Implementation of a nurse-led behaviour change intervention to support medication taking in type 2 diabetes: beyond hypothesised active ingredients (SAMS Consultation Study)

Trial evaluations of behaviour change interventions in primary care are increasingly common. They have paid considerable attention to processes affecting the validity of the trial design, such as allocation concealment and robust randomisation, and trial parameters, such as inclusion criteria, power and precision of measurement, facilitated by the CONSORT statement [1]. However, trial evaluations rarely include an assessment of the extent to which interventions are delivered and received as planned (fidelity), to what extent they are adapted, and what this means for long-term implementation and impact in routine clinical practice [2, 3]. A review of 202 psychosocial treatment evaluations showed that fidelity was only adequately addressed in 3.5% of the treatments [4]; and a review of 162 evaluations of primary and early secondary prevention programs showed that only 24.1% reported fidelity procedures [5]. Fidelity assessment is also rare for behaviour change interventions to support medication adherence. As an example, in a nurse-led medication adherence intervention based on social-cognitive theory among people with HIV/AIDS, Wickersham et al.[6] assessed adherence and quality of nurse-patient interaction and concluded that intervention delivery was successful. They reported overall adherence scores rather than scores for intervention components and did not report reliability of the measures.

Fidelity assessment of behaviour change interventions is important for several reasons. When a hypothesised mechanism of effect or logic model has been defined, fidelity can be assessed in relation to hypothesised active ingredients (intervention components assumed to facilitate behaviour change) to demonstrate that they were delivered in the intervention group as planned (treatment integrity) but not in the comparison group (treatment differentiation) [7]. This approach lends itself to tests of theory by showing the extent to which fidelity is linked to outcomes. However, interventions may be effective or ineffective due to factors unrelated to the hypothesised mechanism, for example adaptations to the planned intervention, practitioner characteristics (e.g., engagement), patient characteristics (e.g., literacy levels), relationship between practitioner and patient, and context. An assessment of actual implementation in practice through tape-recording or observation may identify active ingredients and ways in which the intervention worked beyond those hypothesised a priori. This could improve practitioner training, fidelity measurement and intervention design, and identify adaptations that may increase faithful implementation and impact in the long term.

Fidelity assessments of behaviour change interventions have focused on delivery, such as proportion of specified components delivered [8], practitioners’ use of communication skills [9] and behaviour change techniques [10, 11]. Assessment of participants’ responses during intervention contacts, e.g., engagement, understanding of intervention principles, and their relationship with practitioners is less common [12]. Studies have relied on self report, but providers’ [10] and recipients’ [13] recall are susceptible to bias and they have weak associations with independent assessment such as observation, audio- or video-recording [14]. To our knowledge no guidance exists for how to identify potential active ingredients beyond those specified a priori from observing implementation in practice.

This paper reports the use of a novel method to develop a coding frame for the independent, in-depth assessment of the implementation of a behaviour change intervention and standard care to support medication taking in type 2 diabetes. The study was motivated by the principal results of the Support and Advice for Medication Study (SAMS) and analysis of the hypothesised mechanism of effect [15, 16]. The trial was designed to improve the documented weak adherence of patients to their oral diabetes medication [17, 18], and was based on extensive development work (see [15] for details including justification of theory selection). The trial evaluated an intervention targeting people with type 2 diabetes, delivered by their usual practice nurse. Participants in the comparison group attended a consultation where the nurses enquired about medications and took a blood sample. Intervention patients additionally received an intervention which targeted two hypothesised causes of suboptimal adherence: weak motivation and forgetting [19]. Nurses were trained in the use of behaviour change techniques which were hypothesised a priori as active ingredients impacting on a theory-based causal pathway. They included motivational techniques which aimed to strengthen patients’ intention to take medication by targeting underlying beliefs based on the Theory of Planned Behaviour (TPB) [20], and an action planning technique which aimed to translate motivation into action, based on implementation intentions [21]. Intervention training also included patient-centred communication skills. These were not hypothesised as active ingredients but deemed important for building rapport with patients and competent delivery of behaviour change techniques.

