considered this issue

2. not delivering SDD after careful consideration

3. not delivering SDD but issue currently being considered

4. delivered SDD in the past but reversed policy and now do not deliver

5. full SDD sometimes delivered but not protocolised, full SDD formally adopted

6. full SDD formally adopted, protocolised and routinely delivered to specific patient subgroups.

Responses are displayed in Figure 10.

FIGURE 10.

SDD policy in participants’ hospitals.

Participants’ self-reported knowledge of the SDD evidence base was assessed with the item ‘I know the SDD evidence base well enough to have an informed opinion regarding its use’. The mean response (1, strongly disagree, to 7, strongly agree) to this item was 4.7 (SD 1.4). Intensive care consultants’ responses to this item had greater spread than the microbiologists’ responses, but the modal score was 5 (signifying slight agreement), as displayed in Figure 11.

FIGURE 11.

Self-reported knowledge of the SDD evidence base (1 = strongly disagree, 7 = strongly agree). Responses to the item, ‘I know the SDD evidence base well enough to have an informed opinion of its use’. (a) Intensivists; and (b) microbiologists.

Free-text comments on content validity of the questionnaire

Four participants made comments on the content validity of the questionnaire. One of these suggested that there should have been clearer differentiation between SDD and selective digestive decontamination. The other three suggested that the questionnaire was insufficiently disease specific and that some questions should have been asked with respect to specific disease categories or patient subpopulations (e.g. trauma patients). A total of 80 participants left a comment and the vast majority of these elaborated on their response to particular items. This information will be considered in the design of any future SDD research, but is not reported further in this chapter.

Views about selective decontamination of the digestive tract

There was a bimodal distribution of responses in both professional groups for the item ‘I am opposed to SDD’, as shown in Figure 12. The clinical microbiologist participants tended to be less favourable towards SDD (i.e. towards the right of the scale; modal score of 6) than the intensive care consultant participants (primary mode of 4), but there was a secondary mode of 2 signifying ‘not opposed’ in both groups.

FIGURE 12.

Opposition to SDD by professional group (1 = strongly disagree, 7 = strongly agree). Response to the item ‘I am opposed to SDD’. (a) Intensivists; and (b) microbiologists.

The association between current SDD delivery and views about SDD is presented in Figure 13.

FIGURE 13.

Box and whisker plot of opposition to SDD by SDD delivery group (larger circles represent mean score).

A one-way analysis of variance on mean score of opposition indicated that there was a difference in mean scores between the groups. Post hoc pairwise comparisons revealed that group 6, the participants working in hospitals where SDD was formally adopted, protocolised and routinely delivered, scored significantly lower on opposition to SDD compared with groups 1, 2 and 4. These results were confirmed using non-parametric tests (Kruskal–Wallis test of medians).

Opposition to SDD was significantly correlated with all seven items relating to the beliefs about the consequences of SDD and with the item ‘I am opposed to the i.v. component of SDD’, as displayed in Table 17. There was no significant association between knowledge of the SDD evidence base and opposition to SDD.

Regression analysis was performed to investigate multiple predictors of opposition to SDD. Nine item variables and five dummy variables (to represent current practice) were entered into the analysis. As shown in Table 18, there were eight significant predictors: current practice (currently delivering full protocolised SDD), opposition to the i.v. component of SDD, self-assessed knowledge of the evidence base and five specific beliefs about the consequences of delivering SDD. The five specific beliefs were SDD increases antibiotic resistance, SDD benefits patients, SDD reduces HAIs, SDD reduces VAP and risks of SDD outweigh benefits.

Acceptability of further selective decontamination of the digestive tract research

Most participants (85%) reported that the current uncertainties in the SDD evidence base should be addressed in a new study. Participants’ views about whether or not it is ethically acceptable to conduct further RCTs evaluating effectiveness are displayed in Figure 14. The mean score was 5.4 (SD 1.5) on the scale range of 1–7, indicating that participants tended to agree that further effectiveness RCTs were ethically acceptable.

FIGURE 14.

Views about ethical acceptability of further effectiveness RCTs. (1 = strongly disagree, 7 = strongly agree) Responses to the item ‘It is ethically acceptable to conduct further RCTs evaluating the effectiveness of SDD’. (a) Intensivists; and (b) microbiologists.

Ethical acceptability of further effectiveness research was significantly correlated with four items relating to the beliefs about the consequences of SDD, as displayed in Table 19.

Regression analysis was performed to investigate multiple predictors of views about the ethical acceptability of further SDD effectiveness research. Seven variables were entered into the analysis. As shown in Table 20, there were two significant predictors: ‘Overall, SDD benefits the patients to whom it is delivered’ and ‘SDD reduces VAP’.

It could be argued that instead of a linear relationship between ethical acceptability and beliefs about consequences, there should be an inverted-U relationship between these factors. That is, if the evidence base clearly shows benefit or harm, then further effectiveness research will be uninformative, and uncertainty (i.e. the neutral response on the Likert scale) should be associated with the highest levels of ethical acceptability. To identify whether or not there was a signal in the data to support this possibility, we inspected the scatterplots for two key beliefs about the consequences of SDD: increase in antibiotic resistance and reduction in mortality. These scatterplots, displayed in Figure 15, indicate a weak signal for a curvilinear relationship as proposed. However, the trend was not strong enough to test for an inverted U–shaped relationship in a multivariate model.

