The cost-effectiveness of exenatide twice daily (BID) vs insulin lispro three times daily (TID) as add-on therapy to titrated insulin glargine in patients with type 2 diabetes

Abstract

Objective: To evaluate the cost-effectiveness of exenatide twice daily (BID) vs bolus insulin lispro three times daily (TID) as add-on therapy when glycemic control is sub-optimal with titrated basal insulin glargine and metformin.

Methods: The analysis was based on the recent 4B Study, which compared exenatide BID and lispro TID as add-on therapies in subjects with type 2 diabetes insufficiently controlled, despite titrated insulin glargine. The Cardiff Diabetes Model was used to simulate patient costs and health benefits beyond the 4B Study. The Swedish healthcare perspective was adopted for this analysis; costs are reported in €EUR to aid interpretation. The main outcome measure was the cost per quality-adjusted life-year (QALY) gained with exenatide BID compared to lispro TID.

Results: Exenatide BID was associated with an incremental cost of €1,270 and a QALY increase of +0.64 compared with lispro TID over 40 years. The cost per QALY gained with exenatide BID compared with lispro TID was €1,971, which is within conventional limits of cost-effectiveness. Cost-effectiveness results were generally robust to alternative assumptions and values for key model parameters.

Limitations: Extrapolation of trial data over the longer term can be influenced by modeling and parameter uncertainty. Cost-effectiveness results were generally insensitive to alternative values of key model input parameters and across scenarios.

Conclusions: The addition of exenatide BID rather than insulin lispro to basal insulin is associated with similar or better clinical outcomes. Illustrated from the Swedish healthcare perspective, analysis with the Cardiff Diabetes Model demonstrated that exenatide BID represents a cost-effective treatment alternative to lispro TID as add-on therapy in type 2 diabetes patients insufficiently controlled on basal insulin.

Keywords:

• Type 2 diabetes
• Cost-effectiveness
• Cost-utility
• Exenatide
• Insulin

Introduction

The overall aim of treatment in type 2 diabetes is the concomitant management of glycemic control, weight gain, and hypoglycemia; however there are typically trade-offs between treatment options in achieving one or a composite of these endpoints¹. These trade-offs are known to vary according to patient phenotype, prior treatment history, and disease status.

The focus of this research is on patients with advanced disease with limited therapeutic options. One of the therapeutic approaches available to type 2 diabetes patients insufficiently controlled on basal insulin is to add bolus insulin. Whilst this approach has been shown to lower glycosylated hemoglobin (HbA_1c), the trade-off is commonly weight gain and an increased incidence of hypoglycemic episodes¹. An alternative treatment strategy to the addition of bolus insulin is the use of the GLP-1 receptor agonist exenatide twice daily (BID). This was evaluated in the 4B Study, a randomized clinical trial of patients with type 2 diabetes and inadequate glycemic control, pre-treated with optimized basal insulin glargine in combination with metformin². Based in part on evidence from this study, the addition of a GLP-1 receptor agonist is now considered an option in those patients on basal insulin with one or more oral agents whose diabetes remains uncontrolled³.

In choosing the most appropriate add-on treatment for patients with uncontrolled diabetes, insulin-associated weight gain is a key consideration as it can be disabling in patients with type 2 diabetes, potentially adversely affecting cardiovascular risk⁴. Weight gain and hypoglycemia are also associated with indirect consequences; notably each can be a barrier to treatment satisfaction and compliance. Studies have shown that lower treatment satisfaction can lead to reduced treatment compliance, and that these factors can reduce treatment effectiveness⁵.

In addition to evidence concerning treatment effectiveness, an accurate estimation of the comparative costs and cost-effectiveness of type 2 diabetes therapies may help to inform treatment decisions and algorithms in the future. Improving understanding and dissemination of the economic consequences of therapy choice among healthcare providers may, therefore, help to support complex clinical decisions and define favorable treatment practices based on potential benefits, risks, and costs.

