Prasugrel vs clopidogrel in patients with acute coronary syndrome undergoing percutaneous coronary intervention: a model-based cost-effectiveness analysis for Germany, Sweden, the Netherlands, and Turkey

Abstract

Objective:

To evaluate the long-term cost-effectiveness of 12-months treatment with prasugrel vs clopidogrel from four European healthcare systems’ perspectives (Germany, Sweden, the Netherlands, and Turkey).

Methods:

In the TRITON-TIMI 38 trial, patients with an acute coronary syndrome (ACS) undergoing percutaneous coronary intervention (PCI) were treated with prasugrel or clopidogrel. Prasugrel reduced the composite end-point (cardiovascular death, MI, or stroke), but increased TIMI major bleeding. A Markov model was constructed to facilitate a lifetime horizon for the analysis. A series of risk equations constructed using individual patient data from TRITON-TIMI 38 was used to estimate risks of clinical events. Quality-adjusted life-years (QALYs) were derived by weighting survival time by estimates of health-related quality-of-life. Incremental cost-effectiveness is presented based on differences in treatments’ mean costs and QALYs for the licensed population in TRITON-TIMI 38, and the sub-groups of UA-NSTEMI, STEMI, diabetes, and the ‘core clinical cohort’ (<75 years, ≥60 kg, no history of stroke or TIA).

Results:

Mean cost of study drug was €364 (Turkey) to €818 (Germany) higher for prasugrel vs clopidogrel. Rehospitalization costs at 12 months were lower for prasugrel due to reduced rates of revascularization, although hospitalization costs beyond 12 months were higher due to longer life expectancy associated with lower rates of non-fatal MI in the prasugrel group. The incremental cost per QALY saved with prasugrel in the licensed population ranged from €6520 (for Sweden) to €14,350 for (Germany). Prasugrel’s cost per QALY was more favourable still in the STEMI and diabetes sub-groups of the licensed population.

Limitations:

Probabilistic analyses of the whole trial population is impractical due to the number of individual patient profiles over which population level results are calculated.

Conclusion:

Among patients undergoing PCI for ACS, treatment with prasugrel compared with clopidogrel resulted in favourable cost-effectiveness profiles from these healthcare systems’ perspectives.

Keywords::

• Prasugrel
• Cost-effectiveness
• Acute coronary syndrome
• PCI

Introduction

Among patients with an acute coronary syndrome (ACS) undergoing percutaneous coronary intervention (PCI), prevention of subsequent thrombotic events is currently pursued by use of dual anti-platelet therapy with aspirin and a thienopyridine1–4, which has been shown to be superior to aspirin therapy alone in the setting of PCI5,6 and medical therapy7–9. Despite a benefit over aspirin therapy alone, dual therapy with aspirin plus clopidogrel has several limitations, including variability in the levels of platelet inhibition, as well as a delayed onset of action6,10,11.

The TRial to Assess Improvement in Therapeutic Outcomes by Optimizing Platelet InhibitioN with Prasugrel (TRITON)-TIMI 38 trial)12 evaluated dual anti-platelet therapy with aspirin plus standard dose clopidogrel (300 mg loading dose followed by 75 mg maintenance dose) compared with aspirin plus a thienopyridine, prasugrel (60 mg loading dose and 10 mg maintenance dose), a P2Y12 receptor blocker that exhibits higher and more consistent levels of platelet inhibition than standard dose clopidogrel13. Among patients from 30 countries with an ACS in whom PCI was planned, prasugrel reduced the combined end-point of the incidence of cardiovascular death, non-fatal myocardial infarction, or non-fatal stroke by 19% over a 15-month period compared with clopidogrel, although accompanied by an increased risk of TIMI non-CABG major bleeding12. The trial showed that in patients with previous stroke or transient ischaemic attack (TIA), prasugrel was associated with an increased risk of ischaemic/haemorrhagic stroke. While the net clinical benefit favoured prasugrel over clopidogrel in the overall trial population, three patient sub-groups had either no net clinical benefit from prasugrel (patients 75 years of age or older [hazard ratio = 0.99; 95% CI = 0.81–1.21; p = 0.92], and patients weighing less than 60 kg [hazard ratio = 1.03; 95% CI = 0.69–1.53; p = 0.89]), or net harm from prasugrel (patients who had a previous stroke or transient ischemic attack [hazard ratio = 1.54; 95% CI = 1.02–2.32; p = 0.04]). Patients in these sub-groups had higher rates of bleeding compared with patients without these risk factors. This is reflected in the contraindication for patients with prior TIA or stroke in Prasugrel’s European license for patients with ACS who are undergoing PCI14, together with directions not to treat patients ≥75 years of age or with body weight <60 kg, with a 10 mg maintenance dose. We therefore distinguish in this paper between the overall licensed population in TRITON-TIMI 38 (defined as all patients other than those with prior stroke or TIA and including patients who are now recommended to be treated with a 5 mg maintenance dose), and the ‘core clinical cohort’15,16 — patients with a positive net clinical benefit, that is without prior TIA/stroke, aged under 75 years with body weight ≥60 kg.

