Delayed start to feasibility study
Several challenges delayed feasibility practice recruitment, mainly finalising the practice-level data collection process (which impacted protocol completion) and the NHS cyber attack in May 2017 (causing reduced practice capacity for research, heightened awareness of information governance and delays in testing data extraction). Attempts to improve this included approaching research site initiative practices and federations, and diabetes and research leads and prominent local professors providing letters of endorsement for the study that were sent to practices. The trial manager also tried to visit CCGs, but only one CCG facilitated a visit.
These delays shortened the timeline between the feasibility study and RCT. Consequently, all feasibility study learnings could not be implemented into the RCT, particularly implementing the intervention at provider level. It may have been preferable to maintain the planned timeline between the feasibility study and RCT to allow implementation of feasibility study learnings by extending the programme length or attempting to gain the time back elsewhere.
Study design complexity
Participating practices had sizable requirements regarding study-related activities (e.g. questionnaire postage; enabling data extractions), which was compounded by embedders sometimes undertaking study-related activities, so the distinction was not always clear. This impacted on intervention engagement in some practices that could not separate the two activities, and on the ability of embedders to deliver the intervention as intended. This may have been reduced if the intervention was wholly at provider-level, thus a step removed from the research data collection.
Embedding of the embedders
Ethnography analyses showed that embedders found it difficult to get ‘buy-in’ and commitment for embedding activities, partly due to physical distance (neither embedder were local to where the intervention was implemented) and not being employed by local providers. Therefore, a lot of embedder availability was spent travelling, rather than continuing work and building better relationships. Furthermore, embedders embedded or employed within provider teams may have needed less regular reintroductions to themselves and the toolkit. This highlights the importance of considering study delivery practicalities when recruiting study-specific roles.
Primary outcome choice
During RCT planning, the primary outcome was much discussed. SSME attendance (the intervention’s primary target) would be a natural choice, but had limitations within a wait-list design because local commissioning rules often prevented an individual attending SSME multiple times. Additionally, our findings show that SSME attendance data are only moderately complete, which would affect their validity as a primary outcome.
The primary outcome of HbA1c was chosen because attending SSME is associated with reduced HbA1c, and this outcome could change within each step of the wait-list design. This outcome is, however, a proxy and is one step removed from the intervention’s primary purpose, meaning that it may be insensitive to change.
Potential contamination
Some of the intervention was delivered at provider level, conferring a potential contamination risk (i.e. control practices may have unintentionally received some of the intervention early via their provider). To limit contamination, we requested that the provider-level aspects of the intervention were targeted at practices participating in the Embedding Package as far as possible, and only aimed at wait-list practices when they were receiving the intervention. This in itself may have hindered intervention engagement as it may have been deemed too complicated. A rationale for embedders maintaining a practice-level focus (despite feasibility study findings that provider level would be preferable) was to limit contamination. With more time between the feasibility study and RCT (see Delayed start to feasibility study), the RCT could have been redesigned to facilitate this change without contamination. The likely impact of any contamination would have been to reduce the effect size estimating the intervention versus control difference.
Lack of fidelity assessment
An intervention delivery fidelity assessment was not planned. The data in Appendix 1 partly mitigate this limitation by providing some understanding of delivery.