The behaviour change wheel: A new method for characterising and designing behaviour change interventions

Improving the implementation of evidence-based practice and public health depends on behaviour change. Thus, behaviour change interventions are fundamental to the effective practice of clinical medicine and public health, as indeed they are to many pressing issues facing society. 'Behaviour change interventions' can be defined as coordinated sets of activities designed to change specified behaviour patterns. In general, these behaviour patterns are measured in terms of the prevalence or incidence of particular behaviours in specified populations (e.g., delivery of smoking cessation advice by general practitioners). Interventions are used to promote uptake and optimal use of effective clinical services, and to promote healthy lifestyles. Evidence of intervention effectiveness serves to guide health providers to implement what is considered to be best practice (for example, Cochrane reviews, NICE guidance). While there are many examples of successful interventions, there are also countless examples of ones that it was hoped would be effective but were not [[1], e.g. [2, 3]]. To improve this situation, and to improve the translation of research into practice, we need to develop the science and technology of behaviour change and make this useful to those designing interventions and planning policy.

The process of designing behaviour change interventions usually involves first of all determining the broad approach that will be adopted and then working on the specifics of the intervention design. For example, when attempting to reduce excessive antibiotic prescribing one may decide that an educational intervention is the appropriate approach. Alternatively, one may seek to incentivise appropriate prescribing or in some way penalise inappropriate prescribing. Once one has done this, one would decide on the specific intervention components. This paper examines this first part of this process. We and others are also working on how one identifies specific component 'behaviour change techniques' [4, 5].

In order to identify the type or types of intervention that are likely to be effective, it is important to canvass the full range of options available and use a rational system for selecting from among them. This requires a system for characterising interventions that covers all possible intervention types together with a system for matching these features to the behavioural target, the target population, and the context in which the intervention will be delivered. This should be underpinned by a model of behaviour and the factors that influence it.

Interventions are commonly designed without evidence of having gone through this kind of process, with no formal analysis of either the target behaviour or the theoretically predicted mechanisms of action. They are based on implicit commonsense models of behaviour [6]. Even when one or more models or theories are chosen to guide the intervention, they do not cover the full range of possible influences so exclude potentially important variables. For example, the often used Theory of Planned Behaviour and Health Belief Model do not address the important roles of impulsivity, habit, self-control, associative learning, and emotional processing [7].

In addition, often no analysis is undertaken to guide the choice of theories [8]. Useful guidance from the UK Medical Research Council for developing and evaluating complex interventions advocates drawing on theory in intervention design but does not specify how to select and apply theory [9]. It should also be noted that even when interventions are said to be guided by theory, in practice they are often not or are only minimally [10].

Thus, in order to improve intervention design, we need a systematic method that incorporates an understanding of the nature of the behaviour to be changed, and an appropriate system for characterising interventions and their components that can make use of this understanding. These constitute a starting point for assessing in what circumstances different types of intervention are likely to be effective which can then form the basis for intervention design.

There exists a plethora of frameworks for classifying behaviour change interventions but an informal analysis suggests that none are comprehensive and conceptually coherent. For example, 'MINDSPACE' an influential report from the UK's Institute of Government, is intended as a checklist for policymakers of the most important influences on behaviour [11]. These influences provide initial letters for the acronym MINDSPACE: messenger, incentives, norms, defaults, salience, priming, affect, commitment, and ego. The framework does not appear to encompass all the important intervention types. Moreover, the list is a mixture of modes of delivery (e.g., messenger), stimulus attributes (e.g., salience), characteristics of the recipient (e.g., ego), policy strategies (e.g., defaults), mechanisms of action (e.g., priming), and related psychological constructs (e.g., affect). In that sense it lacks coherence. The report recognises two systems by which human behaviour can be influenced -- the reflective and the automatic -- but it focuses on the latter and does not attempt to link influences on behaviour with these two systems.

A second example comes from the Cochrane Effective Practice and Organisation of Care Review Group (EPOC)'s 2010 taxonomy [12]. This categorises interventions to change health professional behaviour into professional, financial, organisational, or regulatory, covering many of the key intervention types. However, the categories are very broad and within each is a mixture of different types of interventions at different conceptual levels. For example, 'professional' includes individual behaviour (distributing educational materials) and organisational interventions (local consensus processes); 'financial' includes individual and organisational incentives and environmental restructuring (changing the available products); 'organisational' includes input (changing skill mix), processes (communication) and effects (satisfaction of providers); and 'regulatory' includes legal (changes in patient liability) and social influence (peer review). Professional, financial, and organisational interventions are found across all categories.

