Long-term projections of clinical and cost outcomes were performed from a healthcare payer perspective using the IQVIA CORE Diabetes Model (version 9.0), a proprietary, validated, internet-based, interactive computer model developed to determine the long-term health outcomes and economic consequences of implementing interventions in the treatment of type 1 and type 2 diabetes mellitus (accessible at http://www.core-diabetes.com) [16, 17]. The architecture, assumptions, features and capabilities of the model have been previously published . Validation studies of the model have been published both in 2004 and more recently in 2014 [17, 18].
Model outputs include time to onset and cumulative incidence of complications, life expectancy, quality-adjusted life expectancy (QALE; expressed in quality-adjusted life years [QALYs]), direct costs and, where required, incremental cost-effectiveness ratios (ICERs), which describe the cost per additional unit of effectiveness gained for the intervention versus the comparator. In comparisons where an intervention is associated with cost savings while providing greater clinical benefits, no calculation of an ICER is required and the intervention is considered to be dominant versus the comparator.
Analyses were performed over patient lifetimes (up to 50 years), as recommended in the guidelines for the cost-effectiveness assessment of interventions for type 2 diabetes, to ensure all relevant diabetes-related complications and their impact on clinical and cost outcomes were captured . The UKPDS 68 risk equations were applied to predict model outcomes. Background mortality was captured based on UK-specific life tables published by the World Health Organisation (Electronic Supplementary Material [ESM] Table S1) . Health-state utilities and event disutilities were based on published sources (ESM Table S2) [21,22,23,24,25,26,27].
This article is based on previously conducted studies and does not contain any studies with human participants or animals performed by any of the authors.
Baseline cohort characteristics and treatment effects were sourced from the PIONEER 2, 3 and 4 trials for comparisons of oral semaglutide 14 mg with empagliflozin 25 mg, sitagliptin 100 mg and liraglutide 1.8 mg, respectively (ESM Table S3; Table 1). PIONEER 2 enrolled people with type 2 diabetes with HbA1c values between 7.0 and 10.5% (53–91 mmol/mol) who were receiving metformin; PIONEER 3 enrolled people with type 2 diabetes with HbA1c values between 7.0 and 10.5% who were receiving metformin with or without a sulfonylurea; and PIONEER 4 enrolled people with type 2 diabetes with HbA1c values between 7.0 and 9.5% (53–80 mmol/mol) who were receiving metformin with or without an SGLT2 inhibitor. The PIONEER trial programme used two estimands, namely the treatment policy estimand and the trial product estimand, to address two different efficacy questions. The treatment policy estimand reflected the intention-to-treat principle by including all study participants randomly assigned to each treatment, using data regardless of discontinuation of study medications and/or use of additional anti-diabetic medications during the trial [28, 29]. In contrast, the trial product estimand assessed treatment effects under the assumption that patients received the study drug for the duration of the trial and did not receive any additional anti-diabetic medications, aiming to reflect the effects of the study medications without the confounding effects of rescue medication or any other changes in glucose-lowering medication . To match the annual cycle length of the model, and to avoid the confounding impact of additional anti-diabetic medications on clinical and cost outcomes, the analyses were performed using the 52-week data evaluated by the trial product estimand. The impact of using data evaluated by the treatment policy estimand was explored in a sensitivity analysis.
Treatment Switching and Long-Term Parameter Progression
Following application of the treatment effects in the first year of the analysis, HbA1c was modelled to follow the UKPDS progression equation, and patients were assumed to receive oral semaglutide or comparator treatment until HbA1c exceeded 7.5% (58 mmol/mol), which is the threshold for treatment intensification defined in the NICE guidelines . At this stage, treatment with oral semaglutide or the comparator was discontinued, and patients were assumed to intensify treatment to basal insulin, with a reduction in HbA1c based on an insulin-naïve population derived from the “Core” multivariate equations estimated by Willis et al. . HbA1c was subsequently modelled to follow the UKPDS progression equation for the remainder of patient lifetimes. This approach was chosen to mirror the HbA1c progression used by NICE for evaluating SGLT2 inhibitors as monotherapy in the UK and to reflect common clinical practice in which, due to the progressive nature of type 2 diabetes, glycaemic control cannot be maintained indefinitely by the addition of one medication [7, 31]. Variations in the thresholds for treatment switching and further treatment intensification to basal–bolus insulin were explored in sensitivity analyses.
