Cost-effectiveness analysis of metformin+dipeptidyl peptidase-4 inhibitors compared to metformin+sulfonylureas for treatment of type 2 diabetes

Background Patients with type 2 diabetes (T2D) typically use several drug treatments during their lifetime. There is a debate about the best second-line therapy after metformin monotherapy failure due to the increasing number of available antidiabetic drugs and the lack of comparative clinical trials of secondary treatment regimens. While prior research compared the cost-effectiveness of two alternative drugs, the literature assessing T2D treatment pathways is scarce. The purpose of this study was to evaluate the long-term cost-effectiveness of dipeptidyl peptidase-4 inhibitors (DPP-4i) compared to sulfonylureas (SU) as second-line therapy in combination with metformin in patients with T2D. Methods A Markov model was developed with four health states, 1 year cycle, and a 25-year time horizon. Clinical and cost data were collected from previous studies and other readily available secondary data sources. The incremental cost-effectiveness ratio (ICER) was estimated from the US third party payer perspective. Both, costs and outcomes, were discounted at a 3% annual discount rate. One way and probabilistic sensitivity analyses were performed to evaluate the impact of uncertainty on the base-case results. Results The discounted incremental cost of metformin+DPP-4i compared to metformin+SU was $11,849 and the incremental life-years gained were 0.61, resulting in an ICER of $19,420 per life-year gained for patients in the metformin+DPP-4i treatment pathway. The ICER estimated in the probabilistic sensitivity analysis was $19,980 per life-year gained. Sensitivity analyses showed that the results of the study were not sensitive to changes in the parameters used in base-case. Conclusions The metformin+DPP-4i treatment pathway was cost-effective compared to metformin+SU as a long-term second-line therapy in the treatment of T2D from the US health care payer perspective. Study findings have the potential to provide clinicians and third party payers valuable evidence for the prescription and utilization of cost-effective second-line therapy after metformin monotherapy failure in the treatment of T2D.


Background
Diabetes mellitus is one of the most prevalent and costly chronic diseases in the United States (US). In 2012, 9.3% of the US population had diabetes mellitus [1]. In that year 2012, the health care cost of diagnosed diabetes in the US totaled $245 billion [2]. The US market of antidiabetic products reached $43.9 billion in 2015 (a 109.0% increase from $21.0 billion in 2011) [3]. The number of prescriptions for antidiabetic drugs totaled 211 million in 2015 (compared to 174 million in 2011) [3]. In 2015, insulin glargine recombinant was the top fifth drug by sales in the US totaling $5.8 billion (241.2% increase compared to 2011) [3]. Sitagliptin was the top tenth prescription drug by sales reaching $4.2 billion in 2015 (a 90.9% increase compared to 2011) [3]. As of December 31, 2015, there were 27 unique non-insulin antidiabetic drugs, belonging to 12 therapeutic classes, including 5 modified formulations and 18 fixed-dose combinations of active ingredients, available in the US market [4].
Metformin has a well-established long-term postmarketing evidence of effectiveness and safety [5][6][7]. While there is a general consensus about the use of metformin as first-line therapy for type 2 diabetes (T2D) [5][6][7]; there is a vigorous debate about best second-line treatment regimen [8]. Sulfonylureas (SU) are a common second-line therapy due to their fast onset on blood glucose lowering [9,10]. However, safety related concerns, including risk of hypoglycemia and weight gain, have been raised [9,10]. Dipeptidyl peptidase-4 inhibitors (DPP-4i) are newer drugs with lower risk of hypoglycemia and weight gain but lower glycemic lowering effect than SU [10,11]. In addition, DPP-4i are costlier than SU.
Two previous studies explored the cost-effectiveness of SU compared to DPP-4i as second-line therapy after metformin failure in the US. Study findings were inconclusive. Bergenheim et al. (2012) [12] assessed the lifetime cost-effectiveness of metformin+SU and metformin +DPP-4i in T2D using data from 52-week randomized controlled trial [9]. The authors concluded that DPP-4i was a cost-effective second-line therapy after metformin failure in the US. Zhang et al. (2014) [8] compared the medication cost and effectiveness of metformin+SU, DPP-4i, and glucagon-like peptide-1(GLP-1) receptor agonists as the second-line therapy until first diabetesrelated complication or death. The authors found that metformin+SU resulted in similar outcomes but lower drug costs compared to other two comparators.
Bergenheim et al., (2012) did not consider insulin treatment after second-line failure; whereas, Zhang et al., (2014) included insulin treatment as third-line in their analyses. Furthermore, Bergenheim et al., (2012) included drug cost and diabetes related health care costs in their economic evaluation; whereas, Zhang et al., (2014) did not assess health care costs associated with diabetic complications, which often pose a significant economic burden on patients with T2D [1].
Prior research compared short-term cost-effectiveness of two alternative drugs for treatment of T2D. To the best of authors' knowledge, no previous US studies assessed the cost-effectiveness of alternative T2D treatment pathways over a patient's lifetime. Thus, this study assessed the long-term cost-effectiveness of dipeptidyl peptidase-4 inhibitors compared to sulfonylureas as second-line therapy for the treatment of T2D. This study has the potential to provide clinicians and third party payers with new perspectives on the cost-effectiveness of long-term treatment pathways for T2D.