Aside from specific frameworks, there are some broad distinctions that have been widely adopted. One such distinction is between population-level and individual-level interventions [13]. While superficially appealing, there are many interventions that this distinction cannot readily classify and it has not been possible to arrive at a satisfactory definition of the distinction that does not contain inconsistencies. For example, if wide reach is a feature of population level interventions, routine general practitioner (GP) smoking assessment and advice (given to all patients) should fall into that category; yet it is delivered specifically to individuals and can be tailored to those individuals. Indeed, the NHS Stop Smoking Services might be considered a typical case of individual-level interventions, but they reach more than 600,000 smokers each year [14]. We do not consider these broad distinctions further in this paper.

It appears that most intervention designers do not use existing frameworks as a basis for developing new interventions or for analysing why some interventions have failed while others have succeeded. One reason for this may be that these frameworks do not meet their needs. In order to choose the interventions likely to be most effective, it makes sense to start with a model of behaviour. This model should capture the range of mechanisms that may be involved in change, including those that are internal (psychological and physical) and those that involve changes to the external environment. In general, insufficient attention appears to be given to analysing the nature of behaviour as the starting point of behaviour change interventions [15], a notable exception being intervention mapping [16]. 'Nature of the behaviour' was identified as one of 12 theoretical domains of influence on implementation-relevant behaviours [9]. Whilst this framework of 12 theoretical domains has proved useful in assessing and intervening with implementation problems [9], the domain of behaviour has remained under-theorised and therefore underused in its application.

There are a number of possible objections to attempting to construct the kind of behavioural model described and link this to intervention types. The most obvious criticism is that the area is too complex and that the constructs too ill-defined to be able to establish a useful, scientifically-based framework. Another is that no framework can address the level of detail required to determine what will or will not be an effective intervention. The response to this is twofold: these are empirical questions and there is already evidence that characterising interventions by behaviour change techniques (BCTs) can be helpful in understanding which interventions are more or less effective [6, 17]; and not to embark on this enterprise is to give up on achieving a science of behaviour change before the first hurdle and condemn this field to opinion and fashion.

To achieve its goal, a framework for characterising interventions should be comprehensive: it should apply to every intervention that has been or could be developed. Failure to do this limits the scope of the system to offer options for intervention designers that may be effective.

In addition, the categories should be able to be linked to specific behaviour change mechanisms that in turn can be linked to the model of behaviour. These requirements constitute three criteria of usefulness that can be used to evaluate the framework: comprehensiveness, coherence, and links to an overarching model of behaviour. We limited the criteria to those we considered to form a basis for judging adequacy. There are others, e.g., parsimony, that are desirable features but do not lend themselves to thresholds. Other criteria can be used to evaluate its applicability, e.g., reliability, ease of use, ease of communication, ability to explain outcomes, usefulness for generating new interventions, and ability to predict effectiveness of interventions

In light of the above, this paper aims to:

Prior to reviewing the literature on intervention frameworks, we needed to establish a set of criteria for evaluating their usefulness. Following this, our method involved three steps: a systematic literature review and evaluation of existing behaviour change intervention frameworks, development of a new framework, and a test of the reliability of the new framework.

Within 19 frameworks for classifying behaviour change interventions, nine intervention functions and seven policy categories could be discerned. None of the frameworks covered all of these. Only a minority of the frameworks could be regarded as coherent or linked to an overarching model of behaviour. However, it was possible to construct a new BCW framework that did meet these criteria from the existing ones. This framework could be reliably applied to classify interventions within two important areas of public health.

We believe that this is the first attempt to undertake a systematic analysis of behaviour intervention frameworks and apply usefulness criteria to them. This is also the first time that a new framework has been constructed from existing frameworks explicitly to overcome their limitations. Moreover, we are not aware of other attempts to assess the reliability with which a framework can be applied in practice.

It must be recognised that there are a near infinite number of ways of classifying interventions and intervention functions. The one arrived at here will no doubt be superseded. But for the present, it has the benefits of having been derived from classifications already available and therefore covering concepts that have previously been considered to be important, and using an overarching model of behaviour to link interventions to potential behavioural targets. The most important test of this framework will be whether it provides a more efficient method of choosing the kinds of intervention that are likely to be appropriate for a given behavioural target in a given context and a given population.

Just by identifying all the potential intervention functions and policy categories this framework could prevent policy makers and intervention designers from neglecting important options. For example, it has been used in UK parliamentary circles to demonstrate to Members of Parliament that the current UK Government is ignoring important evidence-based interventions to change behaviour in relation to public health [43, 44]. By focusing on environmental restructuring, some incentivisation and forms of subtle persuasion to influence behaviour, as advocated by the popular book 'Nudge' [45], the UK Government eschews the use of coercion, persuasion, or the other BCW intervention functions that one might use.