The aim of this study was to undertake an applied economic evaluation to estimate the cost-effectiveness of exenatide BID vs insulin lispro TID as add-on treatment options based on the 4B Study results. We illustrate the costs and benefits in the Swedish healthcare setting.

Patients and methods

Cost-effectiveness model

The Cardiff Model, a stochastic simulation model, was utilized for this evaluation⁶. The model is designed to evaluate the impact of therapies in a population of type 2 diabetes patients, modeling disease progression through implementation of the UK Prospective Diabetes Study (UKPDS) 68 outcomes equations⁷. The model has been the basis of a number of applied evaluations of type 2 diabetes therapies^8,9, and has been subject to peer-reviewed validation¹⁰.

The model required specification of age, sex, ethnicity, smoking status, and duration of type 2 diabetes and model changes to modifiable risk factors: HbA_1c, total cholesterol, high-density lipoprotein (HDL) cholesterol, systolic blood pressure, and weight. Time-dependent risk factor profiles were simulated through implementation of equations reported in the UKPDS 68 study⁷. On entering the simulation, patients were initialized to the selected therapy. Following the modification of each patient’s HbA_1c in line with treatment effect, the model projected HbA_1c over time. Pre-specified HbA_1c threshold values were used to invoke escalation to the next line of therapy. Costs were applied to all predicted complications in the year of occurrence. Healthcare maintenance costs were applied in all subsequent years following non-fatal events. Baseline health state utility was modeled using age-dependent mean EQ-5D values (a standardized measure of health outcome) in subjects with no major complications, obtained from the Health Survey for England 2003¹¹. The occurrence of diabetes-related complications is associated with reductions in health-related quality-of-life, which were applied through utility decrements.

Model output included the incidence of microvascular (blindness, nephropathy, amputation) and macrovascular complications (congestive heart failure, myocardial infarction, stroke, ischemic heart disease), hypoglycemia, diabetes-specific mortality and all-cause mortality, point estimates, and probabilistic output for cost-effectiveness. Mortality relating to cardiovascular disorders and diabetes has already been accounted for in the above risk equations. Therefore, cardiovascular and diabetes-related deaths were subtracted from the all-cause mortality. Swedish-specific mortality rates were used.

The model was run over a 40-year time horizon (lifetime perspective) in half-year cycles. Costs (€EUR) and benefits were discounted at a rate of 3% per year. The chosen measure of health benefit was the quality-adjusted life-year (QALY). Cost-effectiveness was expressed as the cost per QALY gained with exenatide BID compared with lispro TID.

Base case analysis

The base case analysis considered the use of exenatide BID in patients with type 2 diabetes inadequately controlled on insulin glargine therapy in combination with metformin, as an alternative to the addition of lispro TID. Baseline characteristics for the economic analysis were sourced from the 4B Study and are presented in full in Table 1. The 4B Study was a randomized clinical trial of 627 patients with type 2 diabetes and inadequate glycemic control (HbA_1c >7.0%, or >53 mmol/mol), pre-treated with optimized basal insulin glargine in combination with metformin. The study compared the efficacy and safety of exenatide BID to mealtime insulin lispro TID over 30 weeks, amongst inadequately controlled patients. Patients had a 12-week run-in period to optimize basal insulin dose to exclude patients who could reach goal with basal insulin alone. Patients were concomitantly treated with basal insulin glargine and metformin in the randomization period². Characteristics at randomization were similar across arms, with patients having a mean age of 59.8 years, mean HbA_1c of 8.2% (66 mmol/mol), and median diabetes duration of 12 years².

Table 1. Data inputs: Baseline characteristics and treatment effects (mean, SE*).