Therapeutic alternatives
Most patients with T2D take one or more drugs in addition to metformin monotherapy to control their blood glucose levels and eventually, will initiate insulin therapy alone or in combination with other non-insulin antidiabetic drugs when previous alternatives fail. In the scenario analysis, there were two different treatments pathways. Both pathways started with metformin monotherapy, the most common treatment. Patient used metformin+DPP-4i or metformin+SU as second-line therapy when metformin monotherapy failed. In addition, both treatment pathways added basal insulin therapy in patients with T2D when combination therapy failed.

Markov model
A Markov model constructed in Microsoft Excel 2013 software was based on current T2D treatment guidelines [5,6]. The Markov model had four states (Fig. 1). In the first state, patients used metformin monotherapy. Patients could remain in the first state or transition to the second state. In the second state, either DPP-4i or SU was added to metformin as second-line therapy. Likewise, patients could remain in the second state or transition to the third state where basal insulin was added to their current therapy as third-line therapy. Patients in those three states could transition anytime to death (i.e., absorbing state). We assumed that patients initiated metformin monotherapy at age 60 years with a time horizon of 25 years (i.e., 60-85 years old). The cycle duration was 1 year [15].

Health outcomes and cost
Health outcomes data were collected from the literature ( Table 1). Treatment failure rates obtained from previous studies were used to determine the annual probability of transitioning from metformin monotherapy to oral antidiabetic drug (OAD) dual therapy and from OAD dual therapy to OAD dual therapy+basal insulin states. Kahn et al., (2006) found that 21% of patients failed to achieve their therapeutic goals after 5 years in metformin monotherapy [16]. In addition, Rascati et al., (2013) estimated that 23.6% of patients using metformin+SU dual therapy progressed to OAD dual therapy+basal insulin after 59 months in treatment [17]. Parchman and Wang, (2012) found that the rate of insulin initiation had a statistically significant and positive association with the A1c increasing rate [18]. More specifically, Bergenheim et al., (2012) [12] found that patients using metformin+DPP-4i had four times lower A1c increasing rates than those using metformin+SU. Thus, we assumed that the annual treatment failure rate of metformin+DPP-4i was four times lower than metfor-min+SU.
Death rates for 60 to 70 years old, 71 to 80 years, and 81 to 85 years old groups were derived from the literature [19]. The hazard ratios of death in patients using metformin+SU and metformin+DPP-4i were also drawn from the literature [20]. Hypoglycemia probabilities in patients using metformin+SU and metformin+DPP-4i were extracted from the results of a 52-week randomized clinical trial [9]. We considered only severe hypoglycemia and hypoglycemia events requiring medical assistance to estimate direct health care costs. The probability of severe hypoglycemia for a patient using OAD + basal insulin was collected from the ORIGIN trial [21]. Weight gain data in patients using SU were derived from a previous study [12]. We assumed that use of metformin, DPP-4i and basal insulin were not associated with a significant weight gain [10].
