Comparison of costs and outcomes of dapagliflozin with other glucose-lowering therapy classes added to metformin using a short-term cost-effectiveness model in the US setting

Abstract

Objective: To compare 1-year costs and benefits of dapagliflozin (DAPA), a sodium-glucose cotransporter-2 (SGLT-2) inhibitor, with those of other treatments for type 2 diabetes (T2D), such as glucagon-like peptide-1 receptor agonists (GLP-1RAs), sulfonylureas (SUs), thiazolidinediones (TZDs), and dipeptidyl peptidase-4 inhibitors (DPP-4i), all combined with metformin.

Methods: A short-term decision-analytic model with a 1-year time horizon was developed from a payer’s perspective in the United States setting. Costs and benefits associated with four clinical end-points (glycated hemoglobin [A1C], body weight, systolic blood pressure [SBP], and risk of hypoglycemia) were evaluated in the analysis. The impact of DAPA and other glucose-lowering therapy classes on these clinical end-points was estimated from a network meta-analysis (NMA). Data for costs and quality-adjusted life-years (QALYs) associated with a per-unit change in these clinical end-points were taken from published literature. Drug prices were taken from an annual wholesale price list. All costs were inflation-adjusted to December 2016 costs using the medical care component of the consumer price index. Total costs (both medical and drug costs), total QALYs, and incremental cost-effectiveness ratios (ICERs) were estimated. Sensitivity analyses (SA) were performed to explore uncertainty in the inputs. To assess face validity, results from the short-term model were compared with long-term models published for these drugs.

Results: The total annual medical cost for DAPA was less than that for GLP-1RA ($186 less), DPP-4i ($1,142 less), SU ($2,474 less), and TZD ($1,640 less). Treatment with DAPA resulted in an average QALY gain of 0.0107, 0.0587, 0.1137, and 0.0715 per treated patient when compared with GLP-1RA, DPP-4i, SU, and TZD, respectively. ICERs for DAPA vs SU and TZD were $19,005 and $25,835, respectively. DAPA was a cost-saving option when compared with GLP-1RAs and DPP-4is. Among all four clinical end-points, change in weight had the greatest impact on total annual costs and ICERS. Sensitivity analysis showed that results were robust, and results from the short-term model were found to be similar to those of published long-term models.

Conclusion: This analysis showed that DAPA was cost-saving compared with GLP-1RA and DPP-4i, and cost-effective compared with SU and TZD in the US setting over 1 year. Furthermore, the results suggest that, among the four composite clinical end-points, change in weight and SBP had an impact on cost-effectiveness results.

Keywords:

• Short-term
• cost-effectiveness
• dapagliflozin
• economic benefits
• type 2 diabetes

Introduction

Diabetes mellitus is a chronic, progressive metabolic disease characterized by elevated levels of blood glucose (hyperglycemia). This disease has reached epidemic proportions, with millions of people affected across the globe. According to the World Health Organization, the number of patients with diabetes has almost quadrupled from 1980 to 2014, accounting for nearly 422 million people with diabetes worldwide¹. National Diabetes Statistics Report estimates showed that the US had ∼29 million patients with diabetes (9.3% of the total population as of the year 2014)².

Type 2 diabetes mellitus (T2D) is the most common form of diabetes, accounting for 90–95% of the total diabetic population³. It is caused by relative insulin deficiency and insulin resistance³. Patients with T2D have a high risk of developing both macrovascular and microvascular complications that have a major impact on healthcare resources⁴.

The management of T2D, along with its comorbidities and vascular complications, has become a major healthcare challenge. According to the National Diabetes Statistics Report (2014), the estimated direct and indirect medical costs in the US for the year 2012 were found to be $176 billion and $69 billion, respectively^2,4.

The pharmacological treatment of T2D includes various glucose lowering therapy classes. Following lifestyle modifications, metformin is recommended as first-line therapy⁵. As per American Diabetes Association (ADA)⁶ guidelines, metformin has the highest long-term safety profile compared to other glucose lowering therapies. Depending on a patient’s characteristics and blood glucose levels, other glucose lowering therapy classes, such as glucagon-like peptide-1 receptor agonists (GLP-1RAs), sodium-glucose cotransporter-2 inhibitors (SGLT-2is), dipeptidyl peptidase-4 inhibitors (DPP-4is), thiazolidinediones (TZDs), and sulfonylureas (SUs), may be used in the aforementioned order as monotherapy or in combination with other glucose lowering therapy classes, as recommended in the AACE T2D management algorithm⁵.

The fact that sodium-glucose cotransporter-2 inhibitors (SGLT-2is) control glucose levels through an insulin-independent mechanism makes it different from other glucose lowering therapy classes. Drugs from the class of SGLT-2is can be administered with any other glucose lowering therapies including insulin due to their unique mechanism of action. SGLT-2is have shown minimal risk of hypoglycemia and substantial reduction in weight gain, even though it is administered as a combination with insulin therapy^5,7. Dapagliflozin (DAPA) was the first drug from the SGLT-2i class to reach the global market. DAPA acts by inhibiting sodium-glucose cotransporter sub-type 2 proteins (SGLT-2) that play a key role in renal glucose re-absorption by absorbing 90% of glucose filtered by the kidney. DAPA lowers plasma glucose levels primarily by increasing urinary glucose excretion⁸. The efficacy and safety of DAPA have been evaluated in multiple phase III randomized controlled trials^9–12. DAPA was approved by the FDA in January 2014 for adults with T2D as monotherapy, as add-on combination therapy, or as an adjunct to diet and exercise to improve glycemic control^9,11,13–15. Additional advantages of treatment with DAPA include modest weight loss, decreased blood pressure, and low risk of hypoglycemia⁸.