An explanatory trial among 211 patients was carried out over 12 weeks after the intervention and comparison consultations, and medication adherence was measured objectively during that period. The intervention group showed a significant increase in the percentage of days when they took the prescribed medication, compared with the comparison group (77.4% of adherent days in the intervention group and 69.0% in the comparison group) [16]. However the intervention did not strengthen self-reported intentions or habits as hypothesised. Thus, either the measures of intention and habit did not capture the mechanism of effect, or the motivational and action-planning techniques improved adherence through a different mechanism, or the nurses did not deliver the intervention as specified. This study was designed to understand how the intervention might have worked, based on a coded analysis of tape-recorded consultations, and additionally using grounded theory to identify any other factors that might have increased medication adherence from listening to tape recordings. Our objectives were (1) to develop a reliable coding frame for tape-recorded consultations, assessing both a priori hypothesised and potential active ingredients observed during implementation of intervention and standard care, and (2) to describe the delivery and receipt of intervention and standard care components to understand how the intervention might have worked.

This study shows that it is feasible to develop a reliable coding frame to assess the implementation of behavioural interventions such as those used in the SAMS intervention and standard care groups. Fidelity of intervention delivery and receipt were good. The findings provide insight into how the intervention might have worked and highlight challenges when nurses deliver behaviour change interventions. These are discussed in turn below.

We succeeded in developing a reliable coding frame for in-depth assessment of implementation of the intervention and standard care in the full trial sample. The coding frame was informed by both a priori hypothesised ingredients and also how the intervention was actually delivered in clinical practice. We are not aware of other studies that have used this combined method for in-depth assessment, drawing on the traditions of both sociology and psychology. Using components of the grounded theory approach, we identified nurse communication, nurse-patient relationship, and patient responses to the intervention as potential active ingredients beyond psychological mechanisms hypothesised a priori. These were congruent with the contents of the intervention and training and may affect patient outcomes. Patient-centred communication skills are considered foundation competencies for facilitating behaviour change [27]. Immediate patient responses to interventions may influence treatment adherence, as adherence is unlikely if patients cannot recall or understand behaviour change techniques taught [2].

The development of reliable measures for practitioner communication, relationship, and participant responses was challenging. Existing tools were designed to assess medical consultations, e.g., RIAS [9], or specific interventions such as motivational interviewing [28], and are difficult to generalise to highly specified interventions. It proved more challenging to reliably assess the delivery of standard care than the intervention, perhaps because it was brief. Use of the grounded theory approach led to a comprehensive assessment that provided a rich insight into how the intervention was actually implemented, and helped to understand how it might have increased medication adherence.

We showed that the intervention and standard care were delivered and received well. The nurses were judged to show good adherence to scripted intervention protocols and contamination was virtually absent. Quality assurance may have played an important role in producing high adherence and low contamination in this study. The literature suggests that 80% to 100% represents high protocol adherence and < 50% low adherence [14]. However, few studies obtain fidelity levels over 80% and positive outcomes have been obtained with levels around 60% [12]. We assessed adherence on a Likert-type scale, but our measure indicates high protocol adherence when using these thresholds. Nurse competence of intervention delivery was rated lower than protocol adherence, confirming that high protocol adherence is not sufficient for competent delivery [29].

On average, the nurses were judged to use good communication skills during intervention delivery. They only occasionally facilitated rapport and partnership, which is likely to be due to the scripted protocol. The tape-recordings revealed that nurses delivered standard care in their own style, but use of the scripted intervention protocol resulted in greater formality and less fluency. This may have affected patients’ responses and interaction. Indeed, nurses used less patient-centred communication with intervention than comparison patients. A key reason may be that each nurse saw few intervention patients. They thus had little opportunity to become familiar with the intervention, deliver it faithfully in their own style, and tailor it to individual patients and medication taking problems. Other studies among practice nurses identified similar challenges with the use of scripted protocols, patient-centred communication and supporting patient self-management. The use of checklists by practice nurses in consultations with type 2 diabetes patients tended to decrease the flow and effectiveness of the consultation, particularly when nurses were less experienced and skilled [30]. Macdonald et al. found that nurses lacked resources to support patient self-management beyond personal experience and common sense approaches, such as giving information [31]. Kennedy et al. found that providing patient-centred communication and supporting patient self-management was challenging because practices did not perceive this as a priority, and nurses faced competing demands of achieving financially incentivised targets [32]. Patients were engaged and showed understanding of the intervention’s principles, but they rarely asked questions. It remains challenging in efficacy trials of behavioural interventions to achieve an optimal balance between standardisation across patients and practitioners and flexibility in delivery.

