Development of a behaviour change intervention: a case study on the practical application of theory

This model, developed by Grol and Wensing [16], was selected to inform the overall approach to developing the behaviour change intervention. It comprises five steps: first developing a ‘concrete proposal’ for the desired change, one that is clearly defined and easily understandable; second undertaking an analysis of current practice, and barriers and incentives for change, in the group in which change is desired; third developing and selecting ways to change practice; and finally (steps 4 and 5) undertaking and evaluating the implementation plan (Table 1). Detailed guidance is available on how to approach the tasks needed for each step with reference to the underpinning evidence [16], and was selected as, in addition to its logical approach, it provides guidance on the answers to three very practical questions during the planning of change: ‘where do we want to be?’ (step 1), ‘where are we now?’ (step 2), and ‘how do we get there?’ (step 3).

At step 2, a key task was to understand which factors, or ‘determinants,’ would impede or facilitate the intended change, and many psychologically-oriented models have been proposed to inform this task. Many of these models overlap, and each tends to focus on different aspects of the change process [16]. One challenge for those facilitating change is how to select the most appropriate model when undertaking an analysis of these factors in a particular set of circumstances. Michie et al. addressed this problem by undertaking a consensus exercise to develop a model that encompassed 128 theoretical constructs (or determinants) included in 33 psychological theories - the Theoretical Domains Framework (TDF) [18]. The TDF consists of 12 domains (Table 2), such as knowledge, skills, beliefs about consequences, motivation and goals, with each domain having a set of theoretical constructs that had been identified as components in the models included in the consensus exercise. A total of 11 out of the 12 domains concern characteristics of the people for whom change is desired, with the 12^th concerning the attributes of the change or desired behaviour itself. The TDF has been used to identify determinants of behaviour change for an extensive range of conditions and clinical situations, for example, mobilisation of older patients in hospital [19], utilisation of a rule for the use of CT scans for head trauma [20], and management of chronic obstructive airways disease [21], and its development and use in a range of other studies has been reviewed [22]. The TDF has been recently validated and refined: experts were asked to re-sort the constructs included in the TDF and to re-develop the domains, with and without reference to the original domains [23]. The refined framework consists of 14 domains, 8 unchanged from the original, 6 derived from a more specific grouping of the constructs underpinning 3 of the domains (beliefs about capabilities, beliefs about consequences, and motivation and goals), with 1 of the original domains omitted (nature of the behaviour). The 12-domain TDF model was selected as the domains in this framework provided a practical and comprehensive list of possible determinants of behaviour change (the 14-domain model had yet to be developed at the time of this study), and the TDF was utilised to identify relevant determinants of behaviour change in this study.

At step 3, one of our tasks was to develop or select techniques to effect behaviour change. Michie et al. developed a model to inform the selection of behaviour change techniques that target the determinants described in the TDF [24]. They identified, and defined, a set of behaviour change techniques described in the literature and mapped them to the domains in the TDF described above (barring the 12^th domain): the techniques that they judged to be effective in changing behaviour for each domain [24]. The approach to mapping behaviour change techniques to TDF domains has been incorporated into protocols for the development of complex interventions, for example for tobacco counselling in dentistry [25] and management of low back pain [26]. This mapping process provides a practical tool for selecting appropriate behaviour change techniques as the components of a behaviour change intervention and was utilised at step 3.

At step 3, the principles of adult learning theory were also utilised; that adults are internally motivated and self-directed, bring life experiences and knowledge to learning experiences, are goal and relevancy oriented, are practical and like to be respected [27]. Adult learning theory was selected to inform the educational process of the behaviour change intervention as it has a well-established role in development of courses to support continuing professional development [27], including interventions such as the one developed in this study.

The model OA consultation, developed by the consensus exercise, consisted of 25 tasks addressing: i) assessment of chronic joint pain, ii) patient’s ideas and concerns, iii) exclusion of red flags, iv) examination, v) provision of the diagnosis and written information, vi) promotion of exercise and weight loss, vii) initial pain management, and viii) arrangement of a follow-up appointment [28].

The advisory group meetings were led by one of the authors (MP) and attended by 15 GPs, 5 practice nurses, and a practice manager. The key finding from the meetings on the characteristics of the model OA consultation was that, presented as 25 tasks, it was too complex to explain simply and quickly to GPs or for them to easily understand and translate into day-to-day practice. To simplify the model, tasks were grouped by core elements of a patient-centred consultation [11, 31, 32], for example support for self-care and provision of evidence-based information, and the model succinctly presented as three tasks.

The advisory group meeting transcripts and field notes on current practice, attitudes to recommended best practice, and perceived barriers to, and incentives for, changing practice, were analysed using the TDF as a coding framework. The analysis was discussed by the study team and by a group of expert educational advisors to the study, and seven TDF domains were identified as relevant to changing GP practice in OA consultations (Table 3).

The content of the behaviour change intervention was derived by the study team from the practical requirements of delivering the model OA consultation and from gaps identified in the advisory group meetings, for example lack of knowledge about the impact of OA on the individual, the skills necessary to deliver the model OA consultation, and the credibility of NICE guidelines. The selection of behaviour change techniques was undertaken by the study team and the educational advisors to the study. The starting point was the list of techniques that Michie et al. had judged appropriate to effect change for domains identified in step 2 [24]. The group used their research, clinical and educational experience to decide which of these techniques to choose. The content of, and techniques to address, each domain are detailed in Table 4.