Patient characteristic	Input
Current age (years)	59.80 (0.379)
Proportion female (%)	50
Duration diabetes (years)	12.80 (0.284)
Height (m)†	1.66 (0.17)
Proportion AC	0.01
Proportion current smokers	0.13
HbA1c (%)‡	8.20 (0.037)
Total cholesterol (mg/dl)§	177.56 (39.76)
HDL cholesterol (mg/dl)§	46.32 (12.55)
SBP (mmHg)	135.80 (0.623)
Weight (kg)	89.60 (0.678)
Treatment effects	Model arm		Rescue therapy
	Exenatide BID	Lispro TID	Exenatide BID	Lispro TID
HbA1c (%)†	−1.10 (0.051)	−1.10 (0.051)	−1.93 (0.19)	−1.93 (0.19)
HbA1c (mmol/mol)∥	−12 (0.5)	−12 (0.5)	−21.1 (2.1)	−21.1(2.1)
SBP (mmHg)	−3.77 (0.906)	0.17 (0.889)	n/a	n/a
Total cholesterol (mg/dl)§	−6.56 (1.814)	−1.54 (1.814)	n/a	n/a
HDL cholesterol (mg/dl)§	−1.16 (0.425)	1.16 (0.463)	n/a	n/a
Weight (kg)	−2.60 (0.250)	1.90 (0.240)	4.10 (0.307)	4.10 (0.307)

AC, Afro-Caribbean; BID, twice daily; HDL, high-density lipoprotein; SBP, systolic blood pressure; SD, standard deviation; SE, standard error; TID, three times daily.

Data sourced from intention-to-treat population from 4B Study at time of randomization and at end of study^2,33 and Rosenstock et al.¹² studies.

* Where SD reported, SE was calculated as SE = SD/SQRT (n).

† SE assumed 10% of mean value for height and HbA_1c rescue therapy (data not reported).

‡ Mean = 66 mmol/mol; SE = 0.4 mmol/mol (calculated based on HbA_1c %)

§ Conversion factor from mmol/l to mg/dl =38.6³⁴.

∥ Calculated based on HbA_1c %.

HbA_1c progression

The HbA_1c effects observed in the 4B Study were applied to baseline HbA_1c in the first year of the model (Table 1). HbA_1c progression was modeled as a gradual increase following the applied treatment effect (year 1) to the time of therapy escalation (estimated in the model to occur when HbA_1c reaches its baseline value; year 4.5). Following therapy escalation, HbA_1c was modeled to gradually increase over the remainder of the time horizon (for the base case and longer duration scenarios)⁷.

Rescue therapy was modeled as a switch to intensified basal-bolus therapy based on the study by Rosenstock et al.¹², where patients previously treated with insulin glargine plus oral agents required escalation of insulin therapy. The HbA_1c treatment effect applied in the model at the start of rescue therapy was made proportional to the difference in HbA_1c at the time of switch (8.20%; 66 mmol/mol) and the baseline value observed in the Rosenstock et al.¹² study (8.90%) and is calculated to be −1.93% [(8.20/8.90) × −2.09%].

Weight progression

A weight treatment effect was applied to each arm in the first year of treatment. Weight change over the first 4.5 years of treatment was modeled differently for the exenatide BID and lispro TID arms. For the exenatide BID arm, weight reduction was assumed to apply for the period a patient remains on therapy (4.5 years in the base case analysis), which itself was determined by the time to rescue therapy. The treatment effect was applied in the first year and maintained over the first 3.5 years of treatment, and weight returned to its baseline value over years 3.5–4.5 (including the assumed effect of natural weight gain, 0.1 kg/year). For the lispro TID group, weight gain was applied over the first year of treatment, and natural weight (0.1 kg/year) is assumed for the remainder of time on therapy prior to switch. The weight treatment effect reported in the study by Rosenstock et al.¹² (+4.50 kg) was made proportional to the model’s estimate of average weight across each model arm at the time of switch (90.90 kg) and the baseline weight profile reported by Rosenstock et al.¹² (99.80 kg), calculated as 4.10 kg [(90.90/99.80) × 4.50]. After applying the treatment effect associated with rescue therapy, an annual 0.1 kg natural weight gain was applied in each year of the remaining modeled time horizon.