During the motivational component, nurses successfully elicited patients’ beliefs about taking medication and reinforced any positive beliefs. Whilst more than 80% of patients mentioned relevant beliefs, this included a majority of patients who said that they could not think of anything. Very few patients mentioned concerns or barriers, although we recruited patients whose elevated HbA1_c levels indicated suboptimal adherence. For these patients with established diabetes, taking medication seems a habitual rather than a reflective behaviour. Alternatively, patients may not have had adherence problems linked to motivation, or felt that sharing problems could damage the relationship with their usual practice nurse. When patients mentioned negative beliefs (e.g., difficulties during the weekends), nurses rarely prompted patients to problem-solve them. This may be due to insufficient training, insufficient emphasis in the scripted protocol, time pressures, nurses failing to spot opportunities for problem solving, and nurse perceptions that problems were irrelevant (e.g., past problems), infrequent (e.g., holidays) or could be addressed by action planning.

Nurses seemed to find it easy to deliver the action-planning component and patients seemed to experience no problems in making action plans for each dose. The action plans were (almost) identical to current medication taking routines. Perhaps patients judged that their current routines were good, or the intervention failed to identify problems with routines. Indeed the plans tended to relate to patients’ normal routines, but not cover situations when their routine was disrupted, such as if they were away from home. There were some protocol deviations without clear reasons. Patients were expected to write down the plan, but nurses did this for a quarter of patients. Whilst nurses were expected to read out the plans, a third of plans were not read out.

The results show how the intervention may have increased medication adherence and generate hypotheses for future research. The intervention increased medication adherence with high fidelity, and was effective despite less patient-centred communication than in the comparison group and infrequent prompting of problem solving. Thus, we can exclude adaptations, patient-centred communication and problem solving as mechanisms of effect. More time discussing medication taking may have increased medication adherence. We hypothesise that another mechanism of increased medication adherence may be the visualisation and/or verbalisation of current routines, combined with formulating these as an action plan (i.e., ‘action planning’). Visualisation maps onto ‘mental rehearsal of successful performance’ in the Behaviour Change Technique Taxonomy v1 [33]. Many patients described their current routines as a chain of behaviours resulting in them taking their tablets. A minority of patients read out their action plan, which could have acted as a commitment to take medication as prescribed.

The results illustrate the challenges for nurses when they deliver even brief behaviour change interventions in clinical practice. Although communication skills in intervention consultations were judged to be good, nurses showed less patient-centred communication than in the comparison group. More on-the-job training and rehearsal might facilitate the delivery of behaviour change techniques in a patient-centred, natural communication style. Training also needs to focus on problem solving techniques.

Our participant characteristics indicate that they were typical of people seen in routine primary care with type 2 diabetes, and our nurses reflect the kinds of nurses who manage people with type 2 diabetes. Therefore, our findings are likely to be generalisable to routine primary care.

Our recommendations include the identification of effective intervention components, development of fidelity measures, intervention design and potential adaptations. First, researchers need to identify critical, effective components of interventions to support medication adherence. Mental rehearsal of successful performance of the behaviour, action planning and commitment deserve further evaluation. Our next step is to identify critical components of the SAMS intervention by examining which aspects of nurse delivery and patient responses were associated with change in cognitive, behavioural, and clinical outcomes. Fidelity assessments could then focus on critical components and be less resource intensive. Second, we recommend more research on the development of reliable measures of practitioner communication, relationship and participant responses to behaviour change interventions. In order to assess factors beyond those hypothesised that may affect outcomes, we recommend that researchers identify these through recording or observing intervention contacts. Third, intervention developers need to observe usual practice and design interventions which are compatible with routine practice and adaptable to patients and local settings [34]. Such interventions are more likely to be implemented faithfully and have impact in the long term [12]. In this regard, the motivational component of the SAMS intervention included many questions and resembled oral administration of a Theory of Planned Behaviour questionnaire at a cost to patient-centred communication. It could be improved by asking patients on benefits and concerns about, and barriers to taking medication only. Furthermore, given the recommended ‘if…then…’ format in the literature about implementation intentions and supporting evidence, patients could be encouraged to read out their plans in this format.

It is feasible to reliably assess the implementation of behaviour change interventions in clinical practice and this can provide insight into how interventions achieve any effects. Results could not be explained by poor delivery of motivational and action-planning components, definition of new action plans, or by improved problem-solving or patient-centred communication. Possible mechanisms of increased medication adherence include more time discussing it and mental rehearsal of successful performance of current routines, combined with action planning. Delivery of a new behaviour change intervention may lead to less patient-centred communication and possible reduction in overall trial effects.