Cost-effectiveness of once daily GLP-1 receptor agonist lixisenatide compared to bolus insulin both in combination with basal insulin for the treatment of patients with type 2 diabetes in Norway

Abstract

Background:

Lixisenatide is a potent, selective and short-acting once daily prandial glucagon-like peptide-1 receptor agonist which lowers glycohemoglobin and body weight by clinically significant amounts in patients with type 2 diabetes treated with basal insulin, with limited risk of hypoglycemia.

Objective:

To assess the cost-effectiveness of lixisenatide versus bolus insulin, both in combination with basal insulin, in patients with type 2 diabetes in Norway.

Methods:

The IMS CORE Diabetes Model, a non-product-specific and validated simulation model, was used to make clinical and cost projections. Transition probabilities, risk adjustments and the progression of complication risk factors were derived from the UK Prospective Diabetes Study, supplemented with Norwegian data. Patients were assumed to receive combination treatment with basal insulin, lixisenatide or bolus insulin therapy for 3 years, followed by intensification of a basal–bolus insulin regimen for their remaining lifetime. Simulated healthcare costs, taken from the public payer perspective, were derived from microcosting and diagnosis related groups, discounted at 4% per annum and reported in Norwegian krone (NOK). Productivity costs were also captured based on extractions from the Norwegian Labor and Welfare Administration. Health state utilities were derived from a systematic literature review. Sensitivity and scenario analyses were performed.

Results:

Lixisenatide in combination with basal insulin was associated with increased quality-adjusted life years (QALYs) and reduced lifetime healthcare costs compared to bolus insulin in combination with basal insulin in patients with Type 2 diabetes, and can be considered dominant. The net monetary benefit of lixisenatide versus bolus insulin was NOK 39,369 per patient. Results were sensitive to discounting, the application of excess body weight associated disutility and uncertainty surrounding the changes in HbA1c.

Conclusions:

Lixisenatide may be considered an economically efficient therapy in combination with basal insulin in the Norwegian setting, due to cost savings, weight loss and associated gains in health-related quality of life.

Keywords::

• Cost-effectiveness
• Lixisenatide
• Modeling
• Norway
• Type 2 diabetes

Introduction

Diabetes is a chronic non-communicable condition that can lead, if left unmanaged, to serious health conditions. These include cardiovascular disease, kidney disease, eye diseases including blindness, nerve damage and lower-extremity amputation. Diabetes mellitus contributes to premature mortality and a lower quality of life, and leads to increases in medical costs for patients and healthcare systems. Diabetes mellitus has resulted in a large healthcare burden in many countries in Scandinavia, including Norway, where the prevalence of the disease was estimated to be 5.9% for 20131. The use of blood glucose lowering drugs, including insulin and oral antidiabetics (OADs), appears to be increasing gradually in Norway, with 3.2% of the population having blood glucose lowering drugs dispensed in 20112. Associations between blood glucose levels and complications have been demonstrated in clinical trials3, and intensive blood glucose control results in a reduction in diabetes complications4.

The clinical and economic burden of diabetes is substantial. Approximately 6–12% of all healthcare spending in five European countries is attributable to the disease5. In Norway, the costs of the disease were estimated to be €293 million (or 1.4% of total health expenditure) in 20056. More recent estimates are that the cost of diabetes comprises 6.8% of total health expenditure in Norway7. Costs of the disease may be related to the degree to which blood glucose levels are controlled and the complications of the disease are avoided. With the availability of ubiquitous and economical treatments, such as metformin for type 2 diabetes, overall cost savings may result from intensive glycemic control8. The economic burden of diabetes may be enhanced in patients who have comorbid obesity, which is common in type 2 diabetes patients. In the Spanish setting, a direct relationship between body mass index (BMI) and healthcare costs was observed in patients with type 2 diabetes, where a one-unit gain in BMI was significantly associated with a 2.4% cost increase9. The relationship between obesity and overall healthcare costs has been well documented in US payer settings10, but the extent to which BMI impacts overall costs independently from the risk of complications may be unclear. While longitudinal data collected in Australia have indicated that weight loss is associated with reduced cost of diabetes medications11, it is also likely that both the number of complications and the cost per complication decrease with patient weight loss. Thus, in addition to glycemic control, weight loss may be an important consideration when selecting treatment approaches in patients with type 2 diabetes, from a cost perspective, and may reduce the economic burden of disease. The impact of obesity on health-related quality of life in type 2 diabetes patients is well known12.

In type 2 diabetes, patients typically commence therapy with one or two OADs, such as metformin and/or a sulfonylurea, to both improve insulin sensitivity and enhance insulin secretion. Based on the observations of the UK Prospective Diabetes Study (UKPDS), progressive decline in beta-cell function leads to decreased endogenous insulin production over time, necessitating the initiation of insulin therapy13^,14. Glucagon-like peptide-1 (GLP-1) receptor agonists were initially introduced as an alternative to insulin therapy in patients failing one or two OADs; a variety of other agents have also been approved as potential second- or later-line treatments. GLP-1 receptor agonists have demonstrated efficacy versus initial insulin therapy15, and have recently demonstrated efficacy in patients already receiving an insulin regimen16^,17, based on meta-analyses18. GLP-1 receptor agonists may be of particular value for patients already receiving insulin due to their effect on weight loss and more favorable hypoglycemia profile versus intensified insulin therapy. Frequency of injections, a particular inconvenience with fully exogenous insulin therapy (i.e. basal–bolus regimen), may be of concern to patients once basal insulin fails to achieve adequate glycemic control.