There continues to be a growing emphasis on establishing the cost-effectiveness of new medicines before their widespread use. Cost-effectiveness analyses have shown prasugrel to be cost-saving in a US cost-setting17,18, based on substantial reductions in re-hospitalization costs. Healthcare provision differs between the US and many European countries, where aggregate cost-savings with prasugrel may be less likely due to lower unit costs for cardiac procedures, including repeat PCI. In the present analysis, we sought to evaluate the long-term cost-effectiveness of treating with prasugrel for 12 months compared with clopidogrel, in the population for which prasugrel is licensed, in Germany, Sweden, the Netherlands, and Turkey.

Methods

Overview

The analysis assessed the cost-effectiveness of treatment with prasugrel compared with clopidogrel among patients with an ACS undergoing PCI. Treatment was assumed to continue for 12 months, as this is the European licensed treatment period for prasugrel, with no additional treatment effects after study drug discontinuation. The perspective of each country’s national health service was adopted. Unit costs for drugs and re-hospitalizations were entered in the model in local currencies, but are expressed as Euros based on purchasing power parities19 relative to Germany; costs are indexed to December 201120. Health outcomes were assessed in terms of quality-adjusted life-years (QALYs). The framework for the analysis was a Markov state transition model constructed to facilitate a time horizon for the analysis of patients’ lifetimes, with assumed maximum survival duration of 40 years after the index ACS event.

The risks of clinical events were estimated using a series of risk equations constructed using individual patient data from TRITON-TIMI 38. The equations predict the risks of cardiovascular death, non-fatal MI or stroke, and bleeding, in each case conditional on treatment allocation, patients’ clinical characteristics at baseline (including qualifying MI type), and treatment interaction effects with prior stroke or TIA and diabetes. The model is thus designed to account for the contra-indication of prior stroke and is capable of addressing heterogeneity in prasugrel’s cost-effectiveness, including across sub-groups. This can be important in aiding decision-makers to maximize health outcomes and minimize costs, and is increasingly expected by decision-making bodies21,22.

Future costs and QALYs were discounted using discount rates summarized in Table 1 together with other data entered in the model.

Table 1.  Summary of key base-case model inputs.

Input	Value		Source
General model data
RRs for post-trial all-cause mortality^a
UA/NSTEMI	1.55 (1.31–1.84)		Allen et al.32
STEMI	1.84 (1.52–2.20)		Allen et al.32
recurrent NSTEMI	2.93 (2.34–3.66)		Mueller et al.33
recurrent STEMI	3.48 (2.77–4.37)		Mueller et al.33
stroke	2.39 (1.44–3.97)		Taneja et al.34
HRQoL decrements:
ACS	0.0409 (±0.0002)		Sullivan et al.36
stroke	0.0524 (±0.0002)		Sullivan et al.36
TIMI major bleed^b	25% decrement to population norms over 14 days		Assumption based on Wechowski et al.37
Country-specific data	Germany	Sweden	Netherlands	Turkey
Discount rates for outcomes	3%45	3%46	1.5%47	3%48
Discount rates for costs	3%	3%	4.0%	3%
Clopidogrel^c	€0.64	€0.08	€0.06	€1.02
Prasugrel	€2.92	€1.50	€1.58	€2.03
Aspirin (low dose)	€0.04	€0.08	€0.04	€0.04
Clopidogrel re-hospitalization^d	€4463	€5945	€3336	€2068
Prasugrel re-hospitalization^d	€4259	€5385	€3164	€2105

HR, hazard ratio; RR, relative risk.