Likewise, annual cardiovascular complication rates (i.e., myocardial infarction, heart failure and stroke) in patients using metformin monotherapy, OAD dual therapy, and OAD-basal insulin therapy for each health state described in the Markov model were derived from previous published clinical trial studies [11,16,21]. We assumed that the probability of treatment failure, hypoglycemia and cardiovascular complications remained constant through the study period with the exception of death rates which gradually increase with age. The proportion of patients in each state and cycle was calculated using the transition matrix (Tables 2 and 3).
The main study outcome was the number of life-years gained over the study time horizon. In order to estimate life-years gained, all life-years for patients in every state, with the exception of death, were aggregated by year and discounted. The incremental life-years gained were estimated as the difference in life-years gained between the two interest therapeutic alternatives metformin+DPP-4i and metformin+SU.
Direct health-care costs related to T2D, which included drug costs and treatment costs for diabetes related medical events such as hypoglycemia, weight gain and cardiovascular events were obtained from the US health care payer perspective. Direct health care cost input data were derived from the literature ( Table 4). Indirect and intangible costs related to the disease were not included in the study model.
Antidiabetic drug costs data were collected from the National Average Drug Acquisition Cost (NADAC) dataset [22]. We used generic NADAC for metformin and SU (glipizide). The cost of DPP-4i was estimated as the average NADAC of sitagliptin, saxagliptin, linagliptin, and alogliptin. We also used the NADAC for basal insulin glargine pen type. Needle cost for insulin glargine pen was estimated at 80% of the average wholesale price (AWP). AWP data were derived from the online version  [23]. Annual prescription drug cost was calculated based on the defined daily dose (DDD) from WHO Collaborating Centre for Drug Statistics Methodology [24]. The direct health care cost in each state, with the exception of death, was estimated for each year through the study time horizon. Base-case health care costs in each state was calculated multiplying the probability of each episode and unit cost ( Table 5). All costs were adjusted to 2015 US dollars using the all urban consumers, not seasonally adjusted, US city average, all items, consumer price index (CPI) [25]. Both costs and outcomes were discounted at a 3% annual discount rate.

Cost-effectiveness analysis
The cost-effectiveness analysis (CEA) of metformin +DPP-4i vs. metformin+SU in patients with T2D was conducted from the US health care payer perspective. The cost and life-years gained over the 25-year time horizon were estimated for each treatment pathway. A cost-effectiveness ratio (CER) was employed to calculate the cost per life-year gained for each treatment strategy. The lowest cost per life-year treatment was considered as the reference therapy. When a treatment had a greater cost and effectiveness in relation to the reference an incremental cost-effectiveness ratio was performed to determine the additional cost to obtain one life-year. Incremental cost-effectiveness ratio (ICER) was estimated for metformin+DPP-4i compared to metformin+SU.

