‘Around the edges’: using behaviour change techniques to characterise a multilevel implementation strategy for a fall prevention programme

Documents describing the fall risk assessment clinics or their implementation were provided by the implementation steering group. This multidisciplinary group of health professionals and health service managers were responsible for planning, managing and overseeing implementation of clinics. We also collected documents relating to the broader fall prevention pathway, promotional material, referral forms and documents used by staff during the fall risk assessment clinic (e.g. risk assessment tool) (see Additional file 3 for an outline of the documents included).

Semi-structured face-to-face interviews were conducted with a purposive sample from the 12-member implementation steering group. We followed the principles of critical case sampling to select participants who would yield the most information about day-to-day implementation [23]. As our aim was to examine whether the content of an implementation strategy could be characterised using the BCT Taxonomy as opposed to evaluating its acceptability among the target population or its impact on behaviour change, we did not include interviews with staff and potential referrers in this analysis. Potential participants were invited via email, telephone or in person and were provided with an information sheet outlining the study objectives. As the research lead on the steering group, the lead researcher (SMH) was known to participants.

During interviews, participants were asked to describe what they did to support the set up and running of the fall risk assessment clinics. The interview topic guide was informed by TIDieR reporting tool for complex interventions [15] and included probes on the implementation processes, activities, their purpose and timing. Prompts were added to the topic guide, informed by concurrent observation and document analysis. Interviews were digitally recorded and transcribed verbatim (ER).

One researcher (SMH) conducted unstructured observations of training sessions with multidisciplinary teams (MDTs) delivering fall risk assessment clinics and meetings with health professionals eligible to refer to the clinics [24]. As the timing of sessions was determined by those delivering and receiving the training a convenience sample of interactions was selected. Observations took place at two primary care sites where training and the risk assessment clinics were being delivered (April–June 2016). The researcher, who previously conducted interviews with team members, adopted a ‘complete observer’ role [25], attending with the explicit purpose of collecting data and did not participate in the sessions.

Observation notes were taken to record the implementers’ actions and health professionals’ behaviours and responses in context. Notes were written up in full at the end of each period of observation.

All data sources (documents, interview transcripts and observation notes) were imported into NVivo software to support data management and analysis. Interviews and observation notes were anonymised prior to analysis and reporting.

Data analysis was guided by the principles of directed content analysis, a deductive approach in which coding can begin with predetermined codes [26]. Initially, data were coded deductively (by SMH, trained in health services research) using the TIDieR framework which helped to clarify the underlying rationale for the implementation strategy [15]. Where possible, we also coded the population and behaviour being targeted [21]. Data were coded to identify what was carried out, at what level (organisational, professional, patient), by whom, how, where, when and how much and any planned tailoring or modification during the study period.

Data were then coded by one researcher (SMH) using the BCT Taxonomy (version 1) of 93 techniques which are organised into 16 categories. First, data were coded to label potential BCTs and these segments of text were checked carefully against BCT definitions to determine whether a BCT was present or absent [16]. Once this coding was complete, each BCT definition in the taxonomy was reviewed and the researcher (SMH) assessed all three data sources for presence/absence of this BCT. Implementation activities that did not fit with a BCT taxonomy description were retained and coded using in vivo codes (words and phrases used by participants in the data).

A selection of the data was coded by two other researchers, one familiar with the service through a wider evaluation (ER—trained in public policy) and one who was independent of the project (CS—primary care physician and clinical researcher). Each researcher coded a document, field notes from one observation and an interview transcript deemed (by SMH) to include the richest description of the implementation strategy. All coders had participated in online BCT training and had experience coding qualitative data. Coding discrepancies were listed (by SMH) and reviewed by the coding team (CS, ER). Researchers outlined their rationale for coding a BCT, and the group reviewed the clarity and volume of data available to support its presence. An accumulation of segments of text led to greater certainty about the presence or absence of a BCT rather than a single line of text, so when an interview transcript was taken in its entirety, researchers tended to code the same BCTs. In a small number of cases where researchers identified different BCTs, this was typically due to different interpretations of the target or purpose of the implementation activity. These discrepancies were reviewed together and resolved by discussing our individual coding rationale, consulting other sources of data and drawing on the lead researcher’s knowledge of the implementation context. Outstanding discrepancies and uncertainties were reviewed by a behavioural scientist with experience in the use of BCTs (MB). Any resultant changes to coding were applied across the data.