Systolic blood pressure and cholesterol treatment effects

Treatment effects for systolic blood pressure (SBP), total, and HDL cholesterol were applied to each arm in the first year of treatment (Table 1).

Healthcare costs

The model considered direct medical costs for medications, consumables (i.e. test strips, lancets, and needles), and type 2 diabetes-related complications (including discontinuations and episodes of severe hypoglycemia) (Table 2). All costs were sourced with the Swedish healthcare perspective and expressed in 2015 SEK using the Swedish Consumer Price Index (CPI)¹³. Conversion into €EUR was undertaken using the rate of exchange in 2015¹⁴ of SEK:EUR (0.108).

Table 2. Data inputs: Annual cost of medications and complications (€EUR) and health state utility changes.

Cost of medications	Model arm		Rescue therapy (cost/unit/kg/day)
	Exenatide BID	Lispro TID	Exenatide BID	Lispro TID
Exenatide BID	1,254	—	—	—
Glargine	808	732	0	0
Lispro TID	—	350	0	0
Metformin	47	47	47	47
Consumables (annual)	328	917	917	917
Total	2,437	2,046	—	—
Cost of complications	Fatal	Non-fatal		Maintenance
Ischemic heart disease		12,310		544
Myocardial infarction	5,478	10,380		244
Congestive heart failure	8,139	8,139		770
Stroke	9,078	9,331		394
Amputation	17,763	17,763		288
Blindness		2,338		753
ESRD		13,559		884
Moderate hypoglycemia		374
Discontinuations		142
Health state utility changes	Value		Source
Age dependent baseline	Time dependent		DoH^11
Ischemic heart disease	−0.09		Clarke et al.^35
Myocardial infarction	−0.055		Clarke et al.^35
Congestive heart failure	−0.108		Clarke et al.^35
Stroke	−0.164		Clarke et al.^35
Pre-blind	−0.029		Clarke et al.^35
Blind	−0.074		Currie et al.^20
ESRD	−0.263		Currie et al.^20
Transplant	−0.075		Clarke et al.^35
Amputation	−0.280		Clarke et al.^35
Nausea	−0.04		Matza et al.^36
BMI (utility per unit decrease)	+0.0171		Lane et al.^21
BMI (utility per unit increase)	−0.0472		Lane et al.^21
Discontinuation	0		Assumption

BID, twice daily; BMI, body mass index; ESRD, end-stage renal disease; TID,: three times daily.

Medications

Cost of medications were based on current unit prices¹⁵. Medication costs for exenatide BID and lispro TID were based on usage during the 4B Study². The insulin requirement at the time of switch to rescue therapy is a function of patient weight. Hence, adjusted insulin units per kg per day (IU/kg/day) was calculated [(a) × (b), see below] to be proportional to the difference between the model’s estimate of average weight across each model arm at the time of switch and the baseline weight profile reported in the Rosenstock et al.¹² study [(90.90/99.80), (a)], and the end insulin doses reported in the Rosenstock et al.¹² study (b). This adjustment ensures the absolute costs associated with insulin use and rescue therapy were accurate; the same daily insulin requirement (IU/kg/day) was applied to each arm of the model, from which the total annual insulin cost associated with rescue therapy was calculated (Table 2).

Consumables

Consumable costs were applied to each arm as the sum of costs for test strips, lancets, and needles. Insulin glargine therapy was assumed to include the use of one self-monitoring of blood glucose (SMBG) test strip per day, one lancet for SMBG, and one needle per day. In addition to these, lispro TID therapy was assumed to include the use of three SMBG test strips per day, three lancets for SMBG, and three needles per day. Exenatide BID was assumed to include an additional two needles per day, but no additional SMBG requirement above that of insulin glargine. The cost per unit for test strips, lancets, and needles were €0.46, €0.04, and €0.13, respectively (Table 2)¹⁶.

