‘Sepsis’ is common and deadly and is a major cause of death from infection through shock and multiple organ failure. It has a mortality rate of 36 % [1], which rises 8 % for every 1 h delay in treatment [2], and a worldwide incidence of 300 per 100,000 population [3]. It consumes 50 % of critical care resources in the UK [4]. Mortality can be halved if sepsis is treated within the hour of presentation by implementing the ‘Six Steps of Sepsis Treatment’, known as the ‘Sepsis Six’ clinical care bundle [5]. This consists of giving high flow oxygen, administering intravenous fluids and antibiotics, and measuring blood cultures, lactate levels and urine output [2, 6]. For every five patients treated with the Sepsis Six, one life is saved [4].

Implementing the care bundle and reducing the mortality of sepsis has become a UK and international standard and priority [7–9]. The UK Sepsis Trust has estimated that each patient treated this way results in a cost saving of £3600, and a 3.4-day reduction in length of stay (including two fewer days spent in intensive care). If the Sepsis Six bundle were implemented for 80 % of septic patients in the UK, this could save 10,000 lives and release £170 million of cost savings for the National Health Service (NHS) each year [9]. However, implementation in the UK is poor, with a national audit of practice in 4500 cases (160 hospitals) reporting median implementation rates of 27–47 % for individual components of the Sepsis Six [10].

Published reports of interventions to improve implementation of the Sepsis Six show small and un-sustained effects. One programme achieved implementation levels of 39 % [6] which fell to 23 % a year later [11]. In 2010, the board of an NHS hospital in London set a target of implementing the bundle for 95 % of septic patients and made this one of its three top quality improvement areas. They created a Sepsis Board and appointed a Patient Safety Facilitator to lead implementation and deliver the intervention in three pilot areas (Accident & Emergency, Renal and Medical Assessment). Once the 95 % target was reached, it was intended to extend the implementation of this evidence-based practice to the rest of the hospital.

Over a 4-year period, this multi-component Sepsis Six implementation intervention was designed through trial and error without the use of theory. Content was not fully reported, but broadly the intervention took the form of introductory group education and training, target setting, audit, group feedback of audit results, individual personalised feedback to staff involved in incidents when the bundle was not fully implemented, and environmental changes including promotional documents and materials to aid implementation such as a ‘sepsis trolley’ or ‘sepsis bag’ containing the necessary equipment. Implementation rose from 20 % at baseline to between 80 and 90 % at the time of data collection for this study, and sepsis mortality dropped from 22 to 12 % (unpublished data c/o Royal Free Hospital Patient at Risk Team, Devaney, Stapleton, Stanley, 2013). This took 4 years and the 95 % target was not reached for all pilot areas, raising the question of how best to develop the current intervention to achieve and sustain the 95 % target before extending implementation elsewhere to the rest of the hospital.

The experience of implementing the Sepsis Six is consistent with the experience of interventions to translate research into practice in other areas which generally demonstrate limited and variable success [12, 13]. This may be in part due to a lack of explicit rationale for intervention choice [14, 15] and precise reporting of what is delivered [16], despite calls for use of theory to understand and target mechanisms of action and comprehensive reporting [17–19]. The lack of theoretical rationale and detailed reporting both hinder efforts to design, replicate or improve interventions and thus derive maximum benefit from advances in research [20, 21].

Tools have been developed by behavioural and health services research scientists to support the detailed characterisation of interventions in terms of their content, and theoretical mechanisms of action, i.e. how they produce their effects.

The Template for Intervention Description and Replication (TIDieR) is a checklist for reporting and understanding the general content of behaviour change interventions including what is delivered, to whom, by whom, when and by what mode of delivery.

The Behaviour Change Wheel (see Fig. 1) is a comprehensive framework for designing interventions [22] derived from an integration of 19 behaviour change frameworks. The wheel consists of three layers. At the centre, the Capability (physical and psychological), Opportunity (social and physical), Motivation (automatic and reflective) model identifies the conditions required in order for behaviour to occur and provides a method for understanding what needs to change in order for a desired behaviour to occur. This layer can be divided into a further 14 domains comprising the Theoretical Domains Framework (TDF) [23] (see Additional file 1). The TDF is derived from a synthesis of 33 theories of behaviour and was designed to make psychological theory more accessible to those working in the field of implementation. It suggests 14 key theoretical domains or determinants of behaviour change where interventions might focus; these are an elaboration of the six COM-B segments, especially the reflective motivation segment.

The second layer of the Behaviour Change Wheel comprises nine intervention ‘functions’ (see Fig. 1) which are the general means by which an intervention might change behaviour. Examples are ‘persuasion’ and ‘incentivisation’. The outer layer of the wheel identifies the policy categories that can be used to support the delivery of these functions.

The Behaviour Change Technique (BCT) Taxonomy (v1) has been developed to standardise the reporting of intervention content, that is their potentially ‘active ingredients’ [24]. It is a structured list of 93 irreducible and discrete intervention components. A BCT is the smallest component of an intervention that may have the potential to change behaviour. Each intervention function is likely to consist of several BCTs; on the other hand, any one BCT may serve several functions. For example, the BCT to give information about the health consequences of performing a behaviour may serve the function of ‘education’ and/or ‘persuasion’ depending on the context and specific message.