Lixisenatide is a potent, selective and short-acting once-daily prandial GLP-1 receptor agonist for patients with type 2 diabetes. In patients receiving basal insulin, lixisenatide has demonstrated improvements in glycemic control versus placebo, mainly due to significant post-prandial glucose reductions, without significantly increased rates of hypoglycemia19. A meta-analysis of the three GetGoal trials investigating lixisenatide as add-on to basal insulin reported that lixisenatide significantly reduced HbA1c from baseline to week 24 and patients were more likely to achieve HbA1c <7%. While the GetGoal program has demonstrated the efficacy and safety of lixisenatide, the economic characteristics have not been defined in Norway. To perform an economic analysis in the Norwegian setting, indirect evidence must currently be relied on, as bolus insulin is the most likely comparator therapy in Norway for patients sub-optimally controlled on basal insulin, and bolus insulin was not included as a comparator in the GetGoal program. The aim of the present analysis was to estimate the long-term cost-effectiveness of lixisenatide in combination with basal insulin versus three times daily bolus plus basal insulin in patients with type 2 diabetes in Norway, based on data from the GetGoal-L study and indirect evidence.

Subjects, materials and methods

Model

The published and validated IMS CORE Diabetes Model20 was used to estimate the long-term cost-effectiveness of lixisenatide in combination with basal insulin versus bolus insulin in combination with basal insulin in patients with type 2 diabetes in the Norwegian setting21. This model was developed to determine the long-term health outcomes and cost consequences of interventions in type 1 and type 2 diabetes. The model is a non-product-specific, diabetes policy analysis tool that performs simulations taking into account intensive or conventional insulin therapy (type 1 diabetes), concomitant OADs and lipid-lowering therapies, aspirin and angiotensin-converting enzyme (ACE) inhibitor usage, and screening and treatment strategies for microvascular complications and end-stage complications. Modern type 2 diabetes therapies, including GLP-1 receptor agonists, have been simulated in the model. Disease progression is based on a series of inter-dependent Markov sub-models that simulate diabetes-related complications (angina, myocardial infarction, congestive heart failure, stroke, peripheral vascular disease, diabetic retinopathy, macular edema, cataract, hypoglycemia, ketoacidosis, nephropathy and end-stage renal disease [ESRD], neuropathy, foot ulcer, depression, amputation) and background mortality. Each sub-model uses time-, state- and diabetes-type-dependent probabilities derived from published sources, utilizing tracker variables. Analyses, including first- and second-order Monte Carlo simulations, can be performed on patient cohorts with type 1 or type 2 diabetes, defined in terms of age, gender, baseline risk factors and pre-existing complications. Relevant physiological parameters (e.g. glycohemoglobin [HbA1c] and systolic blood pressure) are captured, and their extrapolation influences simulated event risk. The model is adaptable, thus allowing the inclusion of setting-specific data as it becomes available. The reliability of simulated outcomes has been tested, with results validated against those reported by clinical trials and epidemiological studies20^,22.

Data specific to the Norwegian setting were captured in the modeling analysis. Life table data for Norway derived from the World Health Organization (WHO) were used for the calculation of other-cause mortality23, which captures all deaths not caused by diabetes and cardiovascular disease. The proportions of type 2 diabetes patients receiving aspirin, statins and ACE inhibitors in the modeling simulation, for either primary or secondary prevention of cardiovascular events, were based on Norwegian data. These Norway specific inputs influence the underlying UKPDS and Framingham Heart Study based risks of diabetes complications in the cohort simulated through the IMS CORE Diabetes Model. The proportions of type 2 diabetes patients receiving foot, eye and renal screening were also based on Norwegian data. Data from a Norwegian Renal Registry were applied to predict the likelihood of ESRD treatment modalities and subsequent survival24.

Cohort characteristics and patient management practices

A cohort of patients was defined for Norway and was based on data from patients receiving basal insulin in the GetGoal-L trial (Table 1). Summary statistics from GetGoal-L for physiological, demographic and disease history characteristics at enrolment were used to define the cohort, in terms of means and standard deviations for continuous variables, and in terms of percentage of patients with the characteristics, for binary variables. Both intention-to-treat and safety data from the multicenter phase III GetGoal-L were used to define the cohort. Patients had suboptimal glycemic control (HbA1c ≥7.0%, ≤10.0%) at enrollment and were receiving basal insulin at a stable dose for at least 3 months. The simulated cohort had a moderate prevalence of diabetes-related complications at baseline, including cardiovascular, renal and ocular complications.

Table 1. Cohort characteristics and patient management practices.

	Mean (SD)	Reference
Patient demographics
Baseline age (years)	57.2 (9.6)	GetGoal-L
Duration of diabetes (years)	12.5 (6.8)	GetGoal-L
Proportion male	46.0%	GetGoal-L
Baseline risk factors
HbA1c (% points)	8.4 (0.9)	GetGoal-L
Systolic blood pressure (mmHg)	130.4 (14.4)	GetGoal-L
Total cholesterol (mg/dL)	182.2 (41.7)	GetGoal-L
High-density lipoprotein cholesterol (mg/dL)	44.9 (12.3)	GetGoal-L
Low-density lipoprotein cholesterol (mg/dL)	104.9 (37.6)	GetGoal-L
Triglycerides (mg/dL)	176.0 (111.9)	GetGoal-L
BMI (kg/m^2)	32.1 (6.2)	GetGoal-L
Proportion smokers	10.0%	GetGoal-L
Cigarettes per day	1.0	GetGoal-L
Alcohol consumption (Oz/week)	5.0	GetGoal-L
Racial characteristics
Proportion White	52.3%	GetGoal-L
Proportion Black	3.9%	GetGoal-L
Proportion Hispanic	27.0%	GetGoal-L
Proportion Native American	0.2%	GetGoal-L
Proportion Asian/Pacific Islander	16.6%	GetGoal-L
Cardiovascular medication
Aspirin use for primary prevention	31.4%	Expert Opinion
Aspirin use for secondary prevention	86.0%	25
ACE inhibitor use for primary prevention	64.4%	26
ACE inhibitor for secondary prevention	94.0%	25
Statins use for primary prevention	45.9%	Expert Opinion
Statins use for secondary prevention	73.0%	25
Complications management
Proportion on foot ulcer prevention program	10.9%	27
Proportion screened for retinopathy	85.0%	28
Proportion screened for renal disease	59.7%	27
Proportion receiving intensive insulin after MI	87.7%	29
Proportion treated with extra ulcer treatment (Regranex)	57.0%	30
Proportion screened for depression – no complications	83.0%	31
Proportion screened for depression – complications	83.0%	31

ACE = angiotensin-converting enzyme; BMI = body mass index; HbA1c = glycohemoglobin; MI = myocardial infarction; Oz = fluid ounces.