Complications

The costs of type 2 diabetes-related complications were extracted from a Swedish study by Gerdtham et al.¹⁷ using administrative hospitalization data. For fatal congestive heart failure and amputation events, the cost of a non-fatal event was assumed. Further, event costs in “subsequent years” were used to characterize annual maintenance costs in the model.

The cost associated with the occurrence of blindness was sourced from the study by Henriksson et al.¹⁸. Maintenance costs were characterized as the costs associated with “annual follow-up after first 12 months”.

Regarding hypoglycemia, only the cost of moderate hypoglycemic events was included in the model. The cost for moderate hypoglycemia obtained from a Swedish study¹⁹ was used as it most closely reflected the definition of severe hypoglycemia used in the Cardiff model, the 4B Study and the Rosenstock et al.¹² study.

Finally, drug discontinuations due to adverse events (AEs) that occurred in the first 6 months were attached a standard cost equal to the cost of a visit to a general practitioner (Supplementary Information; Table S1).

Health state utility data

The health state utility decrements associated with age, type 2 diabetes-related complications, and hypoglycemia are reported in Table 2.

The UKPDS 62 data were used to characterize the impact of type 2 diabetes-related complications on health-related quality-of-life. Utility decrements associated with episodes of hypoglycemia were extracted from the study by Currie et al.²⁰; equations characterizing the relationship between the fear of hypoglycemia and health-related utility using the EQ-5D were hard coded into the model. The utility change associated with a one-unit change in body mass index (BMI) was sourced from Lane et al.²¹.

Treatment-related adverse events

A discontinuation rate due to AEs was applied to each arm of the model, based on data from the ITT population of the 4B Study². The same study was used to characterize the number and probability of hypoglycemia episodes. For rescue therapy, numbers were based on data reported in Rosenstock et al.¹².

Sensitivity analyses

Deterministic sensitivity analyses

A number of one-way sensitivity analyses were undertaken to explore the impact of alternative parameter values for key model inputs on estimates of cost-effectiveness: variation in discount rate (discounting 0 to 6%); variation in time horizon (20 years); variation of complication costs (± 20% mean value); variation of complication disutility (± 20% mean value); variation in the utility change associated with a one-unit change in BMI (a 0.006 utility change associated with a one-unit change in BMI²²); variation in the utility change associated with a one-unit change in BMI (a 0.014 utility change associated with a one-unit change in BMI²³).

HbA_1c reduction during the 4B Study was −1.10% in the exenatide group and −1.07% in the lispro TID group (ITT population)². Based on the demonstration of non-inferiority during the 4B Study, HbA_1c treatment effects for exenatide BID and lispro TID treatment arms were considered the same (−1.10%) for the base case analysis; hence, values for these parameters were not assessed in one-way sensitivity analysis. However, the impact of an alternative weight progression profile was explored in the following model analyses for each successive year (Figure 1).

Figure 1. Modeling weight progression.

Duration of treatment. The base case analysis evaluated a switch to rescue therapy after ∼4.5 years of treatment based on a return to baseline HbA_1c of 8.20% (66 mmol/mol). An alternative switching rule was considered where the switch to rescue therapy occurs at 8.90%, corresponding to average treatment duration of 9 years. The HbA_1c and weight treatment effects were then applied as reported in Rosenstock et al.¹². The same assumptions were made; the treatment effect associated with the exenatide BID arm is maintained over the duration of therapy (treatment reduction in year 1, weight effect maintained to year 8, and a gradual return to baseline weight from years 8–9); and the lispro TID arm experiences weight gain over the first year of treatment, and natural weight (0.1 kg/year) is assumed for the remainder of time on therapy prior to switch.

After applying the treatment effect associated with rescue therapy, an annual 0.1 kg natural weight gain is applied in each year of the remaining modeled time horizon for each arm (Figure 1).