^aIn addition to cited source relative risk is derived using Rosengren et al.30

^bPopulation norms according to Kind et al.35

^cDrug cost sources: Germany: Lauer Taxe; Sweden: Public Prices, TLV, August 2011; Netherlands: Taxe, Z-Index, August 2011; Turkey: Ministry of Health General Directorate of Pharmaceuticals and Pharmacy Price List. German drug prices paid by sickness funds, excludes patient co-payment and mandatory rebates (manufacturer and pharmacy).

^dCalculated from local DRGs: Germany: G-DRG-System 2009. Institut für das Entgeltsystem im Krankenhaus 2009, Sweden: Price list Region skåne-Skånes University Hospital; Nord DRG classics, August 18, 2011, The Netherlands: http://ctg.bit-ic.nl/Nzatarieven/top.do; Price level June 2009, Turkey: DRG costs from 20 hospitals in Turkey Price level 2008. Purchasing power parities (December 2011), Germany 1.00, Sweden 12.772, Netherlands 1.188, Turkey 1.368 [http://stats.oecd.org/Index.aspx?DataSetCode = SHA accessed 23 January 2012].

Markov model

Model pathways for the Markov model are shown in Figure 1. The model consists of a single 3-day acute period and 12 monthly cycles, plus an extrapolation period which completes the 40-year model time horizon. Patients entered the model at the point of study treatment allocation, and faced risks of cardiovascular death, non-fatal MI, non-fatal stroke, and bleeds. These risks are implemented in two stages: first, the risk of the composite end-point (the primary end-point in TRITON-TIMI 38) according to the baseline characteristics of each individual patient21; second, the composite event is then disaggregated into each of the individual event types (CV death, MI, or stroke). Similarly, the risk of bleed events (fatal, major, and minor) was modelled over the course of the 12-month treatment period. The bleeding risk equations reflect all TIMI major and minor, CABG and non-CABG bleeds. Although patients may suffer a sequence of events in practice, the model predicts a maximum of one primary end-point and one bleed event per patient. All cardiovascular deaths and fatal bleeds are modelled, and all re-hospitalizations are captured through a re-hospitalization risk equation. Only the first non-fatal event is modelled to impact on quality-of-life. No clinical events are modelled beyond 12 months; however, non-fatal ischemic events that are modelled to have occurred by 12 months are considered prognostic and are assumed to increase long-term all-cause mortality risks in the extrapolation period beyond 12 months. Although further non-fatal ischaemic events are not predicted by the model, the impact of such events is reflected through the relative risks for mortality associated with the first event, which reflect the impact of repeat events in the observational studies on which they are based. In contrast, non-fatal bleeding events in the model are treated as temporary events conferring no additional ongoing mortality risk during this period (fatal bleeds are captured directly in the model’s trial based risk equations), for reasons discussed below.

Figure 1.  Model pathways for the cost-effectiveness model.

Risk equations

Risk equations were estimated using data from all patients in TRITON-TIMI 38, irrespective of baseline stroke/TIA status. Separate risk equations were generated for the UA/NSTEMI and STEMI populations due to their different risk profiles. The equations adjusted for baseline characteristics and accounted for potentially important treatment interaction effects. Potential baseline covariates for the equations included demographic, clinical and disease history, and concomitant medication use. The selected covariates in each of the final equations reflected their importance in terms of magnitude of effect, statistical significance, and clinical rationale. Initially, model selection used backward step-wise elimination using p-values of <0.10. Baseline variables were included in the models on this basis with a check that forward step-wise selection did not yield alternative variables for inclusion. In addition to variables selected in this way, age ≥75 years, weight <60 kg, and the main effects of prior stroke or TIA and diabetes, and the potential interaction of each of these with randomized treatment were also included. The age and weight characteristics were included as TRITON-TIMI 38 demonstrated that age ≥75 years and weight <60 kg, in addition to prior stroke/TIA, were independent risk factors for non-CABG-related TIMI major bleeding for patients undergoing prasugrel treatment16.