The simulated cohort was subject to Norway-specific patient management practice during the simulations, which were ascertained from Norwegian literature and expert opinion (Table 1). Patient management practices influence the likelihood of receiving efficacious cardiovascular and antihypertensive medications and complication screening procedures in IMS CORE Diabetes Model simulations. A national diabetes survey informed the likelihood of receiving cardiovascular and antihypertensive medications for secondary prevention25. Other surveys were referred to for diabetic foot, retinopathy and renal management practices26^,27.

Treatment effects

To estimate the long-term cost-effectiveness of lixisenatide versus bolus insulin in patients with type 2 diabetes sub-optimally controlled on basal insulin in Norway, treatment effects were derived from an indirect comparison, in the absence of the availability of head-to-head data from the phase III GetGoal clinical trial program. The indirect comparison considered data from five clinical trials identified via systematic literature review32–35. Bolus insulin was chosen as the most relevant comparator based on National Treatment Guidelines and clinical practice in Norway. Treatment effects for lixisenatide were derived directly from the GetGoal-L trial. For lixisenatide, change in HbA1c from baseline to week 24 was −0.74% and change in BMI from baseline was −0.66 kg/m². For bolus insulin, treatment effects were estimated based on mean difference estimated via indirect comparison. Indeed, no head to head data were identified from the published literature at the time of the analysis so the Bucher method36 was used to combine treatment effects from the different studies and indirectly compare treatment effects. Using this method, for bolus insulin the change in HbA1c from baseline was −0.93% and the change in BMI from baseline was 0.62 kg/m². Major and minor hypoglycemic events were included, and were 2.9 and 219.3 events per 100 patient-years, respectively, for lixisenatide, and were 3.5 and 262.7 events per 100 patient-years, respectively, for bolus insulin, based on a relative risk of 0.8349 for hypoglycemia for lixisenatide versus bolus insulin, derived from indirect comparison. Hypoglycemic episodes were classified as minor (signs or symptoms associated with hypoglycemia that were either self-treated or resolved on their own) or major (resulting in loss of consciousness or seizure from which the participant promptly recovered in response to glucagon or glucose, or presumed hypoglycemia requiring the assistance of another person because of severe impairment of consciousness or behavior).

All relevant comparative treatment effect data available from the indirect comparison, even if statistically non-significant, were included in the computer simulation model. Non-significant data may legitimately be used as a source of information in decision analytic models37. As treatment effects for SBP, lipids and triglycerides were not captured in the indirect comparison, they were excluded from the modeling analysis. The treatment effects for lixisenatide and bolus insulin were applied in the model as immediate initial changes in the physiological parameter, followed by the simulation of a non-linear progression according to the HbA1c equation derived from the UKPDS Outcomes Model38, which predicts gradual convergence in HbA1c as benefits are lost over a period of several years. The risk equation derived from the UKPDS Outcomes Model was based on patients enrolled in the UKPDS who were receiving a range of glycemia lowering therapies, including metformin, sulfonylureas and insulin. No long-term progression of BMI was modeled in the simulations. Initially assigned therapy was assumed to be taken for 3 years in the simulations, followed by discontinuation of GLP-1 receptor agonist therapy and intensification of insulin therapy, with commensurate increase in hypoglycemia to 12 and 1782 major and minor hypoglycemic events per 100 patient-years, respectively.

Costs and perspective

Costs of diabetes-related complications, primary interventions and patient management practices were captured in the cost-effectiveness analysis, and were based on a societal perspective, with costs expressed in 2012 Norwegian krone (NOK; Table 2; NOK 1 = US$ 0.127 or EUR 0.116 at 25 March 2015). A societal perspective is the default perspective to be taken in cost-effectiveness analyses in Norway37^,39. Most complication costs were derived from the National Directorate of Health: Innsatsstyrt finansiering (ISF) 201340, which are based on diagnosis related groups, and the 2012–2013 Norwegian Medical Association Fee Schedule for General Practitioner/Specialist Consultations41. For cost data not available as 2012 costs, values were inflated using the health component of the national consumer price indices reported by Statistics Norway42. The costs relevant to each simulated complication were estimated for the year of that complication and for subsequent years, where the complication cost may differ over time. Additionally, indirect costs per complication were captured in the analysis based on the human capital approach. Time off paid employment associated with diabetes and its complications has been investigated in a number of settings, including Norway6. The large number of complications captured in the IMS CORE Diabetes Model necessitated a bespoke investigation of time off paid employment per complication and event for the present analysis. Official sick leave data was requested from the Norwegian Labor and Welfare Administration (NAV). Relevant International Classification of Primary Care (ICPC-2) codes for 20 diabetes related complications and events captured in the IMS CORE Diabetes Model were identified in cooperation with NAV. Data were extracted from the NAV database in December 2012, and pertained to all episodes of sick leave that took place during 2011. The time off work per complication and event experienced by patients simulated in the analysis is displayed in Table 2.

Table 2. Simulated diabetes complication and intervention costs.