Weight convergence. An alternative weight progression profile was considered for the periods following switch to rescue therapy (year 4.5+). This analysis allowed the weight curves describing weight progression in the exenatide BID and lispro TID arms to converge over time such that the difference is reduced.

Probabilistic sensitivity analyses

The impact of the joint uncertainty in the model parameters was explored in a probabilistic sensitivity analysis of the model. A non-parametric bootstrapping approach was used to sample from distributions around the means of input parameters in the model. Cohort characteristics and treatment effects were based on means and standard errors reported in the 4B Study or the source referenced in the base case analysis. Complication costs and utility decrements were based on variation of ±10% of mean value. The results of these analyses are presented as incremental cost-effectiveness ratio (ICER) scatterplots, illustrating the distribution of cost and effect pairs for each scenario, and cost-effectiveness acceptability curves (CEAC), illustrating the proportion of the cost and effect pairs where the intervention is cost-effective for a given willingness to pay per QALY threshold.

Results

Base case cost-effectiveness

In the base case analysis, exenatide BID was associated with an incremental cost of €1,270 and an incremental QALY of +0.64 over 40 years. The cost per QALY gained with exenatide BID compared with lispro TID was €1,971 (Table 3).

Table 3. Expected costs, QALYs, and incremental cost-effectiveness (1000 people over 40 years): base case and deterministic sensitivity analysis (1000 people over 40 years).

Parameter	Costs (€EUR)		QALYs		Incremental		ICER
	Exen. BID	Lispro TID	Exen. BID	Lispro TID	Costs	QALYs	Cost/QALY
Base case	44,526	43,256	11.51	10.86	1,270	0.64	1,971
Discounting 0%	64,850	63,665	15.99	15.11	1,186	0.88	1,344
Discounting 6%	32,968	31,678	8.87	8.37	1,290	0.50	2,557
Time horizon 20 years	35,886	34,563	9.78	9.23	1,324	0.55	2,415
Complication utility (−)	44,526	43,256	11.79	11.25	1,270	0.53	2,394
Complication utility (+)	44,526	43,256	11.23	10.47	1,270	0.76	1,675
Complication costs (−)	43,359	42,091	11.51	10.86	1,268	0.64	1,968
Complication costs (+)	45,693	44,421	11.51	10.86	1,272	0.64	1,974
Cost of SHE^19	44,458	43,282	11.51	10.86	1,265	0.64	1,963
BMI-related utility^23	44,526	43,256	12.30	12.03	1,270	0.27	4,722
BMI-related utility^22	44,526	43,256	12.47	12.31	1,270	0.16	8,076
Weight convergence	44,620	43,256	11.40	10.86	1,365	0.53	2,568
Longer duration of treatment	44,993	43,326	12.04	11.24	2,667	0.80	3,343

Discounted costs, QALYs and cost per QALY per patient. Costs reported in EURO.

BID, twice daily; BMI, body mass index; Exen, exenatide; ICER, incremental cost-effectiveness ratio; QALY, quality-adjusted life-year; SHE, severe hypoglycemic event; TID, three times daily.

Sensitivity analysis

Deterministic sensitivity analyses

Results of the deterministic sensitivity analyses are reported in Table 3. Across analyses, the ICER ranged from €1,344–€8,076. Model estimates were relatively insensitive to individual changes in parameter values for discounting, complication costs, and disutility, holding all else constant. Model estimates were most sensitive to the utility change associated with a one-unit change in BMI.

Probabilistic sensitivity analyses

Results of the probabilistic sensitivity analyses are reported in Supplementary Information Figures S1 and S2. In this analysis, the exenatide BID arm generated a mean QALY gain of 0.75 over a patient’s life-time, at an average incremental cost of €1,102. These estimates correspond to a mean cost per QALY-gained of €1,471 with exenatide BID compared to lispro TID.