FIGURE 15.

Scatterplots indicating a curvilinear relationship between ethical acceptability of further SDD effectiveness research (on y-axis) and (a) ‘SDD increases antibiotic resistance’ and (b) ‘SDD reduces mortality’ (on x-axis). (1 = strongly disagree, 7 = strongly agree.)

The multimethod nature of this study afforded the opportunity to examine the question of the relative importance of different beliefs about the consequences of delivering SDD, using a multimeasure approach to check whether or not three indices of importance converged. The three indices of importance were:

1. the frequency with which each belief was elicited in the Delphi round 1 interviews (as discussed in Chapter 3)

2. importance ratings (possible range 1–9) made by the Delphi participants in round 3 (reported in Chapter 4)

3. the size of the standardised regression coefficients when the beliefs were used to predict scores representing opposition to SDD (in the national survey, reported in Table 18).

The three indicators of importance for the Beliefs about consequences items in the national survey are presented in Table 21 in order of apparent importance from round 1 data. The table shows that, on all three measures, the potential effect of SDD on antibiotic resistance was the most important clinical consequence of SDD. Furthermore, the rank orders of the importance indicators based on round 1 data and regression coefficients were consistent (except for one item with changed wording for the national survey). The least convergent index of importance was the mean rating of importance at round 3.

Feasibility of further selective decontamination of the digestive tract research

Participants were asked to state whether they would be prepared for their patients or centre to be randomised in an effectiveness RCT and the majority of participants indicated that they would (78.4%), as displayed in Table 22. Furthermore, almost all participants (94.2%) reported that they would go along with an effectiveness RCT if their colleagues supported it. Almost two-thirds of participants (63.5%) were willing to support an implementation study. Only 14.3% of participants were unwilling to participate in either an effectiveness RCT or an implementation study. The majority of participants (58.1%) indicated they were willing to participate in either type of study design. A lower proportion of microbiologists than intensive care consultants were prepared to participate in an effectiveness RCT (42.9%) or an implementation study (34.7%).

Multiple regression analyses were conducted to examine the predictors of participants’ willingness to participate in (1) an effectiveness RCT and (2) an implementation study. The results for the first regression analysis are displayed in Table 23.

Table 23 shows that willingness to participate in an effectiveness trial was predicted by participants’ responses to the items asking whether or not it is ethically acceptable to conduct further RCTs evaluating the effectiveness of SDD and participants’ overall opposition to SDD. However, the most powerful predictor of this item was whether or not a new study should address current uncertainties in the evidence base.

The results of the regression analysis of willingness to participate in an implementation study are displayed in Table 24. Willingness to participate in an implementation study was predicted by overall opposition to SDD and responses to the item ‘Current uncertainties in the evidence base should be addressed in a new study’.

Participants were further asked to specify the components of a control group for a possible future effectiveness RCT of SDD. Responses are summarised in Table 25. The greatest proportion of participants (90.4%) stated a desire for a control group that delivered VAP bundles. Slightly fewer participants preferred the control group to include chlorhexidine mouthwash (85.7%) and a smaller proportion of participants felt that the control group should include a ‘standard practice’ intervention that would reflect variations in current practice. In terms of outcome measures, the majority of participants (91.9%) agreed that an effectiveness RCT should include a measure of antibiotic resistance (details are presented in Table 25).

Answers to research questions

In terms of current SDD practice, around 10% of the sample reported currently delivering SDD whereas around 40% have not yet considered SDD.

The distribution of scores reflecting opposition to SDD was bimodal and this bimodality was evident among both intensive care consultants and clinical microbiologists. In other words, both professional groups included a substantial proportion (approximately 20%) who were not opposed to SDD. The other, primary, mode was at the mid-point of the scale (reflecting uncertainty) for intensive care consultants and at the ‘opposed’ end of the scale for microbiologists. Opposition to SDD was significantly predicted by all the beliefs about consequences items in the questionnaire but was not predicted by self-assessed knowledge of the SDD evidence base.

A large majority of the participating clinicians reported that uncertainties should be addressed in a new study, that further SDD research would be ethically acceptable and that they would be prepared to participate in further SDD research. Hence, further research appears to be appropriate, acceptable and (from the perspective of recruiting ICUs for a trial) feasible.

As would be expected, there was a strong association between current provision of SDD (with full protocol) and support for SDD (see Table 18). In terms of participants’ beliefs, the strength of belief about whether or not SDD increases antibiotic resistance was the strongest predictor of two key opinions: opposition to SDD and the belief that further SDD research would be ethically acceptable. The strength of belief that current uncertainties in the evidence base should be addressed in a new study was the strongest predictor of two intentions: willingness to participate in future effectiveness research and willingness to participate in future implementation research. There was strong support for the following trial design features: antibiotic resistance to be measured as an outcome and control group to receive VAP bundles or chlorhexidine mouthwash.

Some differences were evident between intensive care consultants and clinical microbiologists, a finding that is not surprising as many microbiologists see their professional role as centred around antimicrobial stewardship. Hence, the profession repeatedly advances the rationalisation of antibiotic use as a Department of Health priority. 64 However, the current study suggests that not all clinical microbiologists are certain of the evidence base to support this view in the context of SDD.