Although awareness of the full range of interventions and policies is important for intervention design, the BCW goes beyond providing this. It forms the basis for a systematic analysis of how to make the selection of interventions and policies (as in Tables 2 and 3). Having selected the intervention function or functions most likely to be effective in changing a particular target behaviour, these can then be linked to more fine-grained specific behaviour change techniques (BCTs). Any one intervention function is likely to comprise many individual BCTs, and the same BCT may serve different intervention functions. An examination of BCTs used in self-management approaches to increasing physical activity and healthy eating [46], and in behavioural support for smoking cessation [47, 48], shows that these BCTs serve five of the intervention functions: education, persuasion, incentivisation, training, and enablement. The other four intervention functions (coercion, restriction, environmental restructuring, and modelling) place more emphasis on external influences and less on personal agency. Reliable taxonomies for BCTs within these intervention functions have yet to be developed.

One of the strengths of this framework is that it incorporates context very naturally. There is a general recognition that context is key to the effective design and implementation of interventions, but it remains under-theorised and under-investigated. The 'opportunity' component of the behavioural model is the context, so that behaviour can only be understood in relation to context. Behaviour in context is thus the starting point of intervention design. The behaviour system also has automatic processing at its heart, broadening the understanding of behaviour beyond the more reflective, systematic cognitive processes that have been the focus of much behavioural research in implementation science and health psychology (for example, social cognition models such as the Theory of Planned Behaviour).

An existing framework that has made an important contribution to making intervention design more systematic is 'intervention mapping' [16]. A key difference between this and the BCW approach is that intervention mapping aims to map behaviour on to its 'theoretical determinants' in order to identify potential levers for change, whereas the BCW approach recognises that the target behaviour can in principle arise from combinations of any of the components of the behaviour system. It may appear that some components are more important than others because of a lack of variance in (including absence or universal presence of) the variables concerned in the population under study. This can be illustrated by a study of GP advice to smokers, which found that a single variable -- degree of concern that it would harm the doctor-patient relationship -- accounted for significant variance in the rate of advice-giving [49]. 'Intervention mapping' would suggest that concern be the target for an intervention (as long as a judgement were made that this could be modified using interventions that were realistically applicable). The BCW would analyse the target behaviour in context and note that, regardless of what covariation might currently exist, the target behaviour consists of an activity in which capability is not at issue, and the reflective motivation is broadly positive. The problem arises because automatic motivational factors are currently working against the behaviour (e.g., lack of emotional reward for giving advice or punishment for not giving it and lack of cues to action). Moreover, the physical opportunity is limited (lack of time) and the social opportunities are also somewhat limited. It would then consider the full range of ways in which the frequency of advice-giving could be increased. Because the target behaviour is part of a 'system,' a single intervention may have consequences for other parts of the system - these might work against sustainable change or in favour of it.

Thus, the BCW approach is based on a comprehensive causal analysis of behaviour and starts with the question: 'What conditions internal to individuals and in their social and physical environment need to be in place for a specified behavioural target to be achieved?' The 'intervention mapping' approach is based on an epidemiological analysis of co-variation within the behavioural domain and starts with the question: 'What factors in the present population at the present time underlie variation in the behavioural parameter?' When it comes to theoretical underpinnings, the BCW approach draws from a single unifying theory of motivation in context that predicts what aspects of the motivational system will need to be influenced in what ways to achieve a behavioural target, whereas the 'intervention mapping' approach draws on a range of theoretical approaches each of which independently addresses different aspects of the behaviour in question.

The BCW is being developed into a theory- and evidence-based tool allowing a range of users to design and select interventions and policies according to an analysis of the nature of the behaviour, the mechanisms that need to be changed in order to bring about behaviour change, and the interventions and policies required to change those mechanisms. An ongoing programme of research is developing an 'intervention design tool' based on the BCW. It starts with a theoretical understanding of behaviour to determine what needs to change in order for the behavioural target to be achieved, and what intervention functions are likely to be effective to bring about that change. It is being field tested by a range of staff involved in policy and intervention work applying the framework to develop prototype strategies for specific implementation targets. Data are being collected about ease of use and the potential of the BCW to generate new insights.

There are a number of limitations to the research described in this paper. First, it is possible that the systematic review missed important frameworks and/or intervention functions. Second, judgement is inevitably involved in conceptualising intervention functions and policy categories. There are many different ways of doing this, and no guarantees that the one arrived at here is optimal. Indeed, different frameworks may be more or less useful in different circumstances. Third, even though the proposed framework appears to be comprehensive and can be used reliably to characterise interventions, it is possible that it may prove difficult to use. However, the systematic way in which development of the BCW has been approached should enable it to provide a more robust starting point for development of improved frameworks than has hitherto been possible.