The ICER scatterplot (Supplementary Information; Figure S1) illustrates that the distribution of cost and effect pairs from the probabilistic analysis were primarily in the upper right (North East) quadrant of the cost-effectiveness plane; this suggests that exenatide BID provides additional benefit compared to lispro TID at higher cost when used as an add-on treatment to basal insulin in the context of this analysis. The CEAC (Supplementary Information; Figure S2) from this analysis suggests the probability that the exenatide BID arm is cost-effective compared to the lispro TID arm is very high, even at willingness-to-pay thresholds of €5,000, €25,000, and €50,000 (99.9%, 100%, and 100%, respectively), which is considerably below conventional limits of cost-effectiveness: in the Swedish context, no formal willingness-to-pay threshold exists, but interventions less than 500,000 SEK (€54,000) per year QALY-gained may be considered cost-effective²⁴.

Discussion

Informed clinical decisions require the consideration of complex and inter-related factors including patient socioeconomic and medical history, expected therapeutic response to treatment, and the costs associated with treatment. Health outcomes and costs can occur in both the short and longer terms, and, whilst practicing clinicians may not be able to directly observe these consequences, health outcomes and costs are, nonetheless, relevant consequences that follow from treatment decisions. For example, total costs and outcomes of diabetes management will depend on the type of patient and the choice and frequency of use of treatments; in this context, the health and economic consequences of different approaches to patient management included in treatment guidelines may be of particular interest to clinicians. We, therefore, undertook an analysis using an established diabetes model to support complex clinical decisions and define favorable treatment practices based on potential benefits, risks and costs, in patients uncontrolled on basal insulin who have limited therapeutic options. Patients treated with exenatide BID were compared to the strategy of adding bolus insulin for uncontrolled diabetes. The analysis was based on the results of the 4B Study that demonstrated similar glycemic control with less hypoglycemia with exenatide BID when compared with lispro TID. Furthermore, exenatide BID was associated with a weight loss of 2.6 kg and lispro TID with a weight gain of 1.9 kg (ITT population)².

Over the projected lifetime of patients, the exenatide BID arm was associated with more health benefits (0.64 additional QALYs). The QALY gains associated with exenatide BID were driven by the superior weight profile associated with exenatide BID vs lispro TID regimens over the modeled time horizon and the avoidance of the disutility associated with weight gain. The incremental cost per QALY-gained estimate was €1,971, which is within conventional limits of cost-effectiveness in Sweden as well as in other European countries and the US. The cost-effectiveness of exenatide BID based on the 4B Study has also been evaluated in the Spanish and Turkish healthcare settings, where exenatide BID was predicted to be a cost-effective alternative to lispro TID^25,26.

The sensitivity of estimated cost-effectiveness was evaluated for alternative weight profiles for exenatide BID and lispro TID. Where the weight progression benefit associated with the exenatide BID arm was reduced by allowing weight to converge between the two groups over time, cost-effectiveness was reduced from €1,971 to €2,568 per QALY-gained, but remained well within conventional limits of cost-effectiveness. In the scenario where treatment duration with exenatide BID and lispro TID regimens was increased from 4.5 to 9 years, the estimated cost-effectiveness ratio increased from €1,971 to €3,343 due to a longer duration on these regimens.

Cost-effectiveness results were generally insensitive to alternative values of key model input parameters. Model estimates were most sensitive to the change in utility associated with changes in BMI. Nonetheless, the ICERs were all within acceptable limits of cost-effectiveness for all modeled BMI-related utility values. In the probabilistic sensitivity analysis the probability that the exenatide BID arm is cost-effective compared to the lispro TID arm was estimated to be 99.9% at a willingness-to-pay threshold of €5,000 and 100% at €50,000. Hence, the base case result that exenatide BID is a cost-effective treatment alternative to lispro TID as an add-on to basal insulin was robust to alternative assumptions and values for key model parameters. Nonetheless, there remain areas of uncertainty that may impact estimated cost-effectiveness. The Cardiff model uses the UK Prospective Diabetes Study (UKPDS) 68 outcomes equations to predict disease progression through type 2 diabetes⁷, similar to other widely used models of type 2 diabetes¹⁰. Although BMI and weight are associated with increased risk of complications and mortality in patients with type 2 diabetes²⁷, BMI is only identified as a risk factor in one of the equations used to estimate the probability of diabetes-related complications (congestive heart failure)⁷. Thus, models of disease progression using these equations may under-estimate the impact of reduced risk of complications from treatment strategies with a proven beneficial effect on weight in terms of reduced healthcare resource utilization and improved quality-of-life.

