A cost–utility analysis of dulaglutide versus insulin glargine as third-line therapy for Type 2 diabetes in Canada

A systematic literature review was conducted to identify randomized controlled trials (RCTs) in adult patients diagnosed with T2DM and inadequately controlled with metformin and sulfonylurea combination therapy. The search focused on identifying RCTs of all diabetes interventions currently approved for use in Canada, the USA or the EU. Searches were conducted across Medline, EMBASE and the Cochrane Central Register of Controlled Trials; the search strategy for Medline and EMBASE is presented in Supplementary Table 1. All abstracts identified from the searches were assessed for potential eligibility in duplicate, and any disagreement in eligibility was resolved by discussion or by consulting a third adjudicator. Potentially eligible full-text articles were similarly assessed in duplicate and any disagreement resolved by a third adjudicator. A network diagram was constructed based on the studies identified (Supplementary Table 2) in the literature review, with comparators grouped into therapeutic classes (Supplementary Figure 1). A Bayesian NMA was also performed around HbA1c, BMI and severe hypoglycemia, but since the literature review identified only one trial directly comparing insulin glargine with dulaglutide 1.5 mg (AWARD-2) [12], the AWARD-2 data were ultimately used to inform the base case analysis, with the NMA results (Supplementary Material) being used as the basis of a sensitivity analysis. This trial was therefore used to inform the cohort characteristics and treatment effects employed in the analysis.

A patient-level cost-utility model was developed in Microsoft Excel (Microsoft Corporation, WA, USA) closely following the model structure of the UKPDS OM, described as a ‘probabilistic discrete-time illness-death model’ in the original publication [10]. The incidence of complications was estimated based on risk equations from the UKPDS OM, which has been used by the CADTH in their optimal use report [9]. The UKPDS OM structure, risk equation functional forms, and coefficients have been described previously [10]. Specifically, the model captured incidence of ischemic heart disease, myocardial infarction, stroke, congestive heart failure, end-stage renal disease, blindness and amputation (based on the UKPDS risk equations) and nonsevere and severe hypoglycemia. In addition to implementing the Weibull risk equations from the UKPDS OM, the model incorporated a number of adaptations. Notably, the model included the ability to switch patients to a subsequent treatment at a specific time or in response to reaching a specified risk factor threshold. The model also allowed health-related quality of life disutilities to be associated with self-monitoring of blood glucose (SMBG) testing, excess BMI, including the option to avoid double counting the effect of BMI by simultaneously excluding it as a driver of complication risk. Finally, the model could switch between the UKPDS Gompertz proportional hazards model and Canadian life tables to capture background mortality risk [10,14].

The model was used to run a base case analysis comparing dulaglutide 1.5 mg with insulin glargine based on the AWARD-2 study. In the base case, 20,000 simulated patients were run through the model, at which point incremental QALE and cost had converged sufficiently (with two different random seeds of 27,183 and 31,416; e and π × 10⁵, respectively, rounded to the nearest integer) that further simulations would not have contributed useful information to the analyses.

Baseline cohort risk factors were based on the overall population in the AWARD-2 trial (Table 1). AWARD-2 was an open-label, parallel-arm, randomized, multicenter trial comprising three periods: screening/lead-in period of 10–12 weeks, 78 week treatment period and a 4 week safety follow-up period [12]. As the baseline prevalence of all complications was not available from the AWARD-2 population, surrogate estimates were taken from the economic evaluation included in the CADTH therapeutic review of third-line therapies (Table 1), which was in turn derived from the DICE Study or the Ontario and Alberta Diabetes Atlases [15–18]. The cohort characteristics employed by CADTH were relatively well matched to those in AWARD-2 with a similar age (57.7 years vs 56.7 years in AWARD-2), duration of diabetes (9.5 years vs 9.1 in AWARD-2) and HbA1c (8.6% vs 8.1% in AWARD-2).

Treatment effects and rates of hypoglycemia were taken directly from the insulin glargine and dulaglutide arms of the trial at 52 weeks (Table 2), with treatment effects being applied to the baseline characteristics at the start of the analysis. HbA1c and systolic blood pressure (SBP) risk factor progressions over time were modeled in line with the UKPDS OM panel regression equations as previously published [10]. All other risk factors were held constant over time. Treatment intensification to a basal-bolus insulin regimen was modeled at an HbA1c threshold of 9.0% in line with a sensitivity analysis conducted in the CADTH therapeutic review of third-line therapies. At the point of intensification, changes in HbA1c and BMI were applied based on the difference in HbA1c and BMI change in the basal and prandial arms of the 4-T trial (-0.2% and +1.0 kg/m², respectively) [19]. No change in SBP was modeled based on the non-significant difference between the basal and prandial arms in the 4-T trial [19]. Subsequent to intensification, risk factor progressions were modeled as previously; HbA1c and SBP followed the UKPDS panel regression equations while all other risk factors were held constant for the remainder of the simulation [19]. Rates of nonsevere and severe hypoglycemia after intensification were applied from the prandial insulin arm of the 4-T study (5.5 nonsevere events/patient/year and 0 severe events/patient/year) [19], and SMBG test strip use was assumed to increase to four tests/day in both arms when treatments were intensified.

