2.1 Design

2.2 Data analysis

All programme materials were imported into NVivo v12 to support data coding and analysis using a theory-driven content analysis approach.^S10 Initially, materials were coded deductively using the TIDieR framework.³⁰ Materials were then coded for BCTs using the BCTTv1.¹⁵ It should be noted that the content of the programmes was not explicitly described in BCT terms. Therefore, BCTs were coded when ‘present’, ‘probably present’ (requiring a certain degree of inference for coding) and ‘planned to be optionally delivered’ at the discretion of the intervention provider (see Appendix S3 for codebook and coding decisions). While acknowledging the possibility of suboptimal operationalisation of some BCTs, this approach aligns with previous research and aimed to avoid missing BCTs potentially present in the programme materials.^16,S11

Following BCT coding, potential MoAs associated with the BCTs and the intervention functions served by the BCTs were identified. Potential MoAs were coded based on the list of 26 MoAs outlined in the Theory and Techniques tool, using theoretical constructs mentioned in proximity to the BCTs or explicitly indicated as underpinning part(s) of the programme where BCTs were identified.^16,S4 Intervention functions were identified using guidance provided by Michie et al.^S3 When BCTs could serve multiple functions, the function(s) they were more likely to serve were determined based on our knowledge of the programmes. The programmes mode and source of delivery were coded using the Mode of Delivery Ontology v1 and Intervention Source Ontology v1.^S1,S2

The coding process was guided by a codebook developed iteratively by the research team (see Appendix S3 for the complete codebook and further details on coding decisions) and relevant coding guidelines from validation studies of the tools used.^{15, 16,S1,S2} One researcher coded all programme materials. To ensure coding consistency, a second researcher double-coded 10% of the materials from each programme. Both researchers had formal training in using the BCTTv1 (https://www.bct-taxonomy.com/) and had no prior knowledge of the programmes. Additionally, after consultation meetings with DESMOND stakeholders to discuss preliminary findings, two researchers familiar with the DESMOND programme and trained to use the BCTTv1 double-coded 10% of the DESMOND materials to further ensure coding consistency and the validity of the findings. Coding discrepancies were resolved through discussion until consensus was achieved and through consultation with a senior behavioural science expert when necessary. The codebook was iteratively revised when necessary to improve its clarity and ensure coherence.

After data coding, the findings were tabulated, and descriptive analyses were conducted in MS Excel to determine the frequency at which the BCTs were intended to be delivered across the programme components (sessions and participant materials) and the links between BCTs and potential MoAs and intervention functions. Frequencies were calculated based on the presence of BCTs and links in programme components (sessions and participant materials), consistent with previous studies.^S11 Inter-rater reliability for BCTs coding was determined using Cohen's kappa coefficient.^S12

2.3 Stakeholder engagement

Preliminary findings were presented to programme managers, diabetes educators and members of a public advisory panel of people living with type 2 diabetes for discussion and feedback to ensure the validity of the findings. The stakeholder engagement process is presented in further detail in Appendix S4.

3.1 Programme content

Thirteen documents were analysed across the two programmes, seven from the CODE programme, and six from the DESMOND programme (see Appendix S2). Table 1 summarises the key characteristics and content of the programmes based on the TIDieR framework (see Appendix S1 for the full TIDieR).²⁴ Details of the programmes' mode of delivery, source and scheduling are provided in Appendix S5.

3.2 Behaviour change techniques

Substantial agreement was achieved between coders before resolving disagreements, as indicated by Cohen's kappa values ranging from 0.69 to 0.81 (see Table 2 for full kappa values).

Thirty-four of the 93 BCTs listed in the BCTTv1 were identified across the programmes, with a mean number of 27.5 BCTs (SD = 0.71; range 27–28) per programme. Of these, four required some inference to be coded, and six were planned to be tailored or optionally applied (see Appendix S6 for further details). Despite this, no BCT was deleted from the analysis to avoid excluding BCTs potentially present in the programmes. A complete list of the BCTs identified in each programme is presented in Table 3 (see Appendices S7–S9 for further details on BCTs and their definitions). The identified BCTs targeted a range of self-management behaviours, as shown in Appendix S9. More than half of the identified BCTs (22/34, 65%) were coded for both programmes. Most BCTs were coded in the programmes' curriculums or both the curriculums and participant materials, with only a small proportion of the BCTs exclusively identified in the participant materials (7/34, 21%). The identified BCTs belonged to 13 of the 16 taxonomy groups. Across both CODE and DESMOND programmes, ‘goals and planning’ (6/28, 21% in CODE, 4/27, 15% in DESMOND) ‘feedback and monitoring’ (5/28, 18% in CODE, 3/27, 11% in DESMOND) and ‘natural consequences’ (4/28, 14% in CODE, 4/27, 15% in DESMOND) were the most commonly coded BCT groups. BCTs from the ‘regulation’ and ‘self-belief’ groups were only identified in the DESMOND programme. No BCTs from the ‘identity’, ‘scheduled consequences’, and ‘covert learning’ groups were coded in either programme. The three most frequently coded BCTs across both programmes were ‘information about health consequences’ (n = 5 in CODE; n = 9 in DESMOND), ‘credible source’ (n = 5 in CODE; n = 9 in DESMOND) and ‘goal setting (behaviour)’ (n = 5 in CODE; n = 8 in DESMOND) (see Table 3 for further details on the frequency of BCTs across programme materials).