Several long-term models such as the Center for Outcomes and Research (CORE) Diabetes Model, Cardiff Diabetes Model, and the Januvia Diabetes Economic (JADE) model have been developed to compare the economic impact of alternative glucose lowering therapies^16–18. All of these models have a long time horizon, i.e. the costs and benefits are estimated over a patient’s lifetime. The costs and benefits in these long-term models were estimated using projections that extrapolate significantly beyond the duration of the clinical trial data. However, short-term models based on the actual trial length might give more accurate results and be preferable over long-term models for estimating costs and benefits in the short-term. We developed a short-term economic model (1-year time horizon) that estimates the costs and benefits of DAPA vs other glucose-lowering therapy classes added on to metformin (GLP-1RAs, DPP-4is, TZDs, and SUs) from the US third-party payer perspective. Furthermore, we also compared the incremental cost-effectiveness ratios (ICERs) obtained from our short-term cost-effectiveness model for the US with long-term cost-effectiveness models published for various countries^16–24.

Methods

Model description and model comparators

A short-term decision-analytic economic model was developed to evaluate the costs and benefits among patients with T2D treated with DAPA vs other glucose lowering therapy classes added as add-on therapy to metformin. The model compared DAPA with four different drug classes: GLP-1RA (liraglutide), DPP-4i (vildagliptin, linagliptin, saxagliptin, sitagliptin), TZDs (pioglitazone), and SUs (glimepiride, glipizide, gliclazide). Four major clinical end-points were included in the analysis: glycated hemoglobin (A1c), body weight, systolic blood pressure (SBP), and hypoglycemia. A 1-year time horizon was used to assess the differences in costs, quality-adjusted life-years (QALYs), and ICERs. Differences in costs and QALYs were estimated by applying medical costs per unit change in clinical end-points to the corresponding change in clinical end-points due to the treatments. All the model calculations, including sensitivity analyses, were performed using Microsoft Excel VBA version 2016. A schematic diagram of the model inputs and calculations is presented in Figure 1.

Figure 1. Schematic diagram of model inputs and model calculations. NMA, network meta-analysis; QALY, quality adjusted life years; A1c, glycated hemoglobin.

Model inputs and data sources

Differences in clinical end-points

Table 1 provides the changes in A1C, body weight, SBP, and hypoglycemia rate associated with a 52-week treatment with DAPA and other comparators as add-on therapy to metformin in patients with T2D, who were obtained from a published Network Meta-Analysis (NMA) study²⁵. The average age, weight, A1C, and duration of diabetes of the population included in the NMA were 57.5 years, 89.53 kg (197.38 lbs), 7.98%, and 6.02 years, respectively²⁵.

Table 1. Input parameters.

Input parameters	Average	Lower bound	Upper bound	Distribution
DAPA vs GLP-1RA
Change in A1c (%)	0.41	−0.01	0.84	Normal
Change in Weight (kg)*	−0.5 (−1.1)	−3.1 (−6.8)	2.0 (4.4)	Normal
Change in SBP (mmHg)	−1.48	−6.01	3.05	Normal
Change in Hypoglycemia Rate (Hypoglycemia patient proportion per patient-year-%)	NR	NR	NR	Normal
DAPA vs DPP-4i
Change in A1c (%)	−0.11	−0.42	0.22	Normal
Change in Weight (kg)	−2.6 (−5.7)	−4.5 (−9.9)	−0.7 (−1.5)	Normal
Change in SBP (mmHg)	−2.20	−5.97	1.57	Normal
Change in Hypoglycemia Rate (Hypoglycemia patient proportion per patient-year-%)	−1.29	−2.58	4.68	Normal
DAPA vs TZD
Change in A1c (%)	−0.02	−0.45	0.40	Normal
Change in Weight (kg)	−4.8 (−10.6)	−7.3 (−16.1)	−2.2 (−4.8)	Normal
Change in SBP (mmHg)	NR	NR	NR	Normal
Change in Hypoglycemia Rate (Hypoglycemia patient proportion per patient-year-%)	−1.29	−2.76	11.43	Normal
DAPA vs SU
Change in A1c (%)	0.00	−0.29	0.29	Normal
Change in Weight (kg)	−4.7 (−10.4)	−6.4 (−14.1)	−2.9 (−6.4)	Normal
Change in SBP (mmHg)	−5.10	−8.27	−1.93	Normal
Change in Hypoglycemia Rate (Hypoglycemia patient proportion per patient-year-%)	−25.65	−26.73	−21.87	Normal
Cost inputs
Cost per 1% A1c increase	$326.04	$163.42	$543.87	Gamma
Cost per 1 kg Weight increase	$338.69	$169.77	$564.97	Gamma
Cost per 1 mmHg SBP increase	$94.49	$47.36	$157.62	Gamma
Cost per 1% Patient-Year Hypo Risk increase	$1,614.07	$809.04	$2,692.44	Gamma
QALY inputs
QALY per 1% HbA1C decrease	0.0220	0.0110	0.0367	Gamma
QALY per 1 kg weight decrease	0.0149	0.0075	0.0249	Gamma
QALY per 1 mmHg SBP decrease	0.0080	0.0040	0.0133	Gamma
QALY per 1% Patient-Year Hypo Risk decrease	0.0142	0.0071	0.0237	Gamma

Abbreviations. GLP-1RA, glucagon like peptide-1 receptor agonists; DPP-4i, dipeptidyl peptidase-4 inhibitors; SUs, sulphonylureas; TZDs, thiazolidinediones; DAPA, dapagliflozin; T2D, Type 2 Diabetes Mellitus; SBP, systolic blood pressure; A1c, glycated hemoglobin; QALY, quality adjusted life years; A1c, glycated hemoglobin; SBP, systolic blood pressure.