Drug costs were taken from the Ontario Drug Benefit Formulary/Comparative Drug Index, effective December 2016 (Table 3) [20]. The insulin glargine dose was based on the mean dose of 32.12 international unit (IU)/day in the AWARD-2 trial. Costs associated with each diabetes complication were taken from the 2013 update to the CADTH optimal use report and expressed in 2016 CAD by inflating the CADTH costs (in 2012 CAD) using the Health Component of the Canadian Consumer Price Index (Table 4), which was 120.7 in 2012 and 124.7 in 2016 yielding a multiplier of 1.033 [9].

SMBG test strips were included in the analysis, assuming 3.85 tests per week for dulaglutide based on the maximum reimbursement of 200 test strips for nonhypoglycemia inducing medications in patients with T2DM in Ontario, and 2.08 tests per day for insulin glargine based on a 2010 study by Gomes et al. [21]. The maximum reimbursable number of strips was selected for dulaglutide as a conservative estimate of the SMBG test cost ultimately borne by the healthcare payer while the 2.08 tests per day for insulin glargine was based on utilization data from Canada. The cost per strip was taken to be CAD 0.79 based on the 2016 Ontario drug benefit formulary price of CAD 0.729 plus pharmacy fee [20].

Insulin costs after intensification were assumed to be the same in both arms, based on a market-share weighted cost of basal and bolus insulin, assuming total insulin use of 0.4 IUs/kg/day split into 40% basal insulin and 60% prandial insulin, based on the mid-point of the recommended dose range (and the exact dose split) from the Canadian diabetes association clinical practice guidelines (Table 3) [22].

Health state utility values were taken from a US EQ-5D score catalogue in combination with an amputation disutility based on a Tobit model-derived tariff value (Table 5) [23–25]. The ‘baseline’ quality of life was taken to be 0.753, based on a US catalog of EQ-5D scores from Sullivan et al. and utilized by CADTH in the analyses presented in the Optimal use report of third-line pharmacotherapy for Type 2 Diabetes [9,24]. Disutilities associated with hypoglycemia were taken from a study by Harris et al., which employed a time trade-off approach to estimate the utility values associated with nonsevere and severe hypoglycemia [26]. All 1696 patients who completed the analysis were Canadian, with 49.3% coming from Ontario, 13.3% from British Columbia, 9.2% from Alberta and at least one participant coming from each of the remaining ten provinces/territories [26].

A disutility of -0.0061 (standard error 0.001) was applied in the base case analysis for every unit of BMI above 25 kg/m² in line with the CODE-2 study published by Bagust and Beale in 2005 [27]. The study used a time trade-off approach to establish the disutility associated with each unit of BMI over 25 kg/m² in a European population enrolled. The CODE-2 population was 50% male and had a mean age at enrollment of 65.9 years, a mean BMI of 28.7 kg/m² and a mean time since diagnosis of diabetes of 9.3 years [28]. Finally, in addition to the complication, BMI and hypoglycemia-related disutilities, a disutility was also associated with each SMBG test based on an annual disutility of -0.0058 associated with weekly testing, yielding a disutility per test of -0.000111 [29].

The analysis was conducted over a 40-year time horizon, corresponding to patient lifetimes. Future cost and effectiveness outcomes were discounted at a rate of 1.5% per annum in line with CADTH guidelines on economic evaluation of health technologies [30].

A series of one-way sensitivity analyses were conducted around the AWARD-2 base case parameters to establish the magnitude of the contributions of individual model parameters to model outcomes. The influence of the analysis time horizon on projected model outcomes was investigated by running analyses over 10, 20 and 30 years. Cost and effectiveness discount rates were set to 0 and 3% (symmetrically) to establish the importance of discounting values to present-day values. Sensitivity analyses were also performed in which the cost of all diabetes complications was increased and decreased by 10% relative to the base case values. Five simulations were then run to assess the key drivers of clinical benefit associated with dulaglutide; in the dulaglutide arm, all risk factors were set to the same values as the insulin glargine arm with changes in HbA1c, systolic blood pressure, serum lipids, BMI and hypoglycemia rates from the comparator arm applied in turn. Disutilities of -0.01 and -0.0000048 were associated with each severe and nonsevere hypoglycemic event based on the CADTH therapeutic review of third-line therapy for Type 2 diabetes base case value. A second analysis was conducted in which a disutility of -0.0033 was applied in line with a sensitivity analysis from the same report and originally reported by Levy et al. for mild-to-moderate hypoglycemia [31]. The -0.0061 disutility associated with each BMI unit above 25 kg/m² was abolished, and the -0.000111 disutility associated with each additional SMBG test was abolished in a separate analysis and then both the BMI and SMBG disutilities were abolished in the same analysis [27,29]. An insulin glargine dose of 0.53 IUs/kg/day was used to calculate the cost of insulin glargine (in place of the dose from AWARD-2) based on the dose for insulin analogs used in the CADTH therapeutic review of third-line therapy for Type 2 diabetes. The per-kilo dose was converted to a mean dose per patient per day using the mean patient bodyweight in the insulin glargine arm at the end of AWARD-2 (86.31 kg + 1.44 kg = 87.75 kg). Finally, an analysis was run in which the base case assumption of intensifying to basal-bolus insulin therapy at an HbA1c threshold of 9% was changed to intensify therapy after 5 years.