Diabetes Complication	Costs (NOK)	DoW
Myocardial infarction, year of event	161,898	44
Myocardial infarction, subsequent years	3907	0
Angina, year of onset	36,299	31
Angina, subsequent years	6653	0
Heart failure, year of onset	59,000	49
Heart failure, subsequent years	8840	0
Stroke, year of event	208,634	59
Stroke, subsequent years	93,394	0
Stroke, death within 30 days	41,814	0
PVD, year of onset	37,646	37
PVD, subsequent years	2917	0
Hemodialysis, year of onset	560,323	18
Hemodialysis, subsequent years	560,323	18
Peritoneal dialysis, year of onset	553,951	18
Peritoneal dialysis, subsequent years	553,951	18
Kidney transplant, year of procedure	423,897	18
Kidney transplant, subsequent years	41,612	18
Major hypoglycemic event	5042	14
Minor hypoglycemic event	134	0
Laser treatment	22,445	0
Cataract operation	15,739	10
Annual cost following cataract operation	775	0
Blindness, first year	5588	38
Blindness, subsequent years	5588	0
Neuropathy, year of onset	4958	24
Neuropathy, subsequent years	0	0
Uninfected foot ulcer, year of onset	14,433	19
Uninfected foot ulcer, subsequent years	14,433	0
Infected foot ulcer, year of onset	52,053	19
Infected foot ulcer, subsequent years	52,053	0
Amputation, procedure	129,426	27
Amputation, prosthesis	16,730	0
Annual cost following amputation	0	0
Gangrene treatment, year of onset	21,580	37
Annual cost following gangrene treatment	0	0
Cost after healed ulcer	12,559	19
Cost after healed ulcer (post-amputation)	221,827	0
Diabetes intervention	Lixisenatide plus insulin	Basal–bolus insulin
Cost per needle	1.28	1.28
Number of needles per day	2.00	4.00
Cost per lancet and test strip	6.69	6.69
Number of lancets and test strips per day	1.00	4.00
Basal insulin dose per day (IU)	51.73	62.60
Basal insulin cost per IU	0.21	0.21
Bolus insulin dose per day (IU)	0.00	62.60
Bolus insulin cost per IU	0.20	0.20
Lixisenatide cost per day	21.84	0.00
Annual cost (NOK)	15,307	21,103

DoW = days off work; IU = international units; NOK = Norwegian krone; PVD = peripheral vascular disease.

Pharmacy cost data were based on the interventions administered in the GetGoal-L trial and from costs derived from the Norwegian Medicines Agency (NoMA) Price Database January 201343. Drug cost data are the approved pharmacy retail price (PRP) excluding 25% VAT as recommended in the Norwegian Pharmacoeconomic Guidelines 2012. The cost of lixisenatide, however, was based on an estimated pharmacy retail price excluding 25% VAT provided by Sanofi, as the national price application was pending approval by NoMA at the time the analysis was conducted. The cost per day of lixisenatide utilized in the model was NOK 21.84 for the 20 µg maintenance dose. The final NoMA approved price was slightly lower (NOK 21.70) than the price applied in the submitted analysis. The basal insulin formulation administered to patients enrolled in GetGoal-L was not pre-specified and therefore varied between patients, with the majority of enrolled patients receiving insulin glargine. Therefore, a weighted basal insulin cost was applied in the analysis, based on insulin market share data for Norway.

Since neutral protamine hagedorn (NPH) insulin is the only reimbursed basal insulin for patients with type 2 diabetes in Norway, a separate sensitivity analysis on the NPH-subpopulation in the GetGoal-L study was conducted to investigate potential differences in treatment effect between insulins as well as using the cost of NPH only. Due to differences in body weight between patients receiving the lixisenatide and bolus insulin treatments, differences in basal insulin consumption (51.73 IU per day versus 62.60 IU per day), and hence costs, were modeled. Costs of needles associated with drug administration and costs of self-monitoring of blood glucose (SMBG) were also included. Costs of background OAD therapy were excluded, as it was not expected that the costs would differ between the treatment arms included in the simulation (lixisenatide plus basal insulin and bolus insulin plus basal insulin). The complete intervention costs are listed in Table 2. Treatment costs increased to NOK 22,163 per annum following the intensification of insulin therapy, based on insulin dose reported by Raskin and colleagues44. The intensified insulin cost was applied equally to patients who were initially assigned lixisenatide or bolus insulin, after three years of the simulation had elapsed. Costs related to patient management (e.g. cardiovascular medications) were also captured and were derived from the NoMA price database.

Health-related quality of life

Health state utility and disutility values were derived from a systematic literature review of utility values associated with type 2 diabetes complications45. The systematic review included all type 2 diabetes complications included in the IMS CORE Diabetes Model, and was performed on MEDLINE, Embase, EconLIT and NHS Economic Evaluation Database. Publications reporting EuroQol 5-Dimensions (EQ-5D) values were preferred. The utilities derived from the literature review were adjusted to the Norwegian setting, by comparing the overall EQ-5D index (0.785) for type 2 diabetes reported in the literature review with a corresponding Norwegian value (0.850) as reported by Solli and colleagues46. The adjustment factor, 0.850 divided by 0.785 (or 1.083), was used to adjust all health state utility and event disutility values reported by the systematic literature review applied in the simulations. Alongside the health state utility and disutility values reported in the systematic literature review, adjusted for the Norwegian setting, disutility for excess BMI was captured in the simulations. A disutility of −0.0061 per unit of BMI over 25 kg/m² was applied in the simulation, based on data from the CODE-2 study12. The disutilities associated with an additional unit of BMI above 25 kg/m² were applied.

Discounting and time horizon

Discounting of future costs and clinical outcomes was performed to account for time preference, as required for economic evaluations. In the base case analysis both costs and clinical outcomes (e.g. life years) were discounted at a rate of 4% per year, as recommended in Norway47. The Core Diabetes Model (CDM) simulations were performed for time horizons of 45 years, in order to fully capture complications of diabetes and associated mortality.