BCTs were mapped to intervention functions, guided by data on the rationale and purpose of the implementation strategy and links between functions and BCTs outlined in the Behaviour Change Wheel Guide to Designing Interventions [27].

Sampling, data collection and analysis were iterative. As initial data were analysed, participants were recruited using the same sampling and contact methods described above and additional documents were sought to elaborate on specific aspects of the implementation strategy. We purposively chose a sample of critical cases, interactions and documents that provided the best opportunity to gather sufficient depth of information to fully describe the implementation strategy being studied [28]. The stages of analysis and codes were revisited multiple times. We assessed construct saturation, whether the conceptual categories of the BCT Taxonomy and intervention functions were adequately populated, based on the analysis of all data sources in consultation with the definitions in the BCT Taxonomy.

We made a number of assumptions during analysis. These assumptions were adapted from the approach used by Presseau et al. to code multilevel implementation interventions [29]. First, we focused on techniques used to support or enable the implementation of the fall risk assessment clinics (implementation strategy), as distinct from techniques used during the risk assessment itself (clinical intervention).

Second, in line with the premise of the Behaviour Change Wheel approach, we assumed that the implementation of an intervention depends on behaviour change at some point [12]. We assumed that all BCTs (included those at the system level) functioned by targeting the behaviour of health service managers, HCPs (referring to and/or running clinics) and patients, to ultimately influence implementation outcomes.

Third, there were a range of target behaviours which were broadly defined as ‘implementation behaviours’. The specific behaviour varied depending on the actor and level of the strategy being considered. In some cases, the target behaviour was clear and thus BCTs were coded at that level (e.g. referring clients to the clinic). Where the specific target behaviour was not clear, BCTs were coded at a more general level of ‘supporting the implementation of the fall risk assessment clinics’.

Fourth, we considered a BCT to be present based on an accumulation of text segments across a single source of data or multiple sources (document, interview and/or observation field notes). If it was unclear whether a BCT was present or absent, following discussion and consideration of multiple data sources, it was not coded. We did not apply a minimum frequency of occurrence of BCTs; techniques that occurred once were coded. Fifth, planning and initial implementation occurred in tandem with data collection. Therefore, some BCTs were planned while others had taken place at the time of data collection.

The Standards for Reporting Qualitative Research (SRQR) was used to guide the reporting of findings (see Additional file 4) [30].

Table 1 presents the main components of the implementation strategy, stratified by target level (i.e. organisational, professional or patient level) [15]. In summary, at an organisational level, the implementation strategy involved establishing a multidisciplinary steering group led by a clinical project manager, appointing an implementation coordinator and administrator and assembling multidisciplinary teams to deliver clinics. Multiple behaviours were targeted simultaneously at this level making it difficult to identify a specific behavioural target; thus, the target behaviour was defined broadly as ‘supporting implementation of fall risk assessment clinics’.

Most of the implementation strategy targeted either (1) the behaviour of professionals in the multi-disciplinary team responsible for ‘delivering the fall risk assessment clinic’ or (2) the behaviour of professionals ‘referring to the clinic’ (including primary care physicians, public health nurses, physiotherapists, occupational therapists and advanced nurse practitioners in emergency departments). At this level, the strategy focused on training and ongoing ‘coaching’ for multidisciplinary teams, and standardised referral documentation and promotional meetings with referrers. At the patient level, a standardised invitation letter and information leaflet were developed to encourage referred patients to ‘attend the clinic’.

The implementation strategy was tailored depending on existing resources at each site, and communication was ‘tailored to the requirements of different audiences’. Other components were modified during initial implementation in response to the needs identified at each site, for example the number of training sessions varied according to the health professional and his/her experience.

Overall, 21 out a possible 93 BCTs, from 14 of the 16 categories in the taxonomy, were identified (indicated in italics). Table 2 illustrates the BCTs used at each level of the implementation strategy (Additional file 5 describes each BCT identified within each component, with supporting data). The most commonly used BCTs were instruction on how to perform the behaviour and adding objects to the environment.