Further, the pre-specified HbA_1c threshold used to invoke escalation of therapy within this study is higher than that typically recommended by international guidelines on diabetes management^28,29. There is a documented disconnect between guidelines and clinical practice; one UK-based study reported a mean HbA_1c level of 9.4% in patients with type 2 diabetes intensified with insulin, while a systematic review reported that mean HbA_1c at insulin intensification ranged from 7.52–10.7% in identified studies^30,31. The HbA_1c threshold utilized within this study is well within this range. Even though the HbA_1c threshold in this study was above that recommended by guidelines, model sensitivity analysis based on alternative HbA_1c thresholds did not significantly impact the estimated cost-effectiveness of exenatide BID, as seen in Table 3.

The analysis conducted represents a long-term evaluation of costs and effects using a simulation model that has been consistently used in comparable cost-effectiveness analyses of exenatide BID vs insulin across Europe, including Spain and Turkey^25,26. Results of these studies, which apply the same methodology (i.e. use of UKPDS equations, 3% discounting and healthcare payer perspective) support the result presented herein; ICERs with high probability of cost-effectiveness (near 100%) were predicted across studies, with similar QALY gains estimated^25,26. The use of direct medical costs only could be considered a limitation; however, this approach was selected to be transparent and would be considered conservative³². Another potential limitation of the study is the use of utilities that were not collected alongside the pivotal trial, and are, therefore, not as contemporary as the 4B Study data, and are from the UK. This may raise queries around the generalizability of these inputs to the Swedish population; however, a single source for Swedish-specific inputs for all of the diabetes-related events evaluated could not be identified; conversely, the UKPDS utilities have been estimated with a consistent methodology, which makes these utilities comparable across different diabetes-related events. Finally, the inputs underlying the presented cost-effectiveness analysis were based on the findings of the 4B Study, a randomized clinical trial of patients with type 2 diabetes and inadequate glycemic control, pre-treated with optimized basal insulin glargine in combination with metformin²; a systematic review or network meta-analysis of different treatment options for patients who are uncontrolled on basal insulin would be an interesting topic for future research, which is beyond the scope of the objectives of this study.

Conclusion

In conclusion, the practice of adding bolus insulin when glycemic control is inadequate with basal insulin is associated with weight gain and an increased incidence of hypoglycemia, and may be viewed as sub-optimal from the patient and provider perspectives. The use of exenatide BID as a treatment alternative to bolus insulin has been associated with similar glycemic control, whilst reducing patient body weight and the incidence of hypoglycemia. The addition of a GLP-1 receptor agonist is considered an option in patients not achieving optimal glycemic control on basal insulin with one or more oral agents³. With a base case estimated ICER of €1,971, evaluated against commonly used cost-effectiveness threshold values, exenatide BID represents a cost-effective treatment alternative to lispro TID as add-on therapy to titrated insulin glargine in patients with type 2 diabetes.

Transparency

Declaration of funding

AstraZeneca funded the study and preparation of this manuscript.

Declaration of financial/other relationships

JG and PM are employees of HEOR Ltd. US and BK are employees of AstraZenenca. JME peer reviewers on this manuscript have no relevant financial or other relationships to disclose.

Acknowledgments

We acknowledge writing support from Dr Daniel Sugrue and Beverley Jones (HEOR Ltd). The study authors extend our thanks to the participating research physicians, nurses, and patients who made this study possible.