Due to a rapidly declining risk of events following the index procedure, logistic models were applied for events occurring within 3 days of randomization23, and Weibull models24 over the remainder of the 12-month treatment period. Multinomial logistic regressions were used to derive equations that predict the probabilities that, having had an event, the event is fatal, a non-fatal MI, or a non-fatal stroke. The estimated risk equations, which apply to all countries in the analysis, are presented in the Appendix. The effectiveness of clopidogrel and prasugrel, in terms of the relative rate of clinical events over the 12-month treatment phase, was estimated separately within each of the risk functions, but was entirely consistent with the overall published trial results in the modelling population12.

Survival

The model reflects all 12 months cardiovascular and bleed deaths seen in TRITON TIMI-38. In addition, the risk of non-cardiovascular mortality over this period is based on national life tables25–28 stratified by age and sex, adjusted to exclude cardiovascular cause-specific mortality29.

Mortality after 12 months is based on these life tables with adjustment to reflect the impact of revascularized ACS and the modelled 12-month MIs and strokes. A literature search yielded no direct relative risks for mortality among revascularized ACS patients compared to a CHD-free population. CHD brings increased risk of mortality, moderated by revascularization. We applied a relative risk (RR), compared to a CHD-free population, of 1.21 in estimating mortality in the absence of MI or stroke. This is based on a pooling of Rosengren et al.’s30 RRs for mortality 4–8, 8–12, and 12–16 years after a diagnosis of angina pectoris (RR = 1.59, 95% CI = 1.16–2.20), and Mehta et al.’s31 pooled effect estimate for the impact of revascularization (odds ratio = 0.76, 0.62–0.94). PCI carries a short-term risk of mortality, but is performed in order to avoid longer term risks including the detrimental effects on life expectancy of further MIs and strokes. As the majority of patients in TRITON TIMI-38 were revascularized as a result of an MI, these patients’ mortality is modelled using a 10 year Duke study32 that found NSTEMI and STEMI to be associated with RRs of 1.28 (0.95–1.72) and 1.52 (1.06–2.19), respectively. For patients who suffered a repeat MI the lower estimate of further risk from the TIMI-2 re-infarction follow-up study33 (RR = 1.89, 1.11–3.23) is applied; the impact of stroke in this population (RR = 2.39) was modelled based on the UK-PRAIS study34.

QALYs

Quality-adjusted survival, in the form of QALYs, was derived by weighting patients’ survival times by estimates of health-related quality-of-life (HRQoL), which reflect the impact of clinical events. The model applies HRQoL weights to all patients based on population norms (by age) based on the EQ-5D35. These were adjusted by HRQoL decrements, also based on the EQ-5D, from the US Medical Expenditure Panel Survey36 (Table 1). The decrements account for patients’ ACS status at treatment initiation and the impact of any prior non-fatal MI. Further decrements are applied for patients who experience non-fatal stroke after PCI and, in patients without a prior MI, a non-fatal MI after PCI. One decrement is applied, regardless of whether patients are modelled to have experienced single or multiple events. Patients predicted to suffer a stroke as a first event retain the stroke decrement after a subsequent MI as the stroke decrement is higher than that for MI. These HRQoL effects are deemed to be permanent. In addition, a temporary (14 day) decrement of 25% is applied for major bleeds37.

