“How nationwide roll-out informed study design”
The All-Wales Diabetes Prevention Programme (AWDPP) was evaluated to see how it affected average (preceding two-three months) blood sugar levels, measured by HbA1c, in people with prediabetes over 12 months. For this evaluation, participants were also grouped into recognised blood sugar categories indicative of type 2 diabetes risk (26), using HbA1c, to show whether their levels were within the healthy (normoglycaemia), prediabetes, or type 2 diabetes range:
The study also looked at whether participants’ blood sugar category had stayed the same (prediabetes)/improved (normoglycaemia) or worsened over the year (type 2 diabetes).
The study used a stepped-wedge design, with primary care practices switching from usual care to offering the AWDPP at different time points across three stages between June 2022 and June 2023. This phased roll-out was practical for implementation and ensured all practices eventually received the intervention. The design also allowed each practice to act as its own control, enabling comparison of outcomes before and after adoption.
HbA1c was measured at baseline and after 12 months using blood tests processed in NHS laboratories, giving a reliable picture of average blood sugar over the previous two to three months (27).
A total of 1,970 adults with HbA1c levels in the prediabetes range (within the last two to three months) were invited to participate in the AWDPP. These individuals were recruited from 59 primary care practices across Wales, representing six of the seven health boards (one health board excluded due to deviation from AWDPP protocol (8)). This cohort is referred to as the “treatment group”.
For contrast, 1,494 people from the same primary care practices and health boards who were within the prediabetes range and had 12-month follow-up HbA1c before the programme became live in their registered practice were included as a historical “comparison group”.
All analyses used an intention-to-treat approach, examining changes in blood sugar levels and corresponding blood sugar category between the pre-implementation (comparison) and implementation phases. The analysis was structured using Hussey and Hughes’s stepped wedge framework (28), adjusting for group allocation, practice onboarding dates, baseline HbA1c, duration of exposure (how long a practice had been delivering the intervention), and practice-level clustering. Changes in blood sugar category were modelled using generalised estimating equations, while changes in blood sugar levels were assessed using a generalised linear mixed model. Relative risk was calculated to compare the likelihood of these outcomes between the treatment and comparison groups.