*Weight in Kg (lbs).

Medical costs

The medical costs associated with change in clinical end-points (A1C, body weight, SBP, and hypoglycemia rate) were obtained from published literature^26–29. All the medical costs were inflation adjusted to December 2016 using the medical care component of the Consumer Price Index (Appendix 1, Table 1).

Table 2 presents the medical costs associated with each clinical end-point for each comparison pair with DAPA.

Table 2. Differences in medical costs for clinical end-points.

Change in clinical end-points	DAPA vs GLP-1RA	DAPA vs DPP-4i	DAPA vs TZDs	DAPA vs SUs
A1c	$133.68	−$35.86	−$6.52	$0.00
Weight	−$179.51	−$877.21	−$1,612.16	−$1,578.30
SBP	−$139.85	−$207.88	NR	−$481.90
Hypoglycemia risk	NR	−$20.82	−$20.82	−$414.01
Total medical cost difference	−$185.67	−$1,141.77	−$1,639.51	−$2,474.20

Abbreviations. T2D, type 2 diabetes mellitus; DAPA, dapagliflozin; GLP-1RA, glucagon like peptide-1 receptor agonists; DPP-4i, dipeptidyl peptidase-4 inhibitors; TZDs, thiazolidinediones; SUs, sulphonylureas; A1c, glycated hemoglobin; SBP, systolic blood pressure; NR, not reported.

The costs were inflated to 2016; negative numbers indicate lower costs for DAPA.

Aagren and Luo²⁶ reported that, on average, a 1-unit increase in A1c leads to a 4.4% increase in T2D-related medical costs. T2D-related medical costs included costs associated with physician office visits and emergency room visits, hospital inpatient and outpatients’ costs, and costs associated with laboratory tests, rehabilitation facilities, and nursing homes. In this study, the annual total diabetes-related cost for a patient with T2D was $5,680 (unadjusted). After inflation adjustment to December 2016 costs, this resulted in a $326.04 annual increase in costs due to a 1-unit change in A1c.

Incremental cost associated with change in weight was calculated using total medical cost estimated from pharmacy claims and medical claims, respectively²⁹. In this study, Yu et al.²⁹ concluded that healthcare cost in patients who gained weight was significantly higher than in those who did not gain weight. A 1% increase in weight was associated with an ∼3% change in total healthcare cost, which translated into a $338.69 change in total cost (inflation adjusted to December 2016) for every 1 kg (2.20 lbs) change in weight.

The impact of change in SBP on the total cost was based on a study by Kulchaitanaroaj et al.²⁷, where the total cost included costs of provider time, laboratory test costs, and the cost of anti-hypertensive medications. This study found that provider time was a significant factor in determining the average incremental cost between the control and intervention groups. The total costs associated with the control group and intervention group were $445.75 and $774.90, respectively. The reduction in SBP between the intervention and control groups was 9.08 mmHg, which translated into an annual expenditure of $94.49 (as per inflation adjusted to December 2016 costs) per 1 mmHg SBP change²⁷.

In a study by Quilliam et al.²⁸, three cost categories were considered: inpatient admission cost, emergency department costs, and outpatient visits cost. These categories included three types of mutually exclusive costs, i.e. hypoglycemia costs, other diabetes-related costs, and other events costs. In our study, we considered the cost associated with hypoglycemic events. The total hypoglycemia costs were $52,223,675. On average, the population was followed up to 2.26 years. There were 18,657 patients who experienced hypoglycemic events with a follow-up period of 42,168.92 person-years. Hence, the mean hypoglycemia cost per year was $1,238.44²⁸. Annual per patient hypoglycemia cost, after inflation adjustment to December 2016 costs, was $1,614.07.

Drug costs

The drug costs were estimated from annual wholesale list prices as of June 2016³⁰. Drug costs were added to medical costs to estimate the total cost associated with each treatment in the model (Table 3).

Table 3. Average drug prices for each glucose-lowering therapy class.^a

Category	Daily drug price	Annual weighted list price - 2016 cost level
GLP-1RA	$19.95	$7,281.75
DPP-4i	$12.11	$4,420.15
SU	$0.36	$131.94
TZD	$3.50	$1,279.29
DAPA	$13.06	$4,766.90

^aAnaly$ource Online. https://www.analysource.com/

Abbreviations. GLP-1RA, glucagon like peptide-1 receptor agonists; DPP-4i, dipeptidyl peptidase-4 inhibitors; SUs, sulphonylureas; TZDs, thiazolidinediones; DAPA, dapagliflozin.