Medication costs

Clopidogrel was costed for each of the four countries based on the availability of generic and branded clopidogrel. The price of clopidogrel used in the modelling was €0.64 (generic clopidogrel, Germany), SEK1.02 (generic clopidogrel, Sweden), €0.06 (generic clopidogrel, Netherlands), and TL1.40 (branded clopidogrel, Turkey) per day, while prasugrel was assigned costs of €2.92, SEK19.10, €1.88, and TL2.78 based on prices current at the time of analyses. These prices (together with those for daily aspirin) are summarized in Table 1 in Euros based on purchasing power parities relative to Germany. Costs for clopidogrel and prasugrel, in combination with aspirin, were applied over the duration of the 12-month maintenance treatment phase.

Other costs

Costs relating to hospitalizations beyond the index admission were based on hospitalization data collected in an economic sub-study of TRITON-TIMI 38 involving all study patients (n = 6705) enrolled in eight pre-specified countries including Germany17. Economic sub-study patients’ baseline characteristics were similar to those in the main study. The economic sub-study documented only those re-hospitalizations that were considered by a clinician blinded to the treatment arm to potentially relate to the ACS condition, bleeding or the PCI intervention, separate from the index episode. Both planned and unplanned PCIs were included as re-hospitalizations. In total, 2487 re-hospitalizations from the economic sub-study met these criteria (there were substantially more re-hospitalizations than events). Each hospitalization for a cardiovascular or bleeding event was assigned a US diagnosis related group (DRG) by the blinded clinician, based on principal diagnosis and procedure(s) performed17. Assignment of national unit costs to match US DRGs was performed by local health economists. The weighted average cost per re-hospitalization was then calculated using each treatment arm’s distribution of DRGs.

Poisson regression, using panel data methods38, was used to estimate the incidence of re-hospitalization in each model cycle. Many re-hospitalizations occur subsequent to a non-fatal event, and, therefore, the regression includes coefficients for MI and stroke in earlier cycles. Thus, the equation predicts all re-hospitalizations based on the first cardiovascular and bleed event.

As many cardiac events occurred during the index admission the likelihood that early events lead to re-hospitalization is lower than for events that occur later. The model, therefore, applies separate re-hospitalization rates for events occurring before and after 30 days. Randomized treatment arms, and interactions between randomized treatment and the incidence of hospitalizations following earlier myocardial infarctions were also included in the regression. For all countries’ analyses the regression constant reflects the rate of re-hospitalizations in the economic sub-study in northern continental Europe (France and Germany), except in the case of Turkey, for which the lower rate seen in Spain, Italy, and the UK was applied. This background rate of re-hospitalization reflects the probability that an event-free patient is re-hospitalized as a consequence of the underlying ACS rather than an end-point event.

Costs are accrued over the trial period by assigning the average re-hospitalization cost for each trial arm to the expected rates of re-hospitalization estimated through the Poisson analysis (Appendix—Table A1). For costs beyond the 12 month treatment horizon, a common cost per year was applied for all surviving patients in both treatment groups. This was based on the background hospitalization rate given by the Poisson regression (i.e., the rate predicted in the absence of clinical events), costed at the average re-hospitalization cost in the clopidogrel arm.

Cost-effectiveness analysis

Cost-effectiveness analyses are presented with mean costs, life years, and QALYs, together with incremental cost-effectiveness ratios, for the licensed population, and its UA-NSTEMI, STEMI, diabetes, and core clinical cohort sub-groups. The mean incremental cost-effectiveness is presented based on the differences between prasugrel and clopidogrel’s mean costs and QALYs, for all patients in each population (from TRITON-TIMI 38).

Parameters used in the model were entered as probability distributions to facilitate probabilistic sensitivity analysis21. Deterministic sensitivity analyses were also performed to examine the impact of alternative assumptions regarding discount rates, utility decrements for haemorrhage, MI and stroke, relative risks for all-cause mortality, DRG costs, and clopidogrel acquisition costs. Sensitivity analyses were performed by re-running the model for each of the patient profiles in the licensed population under each scenario examined, and taking the mean costs and effects over all patients. Probabilistic analysis is based on a single patient profile for each patient group, selected as the median cost-effectiveness profile in the relevant group (defined as the profile with median net benefit, calculated as the value of prasugrel’s incremental QALYs using a threshold value of €30,000 per QALY, less incremental costs). In the countries included in this analysis, formal threshold values for cost-effectiveness are not established. In the UK, NICE applies a value of £20–30,000 per QALY.

