Development of a theory-informed implementation intervention to improve the triage, treatment and transfer of stroke patients in emergency departments using the Theoretical Domains Framework (TDF): the T3 Trial

Background Theoretical frameworks and models based on behaviour change theories are increasingly used in the development of implementation interventions. Development of an implementation intervention is often based on the available evidence base and practical issues, i.e. feasibility and acceptability. The aim of this study was to describe the development of an implementation intervention for the T3 Trial (Triage, Treatment and Transfer of patients with stroke in emergency departments (EDs)) using theory to recommend behaviour change techniques (BCTs) and drawing on the research evidence base and practical issues of feasibility and acceptability. Methods A stepped method for developing complex interventions based on theory, evidence and practical issues was adapted using the following steps: (1) Who needs to do what, differently? (2) Using a theoretical framework, which barriers and enablers need to be addressed? (3) Which intervention components (behaviour change techniques and mode(s) of delivery) could overcome the modifiable barriers and enhance the enablers? A researcher panel was convened to review the list of BCTs recommended for use and to identify the most feasible and acceptable techniques to adopt. Results Seventy-six barriers were reported by hospital staff who attended the workshops (step 1: thirteen TDF domains likely to influence the implementation of the T3 Trial clinical intervention were identified by the researchers; step 2: the researcher panellists then selected one third of the BCTs recommended for use as appropriate for the clinical context of the ED and, using the enabler workshop data, devised enabling strategies for each of the selected BCTs; and step 3: the final implementation intervention consisted of 27 BCTs). Conclusions The TDF was successfully applied in all steps of developing an implementation intervention for the T3 Trial clinical intervention. The use of researcher panel opinion was an essential part of the BCT selection process to incorporate both research evidence and expert judgment. It is recommended that this stepped approach (theory, evidence and practical issues of feasibility and acceptability) is used to develop highly reportable implementation interventions. The classifying of BCTs using recognised implementation intervention components will facilitate generalisability and sharing across different conditions and clinical settings. Electronic supplementary material The online version of this article (doi:10.1186/s13012-017-0616-6) contains supplementary material, which is available to authorized users.


(Continued from previous page)
Conclusions: The TDF was successfully applied in all steps of developing an implementation intervention for the T 3 Trial clinical intervention. The use of researcher panel opinion was an essential part of the BCT selection process to incorporate both research evidence and expert judgment. It is recommended that this stepped approach (theory, evidence and practical issues of feasibility and acceptability) is used to develop highly reportable implementation interventions. The classifying of BCTs using recognised implementation intervention components will facilitate generalisability and sharing across different conditions and clinical settings.
Keywords: Implementation intervention, Theoretical Domains Framework, Behaviour change techniques Background Evidence-based guideline recommendations are available for the early management of patients with acute stroke. Early diagnosis of stroke in emergency departments (EDs); administration of recombinant tissue plasminogen activator (rt-PA) and endovascular clot retrieval to eligible patients; and management of fever, hyperglycaemia and swallowing difficulties before transfer to a stroke unit are essential elements of evidence-based stroke care and recommended in current clinical guidelines [1]. Yet, inappropriate triage [2] and delays in diagnosis, treatment and transfer of stroke patients from ED to stroke units still occur [1]. The T 3 Trial is a prospective, multicentre, parallel group, blinded, cluster randomised trial that aimed to evaluate the effectiveness of an implementation intervention to improve the triage, treatment and transfer of stroke patients from ED to stroke units on 90-day outcomes and in-hospital processes of care [3]. This paper describes the development of the theorybased implementation intervention for this Trial.
The use of theory in the intervention development process has been identified by the UK Medical Research Council (MRC) as crucial to increase intervention effectiveness by targeting causal determinants of behaviour and facilitate an understanding of what works (i.e. the mechanisms of change) [4,5]. Several approaches have been proposed that integrate the use of theory in implementation intervention development [5][6][7]. Although there are studies that apply these approaches in the process of developing an implementation intervention [8][9][10], frequently, these interventions are still based on intuitive or non-theoretical methods [11]. There is also a lack of detailed reporting of the process of intervention development and the content of the implementation intervention which, if available, would assist replication and advance the knowledge base about the optimum approach for intervention development [12].
The Theoretical Domains Framework (TDF) is a framework of 14 theoretical domains derived from 33 behaviour change theories developed using a process of expert consensus with subsequent validation work [13,14]. The TDF has successfully been applied in a number of healthcare settings to (i) guide intervention development for the implementation of guidelines or clinical interventions [6,9,15], (ii) characterise, according to theory, an existing intervention to implement evidence-based care to facilitate accurate replication [16,17], and (iii) understand factors that may inhibit uptake of an intervention [18,19]. The additional benefit of the TDF is that behaviour change techniques (BCTs) have been preassigned to each of the TDF domains [20]. Two matrices which assign the most appropriate BCTs to each of the TDF domains have previously been developed by Cane et al. [20] and by Michie et al. [5].
The implementation of complex clinical interventions, such as those that have numerous intervention components, as is the case in the T 3 Trial, often involve the use of theory but may also require incorporation of the evidence base and consideration of practical issues such as feasibility and acceptability [21]. Firstly, theory is important to understand the factors influencing clinician behaviours and to guide the use of appropriate behavioural change techniques (BCT), the smallest components of an implementation intervention [6]. Secondly, evidence regarding technique effectiveness can assist the selection of BCTs and the best mode of delivery [6]. This might be generic behaviour change evidence but also might incorporate context-specific evidence, from the stroke or ED literature in the case of this study. Thirdly, an understanding of practical issues (feasibility and acceptability) and expert clinical judgment can guide the selection of the most relevant BCTs for a particular context [6]. Some studies have incorporated stakeholder opinion in the design of implementation interventions to incorporate practical considerations and judgment [6,9,10,22]. As contextual issues have a significant influence on the delivery and impact of complex clinical interventions [4], a theory-based evidence-driven approach which takes into account context should be considered in developing implementation interventions. However, there are very few well reported studies that use this stepped approach of intervention development (theory, evidence and practical issues of feasibility and acceptability).
The aim of this study was to describe the development of an implementation intervention (i) using theory to inform selection of BCTs for the T 3 Trial, (ii) further guided by the evidence of effectiveness of implementation interventions (including that from a previous acute stroke implementation trial [23]) and (iii) consideration of researcher opinion to select appropriate BCTs. This implementation intervention will be subsequently tested under trial conditions [3].