QALY inputs

The QALY gains associated with the clinical end-points were obtained from published literature (Appendix 2; Table 1). Vijan et al.³¹ provided QALY gain estimates for a 1% reduction in A1c for patients, based on their age and disutility from diabetes treatments. The range of disutilities from diabetes treatment were reported as 0.001 for taking daily diabetes treatment pills, to 0.05 for insulin treatment. The estimated utility gain from a 1% reduction in A1c level for a 57.5-year-old patient was 0.022³¹. Matza et al.³² assumed a disutility of 0.04 QALY per 3% gain in weight. This translated into a 0.0149 QALY loss associated with a 1kg (2.20 lbs) weight gain³². A detailed estimation of QALY inputs is given in Appendix 2.

A study by Timbie et al.³³ estimated that patients undergoing blood pressure treatment intensification had an average reduction in SBP of 14.8 mmHg, a gain of 1.35 QALYs, and a mean quality-adjusted life expectancy of 10.1 years. This translated into a QALY gain of 0.008 per 1 mmHg change in SBP³³. Currie et al.³⁴ used the hypoglycemia fear survey (HFS) to measure specific concerns and worries of patients with T2D. The study concluded that a symptomatic hypoglycemic episode results in a 0.0142 decrement in utility (Appendix 2).

Sensitivity analyses

To assess the robustness of results over plausible ranges of inputs, univariate and multivariate sensitivity analyses were performed. One-way sensitivity analyses (OWSA) were conducted to determine the effects of input parameters on total cost differences and ICERs. Changes in clinical end-points were varied within the lower and upper 95% credible interval provided in the NMA report²⁵ (Table 1). The medical costs and QALYs associated with each clinical end-point were varied between 95% confidence interval (CI) using gamma distribution and assuming a standard error of 30%. Probabilistic sensitivity analyses (PSA) were performed by varying difference in clinical end-points, medical cost, and QALYs associated with unit change in clinical end-points for DAPA vs other glucose lowering therapy classes using different statistical distributions.

Comparison of short-term model with long-term models

To evaluate the face validity of the models, we compared the ICERs obtained from our short-term model with the ICERs reported in the published long-term economic models^16–24. We found multiple studies for comparison of DAPA vs SUs, DPP-4is vs SUs, GLP-1RA vs DPP-4is, and DPP-4is vs TZDs. Only one published paper was available for comparison of DAPA vs TZDs, and GLP-1RAs vs SUs. Incremental costs in the published studies were inflated to December 2016 cost levels and converted to US dollars. In a scenario analysis, we adjusted the prices in our short-term model to obtain a drug price ration similar to those in the long-term studies. The drug price ratio of the comparison pair studied in the long-term models was used to calculate the drug prices in the short-term model (see Adjusted Drug Price Ratio in Table 4).

Table 4. Comparison of ICERs from our short-term model with the long-term models.*†

Short-term model					Published long-term models	Published long-term model references
Comparators	ICERs	Country	Adjusted drug price ratio from long-term model	ICERs after adjusting drug prices
DAPA vs SUs	$19,005	UK	16:1	$17,545	$3,807	Charokopou et al.^16
		UK	10:1	$15,973	$3,807	Cummins et al.^19
		Nordic countries
		Denmark	8:1	$14,925	$8,903	Sabale et al.^23
		Finland	4:1	$9,684	$6,089
		Norway	4:1	$9,684	$5,698
		Sweden	6:1	$13,178	$6,806
DPP-4i vs SUs	$53,681	Sweden	10:1	$48,035	$9,990	Granström et al.^21
		Germany	5:1	$39,999	$16,168	Erhardt et al.^20
		European countries
		Austria	6:1	$42,678	$26,836	Schwarz et al.^18
		Spain	5:1	$39,999	$16,674
		UK	9:1	$47,142	$15,743
		Portugal	6:1	$42,678	$7,277
		Finland	8:1	$46,026	$17,515
		Sweden	10:1	$48,035	$14,731
DPP-4i vs TZDs	$80,788	Austria	1:1	−$23,267	−$4,148	Schwarz et al.^18
		Spain	1:1		$188
		UK	1:1		$3,068
		Portugal	1:1		−$13,425
		Finland	1:1		$6,019
		Sweden	1:1		−$4,967
GLP-1RA vs DPP-4i	$40,240	Greece	3:1	$81,870	$15,678	Tzanetakos et al.^24
		Spain	3:1	$81,870	$15,051	Mezquita et al.^22
		UK	2:1	$56,518	$16,617	Davies et al.^17
DAPA vs DPP-4i	−$13,540	UK	1:1	−$19,446	$9,631	Charokopou et al.^35
		UK	1:1		−$10,630	Cummins et al.^19
DAPA vs TZDs	$25,835	UK	1:1	−$22,919	−$197	Cummins et al.^19
GLP-1RA vs SUs	$41,129	UK	19:1	$38,914	$19,828	Davies et al.^17

*The consumer price index is calculated using OECD.statistics⁴⁴.

†Conversions using XE Currency Converter⁴⁵.

Abbreviations. QALY, quality adjusted life years; DAPA, dapagliflozin; GLP-1RA, glucagon like peptide-1 receptor agonists; DPP-4i, dipeptidyl peptidase-4 inhibitors; TZDs, thiazolidinediones; SUs, sulphonylureas.