Body mass index (BMI) benefits were assumed to persist while patients received either oral semaglutide or comparator treatment, before reverting to baseline following intensification to basal insulin therapy. Therefore, no difference in BMI was seen between the patient arms following treatment intensification with basal insulin.
Changes in blood pressure and serum lipids were assumed to follow the natural progression algorithms built into the IQVIA CORE Diabetes Model in all arms, based on the UKPDS or Framingham data (as described by Palmer et al. ), following application of the treatment effects in the first year of the analysis. Hypoglycaemia rates following treatment intensification were based on published data, with non-severe and severe hypoglycaemic events projected to increase to 4.08 and 0.10 events per patient per year, respectively .
Costs were accounted from a UK healthcare payer perspective. Captured direct costs included pharmacy costs, costs associated with diabetes-related complications and patient management costs (ESM Tables S4, S5). The annual acquisition cost of oral semaglutide was assumed to be the same as that of once-weekly semaglutide, based on the similar level of pricing seen between the GLP-1 analogues in the US market. Costs of other included medications and consumables were based on published list prices (sourced in July 2019), while costs of diabetes-related complications were identified through a 2017 literature review and updated or inflated where necessary to the most recent costs available (2018 GBP) using published NHS diagnosis-related groups and the healthcare inflation index published by the Personal Social Services Research Unit [33,34,35,36,37,38,39,40,41,42]. No self-monitoring of blood glucose (SMBG) testing costs were associated with oral semaglutide, empagliflozin, sitagliptin or liraglutide, as all these interventions are associated with low rates of hypoglycaemia and, consequently, little to no SMBG testing would be required. No needles were required for the administration of oral semaglutide, empagliflozin or sitagliptin as these medications are administered orally, but one needle per day was required for the administration of liraglutide. Following treatment intensification to basal insulin (assumed to be insulin Abasaglar®, the most widely used biosimilar of insulin glargine in the UK), patients were assumed to require one SMBG test per day and to use one needle per day for the administration of basal insulin.
The extrapolation of clinical results by modelling the long-term consequences is associated with uncertainty. Sensitivity analyses were therefore performed on key parameters in the modelling analysis to assess the robustness of the base case findings. Sensitivity analyses conducted for all comparisons included: applying only statistically significant differences between the treatment arms; shortening the time horizon of the analyses to 35, 20 and 10 years (for which it should be noted that some patients were still alive at the end of the modelling period and, therefore, not all costs and consequences were captured); applying discount rates of 0 and 6% in separate analyses; applying the upper and lower limits of the 95% confidence intervals for the estimated treatment differences in HbA1c and BMI in separate analyses; maintaining BMI treatment effects for patient lifetimes; altering the HbA1c threshold for treatment intensification to 7.0% (53 mmol/mol) and 8.0% (64 mmol/mol); applying a second treatment intensification step to basal–bolus insulin at an HbA1c threshold of 7.5% (58 mmol/mol); exploring the effect of applying alternative basal insulin costs (insulin neutral protamine Hagedorn [NPH], Semglee® [Mylan, biosimilar of insulin glargine] and Lantus® [insulin glargine]) following treatment intensification; increasing and decreasing the annual acquisition cost of oral semaglutide by 5% in separate analyses; application of the liraglutide 1.2 mg price in the liraglutide arm of PIONEER 4; increasing and decreasing the costs of complications by 10% in separate analyses; applying an alternative cost of stroke in the year of the event and in subsequent years, based on a publication by Patel et al. ; applying the UKPDS 82 risk equations to predict model outcomes; application of alternative disutilities for increases in BMI (based on a publication by Lee et al. ) and hypoglycaemic events (based on publications by Currie et al.  and Lauridsen et al. ); application of the 26-week clinical data; and application of data evaluated by the treatment policy estimand from the PIONEER 2, 3 and 4 clinical trials [13,14,15].
Probabilistic sensitivity analyses (PSA) were also performed using a second-order Monte Carlo approach. Cohort characteristics, treatment effects and complication costs and utilities were sampled from distributions, with cohorts of 1000 patients run through the model 1000 times.