At the organisational level, we found many implementation activities could not be coded using any of the BCTs and that they appeared to target attitudes rather than immediate behaviour change. First, there were activities, described as ‘the hustle’, to generate buy-in and foster goodwill among key stakeholders. These activities appeared to implicitly target stakeholders’ intentions to support clinics in the future for example by releasing staff and space to run clinics (target: heads of discipline) or referring clients (target: health professionals eligible to refer). Selling the clinics served the function of educating health professionals about their ‘existence’ and purpose, but also sought to persuade potential referrers about the benefits and ease of adopting the service.

‘[we tell them] you can’t do without [these clinics] but it’s no trouble [to refer]’ (interview)

Second, piloting activities, which were evaluative and reactive and did not align with goals and planning BCTs such as reviewing behaviour and outcome goals, were used at the first implementation site.

‘What changed was the onward referral form and the process of who we were referring on to, and what form they got and what level of information they got or needed. That changed really on a weekly basis in clinic 1.’ (interview)

We examined the utility of the BCT Taxonomy to characterise the content of a real-world implementation strategy targeting a range of stakeholders across organisational boundaries. Overall, 21 distinct BCTs were used to restructure the environment for implementation and to train, educate, enable and persuade health professionals and patients. Of the nine possible intervention functions, none of the techniques sought to restrict behaviour or incentivise behaviour change. While we identified several techniques at the individual professional level, fewer techniques were identifiable at an organisational level. Furthermore, some implementation activities at this level did not target immediate behaviour change. The utility and application of implementation strategies could be improved by adopting a systematic approach using a theory-based framework which identifies active ingredients and offers the potential to understand mechanisms of change [13, 29, 31]. However, to accurately describe and fully understand a multilevel implementation strategy, the behavioural approach may need to be integrated with organisational theories of change.

Like Steinmo et al., we found that this systematic approach could be applied ‘post hoc’ to an existing implementation strategy that was developed pragmatically based on organisational knowledge and experience. Building on their analysis of an implementation strategy targeting professional behaviour in the hospital setting, our results show that the BCT Taxonomy is also applicable to wider implementation strategies which target multiple groups across different healthcare settings. Similar to Presseau et al., who examined randomised controlled trials of diabetes implementation interventions, we found that a small number of the possible BCTs (21/93) were used to support the introduction of a fall prevention programme. The main techniques used to target professional behaviour, including instruction on how to perform behaviour, adding objects to the environment, prompts and cues and social support, were similar across all three studies. The popularity of certain BCTs and the narrow range being used in implementation strategies highlights the potential for practitioners and researchers to think creatively about implementation by drawing on other techniques listed in the BCT Taxonomy [29].

In this study, fewer BCTs were identifiable at the organisational level compared to the professional level. Techniques at the organisational level tended to be broadly defined BCTs, such as restructuring the social environment, which were not necessarily specific to the clinical context of fall prevention. Our findings highlight some of the more specific techniques, such as action planning and problem solving, used to operationalise this broad BCT at an organisational level. Rather than simply listing the BCT ‘ingredients’ in an implementation strategy, we believe that techniques should be linked to specific components and levels of delivery in the strategy’s description.

Some activities did not have an immediate behavioural target, highlighting the importance of measuring proximal outcomes (such as attitudes and beliefs) as well as distal outcomes (like changes in behaviour) when evaluating implementation strategies. Although the ultimate goal of implementation may be behaviour change, immediate goals include changing organisational determinants of behaviour such as policy and procedures, staffing and resource allocation [32‐34]. It could be argued that the recruitment of dedicated staff to coordinate implementation, an example of restructuring the social environment and a resource which enabled most components of the implementation strategy in this study, is targeting organisational as well as individual determinants of change. Research on implementation climate suggests that when organisations align their policies, practices and resources to support implementation, staff have a clearer understanding of the priorities and value placed on an intervention, and thus are more likely to change their behaviour accordingly [35].

The systematic process of characterising and reporting an implementation strategy may prompt those designing and delivering strategies to think about the intended target(s) and expected outcomes of their activities. Outside of research studies, the process of specifying existing implementation strategies using approaches such as the BCT Taxonomy may be difficult to navigate in practice without support, given the number and granularity of techniques, and the specific terminology. An overarching framework, such as the AIMD (Aims, Ingredients, Mechanism, Delivery), may be a useful and efficient starting point for health care professionals, funders, managers and policy makers seeking to formally describe interventions. This simplified framework is deliberately ‘terminology agnostic’ to promote collaboration across related fields including quality improvement, policy, public health, patient safety and behaviour change [32]. As a meta-framework, it accommodates the use of specific frameworks such as BCT Taxonomy for particular purposes.