Results

Total annual costs for clinical end-points

Compared with other glucose lowering therapy classes, patients receiving DAPA had lower medical costs in the year following the treatment. The medical costs were $186, $1,142, $1,640, and $2,474 less for DAPA compared with GLP-1RAs, DPP-4is, TZDs, and SUs, respectively. Change in weight had the highest impact on difference in medical costs between DAPA and other drug classes (Table 3). The total costs (medical + pharmacy) for DAPA were less than those for GLP-1RAs ($2,701) and DPP-4is ($795). DAPA had higher total costs in comparison to TZDs ($1,848) and SUs ($2,161) (Table 5). The total cost for DAPA vs TZDs and SUs was higher, mainly due to the higher drug acquisition cost of DAPA (Table 2).

Table 5. Differences in costs, QALYs, and ICERs for DAPA vs each comparator.

	DAPA vs GLP-1RA	DAPA vs DPP-4i	DAPA vs TZDs	DAPA vs SUs
Total medical cost difference	–$186	–$1,142	–$1,640	–$2,474
Total cost difference (includes drug cost	–$2,701	–$795	$1,848	$2,161
QALYs gained	0.0107	0.0587	0.0715	0.1137
ICER: Net total cost per QALY gained	DAPA Dominates	DAPA Dominates	$25,835	$19,005

Abbreviations. T2D, type 2 diabetes mellitus; QALY, quality adjusted life years; DAPA, dapagliflozin; GLP-1RA, glucagon like peptide-1 receptor agonists; DPP-4i, dipeptidyl peptidase-4 inhibitors; TZDs, thiazolidinediones; SUs, sulphonylureas.

DAPA dominates because it improves QALYs at a lower total cost. The costs were inflated to 2016; negative numbers indicate lower costs for DAPA.

Incremental QALY for clinical end-points

Treatment with DAPA resulted in an improvement in QALYs compared with all the other drug classes. The QALY gains associated with DAPA were estimated to be 0.0107, 0.0587, 0.0715, and 0.1137 compared with GLP-1RAs, DPP-4is, TZDs, and SUs, respectively. Change in body weight had the highest impact on QALYs among all the four clinical end-points (Table 6).

Table 6. QALYs gained with DAPA vs each comparator associated with clinical end-point differences.

Clinical end-point	DAPA vs GLP-1RA	DAPA vs DPP-4i	DAPA vs TZDs	DAPA vs SUs
Change in A1c	−0.0090	0.0024	0.0004	0.0000
Change in Weight	0.0079	0.0386	0.0709	0.0694
Change in SBP	0.0118	0.0175	NR	0.0406
Change in Hypoglycemia	NR	0.0002	0.0002	0.0036
Total QALYs	0.0107	0.0587	0.0715	0.1137

Abbreviations. QALY, quality adjusted life year; DAPA, dapagliflozin; GLP-1RA, glucagon like peptide-1 receptor agonists; DPP-4i, dipeptidyl peptidase-4 inhibitors; TZDs, thiazolidinediones; SUs, sulphonylureas; A1c, glycated hemoglobin; SBP, systolic blood pressure; NR, not reported.

The costs were inflated to 2016; negative numbers indicate lower costs for DAPA.

Incremental cost per QALY gain

Compared to GLP-1RA and DPP-4i, treatment with DAPA was dominant as it lowered total cost and achieved greater QALY gains. When compared to TZD and SU, treatment with DAPA is considered to be cost-effective, with incremental costs per QALY gained of 25,835 ($/QALY) and 19,005 ($/QALY), respectively (Table 5).

Sensitivity analyses

One-way sensitivity analysis

ICER for clinical end-points of DAPA vs other glucose lowering therapies: The change in weight, QALY per kg weight change, and cost per kg weight change had the highest impact on ICERs for comparison of DAPA with DPP-4is, TZDs, and SUs. Additionally, for DAPA vs DPP-4is and SUs, variation in SBP, cost per mmHg drop in SBP, and QALY associated with change in SBP had a large numerical impact on ICERs. For DAPA vs GLP-1RAs, change in A1c had the highest impact on ICERs (Figure 2(A)). Despite results varying by these parameters, the conclusion that was made from the base case results did not change. DAPA remained dominant (vs DPP-4i, Figure 2(B)) or cost-effective, with ICERs well below $100,000 per QALY gained (vs GLP-1RA, TZDs, and SUs, Figures 2(A,C) and (D), respectively).

Figure 2. Tornado dagrams for ICERs: DAPA vs each comparator. Here the red bars indicate the impact on ICERs by increasing the value of the input parameter up to its upper bound, and the blue bars indicate the impact on ICERs by decreasing the value of the input parameter up to its lower bound. The scales may differ across the panels to show variation in the parameters of the sensitivity analysis within a given panel. Abbreviations. NMA, network meta-analysis; QALY, quality adjusted life years; mmHg, millimeters of mercury; A1c, glycated hemoglobin; SBP, systolic blood pressure; GLP-1RA, Glucagon Like Peptide-1 Receptor Agonists; k = 1,000.

Differences in total costs for clinical endpoints of DAPA vs other glucose lowering therapies: In the model, the change in patient weight had the highest impact on total cost difference for DAPA vs GLP-1RAs and DPP-4is. For DAPA vs TZDs and SUs, the cost associated with a unit change in body weight had the highest impact on total costs. Change in SBP had a significant impact on total costs for comparison of DAPA vs GLP-1RAs, DPP-4is, and SUs. In the case of DAPA vs TZDs, change in A1c had a significant impact on total costs (Figure 3(C)).