One-way and probabilistic sensitivity analyses
The impact of parameter uncertainty was explored by one-way sensitivity analysis on each model parameter. Results of the one-way sensitivity analysis were expressed as tornado charts. Values for treatment failure rates, hypoglycemia events probabilities, weight gain rates, cardiovascular events rates, and costs were changed by ±25% from the base-case. The cost of insulin glargine was changed by ±20%. Death rates and the death hazard ratios were changed by ±10%. The one-way sensitivity analysis was also conducted to compare differences in study results using 20 and 30 year time horizons. The sensitivity analyses also included a scenario in which there was no difference in cardiovascular event rates after 2 years from initiation of metformin+DPP-4i and metformin+SU [26,27].  found that the mean (standard deviation) age at the start of the antidiabetic treatment was 57.4 (11.7) years [28]. The average time to insulin initiation was 1.94-20.7 years depending on basal treatment. Machado-Alba et al., (2015) found that mean age at the start of oral antidiabetic therapy in patients with type 2 diabetes mellitus was 63.4 years [29]. After 5 years, 26.1% initiated insulin therapy. Roussel et al., (2016) found that the average age (standard deviation) of insulin therapy initiation in patients with type 2 diabetes mellitus (T2DM) was 67.5 (14.2) years [30]. In one-way sensitivity analyses, we changed the patient age at start of the antidiabetic treatment metformin monotherapy from 60 years in the base case to 55 and 65 years. A probabilistic sensitivity analysis was conducted to investigate the combined impact of uncertainty of the variables included in the analysis. Random values were drawn from the chosen distributions as a second-order Monte-Carlo simulation of 1000 patients to estimate the mean and 95% confidential intervals (CI) of costs and life-years gained. All parameters in the model had correspondingly appropriate distributions. Costs were randomly drawn from a gamma distribution; hazard ratio were randomly sampled from a lognormal distribution. Likewise, binominal data, such as hypoglycemia probabilities and cardiovascular event rates were randomly drawn from a beta distribution. Multinomial data, such as transition probabilities in the metformin monotherapy and the OAD dual therapy states, were randomly sampled from a Dirichlet distribution [31].
The ICER was recalculated based on the patient age at start of the antidiabetic treatment, the average incremental costs and life-years gained derived from the probabilistic sensitivity analysis. The simulation output was presented using a cost-effectiveness plane. A costeffectiveness acceptability curve (CEAC) was also plotted to summarize the uncertainty in the cost-effectiveness estimates.

Base-case analysis
Diabetic-related annual average costs and life-years gained after discounting were $18,853 and 12.42 years, respectively for patients in the metformin+DPP-4i treatment pathway. Patients in the metformin+SU treatment pathway incurred in a lower annual average costs per patient ($7004) and gained on average a lower number of life-years (11.81 years) ( Table 6). The incremental costs and life-years gained for metformin+DPP-4i compared to metformin+SU treatment pathways were $11,849 and 0.61 years, respectively. Thus, the ICER was $19,420 per life-year gained for patients in the metfor-min+DPP-4i treatment pathway.

One-way and probabilistic sensitivity analyses
One-way sensitivity analysis was conducted by varying the range of values in the base-case to determine potential impacts on the results. The percentage changes in ICER from base-case are presented in the tornado graph (Fig. 2). Time horizon, death hazard ratio and age at start of metformin monotherapy, and death rate parameters in the model had largest impact on the model results. The results of the analysis did not change  significantly when varying estimates used in the basecase scenario (Table 7). Results for the base-case scenario were not sensitive to changes in the costs of insulin glargine, treatment failure rates, costs and rates of cardiovascular events, or the costs and probabilities of severe hypoglycemia, and weight gain. Results for the base-case scenario were not sensitive either to changes in the cardiovascular event rates of metformin+DPP-4i and metformin+SU. The ICER increased to $24,250 per life-year gained (+ 24.3%) when the time horizon decreased to 20 years. Conversely, the ICER decreased to $17,580 per life-year gained (− 9.8%) when the time horizon increased to 30 years. In addition, a 10% decrease in the death hazard ratio resulted in a 21.9% increase in the ICER ($23,760). A 10% increase in the death hazard ratio resulted in a 14.0% decrease in the ICER ($16,760). Assuming that patients start metformin monotherapy at age of 55 increased ICER to $21,360 (9.6%); whereas, starting antidiabetic treatment at age of 65 decreased ICER to $18,120 (− 7.1%).
The average results of the probabilistic sensitivity analysis yielded $11,786 incremental costs and 0.59 incremental life-year gained for patients using metformin +DPP-4i compared to alternative metformin+SU treatment pathway ( Table 8). The ICER in the probabilistic sensitivity analysis was $19,980 per life-year gained. The difference between the probabilistic sensitivity analysis and the base-case strategy was $63 in incremental costs and 0.02 additional life-years gained.
The uncertainty surrounding the expected costs and outcomes associated with metformin+DPP-4i compared with metformin+SU is illustrated in Fig. 3. The incremental cost-effectiveness plane shows the trade-offs in the northeast (i.e., positive costs and positive effects) and southeast quadrants (i.e., negative costs and positive effects).
The CEAC indicates that metformin+DPP-4i and met-formin+SU would have the same probability of being the most cost-effective treatment for a WTP threshold of $12,500 per life-year gained; after exceeding this threshold the probability of metformin+DPP-4i being the most cost-effective treatment pathway approaches to 1 (Fig. 4).