Deterministic sensitivity analysis

A series of one-way deterministic sensitivity analyses were performed to assess the impact on cost-effectiveness results of changes to key analysis input parameters and modeling methodologies.

Time on treatment and time horizon of the analysis

In the base case patients would receive initially assigned lixisenatide or bolus insulin for 3 years, followed by further intensification of insulin therapy. In a sensitivity analysis it was assumed that patients received initially assigned therapy for 5 years, in line with the modeled duration of GLP-1 receptor agonist treatment, as detailed in a published reimbursement submission for the OAD population48. In the base case, the time dependent treatment switch was performed without regard to the HbA1c level of the simulated patient, a sensitivity analysis made the treatment intensification dependent on the HbA1c level of the simulated patient; it was assumed that patients intensified insulin therapy once HbA1c levels reached 8.4%44. Simulations were also performed over a 20 year time horizon (versus 45 years in the base case), to assess the impact of shorter time horizons on the results.

Discount rate

Discount rates were varied in three sensitivity analyses, where rates of 0% (no discounting) and 6% were applied to costs and clinical outcomes, and where a rate of 4% for costs and 0% for clinical outcomes was applied, within the ranges of discount rates recommended in Norway47.

Efficacy outcomes – HbA1c and BMI

To assess the impact of changes in HbA1c on the analysis results, sensitivity analyses were performed where change in HbA1c from baseline was set to the upper and lower limits of the 95% confidence intervals. To assess the impact of changes in BMI on the analysis results, sensitivity analyses were performed where change in BMI from baseline was set to the upper and lower limits of the 95% confidence intervals.

Cohort characteristics

An analysis was performed with cohort characteristics derived from the Norwegian setting, as opposed to the GetGoal-L based population simulated in the base case. Norwegian data were derived from Jenssen and colleagues: this cohort had a mean age of 59.9 years, 6.8 years mean duration of diabetes and mean HbA1c of 7.2%, compared to 57.2 years, 12.5 years and 8.4% in the GetGoal-L population25.

Cost and utility inputs

The costs of complications were increased by 25% and reduced by 25% in two respective sensitivity analyses. In the base case, a weighted average cost of basal insulins was applied based on Norwegian market share data; a sensitivity analysis was performed where it was assumed that all patients were receiving NPH as the basal insulin therapy. The health state utility and event disutility values were increased and decreased by 25% in two separate sensitivity analyses. To assess the impact of BMI-related disutility on analysis results, a sensitivity analysis was performed where the CODE-2 disutilities for BMI levels over 25 kg/m² were not applied (versus a disutility of −0.0061 per unit of BMI in the base case).

Scenario analysis

To assess the cost-effectiveness of lixisenatide versus placebo (instead of versus bolus insulin), a scenario analysis was performed where efficacy data from the GetGoal-L trial were applied to lixisenatide in addition to basal insulin and placebo in addition to basal insulin. This analysis was performed to economically evaluate lixisenatide on a stand-alone basis in the Norwegian setting, and overcomes some of the limitations of relying on data from an indirect comparison as performed in the base case.

Statistical methodology

The cost-effectiveness analyses were performed using a non-parametric bootstrapping approach. For each of the 1000 bootstrap iterations the progression of diabetes was simulated in 1000 patients, to calculate the mean and standard deviation of costs, life expectancy and quality-adjusted life expectancy49. Mean results of each of the 1000 bootstrap iterations in the analyses were used to create a scatterplot diagram comparing the differences in clinical and cost outcomes (cost and effect pairs) for lixisenatide plus basal insulin and bolus insulin plus basal insulin. Probabilistic sensitivity analysis was performed in every scenario, with input values sampled from distributions (typically Gaussian) with replacement. Parameters sampled included baseline cohort characteristics and change in physiological parameters from baseline.

Results

Clinical outcomes

Lixisenatide was associated with improvements in quality-adjusted life expectancy compared to bolus insulin in patients with type 2 diabetes receiving basal insulin in Norway. Quality-adjusted life expectancy was 6.908 QALYs for lixisenatide and 6.842 QALYs for bolus insulin (an increase of 0.066 QALYs; Table 3). Improvements in quality-adjusted life expectancy were a result of lifetime simulated reductions in BMI for patients receiving lixisenatide and associated improvements in health-related quality of life. The analysis suggested slightly less improvement in HbA1c with lixisenatide than with bolus insulin. However, this was outweighed in the QALY calculation by the quality of life gains from BMI. The lifetime cumulative incidence of diabetes related complications was generally higher for patients receiving lixisenatide compared to patients receiving bolus insulin (Table 3). The cumulative incidence of severe vision loss (16.125% versus 15.478%), lower extremity amputation (18.832% versus 18.476%), and myocardial infarction (31.258% versus 31.085%) were higher for patients receiving lixisenatide versus bolus insulin. The cumulative incidence of heart failure was lower for patients receiving lixisenatide (36.556% versus 38.232%), which is a function of the weight loss experienced by patients receiving lixisenatide, where body weight is a key driver of heart failure risk. The mean time to onset of any diabetes related complication was 1.68 years for lixisenatide and 1.76 years for bolus insulin. The mean time to onset of complications for patients experiencing them was generally shorter for patients receiving lixisenatide compared to bolus insulin (data not shown). This can be explained by the benefit of bolus insulin on microvascular complications, driven by greater HbA1c reductions, which were not statistically significant versus lixisenatide.

Table 3. Summary cost, clinical and cost-effectiveness results for lixisenatide plus insulin versus basal–bolus insulin.