Figure 3. Tornado diagram for differences in total cost: DAPA vs each comparator. Here the red bars indicate the impact on ICERs by increasing the value of the input parameter up to its upper bound, and the blue bars indicate the impact on ICERs by decreasing the value of the input parameter up to its lower bound. The scales may differ across the panels to show variation in the parameters of the sensitivity analysis within a given panel. Abbreviations. NMA, network meta-analysis; QALY, quality adjusted life years; mmHg, millimeters of mercury; A1c, glycated hemoglobin; SBP, systolic blood pressure; GLP-1RA, Glucagon Like Peptide-1 Receptor Agonists.

Probabilistic sensitivity analysis

PSA showed that DAPA was cost-effective compared with TZDs and SUs drug classes, and cost-saving compared with GLP-1RAs and DPP-4is in most simulations (Figure 4). Also, the total cost for DAPA was less than that for GLP-1RAs and DPP-4is for 100% and for more than 96% of the simulations, respectively. At a threshold value of $50,000 per QALY, DAPA had 96%, 100%, 99%, and 83% probability of being cost-effective compared with GLP-1RAs, DPP-4is, SUs, and TZDs, respectively (Figure 5). Overall, our results were found to be robust in the sensitivity analysis.

Figure 4. CE planes for DAPA vs. each comparator. (a) CE plane for DAPA vs GLP-1RA; (b) CE plane for DAPA vs DPP-4i; (c) CE plane for DAPA vs SUs; (d) CE plane for DAPA vs TZDs. The red line indicates the maximum acceptable cost effectiveness ratio, which was assumed to be $50,000 per QALY gained. Abbreviations. CE, cost-effectiveness; DAPA, dapagliflozin; GLP-1RA, glucagon like peptide-1 receptor agonists; DPP-4i, dipeptidyl peptidase-4 inhibitors; TZDs, thiazolidinediones; SUs, sulphonylureas.

Figure 5. Cost-effectiveness acceptability curve (CEAC) for DAPA vs each comparator. (a) CEAC for DAPA vs GLP-1RA; (b) CEAC for DAPA vs DPP-4i; (c) CEAC for DAPA vs SUs; (d) CEAC for DAPA vs TZDs. Abbreviations. CEAC, cost-effectiveness acceptability curve; DAPA, dapagliflozin; GLP-1RA, glucagon like peptide-1 receptor agonists; DPP-4i, dipeptidyl peptidase-4 inhibitors; TZDs, thiazolidinediones; SUs, sulphonylureas.

Comparison of short-term model with long-term model

The ICER obtained from our short-term model for DAPA vs SUs was $19,005, and those obtained from the published long-term models ranged between $3,800–$9,000^16,19,23. When the drug price ratio in our model was changed to a similar level as that in the long-term models, the ICERs in our short-term model ranged between $9,600–$18,000. Thus, DAPA is cost-effective compared with SUs in both our short-term model and the published long-term models.

When DAPA was compared with DPP-4is, the results from the published long-term models were either cost-saving or cost-effective, whereas the result was cost-saving for the short-term model^19,35. After keeping a similar drug price ratio in the short-term model, the results remained cost-saving. Thus, both our short-term and long-term models show that DAPA is cost-saving compared with DPP-4is.

For DAPA compared with TZDs, the ICER from the short-term model was $25,835, while that from the long-term models was cost-saving¹⁹. When the drug price ratio in our model was changed to a similar level as in the long-term models, DAPA was cost-saving compared with TZDs in our short-term model. Thus, both models predict DAPA to be cost-saving compared with TZDs.

The ICERs for DPP-4is vs SUs for the published long-term models ranged between $7,000–$27,000, whereas the ICER obtained from the short-term model was $53,681^18,20,21. The ICERs ranged between $39,000–$49,000, even after keeping a similar drug price ratio. Both modeling approaches predicted that DPP-4is are cost-effective compared with SUs. When DPP-4is were compared with TZDs, the results remained either cost-saving or cost-effective for the long-term models, but the ICER from the short-term model was $80,788¹⁸. The ICER became cost-saving in the short-term model after adjustment of the drug prices. Thus, both modeling approaches predicted DPP-4is were cost-saving compared with TZDs.

For GLP-1RAs vs DPP-4is, ICERs from the long-term models ranged between $15,000–$16,000. ICER from the short-term model was $40,240^17,22,24. After adjusting the drug prices, the ICER from the short-term model ranged between $56,000–$82,000. When GLP-1RAs were compared with SUs, the ICER from the long-term models was $19,828, whereas the ICER from our short-term model was $41,161¹⁷. Even after adjusting the drug prices, the ICER in the short-term model remained at $38,914 (Table 4). Even though both our short-term model and the long-term models predict GLP-1RAs to be cost-effective compared with SUs and DPP-4is, there are differences in the numbers. One reason for the difference in ICER values could be that, in the published long-term models, only A1c and change in weight were considered, whereas, in our short-term model, four clinical end-points were considered¹⁷.

Correlation between A1c and weight

We conducted a two-way sensitivity analysis (SA) around A1c and weight to assess the potential impact of the correlation between these two clinical end-points. Because of the concern for double-counting if these end-points are (perfectly) correlated, we assumed no treatment differences in A1c and on the associated A1c-related medical care costs and QALY gains. Results for this sensitivity analysis, shown in Table 7, remained similar to the base case results (Table 5). DAPA continues to be cost-saving compared to GLP-1 and DPP-4i, and cost effective compared to SU and TZD.