Discussion
This study assessed the long-term cost-effectiveness of metformin+DPP-4i compared with metformin+ SU treatment pathways as second-line therapy from the US health care payer perspective. In the base-case results, the total costs and life-years gained were higher for met-formin+DPP-4i than for the metformin+SU treatment pathway.
The results from the probabilistic sensitivity analyses were similar to those of the base-case results. The results of the one-way sensitivity analysis showed that the main factors impacting on the ICER were time horizon   [12]. The authors concluded that metformin +DPP-4i was a cost-effective second-line therapy in the US. Some methodological differences between Bergenheim et al., (2012)    would enter the market in a 10 year-time frame and that generic DPP-4i prices would be 16% of the corresponding brand name drug price. Nevertheless, prices of generic drugs are set up as 94% of brand name drug prices when there is only one generic competitor in the market [32]. A decrease in the price of generic drugs to 16% of the price of brand name drugs is observed only when there are several generic competitors in the market. Therefore, Bergenheim et al., (2012) may underestimate DPP-4i generic drug prices. Zhang et al., (2014) compared metformin+SU, metfor-min+DPP-4i, metformin+GLP-1 receptor agonists, and insulin as the second-line therapy using a Markov model with 10 A1c states [8]. Zhang et al., (2014) only assessed the medication cost; Authors did not include the costs of medical events related with diabetic complications. Zhang et al., (2014) did not include either in the analysis the costs associated with severe hypoglycemia. Furthermore, they assumed that the termination state was either the first diabetes-related complication or death.     Zhang et al., (2014) set up three different A1c goals (i.e., 6.5, 7, and 8%) to evaluate the impact of glycemic control goals on patients with diabetes. Like in our study, they considered insulin triple therapy after dual therapy failure, but they assumed that patients initiated insulin therapy only after exceeding the A1c goals. This assumption lead to the initiation of insulin therapy only after 1.59 to 2.76 years after onset T2D diagnosis. Nevertheless, assuming that insulin initiation is based on patients' A1c levels may lead to overestimate insulin initiating rate due to well document barriers for the patient's and provider's to start insulin therapy [33][34][35]. In addition, the timeline for second line therapy before insulin therapy may not be long enough to show clinically meaningful differences in outcomes associated with the use of alternative second line therapies. Last, Zhang et al., (2014) assumed that the first diabetesrelated complication and death were termination states, resulting in lower rates of diabetic complications and costs than this study estimations.  concluded that the life-years and QALYs until the first event were similar in the four treatment pathways and that metformin+SU had similar outcomes and lower drug costs compared to the assessed treatment alternatives.

