A within-trial cost analysis was conducted using the 2 year data from DiRECT, including both the intervention costs and routine healthcare resource use measured during the time course of the study, for all participants (including those who did not complete the trial or who were unsuccessful in achieving remission). Lifetime cost-effectiveness was then estimated by predicting time to relapse (i.e. re-emergence of diabetes, assumed to be permanent thereafter) among those who were in remission at 2 years, applying the mean management costs for type 2 diabetes incurred within the UK National Health Service (NHS), under clinical guidelines which tend to favour older, less expensive medications. An NHS perspective for costing was adopted for both within-trial and lifetime analyses. Costs are presented in 2018 UK prices (£).
Ethics approval was granted in January 2014 by the West of Scotland Research Ethics Committee 3, with approvals by the NHS health board areas in Scotland and by clinical commissioning groups in Tyneside, UK.
Two year within-trial economic analysis
DiRECT intervention costs over the 2 year trial period included costs for initial training of practitioners (dietitians or nurse practitioners, who each received a total of 16 h face-to-face training from Counterweight-Plus instructors), sachets of low-energy formula diet, practitioner monitoring appointments and the tailored Counterweight-Plus workbooks issued to each participant (Table 1). Costs of the sessions included practitioners’ attendance time and standard Counterweight-Plus materials. The number of sachets issued to each participant and the number and duration of practitioner appointments were collected prospectively throughout the study. Full costs for participant Counterweight-Plus workbooks were applied irrespective of participants’ persistence with the programme, the details of which are described elsewhere [8, 10, 12].
Details of all primary and secondary care visits for each of the participants were obtained directly from the participating general practitioner (GP) practice records. The costs of these were calculated using the recorded duration of contact for each appointment. Medication use was costed based on dose, frequency and start and end dates of individual participants’ medication records in each participating GP practice. Hospitalisation costs were estimated by matching reason for admission and recorded length of stay in DiRECT to the appropriate NHS reference cost (excluding excess bed-days, where length of stay recorded in DiRECT exceeded the national average). Unit costs were obtained from published national sources (Personal Social Service Research Unit , NHS reference costs , or Information Services Division Scotland , and British National Formulary ) and have been reported previously .
Statistical analysis was conducted based on the intention-to-treat principle. Missing data were minimal, as all resource use data were obtained directly from participating GP practices. Five intervention (3%) and three control (2%) participants relocated with loss to follow-up during the study. Their medication use was assumed to continue as at their last available records, and other healthcare resource use was assumed to be zero after relocation. Twenty intervention (13.4%) and six control (4%) participants had no data available for the 2 year remission outcome. In line with the primary outcome analysis, these participants were assumed not to have achieved remission.
Mean costs were calculated for each group, with clustering-adjusted SEs for each cost item. Incremental cost per remission at 2 years was reported as the difference in the groups’ total 2 year costs, divided by the difference in diabetes remission rates. All analyses were undertaken in Stata/MP, version 14.2 (StataCorp LP, USA), with 95% CIs based on 1000 non-parametric bootstrap iterations.
Long-term outcomes were projected for each treatment arm in DiRECT. A three-state model (remission, diabetes, death) was constructed. Individuals enter the model with existing diabetes. After 1 year a proportion achieve remission but are subject to relapse in future years. The proportion remaining in remission over time was estimated based on the rate of relapse observed in year 2; however, all participants were assumed to relapse after a given number of years (a maximum period of 10 years of remission in the base case). Life expectancy was calculated for each potential year of remission by applying rates of mortality for people free of diabetes up to the year of relapse and with diabetes thereafter. Along with life expectancy, quality-adjusted life expectancy and healthcare costs were estimated conditional on each potential year of relapse.
Life expectancy was calculated based on rates of mortality in people free of diabetes (N = 2.75 million) and with diabetes (N = 272,597), based on a recent UK study  which reported mortality rates by sex and 5 year age bands during the period 2012–2014 in Scotland. Reported life expectancy for men aged 55–59 years was 23.0 and 26.0 years with diabetes and free of diabetes, respectively, and 24.4 and 28.7 for women.
QALYs were calculated by applying standard UK age-dependent health state utility population norms . These were assigned directly to people in remission from diabetes. For people not in remission, including those who had relapsed, these age-dependent health state utilities were reduced using a constant multiplier of 0.925 to reflect a decrement due to diabetes. This was estimated based on the mean population score (0.828) and the regression coefficient for diabetes (−0.0621) in the US Medical Expenditure Panel Survey catalogue of UK EuroQol EQ-5D scores . This estimate was employed rather than being based on data from DiRECT, for reasons discussed below.
The lifetime healthcare costs associated with diabetes were compiled first from the measured costs for the first 2 years of DiRECT, to which were added further costs for ongoing weight management for participants remaining in remission and long-term healthcare costs associated with diabetes. Long-term diabetes-related healthcare costs were assumed to increase linearly with duration of diabetes (i.e. time since relapse), over 15 years, from £1250 in the year of diagnosis to £3117 after 15 years, based on a UK cost of diabetes study [2, 21]. No further increase in long-term healthcare costs beyond 15 years was applied. Long-term healthcare costs due to diabetes were applied to the proportion of people projected to be in the diabetes state each year after year 2.
We compared the incremental cost and QALYs for the intervention and control arms over the 2 year follow-up of DiRECT and over a lifetime horizon. Sensitivity analysis was undertaken for the long-term analysis, including exploring the impact of alternative relapse rates and maximum assumed durations of diabetes remission. Probabilistic sensitivity analysis was performed using 1000 bootstrap iterations of the DiRECT data (remission and 2 year costs), and Monte Carlo simulations for other variables (e.g. long-term mortality rates). Model input data are summarised in Table 2; all costs and outcomes beyond year 1 were discounted at the standard UK annual rate of 3.5%.