Strengths and limitations

The response rate was difficult to estimate but was probably in excess of 25%. Thus, although the results may not have been based on a representative sample, they do indicate a substantial mass of support among the intensive care community for further SDD research. Indeed, the number of intensive care consultants and clinical microbiologists who would be prepared to participate in an effectiveness trial or an implementation trial involving SDD is far in excess of the number that would be required to mount such a trial. Furthermore, this study reflects the largest number of clinical microbiologists’ views on SDD recorded to date. However, the response rate remains a substantial limitation to the generalisability of these findings. A further limitation is that only two of the four stakeholder groups from the Delphi study were included in the survey study. However, these two groups (intensive care consultants and clinical microbiologists) were the two groups that Delphi participants identified as potentially holding conflicting views. Although there were some differences between these two groups, there was also considerable overlap in terms of their support of, or opposition to, delivering SDD, as described in Figure 12.

A further limitation may be that the participant group possibly did not fully grasp what is meant by ‘implementation trial’. It is puzzling that so many participants (51%) reported being prepared to participate in both an effectiveness trial and an implementation trial. On the face of it, an effectiveness trial should be considered only when there are uncertainties in the evidence base, whereas an implementation trial should be considered only when the evidence shows that an intervention has clear benefit, but has not yet been adopted. The implementation trial then evaluates the extent to which an intervention succeeds in changing practice. Although a naive view is that implementation of evidence-based interventions into routine practice is a simple matter of publishing a guideline, there is ample evidence that such passive strategies are often insufficient to improve patient care. 65

This survey had a number of methodological strengths. First, the questionnaire items were generated from theory-based interviews with four stakeholder groups (reported in Chapter 3) and were subsequently selected for inclusion on the basis of an iterative process of prioritisation (reported in Chapter 4). The importance of a range of clinical outcomes of SDD was assessed using multiple methods. An extremely robust finding across indices of importance was that antibiotic resistance is the most important issue. Hence, careful attention should be given to measuring this in any further SDD research. Importance ratings did not converge with the other indices of importance. This has implications for Delphi methodology, as ratings of importance are key to the assessment of clinical consensus using this approach. 53

A further strength is that we identified that further trials are feasible (i.e. a sufficiently large number of clinicians would be prepared to participate) and also ethically acceptable. Ethical acceptability could arguably be a function of uncertainty about the evidence base (rather than being a function of certainty that the intervention is effective). There was a small signal in the data to suggest this may be the way in which clinicians think about ethical acceptability and such thinking could be investigated in more detail in future research. Box 8 presents the key findings from the national survey that inform the need for, and acceptability of, a clinical effectiveness RCT or an implementation RCT.

Sample

A list of expert international triallists with known expertise in intensive care trials and/or implementation trials was developed by members of the international research team (MKC, BHC, GB, GSM and one Australian collaborator). As this study sought the views of experts with in-depth experience in their field, the target sample size was relatively small, being set at 10. This number was likely to provide a suitable balance between the expected homogeneity of the sample and the study’s objective of eliciting views from those who might favour a clinical trial and those who might argue for an implementation trial. To ensure that 10 interviews were available for analysis, we identified an initial list of 20 expert triallists to approach. Potential participants from the UK, North America, Europe and Australia were identified from their public profile and the research team’s knowledge of the field, and were invited to take part if they had expertise in (1) randomised clinical trials of intensive care interventions or (2) implementation trials. 66

Data collection

Semistructured one-to-one telephone interviews were conducted using a topic guide developed from expert experience and informed by the identification (from earlier phases of the study) of clinicians’ interest in further SDD research. Telephone interviews were considered a more efficient use of resources (i.e. time, cost, effort) than face-to-face interviews for generating data on this international sample. 67 Topic guide prompts included questions about participants’ preference for an effectiveness or implementation trial and aspects of trial design (e.g. individual vs. cluster randomisation), specification of the components of the SDD intervention and of control group care, outcome measurement, recruitment and ethical considerations. The topic guide for these interviews is included in Appendix 6.

Procedure

The UK clinical lead (GB) sent a personal e-mail to each potential participant informing them that they would be invited to take part in the study and would receive an information sheet and consent form, by e-mail, from the project manager (EMD). Those expressing an interest in taking part were contacted to set up mutually convenient times for a 30-minute recorded telephone interview. A senior triallist (GB) conducted all interviews by telephone and began by giving a summary of the SuDDICU study. Participants were asked to provide their general views of conducting further research about SDD in ICUs and any issues they foresaw arising in undertaking such research. Participants were then asked to consider two possible trial types: a RCT to evaluate the effectiveness of SDD or a RCT to evaluate a behaviour change intervention (aimed at health-care professionals and ICUs) to increase uptake of SDD (i.e. an implementation trial). They were asked to indicate whether or not they had an initial preference for discussing one particular type of trial (effectiveness vs. implementation). The interviewer returned to the other option if time allowed. Prompts were used to elicit trial design features and associated challenges and barriers. Triallists were encouraged to talk freely and prompts were used only if the participants did not cover specific areas relating to the research question. While triallists’ preferences for effectiveness or implementation trials and for specific trial design features (e.g. cluster randomisation) were explored during interview, these were not the primary focus of this stage of the research. Preferences were elicited to provide the context to ask about challenges and barriers to undertaking a large multinational RCT of SDD in ICUs. Interviews were audio recorded and transcribed verbatim.