Methods
The T 3 Trial clinical intervention is an evidence-based care bundle of clinical protocols for triage, treatment and transfer of patients following acute stroke and comprised of 12 different clinical care elements (hereonin referred to as 'target behaviours' [ Table 1]) [3]. As the T 3 trial clinical intervention consisted of 'a number of separate elements which seem essential to the proper functioning of the intervention although the 'active ingredient' of the intervention that is effective is difficult to specify' , it meets the MRC definition of a complex intervention [4]. Four of the five-stepped method proposed by French et al. [6] for developing complex interventions based on theory, evidence and practical issues were undertaken as follows: Step 1: Who needs to do what, differently?
Step 2: Using a theoretical framework, which barriers and enablers need to be addressed?
Step 3: Which intervention components (behaviour change techniques and mode(s) of delivery) could overcome the modifiable barriers and enhance the enablers?
Step 4: 'How can behaviour change be measured and understood?' previously has been reported in our published protocol paper (primary and secondary outcomes with an a priori planned process evaluation) [3].
Step 5: 'How can behaviour change be sustained' is beyond the scope of the T 3 Trial.
Step 1: Who needs to do what, differently?
Twelve evidence-based targeted behaviours were identified by the trial investigators for the triage, treatment and transfer (T 3 ) elements of the intervention. We selected the target clinical behaviours to be addressed, based on documented evidence-practice gaps. As per French et al.'s approach we specified the target behaviours in detail by asking the following questions: What is the clinical behaviour that you will try to change? Who performs the behaviour(s)? And when and where do they perform the behaviour(s)?
Step 2: Using a theoretical framework, which barriers and enablers need to be addressed?
One barrier and enabler multidisciplinary workshop (1-h duration) was conducted at each of the thirteen T 3 Trial intervention hospitals across three Australian states and the Australian Capital Territory between October 2014 and December 2014. Purposive sampling was used to select workshop participants who could provide detailed feedback on barriers and enablers to the T 3 Trial clinical intervention, namely (i) senior healthcare professionals working in ED (e.g. emergency physician, emergency nurses) or in stroke units (e.g. stroke physician, stroke nurses, endocrinologists, speech pathologists and bed managers) and (ii) involved in routine delivery of the target behaviours. The workshops were aimed at identifying the perceived barriers and enablers that may influence the uptake of each of the target behaviours. The workshops were facilitated by SD and SM with assistance from emergency, neurology and endocrine physician T 3 Trial researchers. A standard presentation was given at the workshops to provide consistent information about each of the target behaviours. The workshop participants were asked to nominate specific barriers for each of the behaviours and specific enablers and strategies that could be used to overcome the barriers. The workshops were audio recorded and transcribed verbatim. The interview transcripts were coded using thematic analysis by a single coder (LC) according to the TDF domains [14]. Individual barriers were classified to the relevant domain of the TDF. The constructs, that is the concepts provided for each of the TDF domains, were used to assist interpretation and to ensure accurate assignment of the TDF domains. The coding framework was devised by the lead author (LC) and RP applied this framework to a subset of transcripts (n = 5) to test the interpretation of the codes. A third researcher (NT), with expertise in the application of the TDF to primary data, independently checked the assignment of all transcript data to the TDF domains. It was agreed that should a number of barriers be reflected by more than one TDF domain, the most relevant domain should be selected. Discrepancies in allocation were resolved by reviewing and discussing the transcript dialogue before and after the extract. The TDF constructs and contextual information reported for an individual barrier/enabler were also used to understand and resolve any discrepancies.
Step 3: Which intervention components (behaviour change techniques and mode(s) of delivery) could overcome the modifiable barriers and enhance the enablers?
The Cane et al. matrix [20] which recommended the most appropriate BCTs to each of the TDF domains was primarily used to identify BCTs. Two domains (memory, attention and decision processes and social/professional role and identity) do not have any specified BCTs in the Cane et al. matrix; therefore, a similar matrix previously developed by Michie et al. was used to inform the BCTs for these domains and other domains where the recommended BCTs were considered more appropriate than that recommended by the Cane et al. matrix. The BCT taxonomy [24] was also provided to the researchers as a resource to use where neither of the two matrices identified an appropriate BCT. Whilst these tools are useful for assigning relevant BCTs, they do not incorporate the evidence regarding implementation intervention effectiveness or issues of feasibility or acceptability. To address this, we used a pragmatic approach to selecting BCTs using the T 3 Trial investigators knowledge of the clinical intervention and experience of the clinical context. A panel of five T 3 Trial investigators and researchers (SM, DC, RG [a stroke physician], RP and ES), all of which had experience of applying the TDF in stroke implementation research, independently reviewed the list of matrixassigned BCTs and from this identified the most feasible and acceptable techniques for the T 3 Trial.
It was identified that the selection of BCTs for a single behaviour would be time consuming likely taking up to two and half hours to complete; therefore, due to time constraints of the researcher panellists, it was considered unfeasible for them to complete the selection process for all 12 behaviours. Instead, the panel were instructed to select BCTs to address the barriers associated with only one of the target behaviours Administration of insulin to all patients with BGL > 10 mMol/L within 1 hour by insulin infusion. This behaviour was selected as it represented nearly every TDF domain (11 out of 13) mapped in step 1 which would allow the findings to then be applied to the remaining behaviours. The panel were provided with a number of resources (shown in brackets below) and specifically asked to  [20] were viewed as appropriate; resources 2 and 3 4. Identify further BCTs not selected by the primary Cane et al. matrix [20] e.g. techniques identified in other BCT matrices and taxonomies [5,24] that were considered to be more appropriate; resources 2 and 3 5. Devise strategies to operationalise the selected BCTs; resource 5: table of evidence [9] to present different modes of delivery, e.g. face-to-face education meetings and local opinion leaders, based on Cochrane Effective Practice and Organisation of Care [EPOC] reviews (Additional file 5) [25][26][27][28][29][30] In summary, each of the researchers were asked to select the most appropriate BCT based on the following criteria; feasibility of use in the ED context, personal experience of use, local relevance and acceptability. An evidence table providing effectiveness data for commonly used modes of delivery such as face-to-face education meetings and local opinion leaders was included to assist the researchers in suggesting strategies to operationalize the BCTs in an ED context. Relevant qualitative data extracts generated from the barrier and facilitator workshops were also included to allow the researchers to assess the feasibility of using the technique in ED (further details can be provided on request). An overview of this process is provided in Fig. 1. As the TDF domains mapped in step 2 were also represented in the remaining 11 behaviours, the researchers were advised that their selections would inform the final set of BCTs to be applied across all the behaviours. The researcher completed the task independently with the lead author (LC) available to guide the researchers through the process and respond to any questions to ensure adherence to the instructions provided. A BCT was included in the final set if it was selected by at least one of the researchers. The BCTs were tabulated and the frequency of selection by the researchers was reported. The selections of the researcher panel were applied by LC to the remaining 11 behaviours to create a final set of BCTs for the T 3 Trial implementation intervention. In order to increase the transferability of the reporting of the implementation intervention, the final set of BCTs were classified by LC according to general evidence-based intervention components (BCTs and mode of delivery) commonly reported in the implementation literature [23]. These were as follows: multidisciplinary barrier and enabler workshop [31], interactive and didactic education programme [32,33], use of opinion leaders [25], reminders [34] and site support [34]. Key relevant literature examples which presented BCTs by intervention components were used to classify accordingly [6,9].