3.3 Mechanisms of action

Approximately 68% of the BCTs identified in the programmes (n = 23) were linked to at least one MoA. Across the programmes, 15 MoAs were identified (see Table 4 and Appendices S7 and S8 for further details on the MoAs identified in each programme and their definitions). Of these, eight MoAs were identified in both programmes, whereas seven were identified in only one programme. The three MoAs most frequently identified across the two programmes were ‘knowledge’ (n = 7 in CODE and n = 24 in DESMOND), ‘beliefs about capabilities’ (n = 4 in CODE and n = 19 in DESMOND) and ‘beliefs about consequences’ (n = 4 in CODE and n = 13 in DESMOND).

3.4 Intervention functions

The identified BCTs served six of the nine intervention functions outlined in the BCW (see Table 5 and Appendices S7 and S8 for further details on the functions served by BCTs identified in each programme and their definitions). The intervention functions most frequently served by the BCTs identified in the programmes were ‘education’ (17/28, 61% in CODE and 15/27, 56% in DESMOND), followed by ‘enablement’ (13/28, 46% in CODE and 11/27, 41% in DESMOND), and ‘persuasion’ (10/28, 36% in CODE and 10/27, 37% in DESMOND). A small number of BCTs served the functions of ‘incentivisation’ (2/28, 7% in CODE and 2/27, 4% in DESMOND), ‘environmental restructuring’ (1/28, 4% in CODE and 1/27, 4% in DESMOND) and ‘modelling’ (1/27, 4% in DESMOND). None of the identified BCTs served the functions of ‘training’, ‘restriction’ or ‘coercion’.

3.5 Stakeholder engagement

Stakeholder consultation meetings led to minor adjustments in the coding of the programmes, such as an increase in the frequency of coding of the BCT ‘credible source’. Further details on stakeholder feedback and changes made to the coding of the programmes are available in Appendix S4.

4.1 Relation to existing research

Previous studies have found associations between certain BCTs identified in the programmes and improved outcomes in people with type 2 diabetes.^{15, 16,S13–S17} BCTs such as ‘instruction on how to perform the behaviour’, ‘action planning’ and ‘demonstration of the behaviour’, identified in both programmes, have previously been found to be associated with clinically significant reductions in HbA_1c.^S13 Social support and self-regulatory BCTs (e.g. goal setting, self-monitoring) have also shown effectiveness in people with type 2 diabetes.^S14,S15 Nevertheless, it is important to note that the presence of these BCTs in programme materials does not necessarily result in their delivery and no causality can be established between the presence of some BCTs and programme effectiveness.

Although the programmes share 22 common BCTs, each programme also incorporates unique BCTs, suggesting potential variations in the active ingredients and mechanisms of these interventions. For example, the CODE programme includes more ‘feedback and monitoring’ techniques, while the DESMOND programme described more BCTs from the ‘regulation’ and ‘self-belief’ groups. Programmes also differ in their frequency of use of certain BCTs. These differences may contribute to the variability in the programme outcomes. However, this requires further investigation in future studies.

Furthermore, similar to previous studies on interventions for type 2 diabetes, only a limited number of BCTs (34/93) were identified in the programmes, with some BCTs rarely or never used.^S13–S18 This suggests the potential for exploring alternative BCTs to broaden the range of targeted barriers. Considering the mixed and modest effects of DSMES programmes for type 2 diabetes on various outcomes, as well as the low sustainability of programmes' effects on HbA_1c, exploring alternative BCTs and psychotherapeutic techniques may help address intervention limitations.³ Research has indicated that interventions with more BCTs may be more effective.^S13,S15 Further research is therefore needed to examine the integration of additional BCTs and behaviour or psychotherapeutic strategies to enhance programmes' effectiveness and ensure sustainable outcomes. Ongoing research on post-programme barriers and enablers for sustaining self-management and support needs among CODE and DESMOND participants will be triangulated with the findings of this study to identify potential opportunities for strategically improving the programmes, better enabling long-term behaviour change and improving the sustainability of outcomes.