Table 7. Sensitivity analysis results (assuming no treatment difference on A1c).

Total cost difference	DAPA vs GLP-1RA	DAPA vs DPP-4i	DAPA vs TZDs	DAPA vs SUs
Medical costs only	–$319	–$1,106	–$1,633	–$2,474
Total cost, including drug cost	–$2,834	–$759	$1,856	$2,161
Combined QALY difference	0.0197	0.0563	0.0711	0.1137
Total cost, including drug cost	DAPA dominates	DAPA dominates	$26,111	$19,010

The costs were inflated to 2016; negative numbers indicate lower costs for DAPA.

Discussion

We developed a short-term economic model (1-year time horizon) that estimates the costs and benefits of DAPA vs other glucose lowering therapy classes (such as GLP-1RAs, DPP-4is, TZDs, and SUs) added to metformin from the US third-party payer perspective. The results from our model showed that DAPA was either cost-saving or cost-effective compared with all other glucose lowering therapy classes. In addition, our study showed that the change in patient weight and the associated change in costs and QALYs had the highest impact on ICERs. The sensitivity analysis results showed that DAPA dominates DPP-4is in all the simulations. Similarly, when compared with GLP-1RAs, DAPA was cost-saving, and, when compared with TZDs and SUs, DAPA was also cost-effective. The results from our short-term model were consistent with results in the published long-term models. For instance, when DAPA was compared with SUs, the results were cost-effective from our model under a threshold of $50,000, and so were the results of published long-term models¹⁶.

The efficacy of DAPA could be demonstrated through several published literatures. A systematic review was conducted to compare DAPA with treatments for T2DM within the European Union, and reported that DAPA in combination with SU had better results when compared to SU monotherapy alone in terms of glycemic control, weight reduction, and hypoglycemic events. DAPA was the only treatment that showed favorable results in weight reduction and hypoglycemia profile³⁶. A retrospective study was done on UK primary care data that also showed effectiveness of DAPA in reducing A1c, body weight, and SBP at different time points up to 2 years in the treatment of T2DM patients as dual therapy with metformin and triple or concomitant therapy with other glucose lowering agents³⁷.

One main limitation of our study is that we used published literature to estimate the impact of change in clinical end-points (A1c, body weight, SBP, and hypoglycemia rate) on healthcare costs. We came across very few published studies that estimated this impact, as it is difficult to find the impact on cost attributable to these clinical end-points in isolation. In a future study, we would like to directly estimate these costs from a database study. An alternative approach would be to collect these costs directly in a clinical trial. Another limitation of this study is that QALYs were estimated from published literature. However, a phase III study was conducted to evaluate the impact on health-related quality-of-life (HRQoL) with change in patient weight in T2DM patients treated with DAPA in combination with metformin. This was a double-blinded, randomized, placebo-controlled trial of 24-weeks, with a 78-week extension period. The patients who were treated with DAPA were associated with improvement in overall HRQoL as a result of weight loss at 24, 50, and 102 weeks³⁸. More studies should be conducted to estimate the direct impact of change in clinical end-points on the quality-of-life for patients. In addition, the study does not include adverse events other than hypoglycemia associated with different glucose lowering treatments, e.g. gastrointestinal upset (GI) for GLP-1RA, genital mycotic infection (GMI), and urinary tract infection (UTI) for SGLT-2is. The FDA label for DAPA reported an incidence of 4.3% for both GMI and UTI³⁹. The incidence of GMI and UTI was 7.46% (7.8%) and 5.9% (4.4%) for 100 mg (300 mg) canagliflozin (a SGLT-2i class drug), respectively, as per the FDA label⁴⁰. The estimated cost to treat GMI and UTI was $3.68 and $27.61, respectively, per patient, with a combined cost of $31.29⁴¹. Given the low incidence rate and low cost to treat GMI and UTI, it will have a very low impact on total cost of DAPA. The incidence of GI upset in the patients treated with Liraglutide (a GLP-1RA class drug), as reported in the FDA label, was 32.6%⁴². One of the studies that compared liraglutide and exenatide estimated the total cost for treating GI disorders was $1,113⁴³. This cost would support the finding that DAPA is cost saving compared to GLP-1RA. There are other adverse events associated with SGLT-2is that are not reported in this model, like risk of amputation. Among the SGLT-2is, Canagliflozin is associated with increased risk of amputations. There were two randomized, placebo-controlled trials (CANVAS and CANVAS-R) evaluating the patients with T2DM with either cardiovascular disease or at risk of cardiovascular disease treated with Canagliflozin. The trials reported that 5.9 (CANVAS) and 7.5 (CANVAS-R) out of every 1,000 patients had a risk of lower limb amputation⁴⁰. Since Canagliflozin was not evaluated in the model, we did not include amputation as an adverse event. Despite these limitations, the clinical end-points included in the model addressed goals recommended by the American Diabetes Association, which recommends A1c < 7%, blood pressure <130/80, avoiding hypoglycemic events and maintaining weight if non-obese, or losing weight if obese⁶.

Moreover, observational studies, central to this model, have inherent limitations. These studies only provide associations between the clinical end-points and related healthcare costs, and cannot prove a causal relationship. Additionally, potential problems of unobserved or missing variables cannot be fully addressed by observational database studies. Only randomized controlled studies can prove a causal relationship and tackle the problem of unobserved variables. However, randomized studies are not without their limitations as well. The study population in RCTs are typically highly selective, based on pre-specified criteria, the final sample of which may be limited, and the results of which cannot be extrapolated to the general population. This is the reason observational study results are useful, and can supplement those from RCTs to provide additional real-world evidence for general practice.