Sample characteristics were not requested during interview, as participants’ eligibility for this study was based on information available from participants’ public profiles (e.g. the country in which they work and their clinical and research expertise) rather than on demographic information (e.g. age).

Data management and analysis

Interview transcripts were analysed thematically using a framework approach. 36 The technique involves a systematic process of inductive qualitative data analysis. Familiarisation with the content of each transcript was followed by the creation of ‘charts’ (i.e. Word document tables) to summarise issues identified as relating to potential challenges and barriers to undertaking further research in the field of SDD. The issues identified were grouped into themes, discussed within the research team, refined and checked against the charts to ensure that all relevant issues were accommodated.

One researcher conducted the initial coding of the first six transcripts, working systematically through the transcripts as they became available and highlighting all text referring to (1) trial design/conduct or (2) trial challenges/barriers. In order to retain the context of each quotation, highlighted sections were extensive and inclusive of all text relevant to a particular design feature or challenge. Highlighted text from each transcript was then copied into Word document tables (‘Charting’). 36 A second researcher (MEP) reviewed the coding/highlighting on two transcripts to check that no relevant text had been excluded. MEP conducted the initial coding of the remaining transcripts and undertook the second phase of analysis: identifying initial themes and using the themes to group together quotations from different participants. The headings and content of the tables produced during the charting phase were refined in numerous versions until all relevant data were tabulated without overlap or duplication. Data were then systematically examined and emerging typologies discussed with other members of the research team (GB, JJF, EMD) to ensure clinical sensibility and methodological rigour. In this study (and in the rest of the SUDDICU study) the terms ‘acceptability’ and ‘feasibility’ were defined, respectively, as a willingness to participate in further SDD randomised trials and an ability to participate or to conduct such trials. 1

Ethics approval for this study was obtained from the University of Aberdeen College of Life Sciences and Medicine Ethics Review Board (Ref: CERB/2011/3/610).

Participant characteristics

Twenty-two international triallists were identified as meeting the eligibility criteria and were invited to participate. Sixteen triallists were recruited and 13 were interviewed. However, the audio-recording equipment failed during three interviews, with a resultant loss of data from these participants (T2, T4 and T8). The 10 interviews that generated completed transcripts were included in the analysis and form the data presented below. The interviewer’s notes from the other three interviews are presented at the end of the results section (see Findings based on notes from the three interviews that were not transcribed).

Of the 10 triallist interviews available for analysis, four triallists were from Canada, three from Australia, two from the USA and one was from Europe. All 10 triallists had expertise in randomised clinical trials of intensive care interventions and three of them also had expertise in leading implementation trials.

Context for discussing challenges: effectiveness or implementation trial?

First, we focus on participants’ views about the current evidence base in SDD and, in the light of these, their views about appropriate implementation strategies.

Of the 10 triallists whose interviews were analysed, nine reported being persuaded by the evidence in favour of SDD. These triallists offered their opinions about the interventions that would be effective in changing the behaviour of the relevant clinicians, or suggested research designs to evaluate the effectiveness of behaviour change interventions. These views are presented in Table 26. This table presents multiple rows for some participants, as this helps to retain the context of the linked quotations.

From Table 26, six triallists were persuaded by the evidence of effectiveness but also noted widespread uncertainty with respect to potential harms of SDD, at least at the population level. Four of these participants recommended an effectiveness trial that focused on microbial ecology and one also noted the importance of an economic evaluation. Two participants recommended an implementation trial with clinical measures as secondary outcomes and one recommended a behavioural intervention trial based on the findings from the current series of studies.

Challenges to the acceptability and feasibility of further selective decontamination of the digestive tract research in intensive care units

The first level of challenges related to the acceptability of SDD as an intervention and to the acceptability of any further SDD research. These challenges were seen to represent potential barriers that would need to be overcome before proceeding with any effectiveness RCT (of any design).

It was suggested that gaining sufficient international support for further SDD research could be a challenge (T, triallist participant code):

I am not sure how many people in the world there are that are willing to do that [support further SDD research], which have such a strong belief in this intervention.

T7

Achieving widespread multidisciplinary endorsement and support for further SDD research in ICU was identified as a potential challenge.

. . . not only have you got to overcome the ICU, the surgeons and everyone else, but you’ve also got all these different views in terms of infectious diseases and microbiology about what we should and shouldn’t be doing.

T6

However, such endorsement and support was noted to be a prerequisite for conducting further research to evaluate either the effectiveness of SDD in ICU, or a behavioural intervention to increase SDD uptake.

I think it is getting ICUs that want to play ball and where it goes beyond more than just verbal commitment, that there is commitment from the entire team and the hospital, I think is paramount.

T12

Another suggested barrier to further SDD research focused on perceptions of tension between the individual and societal consequences of delivering SDD in ICU (i.e. the potential benefit to individual patients receiving SDD vs. the concerns about the societal/population impact of antibiotic resistance resulting from the routine delivery of SDD).

. . . the key aspects to me are the resistances seen at the population level whereas the preventing the infection with SDD is more at the individual than population level . . .

T12

These findings suggest that the acceptability of SDD as an intervention may not be related to the perceived effectiveness of SDD in individual patients, but to beliefs that these benefits are outweighed by the potential harms of SDD at a population/societal level.