Summary cost outcomes	Lixisenatide plus insulin	Basal–bolus insulin
Direct costs from payer perspective (NOK)	742,563	750,032
Treatment	226,553	242,034
Patient management	15,143	15,113
Cardiovascular disease	138,693	139,846
Renal disease	25,707	23,132
Diabetic foot and neuropathy	297,674	291,327
Eye disease	12,425	12,030
Hypoglycemia	24,682	24,874
Depression	1685	1675
Indirect costs	508,827	508,227
Combined costs from societal perspective (NOK)	1,251,390	1,258,259
Summary clinical outcomes
Quality-adjusted life expectancy (QALYs)	6.908	6.842
Cumulative incidence of complications
Background retinopathy (%)	22.228	21.405
Proliferative retinopathy (%)	2.146	1.983
Macular edema (%)	18.279	17.517
Severe vision loss (%)	16.125	15.478
Cataract (%)	9.211	9.059
Microalbuminuria (%)	35.179	34.015
Gross proteinuria (%)	12.666	11.875
End-stage renal disease (%)	3.963	3.546
Nephropathy-related death (%)	2.240	2.025
Foot ulcer, first (%)	41.032	40.070
Foot ulcer, recurrence (%)	98.821	97.274
Amputation (%)	18.832	18.476
Neuropathy (%)	62.907	61.660
Peripheral vascular disease (%)	11.856	11.617
Congestive heart failure, onset (%)	36.556	38.232
Congestive heart failure, death (%)	26.097	27.289
Angina (%)	13.727	13.464
Myocardial infarction, event (%)	31.258	31.085
Myocardial infarction, death (%)	16.864	16.720
Stroke, event (%)	19.686	19.729
Stroke, death (%)	6.524	6.477
Summary cost-effectiveness outcomes
ICER based on quality-adjusted life expectancy, payer perspective	Dominant*
ICER based on quality-adjusted life expectancy, societal perspective	Dominant*

ICER = incremental cost-effectiveness ratio; NOK = Norwegian krone; QALY = quality-adjusted life year. Values shown are means.

*More effective and less costly.

Cost outcomes

Lixisenatide was associated with cost savings from a societal perspective compared to bolus insulin in type 2 diabetes patients receiving basal insulin in Norway. Lifetime direct medical costs were NOK 742,563 per patient receiving lixisenatide and NOK 750,032 per patient receiving bolus insulin (a saving of NOK 7,469 per patient; Table 3). Indirect costs were higher for patients receiving lixisenatide versus patients receiving bolus insulin (by NOK 600 per patient), due to reduced cumulative incidence of complications for patients receiving bolus insulin. Combined direct and indirect costs, from a societal perspective indicated that lixisenatide was associated with lifetime cost savings compared to bolus insulin (NOK 1,251,390 versus NOK 1,258,259 per patient). Overall cost savings were primarily driven by the lower lifetime treatment costs of lixisenatide versus bolus insulin (NOK 226,553 versus NOK 242,034 per patient). Lixisenatide was also associated with lower cardiovascular disease costs compared to bolus insulin (NOK 138,693 versus NOK 139,846 per patient).

Cost-effectiveness outcomes

Based on quality-adjusted life expectancy, lixisenatide can be considered dominant versus bolus insulin in type 2 diabetes patients receiving basal insulin in Norway, as lixisenatide was associated with improvements in quality-adjusted life expectancy (by 0.066 QALYs) and cost savings (of NOK 6,869 per patient). A scatter plot indicating the dispersion of outcomes on the cost-effectiveness plane indicates that a large number of points lie within the south-east quadrant, indicating improved quality-adjusted life expectancy and reduced costs for lixisenatide versus bolus insulin from a payer perspective (Figure 1). The likelihood of lixisenatide being considered cost-effective at a willingness-to-pay threshold of NOK 500,000 per QALY gained was 94.5%. Expressed in terms of net monetary benefit (NMB), and based on a willingness-to-pay threshold of NOK 500,000 per QALY gained in Norway, the NMB of lixisenatide was NOK 39,369.

Figure 1. Scatterplot of change in costs and change in quality-adjusted life expectancy for lixisenatide plus insulin versus basal–bolus insulin. NOK = Norwegian krone; QALY = quality-adjusted life year.

Sensitivity analysis results

A series of one-way sensitivity analyses indicated that lixisenatide would remain associated with improved quality-adjusted life expectancy and reduced costs compared to bolus insulin in patients receiving basal insulin in the Norwegian setting in the majority of simulations (Table 4). The results were most sensitive to discount rates, reductions in HbA1c and the application of CODE-2 disutilities for excess BMI. In all scenarios tested, lixisenatide remained associated with improved quality-adjusted life expectancy compared to bolus insulin. When discount rates were set to 0% (versus 4% in the base case), lixisenatide was associated with increased lifetime societal costs compared to bolus insulin (NOK 143 per patient), and an incremental cost-effectiveness ratio of NOK 1422 per QALY gained. When the upper limit of the 95% confidence interval for change in HbA1c from baseline was applied in the simulations, to both the lixisenatide and bolus insulin treatment arms, lixisenatide was associated with an ICER of NOK 76,066 per QALY gained. When the CODE-2 disutilities for excess BMI were not applied in the simulations, the improvement in quality-adjusted life expectancy for lixisenatide versus bolus insulin was reduced to 0.002 QALYs (versus an improvement of 0.066 QALYs in the base case) indicating that excess body weight associated disutility was a key driver of the results. Changes to complication costs, health state utility and event disutility values, time horizons, treatment duration and the use of the upper or lower confidence limits for change in BMI from baseline had little impact on the cost-effectiveness results for lixisenatide versus bolus insulin. The application of Norwegian cohort data and the use of the NPH cost as the basal insulin cost in two separate sensitivity analyses also had little impact on the results.

Table 4. Sensitivity analysis results for lixisenatide plus insulin versus basal–bolus insulin.

