Sources of Data
In both SWITCH trials, degludec was compared to insulin glargine 100 units/mL (glargine U100). Importantly, both trials included participants with a higher risk of hypoglycaemia as determined by trial inclusion criteria (i.e. at least one of the following: experienced at least one severe hypoglycemic episode within the last year; moderate chronic renal failure; hypoglycemic symptom unawareness; exposure to insulin for longer than 5 years; or an episode of hypoglycemia (symptoms and/or blood glucose level ≤ 70 mg/dL) within the previous 12 weeks) [3], and thus are considered to be more representative of real-world clinical practice than the group of randomised controlled trials conducted for regulatory purposes, which excluded such patients (BEGIN® clinical trial programme). Secondly, the SWITCH trials used a two-period (32 weeks each) crossover design in which each participant ultimately received both treatments, thereby minimising treatment bias. Finally, the trials were each double-blinded, so that neither participants nor investigators knew which insulin was being injected. Double-blinding is typically difficult to implement in large clinical trials of insulin analogues because they are marketed in proprietary, pen-type delivery devices unique to each product. In the SWITCH trials, each native formulation was administered using indistinguishable vials and syringes. Trial designs and key findings are summarised in Table 1.
Table 1 Characteristics and primary findings of the SWITCH 1 and SWITCH 2 trials
Economic Model
A cost-effectiveness model with QALYs as the effect measure (also called a cost-utility model) was used to compare degludec with glargine U100 based on clinical data from SWITCH 1 [2] and SWITCH 2 [3]. The framework for the cost-effectiveness model has been published previously [12].
The SWITCH trials [2, 3] were conducted in accordance with the Declaration of Helsinki and ICH Good Clinical Practice. Prior to trial initiation, the protocol, consent form and patient information sheet were reviewed and approved by appropriate health authorities and an independent ethics committee/institutional review board. All patients provided signed informed consent.
The modelling process is shown in Fig. 1. The cost-effectiveness of degludec was analysed over a 1-year time horizon. This represents the average annual cost-effectiveness in a steady state and not necessarily only the results after 1 year. This short-term approach based on the different rates of hypoglycaemia and actual doses of insulin used is appropriate since the treat-to-target efficacy trials required by regulatory bodies do not generally result in differences in glycaemic control between comparators; therefore, modelling long-term glycaemic control would not result in differences other than random variation. The model uses only treatment effects for which a statistically significant difference between the treatment arms is documented, and assumes that all other differences are due to random variation (i.e. equivalent to classical statistical tests where the null hypothesis could not be rejected). However, nonsignificant differences were explored in the sensitivity analysis. As the time horizon for the analysis was 1 year, no discounting (adjustment of future values to the present value) was applied. The analysis was conducted from the perspective of the UK National Health Service.
Insulin Use
Units of basal insulin used per day for the degludec and glargine U100 treatment groups were captured from the clinical trial data, and dose reductions were estimated with log-transformed end-of-trial doses, employing treatment, period, dosing time and visit as fixed effects, subject as a random effect and the log-transformed baseline dose as a covariate. For type 1 diabetes mellitus, the glargine U100 basal dose (40.58 units/day) and degludec/glargine U100 basal dose ratio (0.97 [0.94; 0.99] 95 % confidence interval [CI], p < 0.05) were derived from SWITCH 1. The bolus dose used in the glargine U100 arm (31.93 U/day) and the bolus dose ratio for the two arms (degludec/glargine U100) (0.97 [0.94; 1.01] 95 % CI, p > 0.05) were also derived from SWITCH 1. The dose ratios, adjusted for covariates, were used to calculate the corresponding doses in the degludec arm, as can be seen in Table 2. For type 2 diabetes mellitus, only basal insulin was used (BOT, basal-only therapy). The glargine U100 basal dose (82.66 units/day) and degludec/glargine U100 basal dose ratio [0.96 (0.94; 0.98) 95 % CI, p < 0.05] were derived from SWITCH 2.