Results

Risk equations

Details of the estimated risk equations are shown in the Appendix. Table A2 shows the equations relating to the composite ischaemic and haemorrhagic end-points. Table A3 displays multinomial risk equations used to disaggregate these composites into their component events. Consistent with the overall trial results, prasugrel was shown to be associated with a reduction in the primary outcome and an increase in bleeding events. There are several risk equations reflecting the structure of the model. Although treatment effects estimated in the equations are consistent with the overall effects reported in TRITON-TIMI 38, their magnitudes differ slightly due to the separation of early event risks (those within the first 3 days) from those over the remainder of the trial, and the inclusion of interaction effects for prior stroke and diabetes at baseline, with attendant consequences for p-values. Separate treatment interaction effects for age and weight were not fitted within these equations. Characteristics associated with an increase in the primary outcome were older age, a history of diabetes, heart failure, and prior MI, whilst those associated with an increase in bleeding events were older age, low body weight, kidney impairment, and female gender.

Event probabilities

Event rates in the model vary only marginally between countries due to minor differences in short-term non-cardiovascular mortality based on national life tables. The primary end-point for the overall licensed population was modelled to occur in 11.3% of clopidogrel patients compared with 8.9% of prasugrel patients (compared with 11.3% and 9.0% in the trial), with the reduction driven primarily by myocardial infarction (8.5% for clopidogrel vs 6.4% for prasugrel). There was a numerical reduction in cardiovascular death (2.1% for clopidogrel vs 1.8% for prasugrel) and similar rates of stroke were observed (0.7% each) (Table 2). Total bleeding events (TIMI major or minor, CABG and non-CABG) for the overall licensed population were higher at 12 months in the prasugrel group (3.6% with clopidogrel vs 4.8% with prasugrel). There is some variation in modelled event rates across population sub-groups (Table 2). The UA-NSTEMI sub-group represents ∼75% of the total licensed population within the trial, and event rates are generally consistent with the overall population, although the rate of cardiovascular death and bleeds was slightly lower in both arms compared with the licensed population. In the STEMI population the rate of cardiovascular death and bleeding in both arms was correspondingly higher, and MI slightly lower. In the diabetic group rates of all events were higher than in the overall licensed population.

Table 2.  Cost-effectiveness results for overall licensed population and specific sub-groups (event rates at 12 months).