Results
Step 1: Who needs to do what, differently?
The target behaviours for the T 3 trial intervention arose from an extensive literature review and the stroke guidelines. These targeted behaviours along with who performs the behaviour, the timepoint and the location that the behaviour occurs are listed in Table 1. These target behaviours were chosen because they had supporting evidence and were potentially modifiable at a clinician level.
Step 2: Thirteen workshops were conducted with 105 staff from 13 hospitals. Workshop group size ranged from minimum of five participants to maximum of 11 participants. These multidisciplinary workshops were facilitated by the researchers and included senior nurses from ED and stroke units, medical practitioners (ED physicians, neurologists, endocrinologists and their junior doctors), Fig. 1 Selecting behavioural change techniques and strategies to inform the T 3 Trial implementation intervention speech pathologists, and nurse managers. Seventy-six barriers were identified by participants ( Table 2). All barriers were mapped to at least one TDF domain. For example, barriers relating to the knowledge domain were associated with gaps in staff's understanding, awareness or content knowledge for certain target behaviours. This resulted in clinical uncertainty and the conduct of practices not compliant with national guidelines. Barriers relevant to the skills domain were related to the conduct of a specific task such as a swallow screen or a lack of skill development opportunity. Barriers assigned to the social/professional role and identity domain were related to professional boundaries, i.e. limited prescribing rights for nurses and professional identity, i.e. opposition to blurring of roles. Barriers relevant to the beliefs about capabilities domain were related to professional confidence, i.e. decision-making or low self-esteem to perform tasks related to the target behaviour. The main area of disagreement between the researchers conducting the mapping related to overlap between two domains, beliefs about capabilities and social/professional role and identity. For example, one researcher understood a barrier to conducting a clinical task to be related to a clinician's self-confidence so mapped the extract to beliefs about capabilities domain whilst the other researcher understood the barrier to be related to professional boundaries, i.e. the task was not traditionally undertaken by that professional group so mapped the extract to social/professional role and identity. Areas of disagreement were resolved by discussion between the two researchers. No barriers were assigned to the intentions domain. Furthermore, the same barrier was often reported for more than one of the behaviours, for example, competing priorities in a busy emergency department environment was reported for four different behaviours. Overall, the same nine barriers were reported for the different target behaviours.
Step 3: Which intervention components (behaviour change techniques and mode(s) of delivery) could overcome the modifiable barriers and enhance the enablers?
The panel selected appropriate BCTs for the 11 barriers identified for the target behaviour provided (Table 3). There was no TDF domain where the assigned BCTs based on the Cane et al. matrix [20] were viewed as inappropriate by the researchers. Overall, 22 of the selected BCTs for all 11 barriers were based on the Cane et al. matrix [20], with the remaining five selected BCTs based on the Michie et al. matrix [5].