Challenges to the acceptability and feasibility of conducting a definitive effectiveness randomised controlled trial comparing selective decontamination of the digestive tract with a control group

It was suggested that even if clinicians were convinced of the benefit of SDD as an intervention, and that support for further SDD research in ICUs could be achieved, there were acceptability and feasibility issues that present further barriers/challenges.

Triallists reported that, were a definitive effectiveness RCT to be planned, it would have to be large.

. . . it would be a trial of that order, 6000 or 7000 people, ideally executed across the world.

T7

The size and international nature of such a trial was suggested by participant T7 to present acceptability and feasibility issues, for example the level of personal commitment required from a chief investigator to run such a trial and the challenge of gaining international funding:

I hope you have sold your children to medical science because you are not going to have time to do anything else and just huge personal burdens. It is a big deal; I would not take it on; it would have to be someone younger with more energy.

T7

Barriers relating to specific design features of an effectiveness randomised controlled trial comparing selective decontamination of the digestive tract with a control group

In addition to the general challenges of conducting a definitive multinational effectiveness RCT, triallists identified potential barriers relating to the specific design features of such a trial (e.g. unit of randomisation).

Triallists’ views on the specific design features of a definitive effectiveness RCT varied, as did their views on the barriers related to specific features. The majority of discussion focused on triallists’ perceptions of the challenges associated with the choice of clinical intervention and comparison/control, when the intervention would be started, outcome measures, unit of randomisation and level of consent.

Specification of the clinical intervention and control

The lack of a formal and universally accepted SDD regimen was suggested by triallists to pose potential acceptability issues for a trial.

The discussion of what you use as antibiotics in such a protocol . . . would you use some sort of so-called ‘old fashioned’ antibiotic like cefotaxime or do we now need to go to something else . . .

T6

One triallist stressed the importance of transparency in the intervention development phase and consultation with stakeholders to address this issue and optimise acceptability of both the trial and the intervention.

I think in developing what the intervention is, the different aspects of the intervention for SDD, it obviously has to be built from consensus . . . amongst intensivists in terms of their buy-in, so going beyond the current group of co-investigators and . . . testing it on people who are outside that group would be really important . . . documenting . . . the process of determining what the . . . intervention is and even publishing that would be important too.

T12

Establishing current international standard care for ICU infection prevention and choosing an appropriate comparison/control for a SDD effectiveness trial were also identified as problematic. It was suggested that the widespread use of oral antiseptics (e.g. chlorhexidine) and/or the use of other procedures associated with SOD or ventilator bundles (e.g. semi-recumbent positioning, subglottic suctioning) present a challenge to determining the components of the intervention and control arms in a definitive SDD effectiveness trial.

If people are using chlorhexidine universally and I cannot really see why not except for the usual, that it takes forever to get people to change behaviour, given the low cost, given the benefit of chlorhexidine and its availability, then both groups should have chlorhexidine . . . So I think we need to be very, very careful about how we handle oral antiseptics when we’re considering an SDD trial.

T5

If you/I are going to develop a trial, I think that there is now SOD and this is complicating everything . . . how you’re going to involve SOD? In other words, are we going to have three arms or are you going to leave out SOD . . . you have to address the problem of SOD.

T9

. . . the subglottic suctioning and . . . the positioning of the paste and all that kind of stuff [is] now considered basically standard care in most places . . . I think it might be difficult to randomise people to doing that or not doing that.

T1

One suggestion to overcome the variation in usual care was to ignore it and accept a lack of standardisation in the care received by patients in the trial control group (i.e. assume a completely pragmatic design).

I would do absolutely nothing, I would just say this is a large multicentre individual randomisation base pragmatic trial . . . Let’s not create a world where we know everything about a place that doesn’t exist. Let’s keep the world dirty, messy, chaotic, random, insane, etc., as it is, and within that world, let’s just change one thing.

T7

The potential benefit of conducting a double-blinded trial using placebo oral pastes, mouthwashes and i.v. components was mentioned during the interviews. However, the associated ethical implications and practical challenges were not discussed, although one triallist reported experience of using placebo oral pastes during a trial to treat, rather than prevent, infections in an ICU context.

We just did an intervention of SDD. We tried breaking an outbreak. We got a neutral paste. That’s how we blinded it. In the placebo arm we used a neutral paste.

T9

Outcome measures

Triallists suggested a number of outcome measures that could be used to evaluate the effectiveness of SDD. The most commonly discussed measures were mortality, VAP and antibiotic resistance patterns. However, triallists reported potential challenges with the use of each of these measures. Mortality (timescale unspecified by participants) was proposed by some triallists to be the optimal primary outcome measure as it is straightforward to measure and a large, multinational SDD effectiveness trial that demonstrated a significant effect on mortality would be persuasive and conclusive.

. . . giving clarity around mortality outcome would be crucial. So international, multinational, large, focused on mortality.

T5

Nevertheless, it was suggested by others that the low incidence of mortality limits its suitability as a primary outcome measure to evaluate the effectiveness of SDD.

So it [SDD] might actually be an impossibly high hurdle to change mortality . . . if you enrolled a thousand patients . . . cut 100 VAPs to 80 VAPs and so you’ve changed 20 VAPs, of those 20 VAPs, there might only have been one extra death in the 20 VAPs, so you’ve enrolled a thousand patients and [avoided] one extra death.