The identified BCTs were linked to multiple MoAs, consistent with prior research.¹⁶ Some BCT-MoA links align with existing literature.^16,S4,S19 Similarly, our findings also align with previous research on DSMES programmes for type 2 diabetes regarding their characteristics and theoretical underpinnings.^{2, 4, 8-10,S20} Consistent with previous studies, neither programme provided a logic model explaining how the programme was expected to work nor detailed the theory selection process in the programme materials. The connections between the planned BCTs and the targeted constructs were also not always explicitly described. This is in line with recent studies that have indicated the need for improved reporting of theoretical underpinnings and content in behavioural interventions for diabetes.^17-20

4.2 Strengths and limitations

This study employed a systematic, evidence-based approach to retrospectively describe the behaviour change content of two well-established DSMES programmes for type 2 diabetes in Ireland. A key strength of this study is the independent nature of the evaluation of the programme's content, with the analysis being led by behavioural science experts independent from the programme's development team. This independence, coupled with collaboration with key programme stakeholders for the selection, coding and analysis of programme materials, contributes to enhancing the credibility of the study's findings. The methods used in the study were outlined transparently, offering guidance for researchers interested in retrospectively analysing other diabetes or behaviour change established interventions. Another strength of this study is the examination and comparison of the potential programmes' MoAs contributing to the growing body of evidence linking intervention content to MoAs. However, it is important to note that the lack of explicit reporting of theory use in the programmes made it challenging to identify links between BCTs and MoAs, particularly in the CODE programme, which may explain the lower number of MoAs identified in this programme. Nonetheless, by reporting the potential MoAs of the programmes, providers will be able to consider BCTs and MoAs when evaluating programme effects.

Although the inter-rater reliability may have been inflated by the agreed absence of many BCTs, substantial to perfect agreement was found between coders for the BCT coding.^S11 However, the lack of clear and explicit descriptions of BCTs in programme materials made coding challenging, especially because the research team was not involved in the intervention design. Measures were taken to mitigate this challenge, including adopting an inclusive coding approach, double coding parts of the materials, and seeking feedback from key programme stakeholders. However, coding assumptions may have led to the over- or under-reporting of certain BCTs. It is also important to acknowledge that the BCTTv1 is not exhaustive, and alternative tools may have identified additional BCTs by better capturing the active role of participants in the enactment of BCTs.^S21

Similarly, determining the specific self-management behaviours targeted by identified BCTs was challenging because of poor reporting of this information in the materials. A further limitation is that the Mode of Delivery Ontology v1^S1 was employed solely to specify the delivery mode of the programmes and not the delivery mode of single BCTs. Furthermore, the involvement of stakeholders in the coding process was restricted to the DESMOND programme. The public advisory panel also provided feedback exclusively for DESMOND's analysis. This was because the panel had been established before the CODE programme was included in the study, and CODE stakeholders had not formal BCTTv1 training. Finally, the comparison of the frequency of BCTs between programmes warrants caution due to differences in the programmes' structure and organisation. Future studies that aim to compare evidence across studies should consider measures to address the challenges posed by the heterogeneity of the interventions.

4.3 Implications for practice

Our findings highlight the need for improved descriptions and reporting of the programmes' content. Specifically, programmes should explicitly describe how underpinning models/theories were used, which BCTs should be delivered in each session, and specify target behaviours. Based on the new Medical Research Council (MRC) framework guidance, we recommend that providers develop logic models to illustrate how and why programmes are expected to work and use behavioural science taxonomies to guide reporting of the programmes' content.^S22 Using logic models and taxonomies can be beneficial for interventions during commissioning processes and can facilitate intervention implementation, provider training, fidelity assessments, contextual adaptations and future intervention refinements for better clinical outcomes.

4.4 Implications for research

This study provides a foundation for further research on programmes' fidelity, specifically regarding how intervention content is delivered, understood and used by participants.¹⁸ Additionally, programmes can build on these findings to explore the extent to which the BCTs and MoAs identified influence programmes' effectiveness. Similarly, programmes can build on this study to explore the extent to which the mode of delivery of BCTs influences their effectiveness. This can be done by using the Mode of Delivery Ontology v1 to specify the mode of delivery of each BCT identified in the programmes.^S1 Doing so would contribute to enhancing our understanding of which modes of delivery are the most effective in delivering a given BCT for this population. Innovative methodologies such as the Multiphase Optimization Strategy (MOST) or Sequential Multiple Assignment Randomized Trial (SMART) can be employed to investigate the contribution of individual components to programmes' effectiveness.^S23 Finally, to advance the evidence base, a systematic review and meta-analysis of BCTs and intervention characteristics of DSMES for type 2 diabetes should be conducted, along with an exploration of the optimal BCT dosage and feasibility, acceptability and effectiveness of additional BCTs.