Health professional behaviour in a range of clinical areas has been investigated using the TDF, including diagnostic tests and imaging [25–27], prescribing errors [28], transfusions [29] and hand hygiene [30, 31]. One systematic review by Laba and colleagues [32] used both the Behaviour Change Wheel and an earlier version of the BCT Taxonomy to classify strategies used to improve adherence to cardiovascular medications across 14 studies [33]. Examples of using the BCT Taxonomy to guide intervention development on the basis of TDF analyses are seen in antibiotic prescribing [34] and organ donation [35]. A common scenario in clinical practice is that of having interventions in place which have achieved some level of success, but are not fully reported or understood and may require improvement. The usefulness of these tools for describing and modifying existing interventions has yet to be investigated.

The aims of this study, therefore, were (a) to use the Behaviour Change Wheel functions and the BCT Taxonomy (v1) to report the content of an existing hospital intervention to implement the Sepsis Six clinical care bundle and (b) to use the TDF to characterise its potential theoretical mechanisms of action. This work is a first step in designing a more effective intervention for wider implementation using these tools. It will be proceeded by a TDF-based qualitative study which will explore experiences of health care professionals receiving the intervention including barriers and levers to Sepsis Six implementation and modification of the intervention’s content based on these data.

We have illustrated a systematic, theory-based approach to specifying the content and possible mechanisms of action of an implementation intervention using behavioural science methodology and triangulation from different data sources. The intervention consisted of 19 BCTs and seven intervention functions that may have stimulated behaviour change through several mechanisms of action, especially ‘beliefs about consequences’ and ‘beliefs about capabilities’ (reflective motivation) and ‘knowledge’ (psychological capability).

There are limitations to the method used. First, since the intervention was not explicitly guided by a formal theory, we did not specify causal pathways between barriers and levers and its content. Though we used published guidance on suggested BCTs for intervention functions and the TDF domains [36–38], we were essentially ‘retrofitting’ the intervention to the frameworks rather than using them at the design stage. Using them at the design stage would involve identifying barriers and levers first and then using this analysis to select BCTs. In order to ‘retrofit’ intervention functions from BCTs, we drew on our own knowledge and understanding of the intervention and its context to link BCTs to the functions we judged them to serve and to their potential theoretical mechanisms of action. Anecdotally, this ‘retrofitting’ approach is being used by several UK Government departments to evaluate their current policies and intervention strategies in terms of identifying gaps. This retrospective approach demonstrates that these tools can be used to analyse interventions ‘post-hoc’ regardless of whether they have been used at the design stage, and may prove useful for analysing other existing interventions in need of improvements. Mapping between the tools may need further refining, perhaps for specific behaviours and contexts.

A second limitation is that intervention content and delivery are likely to vary across wards and between education and feedback sessions. For example, some wards were provided a sepsis trolley, which held all the materials needed to implement Sepsis Six while others with fewer septic patients were provided sepsis bags only, which did not include bags of fluid or antibiotics. The resource constraints of the study did not allow us to investigate this variation. In particular, we were not able to provide information about the frequency, dose or ‘weight of importance’ of each component. This complex issue of ‘amount’ as well as type of intervention requires the development of methods for measuring dimensions of intervention dose, which may also change over time [39]. The full reporting of the content in real-time, however, is a first step towards ensuring that what is intended to be delivered is in fact delivered.

A mixed method approach to data collection [36] was a strength of the study. Observation identified the largest number of BCTs and functions, and this data source revealed substantial intervention content that was not mentioned in interview. This may reflect poor recall and/or the lack of awareness people have of behaviours that have become habitual and underlines the limitations of implementation research relying only on self-reported behaviour [40, 41]. Documents revealed the least data, possibly because incomplete documentation of improvement programmes is typical of professional practice in a working hospital with competing clinical priorities and schedules. Our study clearly demonstrates the utility of direct observation of behaviour in its natural setting for collecting real-time, real-world data. Though the other two data sources did not substantially add to the data, triangulation of the three data collection methods increased our confidence in the comprehensiveness of the description.

Insufficient description of content occurs in research as well as in clinical practice, and is well recognised in the literature on behaviour change interventions [16]. The primary strength of the current research is that we were able to describe systematically this intervention using a common language and make theoretically informed inferences about how it might bring about its effects. This is a useful first step in improving the intervention and its implementation to achieve better clinical outcomes. In the next stage of the research, this understanding of the content delivered and its potential theoretical mechanisms of action will be combined with a TDF analysis of health professionals’ experience of the intervention and barriers and levers to implementation to identify gaps and enhance the existing intervention. A full description of the updated intervention using these tools will be created so that fidelity of delivery can be assessed. This approach is in keeping with the importance of providing a transparent and robust rationale for developing intervention content and for clear and full reporting.

These tools have been used across many clinical domains for diagnosing implementation problems and informing intervention development, as described in the ‘Background’. To the authors’ knowledge, this is the first report that describes a single existing intervention using the Behaviour Change Wheel, the BCT Taxonomy and the TDF, together with the recent TIDieR statement [21]. This approach is in line with calls for improved methods for applying theory to intervention design including increased understanding of how BCTs exert their influences [42, 43]. This work therefore contributes to accumulating evidence on the linking of intervention content to mechanisms of action [36–38, 44]. Further research should evaluate the usefulness of these tools for ‘retrospective’ analyses of different interventions targeting different behaviours in different settings.

This study demonstrates how the use of a variety of information sources, particularly observation, and tools developed to make behavioural theory and implementation accessible to non-specialists can be used to specify the content and possible mechanisms of action of existing behaviour change interventions which, although designed without the use of theory, have achieved some level of success in clinical practice but require improvement. This enables comprehensive thinking about current practice and the drivers of behaviour and thus provides a sound platform for intervention improvement and replication.