Table 2 Use of basal and bolus insulin in the SWITCH 1 and SWITCH 2 trials
Hypoglycaemic Events
Hypoglycaemic events were obtained by pooling data from the two full crossover periods in the SWITCH 1 [2] and SWITCH 2 trials [3]. For the purposes of our cost-effectiveness model, rates of hypoglycaemia were divided into three mutually exclusive groups: severe events, nonsevere events occurring during the day (diurnal including unknown time) and nonsevere events occurring during the night (nocturnal). These groups were meaningful in terms of costs and impact on quality of life. By ensuring that the groups were mutually exclusive due to stringent criteria for defining hypoglycemia, double counting of events was avoided. In both trials, non-severe hypoglycaemia was defined as a symptomatic event with a confirmed blood glucose level < 3.1 mmol/L (56 mg/dL). Hypoglycaemic events were pre-specified: nocturnal hypoglycaemia was defined as an event occurring between 00:01 and 05:59 am, and severe hypoglycaemia was defined as an episode of hypoglycaemia requiring medical assistance and/or assistance from a third party [13].
The number of events for the rate ratios were analysed using a Poisson model with logarithm of the exposure time as offset (Table 3). The model included treatment, period, sequence and dosing time as fixed effects and subject as a random effect. Rate ratios and 95% CIs used in the model for various types of hypoglycaemic event are shown in Table 3 for type 1 and type 2 diabetes mellitus. The degludec/glargine U100 rate ratios, adjusted for covariates, were applied to the glargine U100 hypoglycaemia rates per person-year to estimate the degludec hypoglycaemia rates, as can be seen in Table 3.
Table 3 Calculation of hypoglycaemic event rates
Treatment Costs
Cost of insulin was based on the Monthly Index of Medical Specialties (MIMS) [14] for April 2018, which includes the updated price of insulin glargine U100. The unit costs were multiplied by the number of units per day from Table 2. The number of needles and self-measured blood glucose (SMBG) tests were assumed to be the same in both arms (one needle and one SMBG test per injection).
Cost of Hypoglycaemia
The costs of nonsevere nocturnal, nonsevere diurnal and severe hypoglycaemic events were derived from two sources. The cost of a severe hypoglycaemic event (type 1 diabetes mellitus, £178; type 2 diabetes mellitus BOT, £427) was calculated based on a study specifically designed to evaluate the cost of severe hypoglycaemia across Germany, Spain and the UK [15]. The cost estimate from the UK (in pounds sterling or GBP) was derived by excluding the indirect costs and adjusting for inflation using the estimates from the Hospital & Community Health Services index [16]. The cost of a nonsevere hypoglycaemic event (daytime: type 1 diabetes mellitus, £2.44; type 2 diabetes mellitus BOT, £3.48; nocturnal: type 1 diabetes mellitus, £3.04; type 2 diabetes mellitus BOT, £5.56) was calculated based on a real-world study that investigated the frequency of self-reported nonsevere hypoglycaemic events across 11 European countries, including the UK (the Hypoglycaemia in Insulin-Treated Patients (HIT) study) [17].
Impact of Hypoglycaemia on QALYs
Quality-adjusted life-years (QALYs) were calculated by applying a disutility per hypoglycaemic event. A large-scale time trade-off (TTO) study was used to obtain the disutility incurred per hypoglycaemic event, and the following disutilities were obtained: 0.0054 and 0.0077 for nonsevere daytime and nonsevere nocturnal events, respectively, and 0.0623 for severe events based on preferences of the UK general population [18]. Severe events were not divided by the time of day at which they occurred. This was because there were no significant differences reported in the cost or utility for these events, so the impact of a severe hypoglycaemic episode was considered the same irrespective of the time at which it occurred. To estimate the incremental impact of degludec, the disutility per hypoglycaemic event was multiplied by the number of events observed in each treatment group and subtracted from a baseline level.
Outcomes
Results are presented as costs in GBP and a breakdown of the costs, effects in QALYs and incremental cost-effectiveness ratio (ICER) as cost per QALY. Two types of sensitivity analyses were conducted. The first was a one-way sensitivity analysis in which we analysed the effect of single changes in the most important parameters: hypoglycaemia rates, use of nonsignificant rate ratios, changes in hypoglycaemia cost parameters, SMBG tests per week for degludec, hypoglycaemic event disutility, insulin doses, needles and work loss arising from hypoglycaemia. The second was a probabilistic sensitivity analysis to capture the uncertainty of the results caused by statistical uncertainty with respect to the parameter inputs. Probabilistic sensitivity analysis allows the model parameters to be varied simultaneously, based on the parameters’ standard error distributions. The probabilistic sensitivity analysis was conducted using the uncertainties for all stochastic parameters in the models. Each probabilistic sensitivity analysis was run with 1000 iterations. Parameters, standard errors and distributions can be found in Table S1 of the Electronic supplementary material (ESM).