Generating a final set of BCTs
The findings from the researcher panel were then applied to the remaining 11 behaviours to generate a final set of BCTs to apply to all the behaviours. Two TDF domains (beliefs about capabilities and reinforcement) were not represented by any of the 11 barriers used in the researcher panel; therefore, selection of the most appropriate BCTs was based on the triallists experience [6]. The final set of BCTs (n = 27) are reported by general implementation intervention components in Table 4. Some techniques, e.g. action planning and coping skills, were classified into more than one relevant implementation intervention component.

Discussion
The barriers that hospital staff believed to be likely to influence the implementation of the T 3 Trial clinical protocol were mapped to 13 TDF domains. These domains were used to guide the content of an implementation intervention consisting of 27 BCTs. It is anticipated that by using this systematic, theory-based approach to inform the content of an implementation intervention the potential for effectiveness in changing behaviour will be optimised. Researcher opinion, together with the existing research evidence base, was used to refine the theoretically driven intervention framework by selecting appropriate BCTs and suggesting strategies to operationalise the BCTs in an ED context. One of the few studies that used the TDF and BCTs to define the content of an intervention to change patient's compliance in bronchiectasis used an adapted scoring system from Michie et al. to select BCTs from a longer list [22]. Experts classified BCTs into one of the following categories which then generated the scores: agreed use, agreed non-use, disagreement and uncertain. The validity of using such criteria in the BCT selection process was not reported by the authors, indicating that further research is required in this area. The use of stakeholder opinion has been used in another study to define how the proposed intervention could be delivered as opposed to selecting the BCTs [22]. This process of incorporating the views of researchers with experience of the clinical context and knowledge of the clinical intervention is likely to enhance the clinical acceptability of the implementation intervention. The BCTs selected by the researchers aligned with the BCTs recommended by the matrix, suggesting that this is a valuable tool to use to highlight relevant BCTs.