	Quality-adjusted life expectancy (QALYs)			Lifetime societal costs (NOK)
Sensitivity analysis	Lixisenatide plus insulin	Basal–bolus insulin	Difference	Lixisenatide plus insulin	Basal–bolus insulin	Difference	ICER (NOK per QALY gained)	NMB (NOK)
Base case	6.908 (0.097)	6.842 (0.093)	0.066	1,251,390 (50,965)	1,258,259 (48,538)	−6869	Dominant	39,369
5 year initial treatment duration	6.996 (0.097)	6.927 (0.097)	0.069	1,235,250 (51,104)	1,247,583 (51,936)	−12,333	Dominant	46,833
HbA1c dependent treatment switch	7.101 (0.099)	7.055 (0.100)	0.046	1,208,474 (49,784)	1,230,477 (52,035)	−22,003	Dominant	45,003
20 year time horizon	6.580 (0.087)	6.514 (0.092)	0.066	1,145,715 (48,422)	1,156,894 (50,129)	−11,179	Dominant	44,179
0% discount rates for costs and clinical outcomes	9.808 (0.167)	9.708 (0.162)	0.100	1,868,009 (77,757)	1,867,866 (73,710)	143	1422	49,857
6% discount rates for costs and clinical outcomes	5.953 (0.077)	5.899 (0.073)	0.054	1,055,187 (43,076)	1,063,694 (41,124)	−8507	Dominant	35,507
4% discount rate for costs and 0% for clinical outcomes	9.808 (0.167)	9.708 (0.162)	0.100	1,251,390 (50,965)	1,258,259 (48,538)	−6869	Dominant	56,869
Upper limit of 95% CI for change in HbA1c from baseline	6.935 (0.106)	6.907 (0.104)	0.028	1,244,979 (52,304)	1,242,843 (50,757)	2136	76,066	11,864
Lower limit of 95% CI for change in HbA1c from baseline	6.873 (0.102)	6.770 (0.096)	0.103	1,257,587 (51,083)	1,269,347 (50,551)	−11,760	Dominant	63,260
Upper limit of 95% CI for change in BMI from baseline	6.907 (0.093)	6.767 (0.268)	0.140	1,250,354 (52,981)	1,257,466 (50,780)	−7112	Dominant	39,231
Lower limit of 95% CI for change in BMI from baseline	6.905 (0.096)	6.859 (0.093)	0.046	1,248,525 (51,645)	1,258,256 (50,684)	−9731	Dominant	32,731
Complication costs increased by 25%	6.908 (0.097)	6.842 (0.093)	0.066	1,360,360 (57,233)	1,365,600 (54,813)	−5240	Dominant	38,240
Complication costs decreased by 25%	6.908 (0.097)	6.842 (0.093)	0.066	1,142,420 (44,894)	1,150,919 (42,482)	−8499	Dominant	41,499
Health state utility and disutility values increased by 25%	8.907 (0.124)	8.839 (0.119)	0.068	1,251,390 (50,965)	1,258,259 (48,538)	−6869	Dominant	40,869
Health state utility and disutility values decreased by 25%	4.763 (0.067)	4.697 (0.064)	0.066	1,251,390 (50,965)	1,258,259 (48,538)	−6869	Dominant	39,869
No CODE-2 disutility for BMI >25 kg/m^2	7.292 (0.102)	7.290 (0.098)	0.002	1,251,390 (50,965)	1,258,259 (48,538)	−6869	Dominant	7869
Norwegian cohort analysis	7.244 (0.097)	7.196 (0.094)	0.048	934,391 (40,655)	940,415 (37,908)	−6024	Dominant	30,024
Cost of basal insulin as NPH	6.908 (0.097)	6.842 (0.093)	0.066	1,242,084 (50,907)	1,248,580 (48,487)	−6496	Dominant	39,469

BMI = body mass index; HbA1c = glycohemoglobin; ICER = incremental cost-effectiveness ratio; NOK = Norwegian krone; QALY = quality-adjusted life year; SBP = systolic blood pressure.

Scenario analysis results

When the cost-effectiveness of lixisenatide was evaluated against placebo, lixisenatide was associated with improved clinical outcomes and increased lifetime societal costs. Quality-adjusted life expectancy was 6.888 QALYs for lixisenatide versus 6.819 QALYs for placebo (an improvement of 0.069 QALYs). Lifetime societal costs were NOK 1,249,357 per patient receiving lixisenatide versus NOK 1,236,511 per patient receiving placebo (an increase of NOK 12,846 per patient). The incremental cost-effectiveness ratio for lixisenatide versus placebo was NOK 186,820 per patient, which is less than the willingness-to-pay threshold in Norway of NOK 500,000 per QALY gained.

Discussion

The results of the present analysis have indicated that lixisenatide in combination with basal insulin is associated with improved quality-adjusted life expectancy and reduced societal costs compared to bolus in combination with basal insulin in patients with type 2 diabetes in Norway. The results of the present analysis indicate that lixisenatide should be considered as a treatment option for patients with type 2 diabetes who have sub-optimal glycemic control on basal insulin, particularly as an alternative to bolus insulin, the initiation of which is associated with a number of challenges. As experience with the use of GLP-1 receptor agonists increases in clinical settings, policy and patient level treatment decisions in the basal insulin population may be better informed by the analysis conducted here. It should be noted that the analysis is based on the GetGoal-L trial, which was conducted in advanced patients with a long history of diabetes, and the simulated cohort had a prior disease duration of 12.5 years. Therefore, the results of the analysis are not necessarily generalizable to patients with earlier-stage diabetes for which further economic studies may be needed.