The choice of delivery style was informed by evidence from the Cochrane Effective Practice and Organisation of Care Group on the effectiveness of strategies for changing practice, with a specific emphasis on small group learning with a mixture of didactic and interactive sessions [33] and facilitated by opinion leaders [34]. In addition, the study team drew on evidence on a learner-centred approach, which utilises prior knowledge and experiences of the participants [27] to effect change in behaviour. Specifically, for the delivery of techniques to address the skills domain, we used empirical evidence on techniques for training experienced GPs in communication skills, a method of training known as ‘context-bound communication skills training’ was adopted [35]. In this technique the ‘context,’ in this case the management of OA, is in the foreground and the communication training in the background. A key feature is that participants practise consultation skills when consulting with simulated patients and receive feedback. This had been found to be a feasible, acceptable and effective method of enhancing the consultation skills of experienced practitioners [36] and preferable, for this group, to the approach taken in undergraduate skills teaching, where it is skill and not context that is in the foreground.

The final step was to consider the practical issues in delivering the workshops in four general practices with all the myriad demands on the GPs’ and other practice staff’s time. The final format was developed by the study team and educational advisors, drawing on their professional experience, and in consultation with GPs working in Keele University Medical School. The format was to deliver the behaviour change intervention at general practices’ premises, in four sessions, lasting one or two hours each, and about two to three weeks apart. The final behaviour change intervention with detailed timings is shown in Table 5.

All the GPs, practices nurses, and administrative staff working in the four practices randomised to the intervention arm of the MOSAICS study, were invited to attend the training sessions (see Table 5 for details) [10]. The GPs were invited to participate in the evaluation of the behaviour change intervention. Methods and measures were chosen and developed to evaluate the behaviour change intervention at the four levels (Table 6).

The TDF and behaviour change technique mapping, developed by Michie et al., have both been published within the last 10 years, and a number of studies have reported on utility and outcome in the development of behaviour change interventions for trials [26, 41, 42]. Both models, used sequentially as in this study, have been employed in development of interventions to improve management of low back pain [26], to enhance GP diagnosis of dementia [41], and to reduce antibiotic use for upper respiratory infections [42]. Two of these have resulted in multi-facetted interventions as developed in this study [26, 41], with the other [42] resulting in two interventions, each specifically addressing one of two determinants of behaviour change identified. The research team in the low back pain study, having determined the behaviour change techniques to include in the intervention, and the mode of delivery, took a pragmatic approach to their final selection: what was locally feasible and acceptable. We also took a pragmatic approach on deciding the final format, but this did not result in any changes to our intended delivery other than that the workshops were run at the practices, lasted no more than two hours each, and were about two to three weeks apart. To date, only the low back pain trial has reported and showed a small effect on GP intention to practice but no significant change in actual behaviour [43]. That clinical practice was not observed to change may not have been due to the intervention per se, as there were logistical problems in getting GPs to attend the intervention workshops and methodological problems in assessing outcome. The drive to use theory to inform development of interventions has been questioned [44], as empirical evidence is lacking on effectiveness of interventions developed in this way. Although the low back pain trial did not demonstrate a change in clinical practice, its use of theory does add to empirical evidence on the process of behaviour change.

Developing complex interventions is a complex task in itself, and understanding how to approach it in a systematic way, informed by relevant theory, can be daunting for research teams [13]. The principal strength of the method described in this paper is that it enabled the MRC guidance on developing complex interventions to be operationalized systematically, and in a practical and do-able manner. The guidance on using the Grol and Wensing model to change clinical behaviour is extensive [16] and provided a very usable manual on ‘how to do it.’ The use of the TDF strengthens the approach advocated for the Grol and Wensing model for step 2, and is reflected in the increasing popularity of the TDF by research teams in developing interventions [22]. In addition, the recent validation and refining of the TDF domains has strengthened the rationale for its methodology, as used in this study, and, with a refined structure, strengthened its use in future studies [23].

The use of GP advisory group meetings both to gain views about the proposed change (step 1) and to undertake the target group analysis (step 2) was a practical strength. It provided an efficient method of: i) involving the target group in the development of the change proposal (an activity it its own right that enhances uptake of an intervention [16]), ii) identifying which characteristics of the intervention might hinder or facilitate uptake, and iii) understanding current practice and identifying relevant determinants of change.

One potential limitation was that the topic guide for the advisory group meetings was not specifically developed from the TDF, which could have resulted in some of the TDF domains not being fully explored in the meetings. The topic guide had been developed, and the meetings undertaken, before deciding to use the TDF in step 2. However, the topic guide was broad and covered current management, views about recommended practice, and perceived gaps between current and recommended care and allowed for free discussion by the groups. This has occurred in other studies [21, 45] and, although not used to develop the topic guide, the TDF did give an efficient method for analysing advisory group comments.

The GPs who attended advisory group meetings were not the same GPs who received the behaviour change intervention in the MOSAICS trial, and their views and attitudes may not have been the same as these GPs. Analysis of the actual target group for the behaviour change intervention – the GPs in the four MOSAICS intervention practices – may have identified different determinants to be addressed, but the timescale for developing the behaviour change intervention in the MOSAICS study did not allow for this. However, as the mode of delivery included interactive sessions, and the sessions encouraged reflection on current practice and on the video-recorded consultations, there was ample opportunity for issues specific to the study GPs to be addressed.

The final measure of success, beyond the fact that this methodology has provided the framework for an intervention deliverable in practice, is whether it achieved what it set out to (a change in clinical practice) in a sufficient dose to achieve optimal outcomes for patients in the MOSAICS trial. Both these outcomes (intermediate professional-focused and ultimate patient-focused) will be reported in the future as part of the main results from the MOSAICS study.