Characterising an existing strategy using a well-established framework provides an in-depth understanding of ‘implementation as usual’ in the health system and highlights gaps that can be addressed to refine implementation and maximise effectiveness. For example, in this study, few behaviour change techniques were used to target older peoples’ attendance at the fall prevention programme. Client engagement is a well-established problem for fall prevention programmes; a systematic review of uptake rates in clinical trials of fall prevention interventions reported non-attendance rates as high as 42% [36]. The implementation strategy in this study did not include any monitoring and feedback techniques to target health professionals referring to the service or those health professionals delivering the service. Yet data on referrals and assessment outcomes were collected and analysed to facilitate self-monitoring and action planning among the implementation steering group. Given the availability of valid routine data with mechanisms for audit and feedback from trusted sources within the health service and the substantial evidence of effectiveness for changing professional behaviour [37], there is potential to trial different approaches to and components of audit and feedback within this ongoing organisational initiative [38].

This study identified the behaviour change techniques planned or used during the initial implementation of fall risk assessment clinics in the health service. This is likely only a sample of the techniques that will be used over the course of implementation. Indeed, from the outset, the number and type of behaviour change techniques varied depending on the target group and context; different techniques were used to target referral behaviour among primary care physicians, advanced nurse practitioners in hospitals and public health nurses working in the community. It is widely acknowledged that implementation strategies should be tailored to the local context [39]. Further research is being conducted with the health professionals and clients to explore their experiences of implementation and perceptions of the acceptability and appropriateness of the implementation strategy. We will combine these results with our current understanding of the content and function of the implementation strategy to address gaps or barriers to its success.

The TIDieR framework helped us to clarify the underlying rationale for the implementation strategy and thus the intended target population for some activities [15]. We supplemented the TIDieR framework with a description of the ‘action targets’ [21], that is population and behaviour being targeted by activities, which was necessary before coding behaviour change techniques.

It was not feasible to observe every interaction, particularly at an organisational level, given the dynamic and often unpredictable nature of implementation and the number of actors and settings involved. Therefore, some BCTs at the organisational level may have been missed. While the number of interviews and observations in this study is small, the underlying sampling rationale was to generate a rich and thick description of the content of an implementation strategy. Thus, our sampling was not driven solely by numbers of interviews but also the appropriateness of the data. A purposively chose sample of participants, interactions and documents provided the best opportunity to gather sufficient depth of information to fully describe the implementation strategy being studied [28, 40]. Given our use of deductive content analysis with a pre-determined framework, we focused on construct saturation over the concept of theoretical saturation from grounded theory. Data collection and sampling was iterative; however, the process of judging saturation may have been facilitated and enhanced by the use of explicit a prior stopping criteria which have been used in theory-based interview studies [41]. The identification of BCTs was influenced by the richness of description and the specificity of BCT definitions. As mentioned previously, the definitions of some BCTs are relatively broad which could increase the ease and frequency with which they are coded. Interview participants did not naturally describe their work activities in the language of BCTs. Data from interviews were triangulated with data from documents and observations which provided reassurance about the presence or absence of a BCT [42]. BCTs were typically coded from an accumulation of evidence across data sources rather than relying on a single source.

In line with the Medical Research Council guidance for process evaluations of complex interventions, which warns against forcing an established framework to fit interventions which were not originally designed based on theory [43], we used participants’ own language to code and preserve aspects of the implementation strategy that were not adequately or accurately described using BCT definitions.

In this study, we did not evaluate the performance of behaviour change techniques and therefore cannot comment on the degree to which they were consistently delivered as intended or their impact on behaviour change. Capturing what is delivered in practice is essential for evaluators to identify adaptations that may undermine or enhance effectiveness [43]. Variability in the way professionals deliver an implementation strategy is to be expected as they try to respond appropriately to the needs of the target group and changing context. Hence, systemisation may not always be feasible [44].