T3

The main challenge identified with the use of VAP as an outcome measure related to difficulties with its definition and diagnosis.

I think the major problems are diagnosing a pneumonia you know, which is always been a vexatious issue and so I think because it’s a relatively loose diagnosis, that’s one issue.

T13

It was also suggested that, as with mortality, the low incidence of VAP in ICUs was a potential barrier to its acceptability as an outcome measure and to the feasibility of its use in a trial to evaluate the effectiveness of SDD (particularly in the context of the widespread use of SOD).

. . . we don’t have much in the way of Gram-negative VAP that . . . we’re aware of. So whether or not we need a different approach for more of a Gram-positive VAP, but again, we don’t seem to be having much of that at the moment. So my worry is that it’s a relatively low-grade problem and you’re having to treat everyone or everyone who’s at high risk for a low-grade problem.

T13

Bacteraemia was suggested by some triallists as a viable alternative outcome measure to VAP.

If you really wanted to know something, I would do some hard stuff like bacteraemia. It would have to be logistically simple, objective, easily ascertained, easily monitored, identified without question by somebody else, like bacteraemia would be and that is the end of that.

T7

It was acknowledged by a number of triallists that the inclusion of some measure of antibiotic resistance would be worthwhile to address the current lack of evidence of the effect of SDD on rates and patterns of antibiotic resistant pathogens. However, it was suggested that such an outcome measure would be costly and would impact on power calculations and other aspects of the trial design such as the unit of randomisation.

. . . we cycled antibiotics and we looked at a resistance and it is hard. It’s very hard. You need a hell of a lot of events to . . . get decent numbers out of it. I think that’s just unachievable . . . you know, you’re looking at thousands and thousands of patients and that’s why I said microbiology . . . to me is hideous. No way are you going to overcome that.

T9

. . . it [incidence of multidrug resistant pathogens] would be very difficult . . . what kind of power calculations can you do if you wanted to use incidence of development of multi germ resistance pathogens . . . because you really don’t know what’s going to happen and there’s such a big variation. I think it . . . would be difficult . . . as your primary influence even though it’s really critical issue.

T1

Unit of randomisation

Triallists reported advantages and disadvantages associated with using cluster randomisation or individual patient randomisation in a SDD effectiveness RCT. Two triallists suggested the unit of randomisation and choice of outcome measure are inextricably linked.

. . . you can’t really do an ICU population level outcome if you are randomising at the patient level.

T3

The problem is if you randomising individual patients, you’re ignoring the possibility of a horizontal spread . . . and therefore you’ve got to cluster randomise.

T9

Reasons given in favour of randomisation at the individual patient level, or against cluster randomisation, were:

1. 1. Individual randomisation is more feasible for a worldwide study.

I would think that the cluster randomisation as a worldwide phenomenon for SDD is going to be very difficult so you are going to have to look at individual randomisation.

T9

1. 2. Differing baselines in ICUs.

A cluster RCT is a scary high-risk position because if there’s a difference in baseline severity of illness within the ICU then you’re in trouble.

T3

1. 3. Recruitment challenges to cluster randomisation resulting from lack of willingness of a sufficiently large number of units.

A cluster randomised trial requires dozens and dozens of units agreeing to be allocated to one or the other . . . there may not be sufficient of units who may say yes to agreeing to participate for months or years in a randomised trial whereby either one get it or not. It is hard. So the logistics of doing a cluster randomisation are enormous . . . because this is so loaded and everybody has opinions.

T5

Level of consent

The unit of randomisation and level of consent were sometimes linked by triallists when expressing their views on the potential challenges associated with gaining consent at the individual patient level.

. . . the cluster trial would be a much more efficient design because you could make much more use of all patients. I think at, an individual level, you know that is not always the case because, of those eligible, very few end up being consented or being enrolled and then having full consent . . . the consent not withdrawn I guess I should say in this case. So I think again, . . . from the public health perspective, using a cluster trial is not only efficient, it is just the right thing to do and I am fairly adamant about that.

T12

There were also suggestions that the use of individual patient consent in a cluster-randomised trial was impractical and could potentially affect recruitment.

. . . if you do it by individual form of consent obviously some people will say ‘no’ and then you’ll have a hard time doing your analysis because some people would have been put into it.

T1

Barriers resulting from international differences in models of consent and research governance procedures were also mentioned during interview.

I would imagine there would be some countries if not regions . . . whereby an approach to units being randomised might be more palatable when alternate models of consent have already been endorsed, such as waived or deferred consent, [as in studies in] some parts of Europe and parts of Australia.

T5

It was suggested that it may be difficult to get approval in the USA for a trial that used unit level consent.

On the downside would be the individual consent . . . very hard in the US to configure an alternate consent model other than the standard which would be first person consent which we can’t do much on the ICU or a priori substitute who can legally consent. I think it’s hard for Americans to plot anything other than that model.

T5

In contrast, research governance procedures in Australia were reported to allow delayed consent (i.e. gaining consent from a patient or relative after enrolment into the trial) in order to aid trial recruitment. However, it was suggested that such a model may not be acceptable elsewhere in the world.

