The Swedish Institute for Health Economics (IHE) Cohort Model of Type 2 Diabetes was used for the cost-effectiveness analyses, with Swedish data for development of complications, costs and patient benefits associated with T2DM. A 40-year perspective was used to capture all costs and effects for the remainder of the patient’s life (a societal perspective, including healthcare costs and productivity losses, was used in the main analysis, and a healthcare perspective was used in the sensitivity analysis). A 3% discount rate was applied to costs and effects as per guidelines by the Swedish Dental and Pharmaceutical Benefits Agency (Tandvårds- and läkemedelsförmånsverket [TLV]) . Outcomes measured in cost per quality-adjusted life-years (QALYs) were compared with the thresholds set out by the Swedish National Board of Health and Welfare to determine whether an intervention was cost effective. This board consider a cost per QALY below Swedish krona (SEK)100,000 as low, whereas incremental cost-effectiveness ratios (ICERs) between SEK100,000 and 500,000 per QALY are deemed as moderate . Any assumptions used within the model were made because data were lacking; assumptions are intended to be conservative to maintain objectivity.
The model was developed by the Swedish IHE and used Markov health states to capture key microvascular and macrovascular complications and early T2DM-related death. A schematic overview of the model is presented in Fig. S1 in the Electronic Supplementary Material (ESM). It was designed in Microsoft® Excel 2013 using Visual Basic for Applications and has been externally validated as described in Lundqvist et al. . The model includes multiple variables such as cohort baseline characteristics (e.g. age, sex, ethnicity, biomarkers and complications), unit costs, utility weights and choice of risk equations, with a treatment algorithm linked to HbA1c defined by the user in terms of treatments used and associated expected effects [25, 26]. Annual transition probabilities govern the progress of patients through the different simulated health states linked to, and influenced by, patient demographics, development of complications and other relevant variables.
Costs and utility weights are applied to the cohort in each annual cycle. Outcomes include the number of life-years, QALYs, costs, ICERs, and cumulative incidence of complications and adverse events. The model provides the option of selecting risk equations; the Swedish NDR macrovascular equations  and the UKPDS-OM2 (UK Prospective Diabetes Study Outcomes Model 2) mortality risk equations  were chosen to reflect the Swedish patient population. The model also includes separate analyses for men and women and for smokers and non-smokers, since the risk equations for these groups differ. Non-smoking men were selected for the base-case analysis, whereas sensitivity analyses were performed for the other populations.
Comparators and Clinical Data
Comparators used in the cost-effectiveness analysis included a number of options for treatment intensification: up-titration of basal insulin, basal–bolus insulin and GLP-1 receptor agonist added to basal insulin. The comparators for up-titrated basal insulin were insulin glargine and NPH insulin; the comparators for basal–bolus regimens were insulin aspart added to both insulin glargine and NPH insulin; and the comparators for the GLP-1 analogue were liraglutide added to both insulin glargine and NPH insulin.
As the clinical study programme for IDegLira (Xultophy®) has reported only two comparative studies with insulin degludec and liraglutide in patients with diabetes uncontrolled with basal insulin (DUAL™ II and DUAL™ V) but none with a basal–bolus regimen or combination of GLP-1 receptor analogue and basal insulin, clinical data for the cost-effectiveness evaluation were based on an indirect treatment comparison (ITC)  of several studies [27,28,29,30,31,32]. For information on the ITC methodology, please see the ESM. Clinical data for all NPH insulin comparisons in the analysis were assumed to be the same as for insulin glargine, with the exception of the hypoglycaemia rates for up-titrated NPH insulin only (hypoglycaemia levels for up-titrated insulin glargine and NPH insulin were based on the results of a separate meta-analysis) . Given the lack of data comparing hypoglycaemia levels between insulin glargine and NPH insulin in combination with insulin aspart or liraglutide, we conservatively assumed that these hypoglycaemia rates are the same (Table 1).
Treatment intensification was based on the assumption that treatment will be intensified when the HbA1c level reaches 8.8%. This figure represents the average HbA1c level at which investigators in the DUAL™ II study deemed patients suitable for intensification and is in line with national treatment guidelines for diabetes that recommend patients with HbA1c levels >8.6% be prioritised for intensification . The same HbA1c level was used as baseline both for initiating IDegLira and for switching to basal–bolus therapy. Additionally, we assumed that patients with a blood pressure of ≥140 mmHg would receive an angiotensin-converting enzyme (ACE) inhibitor (20 mg daily) as per diabetes treatment guidelines , which would lower blood pressure by 5%, adjunct statins (40 mg daily) once their serum lipids (low-density lipoprotein [LDL]) reached 2.5 mmol/l and a fibrate once their triglycerides reached 1.7 mmol/l.
Data from DUAL™ II  were used for baseline values for the model (Table 2). Additional baseline variables such as heart rate, white blood cells, and estimated glomerular filtration rate (eGFR) were obtained from the UKPDS-OM2 study . The prevalence of cardiovascular complications (with the exception of atrial fibrillation) was obtained from the NDR cohort on which the risk equations are based (personal communication; Ali Kiadaliri, Lund University).
Efficacy endpoints were taken from the ITC  and only included clinical endpoints where there was a statistically significant difference between treatments. Otherwise, the comparator was assumed to be as effective as IDegLira (Table 1).