Limitations
The Markov model used in this study does not intent to represent the actual clinical progression of patients with diabetes but to assess differences in two alternative therapy pathways under a defined set of assumptions. Thus, study results should be interpreted taking into consideration some limitations. Adult patients may develop T2D at any time during their life. This study assumed patients entered the model at age 60 years old. Study results are not generalizable to other T2D therapy initiation ages.
The Markov model employed in this study, assessed alternative pharmacological treatment pathways for T2D. To define the Markov states, this study model included most common drug therapies for the treatment of T2D instead of conventional health states, such as patient' A1c level or disease progress status [36,37]. Therefore, in this study transition probabilities did not depend on changes in A1c or disease progressions but on treatment failure rates observed in prior studies in patients with T2D.
We set the study time horizon at 25 years for the base-case because the survival data were available only until patients reach 85-years old. The death hazard ratio was estimated based on data drawn from a two-year trial results. Therefore, a more robust model would include death rates data for patients with diabetes for a longer time horizon [38].
Due to scarcity of studies some clinical input data were derived from trials outside of the US. Treatment failure rates for metformin were derived from studies conducted in the US, Canada, and the European Union (EU). Death hazard ratio for metformin+SU was derived from a study conducted in the United Kingdom and cardiovascular complication rates for dual therapy were derived from the studies conducted in the EU. Hypoglycemia data were derived from an international randomized clinical trial. Hypoglycemic events might lead to changes in medication. Future studies may include more treatment alternatives to account for changes in medication. Representativeness of study results may improve in the future using ongoing longterm comparative effectiveness studies conducted in the US in patients with T2D as model inputs [39]. Additionally, the study did not account for changes in clinical practice that have occurred after the publication of some of the studies used to derive the clinical input data.
Probabilities of treatment failure, hypoglycemia and cardiovascular complications were derived from studies with a limited time horizon. In addition, we assumed that the rates of cardiovascular events and treatment failure, and insulin dose remained constant through the study time horizon. Hypoglycemia rate data were derived from a trial which included patients with prediabetes. Thus, the hypoglycemia rate may be overestimated.
Cost-effectiveness estimations included only hypoglycemia, weight gain and cardiovascular eventsrelated costs. While these outcomes have been documented as the main outcomes differences between DPP-4i and SU other differences in outcomes between these treatment alternatives may exist [9,40]. Weight gain caused by insulin glargine was not considered in this study. Including different weight gain rates for each treatment pathway would yield more robust estimations but it would significantly increase the complexity of the Markov model. We conducted a sensitivity analysis for several key study measures including weight gain rates and study results did not change significantly. Microvascular complications, such as amputation, blindness or end state renal disease were not included either in the study because these outcomes are associated with uncontrolled blood glucose level and not with the use of specific drugs. Acute treatment costs for cardiovascular events were included in the CEA. Thus, medical costs for T2D related cardiovascular events could be underestimated.
We assumed that patients were adherent to antidiabetic medications when estimating the outcomes and drug costs. High medication costs may impact on the DPP-4i treatment adherence. Likewise, the risk of hypoglycemia may impact on the adherence of SU and insulin. Fixeddose combination drugs were not considered when estimating medication costs. Self-monitoring of blood glucose related costs were not included in the CEA.
This study assessed metformin+SU and metformin +DPP-4i treatment pathways; other treatment alternatives are marketed in the US such as newly FDA approved sodium glucose cotransporter-2 inhibitors and GLP-1 receptor agonists. Last, study model did not include triple oral or dual oral plus non-insulin injectable treatment alternatives before initiating insulin and prandial insulin option. Future studies may include more complex Markov models to compare the costeffectiveness of all currently available T2D treatment pathways. Future studies might consider the use of quality of life adjusted years as the study outcomes.
In spite of these limitations, this study has important strengths. Our study assessed three alternative treatment pathways during a long-term time horizon to better capture progression of the disease overtime. In addition, this study comprehensively assessed all prescription drug and health care costs related with diabetes complications. Furthermore, this study used insulin initiating rate to reflect both patients and providers decision making process to start insulin therapy.

Conclusions
This study assessed the cost-effectiveness of most commonly recommended in clinical guidelines T2D alternative long-term treatment pathways. The treatment pathway with DPP-4i as the second-line therapy was cost-effective compared to SU from the US health care payer perspective. The results of the one way and probabilistic sensitive analyses indicate that study findings are not sensitive to changes in the parameters used in the model. More studies assessing the cost-effectiveness of all long-term alternative T2D treatment pathways marketed in the US are needed.