Limitations
The final set of BCTs was generated using the researcher panellists' selections based on one target behaviour. This assumes that the BCTs considered appropriate for one behaviour have applicability to target similar barriers associated with the other behaviours. Potentially, appropriate BCTs to target different barriers for the other behaviours may not have been included. However, the       It was considered resource-intensive to apply this selection process for all 12 behaviours in this complex intervention. Acknowledging this limitation, this approach should be considered valid and reliable as a theoretical framework was used to develop the implementation intervention and nearly all the relevant TDF domains (11 out of 13) common to the 12 behaviours were represented by the behaviour used as an example in the researcher panel.
The considerable time and resources required in the intervention development process have been raised elsewhere [6] and have implications for applying for research funding, whereby funding bodies often expect the intervention to be fully developed on application, which is usually not possible without separate pre-trial funding, usually difficult to secure.

Strengths
The barrier and enabler workshops allowed comprehensive identification of relevant barriers and enablers by hospital staff, providing data for each of the T 3 Trial behaviours. This ensured that all the individual elements of the T 3 Trial clinical intervention were addressed, a necessity for complex interventions. Using the TDF framework enabled the classification of barriers to BCTs required for each T 3 Trial behaviour. The TDF is readily being used by researchers to explore behaviour change but few describe the process of how to use theoretical frameworks when developing implementation interventions. Describing the process in steps ensured transparency and replicability of the method that could be used to develop similar implementation interventions for stroke treatments or guideline implementations across different conditions. The use of researchers incorporated well-informed judgment, acknowledged to be an important part of the process, especially for complex interventions [35]. The enabler data provided the researchers with valuable context-specific information to assist in the decision-making process. Reporting the BCT's using a standard taxonomy for by each T 3 Trial behaviour has produced a well-described, reproducible and testable implementation intervention.

Further research
This study used a pragmatic approach to selecting BCTs that involved the T 3 investigators who were familiar with the clinical protocols and clinical context. However, there is little evidence on who is best placed to make these judgments and what methods should be used. Michie et al. has developed the ' APEASE' criteria (A: Affordability; P: Practicability; E: Effectiveness/cost-effectiveness; A: Acceptability; S: Side-effects/Safety; E: Equality). However, it is unlikely that this information will be available for all the BCTs [7]. The T 3 Trial is one of the few studies that have reported BCTs that were considered inappropriate for use (Table 3). Knowing more about the BCTs that are inappropriate for certain clinical contexts or even healthcare settings generally would be useful to make the overall pool of BCTs to choose from more relevant and straightforward. The Cane et al. matrix [20] is limited in suggesting BCTs for  [20] two of the TDF domains (social/professional role and identity; memory, attention and decision process). Therefore, further work should be conducted to assign more BCTs to these two domains. We acknowledge that research is currently underway which aims to link BCTs with theoretical mechanisms to better understand how interventions exert their effect and how to apply theory in implementation intervention development [36]. It is anticipated the approach used in this paper to develop and report an implementation intervention may contribute to the advancement of use of theory to guide intervention development. The TDF was a useful tool to map barriers to theoretical domains; however, as highlighted in this paper, inconsistencies between researchers can arise. Therefore, further work to clarify TDF domain definitions and to provide example barriers for each domain would complement this mapping process and minimise disagreements.

Conclusion
The TDF was successfully applied in all steps of developing an implementation intervention for the T 3 Trial clinical protocols. The use of researcher opinion was valuable for the BCT selection process in terms of incorporating research evidence and well-informed judgment and incorporating the important practical issues of feasibility and acceptability. However, further recommendations are needed to advance understanding of who is best placed to inform implementation intervention development, and how best to incorporate this well-informed judgment. There is also a need to devise criteria for use in this BCT selection process. It is recommended that BCTs are classified by recognised implementation intervention components to facilitate generalisability and sharing across different conditions and settings. Technique may be classified to more than one implementation intervention component, e.g. action planning