All costs, except drugs and consumables, are expressed as SEK, year 2013 values, and adjusted for inflation using the Swedish healthcare consumer price index yields as necessary  (Table S1 in the ESM). The current analysis was conducted in 2014 so we used the 2013 costing year. We acknowledge the costing year might affect the results of the analysis; however, we believe that expressing costs in SEK, year 2013 values, does not affect the results significantly because of the low inflation rate (0.7%)  in 2013 in Sweden compared with 2016. Prices of drugs and consumables were obtained from the TLV price database (Table 3) .
The drug costs were based on pharmacy retail prices (PRPs) in September 2014, and daily doses from the clinical studies were used for IDegLira and its comparators. The price of IDegLira is billed per dose step rather than unit, which accounted for varying doses of insulin degludec and liraglutide. Doses reported by Freemantle et al.  were adjusted using statistical modelling to account for differences in baseline characteristics between the study cohorts. We assumed the dose of human insulin was the same as that of insulin glargine for all treatment arms with a basal insulin when dosing once daily, given there has been no documentation of any difference in dose and NPH insulin is more likely to be administered twice daily with an associated increase in dose .
All patients were assumed to be using metformin 1500 mg daily (highest tolerable dose) as an adjuvant to the study medication as per diabetes treatment guidelines . Patients who received IDegLira, basal insulin or GLP-1 added to basal insulin were assumed to perform one blood glucose test per day, whereas patients receiving basal–bolus regimens were assumed to perform four tests daily. The number of blood glucose tests for each comparator was based on the minimum number of daily doses required, i.e. one daily dose for IDegLira, basal insulin or GLP-1 added to basal insulin and four daily doses for basal–bolus insulin . The unit cost of blood glucose tests (SEK2.58) was obtained by adding up the prices (PRP excluding value-added tax [VAT]) of the test strips using the lowest price per piece and the lancets using the lowest price according to the TLV price database .
Costs for an episode of severe hypoglycaemia were based on figures from a study by Jönsson et al. . A weighted mean cost (SEK1970, year 2013 values) for each severe hypoglycaemia episode was calcualted using the conversion rate €1 = SEK9.21, August 2006 values, and adjusting for inflation. For mild hypoglycaemia, we used data from Geelhoed-duijvestijn et al. , which enabled calculation of the cost per mild hypoglycaemia episode at SEK50.
The costs of diabetes-related complications (e.g. heart failure, stroke, etc.) were identified from a published Swedish cost-effectiveness analysis (Table S1 in the ESM)  and adjusted for inflation to 2013 monetary values. The analysis also included indirect costs of diabetes-related complications. Working age was assumed to be 20–65 years, with an average annual income of SEK394,800 for men and SEK340,800 for women . A working year was assumed to consist of 250 working days. Data on sick days due to various diabetes complications were obtained from a Danish registry data analysis that comprised 34,882 patients with diabetes . If we were unable to determine the number of sick days, a conservative assumption was made that the number of sick days was 0. The health economic model does not differentiate between sick days during the ‘first’ and ‘subsequent’ years for each condition; however, this was reported in the Danish study. We therefore used the ‘subsequent years’ values to ensure a conservative analysis, as these were consistently lower. The analysis used a human capital approach.
Utility values for various diabetes-related complications used in this cost-effectiveness analysis were derived from published studies (Table 4). Utilities associated with treatment complexity and utilities associated with the short-term effects of changes in HbA1c levels were derived from the results of a time trade-off (TTO) study carried out in Sweden, Denmark and the UK . The disutility of blood glucose testing was obtained from a TTO study carried out in Sweden, the UK and Canada . For the effect of weight loss, as measured by body mass index (BMI), we used the lower of the utility values found in published studies (−0.006) [45,46,47] for the main analysis and the value from the TTO study (−0.021) in the sensitivity analysis. Data on the extent to which patients’ perceived utility is affected by hypoglycaemia episodes were derived from results for Swedish patients with T2DM in an extensive web-based TTO study [48, 49].
One-Way Sensitivity Analyses
As the base case used non-smoking men to model the cost effectiveness of IDegLira, we performed sensitivity analyses on patient characteristics with risk equations that differ from those for non-smokers and men, i.e. smokers and women. Various other characteristics, such as a baseline HbA1c of 9% and a baseline BMI of 25 and 35 were also modelled in sensitivity analyses to investigate their effects on the cost effectiveness of IDegLira. Other sensitivity analyses were performed for clinical endpoints (HbA1c drift for GLP-1 of 0.08%, no difference in HbA1c, no difference in HbA1c but different complexity of regimen, depending on the number of daily injections, half the difference in HbA1c, no difference in BMI, no difference in hypoglycaemia for insulin glargine/NPH insulin, lower efficacy for liraglutide 1.2 mg), costs (healthcare perspective and using the price of a higher dose of liraglutide) and patient utilities (impact of BMI using a utility value of −0.021, no impact of complexity vs. basal–bolus). These analyses and the rationale for undertaking them are outlined in Tables S2–S4 in the ESM.
Probabilistic Sensitivity Analyses
Probabilistic sensitivity analyses (PSAs) were conducted to assess the cost effectiveness of IDegLira when uncertainty of input variables were examined simultaneously. The absolute treatment effect on the HbA1c level, the annual absolute drift of the HbA1c level, HbA1c, initial absolute treatment effects on other biomarker levels and absolute drift of the other biomarker levels were varied using a normal distribution. To avoid negative values, event rates for hypoglycemia and other adverse events were varied using a log-normal distribution. Standard errors (SEs) of variables were used in the analyses where available (Table S3 in the ESM). If the SE of a variable was not available, assumptions were made (Table S3 in the ESM). All assumptions were made in consultation with TLV in conjunction with the reimbursement application of IDegLira. Each simulation employed 500 iterations.