All sources of available information at the time of this analysis were utilized in the generation of cost-effectiveness results for lixisenatide versus bolus insulin. Consistent with previous IMS CORE Diabetes Model analyses, any source of data were included in the decision analytic framework, even in the absence of statistical significance, including greater reductions in HbA1c for bolus insulin as ascertained from indirect evidence. However, the key driver of the long-term cost-effectiveness results was weight gain associated disutility, as measured by the CODE-2 investigators12, and applied in previously performed IMS CORE Diabetes Model analyses of GLP-1 receptor agonist therapies50^,51. Given the sensitivity of the cost-effectiveness results to weight gain associated disutility, the persistence of weight benefit on GLP-1 receptor agonist therapy compared to insulin requires further investigation. Klonoff and colleagues have indicated that in type 2 diabetes patients persistent with therapy, GLP-1 receptor agonist therapy is associated with sustained reductions in BMI for periods of up to 3 years52. The persistency of weight loss advantages beyond the initial treatment period (3 years in the current base case) requires further investigation for long-term modeling analyses, particularly for type 2 diabetes patients who intensify to a basal–bolus insulin regimen.

Lixisenatide was associated with substantial cost savings compared to bolus insulin in the present analysis. The complication costs applied in the simulation were equal for both treatment arms in the IMS CORE Diabetes Model, where there is evidence to suggest that costs of treating individual diabetes complications are higher in more obese patients. A number of studies have indicated that body mass index or a diagnosis of obesity are independently associated with higher costs in type 2 diabetes10^,53. Thus, the approach taken in the present analysis, where a complication cost did not differ according to the BMI level of the simulated patient, could be considered conservative. Changes in body weight did impact the utilization of basal insulin in the analysis for the first 3 years of the simulation, where it was assumed a lower dose of basal insulin was required for patients who experienced initial weight loss on lixisenatide therapy, versus patients who experienced weight gain on bolus insulin therapy. The gain in QALYs with lixisenatide was small (0.066 QALYs per patient), and may not be clinically significant for individual patients. Nevertheless, in combination with reduced healthcare costs this QALY gain was sufficient to create a dominant position for lixisenatide versus bolus both in combination with basal insulin in the present cost-effectiveness analysis.

This analysis was subject to limitations common to most diabetes modeling analyses. Firstly, the understanding of the disease, encapsulated in the IMS CORE Diabetes Model in the form of the model structure, the transition probabilities and regression formulae, is based on studies that were conducted many years ago and in non-Norwegian settings. It is perhaps unrealistic to expect studies on the same scale as the UKPDS to be conducted in the multitudinous settings where health economic evaluations of diabetes interventions are to be performed. The IMS CORE Diabetes Model has been externally validated against a number of contemporary diabetes outcomes studies conducted in a range of settings20. Secondly, GLP-1 receptor agonist therapy was not available at the time the UKPDS was conducted. However, the patients enrolled in the UKPDS were receiving therapies along the entire continuum of diabetes care, including insulin sensitizing agents (metformin), secretagogues (sulfonylureas), and exogenous insulin therapy. The long-term upwards drift in HbA1c observed in the UKPDS, representing a gradual decline in beta-cell function on a range of antidiabetic therapies, was assumed to apply to GLP-1 receptor agonist therapy in the present analysis. Thirdly, SBP, lipids and triglycerides were excluded from the model; this may, however, constitute a limitation because these variables are potential drivers for cardiovascular disease. The fact that long term BMI progression was not modeled is also a limitation, since the two treatment arms showed different impacts on BMI. Sensitivity analysis was conducted around impact on BMI and associated disutility to address this. Finally, in line with standard economic modeling practice, it was assumed that current drug prices would continue to operate; therefore, the model does not take into account any potential future impact of changes to drug costs through, for example, the availability of generic alternatives.

The current analysis was based on indirect evidence in the reference case analysis. No head-to-head efficacy data are currently available for lixisenatide and bolus insulin in patients with type 2 diabetes sub-optimally controlled on basal insulin, a scenario that could be tested when the results of the ongoing clinical trial GetGoal-Duo2 are available. The decision was made to base the analysis on the population characteristics from the GetGoal-L trial and to explore the potential impact of different cohort characteristics using sensitivity analyses. This approach is likely to be conservative, as the sensitivity analysis using the Norwegian cohort produced a lower ICER than the base case. A meta-analysis that uses propensity score-matching to compare the efficacy and safety of basal insulin plus insulin glulisine with basal insulin plus lixisenatide previously reported that the lixisenatide combination was at least twice as likely to achieve composite efficacy and safety endpoints54. Systematic reviews, employing meta-analysis techniques to synthesize evidence from several trials, are an essential component of comprehensive decision making for all relevant competing interventions55, even though it may often be argued that indirect comparison does not provide as high quality evidence as that obtained from direct evidence in a head-to-head clinical trial. In any case, uncertainty around the treatment effects, as derived from indirect evidence, was comprehensively assessed in sensitivity analysis, and lixisenatide remained associated with improved quality-adjusted life expectancy and reduced lifetime societal costs compared with bolus insulin in type 2 diabetes patients sub-optimally controlled on basal insulin.

Conclusion

This modeling analysis has collated data from disparate sources to generate conclusions that may aid treatment decision makers in Norway. Based on indirect evidence and the modeling analysis conducted with the IMS CORE Diabetes Model, lixisenatide is associated with improved quality-adjusted life expectancy and reduced lifetime societal costs compared to bolus insulin in type 2 diabetes patients sub-optimally controlled on basal insulin.

Transparency

Declaration of funding

This study was supported by an unrestricted grant from Sanofi.

Declaration of financial/other relationships

P.H., J.L.P. and E.M. have disclosed that they are current or former employees of IMS Health, a consultancy that has received funds from Sanofi. M.F., M.G. and A.L. have disclosed that they are current employees of Sanofi.

Acknowledgments

The authors would like to thank Sanofi for the provision of an unrestricted grant to support this project.