Well this is a study that you probably would allow delayed consent for here in Australia because it so much facilitates patient recruitment . . . In our experience delayed consent confirmation takes about 20 minutes and prospective consent takes about 60 and 70 minutes and again, if you are running a trial of 20 people, that’s okay. If you are running a trial for 10,000 people, that is 10,000 hours of work that you have to add and that’s really hard, and your consent rate will decrease and your recruitment rate will also decrease and then people will develop trial fatigue.

T7

Practical barriers to trial conduct

In addition to the potential challenges associated with conducting further SDD research and the design of a trial to evaluate the effectiveness of SDD, the triallists suggested that there may be a fourth level of challenge associated with practical barriers to the conduct of such a trial. These potential barriers were also examined in earlier stages of this study. The following results should be seen in the light of other study findings, which are integrated in Chapter 7.

Three triallists stated that, in their experience, nurses find administering some SDD components (e.g. oral pastes) unpleasant and/or burdensome.

. . . some of those oral based solutions are gooey, sticky mess that frankly nurses find a pain . . . the chlorhexidine products I think are nicer to use. So there’s a practical barrier around making sure that is not [difficult] to use.

T3

We pretty much used an oral paste as well as GI and the nurses hated it. We put it in several times a day and everyone hated it, paste around the mouth and in the gut. The nurses hated doing it. So to get it implemented, to get compliance is going to be hard . . .

T9

There were also perceptions of increased nursing and/or pharmacy workload associated with the preparation or delivery of SDD components.

. . . it’s a lot of nurse time with all the pastes and all of that . . . [named hospital] make up their own drug but I don’t think it’s generally like off the shelf available; you’ve got to make up the paste which is a lot of work.

T13

The challenges of maintaining intervention fidelity and preventing trial fatigue were also briefly discussed and ideas to overcome such issues were proposed.

In terms of the implementation and compliance from nursing staff, that is an education process . . . you’ve got to have your investigators and research nurses running an education programme for the bedside nurses and providing them with feedback and lunches and breakfast meetings where you tell them all about the trial and how it is going.

T6

Findings based on notes from the three interviews that were not transcribed

The interviewer’s notes from three triallist interviews for which recording failed and hence were not transcribed are summarised below.

Summary of results

In summary, triallists were interviewed from across a wide spectrum both geographically and from their area of specialist expertise. The majority spent most time discussing an effectiveness trial. This was the case for the intensive care clinical triallists, while those with implementation trial expertise discussed both types of trial. The hierarchy outlined in Figure 17 holds as a framework for summarising the responses of all the triallists.

FIGURE 17.

Summary of hierarchy challenges of conducting further SDD randomised trials, as reported by triallists.

Major challenges in future studies

There were differing views from triallists on the need for further research on SDD. Not all the triallists were convinced that further research was justified or feasible although the majority felt that it was warranted and important. The challenges identified by triallists are summarised in Figure 17. There was no consensus amongst the triallists on approaches to dealing with the challenges presented in this hierarchy.

Triallists from both an effectiveness and an implementation background highlighted the complexities of further research. The triallists focused on approaches to an effectiveness study and few views emerged on the appropriateness of, or approaches to, an implementation study. This may be partly due to the lack of clarity about the core components of an implementation trial.

All triallists proposed that an effectiveness trial would be large and demanding and that this could well lead to burnout for the chief investigator and team. They also identified the need for multidisciplinary support to conduct and complete an effectiveness study. All agreed with the requirement for any study to be multinational, although there was not clarity as to which countries could, or should, be included.

Triallists also identified key stakeholders such as clinical microbiologist and infectious disease clinicians as potential facilitators or barriers to the successful conduct and completion of any study. Nurses and pharmacists delivering SDD were suggested as potentially raising sustainability challenges owing to workload acceptability, although this was discussed more as anecdote than fact.

Study design issues

There was a range of views on the primary end points because of perceived problems with the size and cost of the study and the ability to identify funding. There was concern about using VAP as an end point owing to difficulties in establishing an agreed and standardised diagnosis. Many triallists identified antibiotic resistance as an important end point; however, most preferred this as a secondary end point. The nature of the control group was frequently discussed. Triallists acknowledged the widespread use of chlorhexidine and VAP bundles and there were varied views on whether or not there should be clear specification of the control group through to a more pragmatic control group without specification.

There was no consensus on the ideal unit of randomisation. The potential benefits of unit level randomisation were discussed with regard to both individual benefit and harm. The issue of unit of randomisation also brought into focus the question of patient consent. It was clear that different countries would approach this in different ways from an ethical perspective.

The impact of SDD on the local microbiological flora was widely discussed. Most saw this as a parallel issue to the primary end point, yet an important one, and most felt this should be measured to convince the community that resistance patterns were not adversely affected.

A editorial on SDD in 2012 also plays out the conflicts. 68 This editorial discussed study design, outcome measures, the impact of antibiotic resistance, the role of SOD and chlorhexidine and the place for a cost–benefit analysis amongst other issues. This article concluded that a cluster-randomised design should be the basis of future effectiveness studies with unambiguous end points such as mortality. It also identified that this would be a very large study with a requirement for detailed monitoring of resistance patterns and recognised the need to undertake the study in a range of settings with different microbiological ecology. 68 The current findings are thus consistent with recently published opinion.