Conclusions

Comparison of DAPA with four other glucose lowering therapy classes added on to metformin showed that DAPA was either cost-saving or cost-effective in the US setting over a 1-year time period. DAPA was cost-saving and achieved gains in QALYs compared with GLP-1RA and DPP-4i treatments, and was cost-effective compared with TZDs and SUs. The results from our short-term model were found to be robust in sensitivity analyses and similar to those in published long-term models for different countries.

Transparency

Declaration of funding

This research is funded by AstraZeneca.

Declaration of financial/other relationships

AC and MR are employees of Complete HEOR Solutions, India, and PD is an employee of Complete HEOR Solutions, USA. KB was formerly an employee of AstraZeneca, and she is presently working at GlaxoSmithKline, PA, USA.

Previous presentations

Posters on this research were earlier presented at the ISPOR 20th Annual International Meeting, PA, USA, May 16–20, 2015 and at QCOR 2017, VA, USA, April 2–3.

Acknowledgment

No assistance in the preparation of this article is to be declared.

Appendix 1: Sources for cost inputs

Table

S. No	Clinical end-points	Inputs and calculation of cost	Reference
1	A1c	•The diabetes related medical cost was $5,680. It was obtained from a US MCO for T2D patients with A1c ≥6%. •The annual medical costs of the patients increased by 4.4% with every 1% increase in A1c. •Therefore, incremental medical cost was $250 (5,680*0.0440). •Annual cost after inflation adjustment to December 2016 costs was $326.04.	Aagren and Luo^26
2	Weight	•Subjects were divided into weight gainers (n = 234) and non-weight gainers (n = 224). •The average weight for the two groups was 96.81 kg, and the average total healthcare cost was $6,382. •Incremental medical cost increased by $213 for every 1% change in weight or $338.69 for every 1 kg change in weight (inflated to December 2016 cost).	Yu et al.^29
3	SBP	•The study subjects were broadly classified in two groups: the control group and intervention group; based on which the total medical costs incurred were calculated. •Total costs associated with the intervention group and control group were $774.90 and $445.75, respectively. •The drop in SBP was 9.08 mmHg. The incremental medical cost for 1 mmHg drop in SBP was $94.49, inflated to December 2016 cost.	Kulchaitanaroaj et al.^27
4	Hypoglyceamia	•Overall population size was 536,581. •Overall population follow-up duration (person-year) was 1,212,791. •Mean follow-up duration (years) was 2.26. •Hypo population size was 18,657. •Hypo population follow-up patient-year count was 42,168 (18,657*2.26). •Total hypo cost was $52,223,675. •Mean hypo cost per patient year among patients with hypo was $1,238.44. After inflation adjustment to December 2016 costs, the cost became $1614.07.	Quillium et al.^28

Appendix 2: Sources for QALY inputs

Table

S. No	Clinical end-points	Inputs and calculations	Reference
1	A1c	•The disutility associated with treatment burden was assumed to be 0.001 in the Markov model. •The average age was estimated to be 57.5 years, and A1c level to be 8% from the network meta-analysis done for this short-term economic model. •The average QALYs gained for patients of age 55 incorporating the disutility of 0.001 were calculated using QALYs gained at two A1c level intervals, i.e. 7.5% to 6.5% and 8.5% to 7.5%. •Similar calculations were done to calculate QALYs gained at age 65. Thereafter, interpolation was used to calculate the QALYs gained per 100 years of treatment for age 57.5, which was the average age from the NMA report. •The QALYs gained at the age of 57.5 was 0.022.	Vijan et al.^31
2	Weight	•A weight gain of 3%, which typically occurs in patients taking sulfonylureas and thiazolidinedione’s, has a mean attributed disutility of 0.04. •As per the NMA results that were incorporated in our short-term economic model, the mean weight of the study subjects was 89.5 kg. Thus, the disutility associated per 1 kg weight increase was –0.0004469 (–0.04/89.5). •For 3% weight gain, total disutility associated was –0.0149 (–0.0004469/0.03).	Matza et al.^32
3	SBP	•Mean quality-adjusted life expectancy in BP intensification arm as per the study was 10.1 years. Overall QALY gain for unadjusted treatment related harm was 1.35 years. Thus, the total quality-adjusted life expectancy in the year came out to be 11.45 years. •The total drop in SBP was 14.8 units, the incremental QALYs per mm Hg change came out to be 0.091216 (1.35/14.8). Thus, the total QALYs gained per mm Hg was 0.007966 years (0.091216/11.45).	Timbie et al.^33
4	Hypoglycemia	•Pooled value data from two postal surveys were utilized to calculate QALY associated with 1 (per-patient year) PPY of hypo risk. •The surveys were conducted among subjects with diabetes in Cardiff, UK. •Univariate and multivariate analyses were performed to analyze the data from the Hypoglycemia Fear Survey (HFS), along with health-related utility using the EQ5D index. •A 1-unit increase on the HFS resulted in a 0.008 unit decrease on the EQ5D index. •Total change in HFS for one symptomatic hypoglycemic event was 1.773 units. Hence, the QALY gain associated with 1 PPY of hypo risk was 0.0142 (–1*(–0.008 units*1.773 units)).	Currie et al.^34