Strengths and limitations

In drawing these findings together, one should of course distinguish between the participants’ views that are based on their experience and expertise as triallists and their personal views derived from other forms of evidence. Specifically, data from other stages of this study (e.g. whether or not nurses find it burdensome to deliver SDD) or from the published literature (e.g. techniques that are effective in changing professional behaviour) can inform some of the opinion-based evidence presented in this chapter. There was an interesting pattern of findings with respect to triallists’ beliefs about what will change behaviour (see Table 26). Although people may believe that more evidence will change clinical behaviour, the weight of evidence in the field of implementation research is that this not an effective approach. There is also a potential ethical question of randomising patients to a control group merely to persuade clinicians of what existing evidence already shows. Hence, a limitation of this study is that the participants may have gone beyond their level of expertise in presenting their own hypotheses about barriers to SDD delivery or about how best to change clinicians’ behaviour.

A further limitation of the study was that we did not succeed in engaging sufficient numbers of European triallists who have led SDD effectiveness trials. Their specific expertise would have been invaluable. Apart from this, the geographical spread of the sample across three continents and four health systems was a strength of this study.

Box 9 presents the key findings from the triallist interview study that inform the need for, and acceptability of, a clinical effectiveness RCT or an implementation RCT. The majority of triallists either accepted that further SDD effectiveness research should be carried out or were actively interested in participating. At least two were not convinced and a third felt that the scale of the venture was probably too great. However, this is set against the persisting issue of concern about resistance and the potential for antimicrobial resistance to lead to harm for an individual patient, even if the ICU population overall benefits. However, this study was designed to address the research question concerning challenges to conducting a trial, so we did not discuss non-trial designs in these interviews (e.g. studies of microbial resistance patterns). Hence, these findings should not be taken as an indication that a randomised trial is the only appropriate kind of study design for further SDD research.

Implications for health care

• There was a striking level of uncertainty about the effects of SDD on clinical outcomes that are regarded as important. This uncertainty suggests that there is considerable potential for improvement in prevention of HAIs in critically ill patients but further evidence is required to clarify the balance between potential individual-level benefits (e.g. infections, mortality) and potential society-level harms (e.g. antibiotic resistance) related to SDD.

• There was significant confusion apparent in clinicians’ understanding and perceptions of the components that constitute SDD and related interventions, e.g. SOD. The importance of detailed guidance on what constitutes different interventions was clear.

• For those units considering the adoption of SDD, it was apparent from our research that the delivery of SDD is feasible and can be adopted into unit practices. However, a detailed specification of the proposed clinical and behavioural components of the intervention should be developed.

• This study highlighted that the introduction of SDD, whether into routine practice or within a research context, requires consensus across a range of different stakeholders (including ICU colleagues, clinical microbiologists and medical directors/those with decisional authority within units). Our study also highlighted that medical microbiologists appear to be more opposed to SDD than intensivists. Representatives of these stakeholder groups should be engaged early in any discussions around the use/introduction of SDD.

• A substantial minority of participants reported that SDD would be adopted (apparently quite straightforwardly) if adoption was mandated by regulatory bodies.

• There was acknowledgement that the infections that SDD was developed to target (such as VAP) are currently less of a problem owing to other current prophylaxis and treatment regimens.

Recommendations for research

Further SDD research was viewed as important, acceptable and feasible to the key stakeholder groups who participated in this study. However, further effectiveness research would need to be on a scale that raises challenges for trial design and trial conduct. Research priorities, in rank order, are as follows:

1. A study within UK ICU is required to model resistance patterns as a function of SDD use.

2. Further large-scale effectiveness trials of SDD in intensive care practice are required to answer the remaining uncertainties, especially those issues relating to antimicrobial resistance.

3. There is a general willingness to participate in a future effectiveness RCT of SDD; however, support for further research is time-sensitive (owing to the changing context) and is not unanimous. Future research needs to address the substantial barriers to acceptance and participation in any future trial. These barriers should be addressed with reference to the study findings, for example (1) clinicians with lower self-assessed knowledge of the SDD evidence base shifted their opinions following feedback about others’ views, suggesting a role for discussion among clinical colleagues; (2) concerns about antibiotic resistance and other potential harms were of paramount importance, suggesting the importance of emphasising that a UK trial would assess antibiotic resistance patterns; (3) consensus between ICU colleagues was seen as important, suggesting that consensus building and development are key to acceptance and participation; and (4) a substantial proportion of clinicians would be prepared to participate in a trial of SDD if their colleagues were in favour, suggesting that the presence of a SDD champion in an ICU could influence participation.

4. Future trials should include (1) a primary mortality outcome, (2) pretrial, during-trial and post-trial monitoring of antimicrobial resistance, (3) a control group that includes chlorhexidine and/or VAP bundles, (4) a cost–benefit analysis, and (5) a qualitative study to investigate the fidelity of the SDD intervention as delivered.

5. Groups proposing to undertake such a trial need to overcome the following challenges: (1) gaining sufficient acceptance of the trial, (2) gaining adequate participation in a trial, (3) the clear specification of the trial intervention, (4) major methodological issues relating to trial design and conduct, (5) clarification of the acceptability (to ethics committees) of cluster-level consent in the case of a cluster RCT and (6) major funding issues.

6. At this time, there is a much lower level of interest in adoption of SDD, or studies designed to encourage implementation of SDD, into practice.