	Licensed population		UA-NSTEMI		STEMI		ACS diabetes		Core cohort
	(n = 13,090)		(n = 9669)		(n = 3421)		(n = 2947)		(n = 10,804)
	Clopidogrel	Prasugrel	Clopidogrel	Prasugrel	Clopidogrel	Prasugrel	Clopidogrel	Prasugrel	Clopidogrel	Prasugrel
Cardiovascular death	2.05%	1.76%	1.80%	1.66%	2.76%	2.05%	3.59%	2.73%	1.57%	1.36%
Myocardial infarction	8.49%	6.43%	8.60%	6.61%	8.17%	5.91%	10.64%	6.72%	8.15%	6.21%
Stroke	0.74%	0.69%	0.72%	0.54%	0.79%	1.12%	1.23%	1.01%	0.64%	0.58%
Total combined end-point	11.28%	8.87%	11.13%	8.80%	11.71%	9.08%	15.46%	10.46%	10.36%	8.14%
TIMI major bleed	1.71%	2.30%	1.49%	2.18%	2.32%	2.64%	2.21%	2.50%	1.50%	1.99%
TIMI minor bleed	1.93%	2.51%	1.69%	2.40%	2.61%	2.82%	2.70%	2.93%	1.47%	1.96%
Total bleed	3.64%	4.81%	3.18%	4.58%	4.93%	5.46%	4.91%	5.42%	2.97%	3.95%
Germany
Drug cost	226	1044	227	1045	225	1042	224	1037	227	1050
Hospitalization costs ≤12 months	1797	1662	1811	1676	1755	1622	1998	1761	1761	1630
Hospitalization costs >12 months	17,919	18,019	17,952	18,029	17,824	17,989	16,773	17,019	19,440	19,540
Total costs	19,942	20,725	19,990	20,751	19,804	20,652	18,995	19,817	21,428	22,220
QALYs	10.657	10.712	10.661	10.702	10.647	10.740	9.972	10.109	11.524	11.579
Incremental cost per QALY (€)		14,350		18,530		9131		6025		14,487
Sweden
Drug cost	28	535	28	536	28	534	28	531	28	538
Hospitalization costs ≤12 months	2395	2102	2415	2120	2339	2052	2664	2228	2347	2063
Hospitalization costs >12 months	24,580	24,708	24,578	24,674	24,587	24,803	22,941	23,261	26,752	26,880
Total costs	27,003	27,345	27,020	27,330	26,954	27,388	25,633	26,021	29,128	29,481
QALYs	10.945	10.997	10.930	10.968	10.988	11.080	10.214	10.347	11.870	11.923
Incremental cost per QALY (€)		6520		8016		4738		2910		6711
Netherlands
Drug cost	18	564	18	564	18	562	18	559	18	566
Hospitalization costs ≤12 months	1338	1230	1349	1240	1307	1200	1488	1303	1311	1206
Hospitalization costs >12 months	12,290	12,353	12,300	12,348	12,262	12,370	11,543	11,703	13,297	13,359
Total costs	13,646	14,147	13,667	14,152	13,587	14,132	13,049	13,566	14,626	15,132
QALYs	12.919	12.987	12.907	12.959	12.952	13.065	11.988	12.156	14.053	14.122
Incremental cost per QALY (€)		7369		9378		4788		3080		7342
Turkey
Drug cost	361	725	361	726	359	723	356	720	363	729
Hospitalization costs ≤12 months	350	345	353	348	342	337	389	365	343	339
Hospitalization costs >12 months	3079	3097	3083	3097	3068	3098	2845	2890	3368	3387
Total costs	3789	4167	3796	4171	3769	4158	3591	3975	4074	4455
QALYs	9.521	9.573	9.518	9.558	9.531	9.616	8.810	8.937	10.366	10.419
Incremental cost per QALY (€)		7294		9371		4552		3036		7207

Base-case cost-effectiveness

Hospitalization costs with prasugrel were lower than for clopidogrel in the first 12 months, but higher beyond the 12-month treatment period due to prolonged life expectancy. These costs represent an allowance in the analysis for follow-on costs associated with cardiovascular illness both among patients who avoid cardiovascular death over the period of the trial, and those with greater subsequent life expectancy due to the avoidance of higher long-term mortality associated with MI and stroke. Taken together, short-term costs savings and allowance for additional follow-on costs result in negligible cost differences in this analysis (Table 2).

Given the negligible net effect of re-hospitalizations, total incremental costs are almost entirely due to drug acquisition costs. We applied drug acquisition costs for clopidogrel based on generic prices. In the overall licensed population the additional drug cost of prasugrel ranged from €364 in Turkey to €818 in Germany, with incremental cost per QALY with prasugrel of €7294 and €14,350, respectively. Prasugrel was cost-effective in each of the sub-groups presented in Table 2. Cost per QALY ratios were lower in populations at higher risk of ischemic events—the STEMI and diabetic sub-groups, than the overall licensed population, and a little higher in the UA/NSTEMI group, reaching a maximum of €18,530 in Germany. The core clinical cohort population comprised ∼80% of patients in the overall licensed population, and ICERs in this sub-group were similar to those in the overall population.

Sensitivity analysis

The probabilities for prasugrel being cost-effective are shown in Figure 2 and relate to the overall licensed population. The probabilities of prasugrel being cost-effective for patient profiles with median cost-effectiveness are 88% or higher at threshold values of €20,000 and above and close to 100% at a threshold of €30,000. The exception to this is Germany, where a greater price premium for prasugrel sees the product’s cost-effectiveness probability being ∼70% at the €30,000 threshold.

Figure 2.  Cost-effectiveness acceptability curves showing the probability of prasugrel being cost-effective (compared to clopidogrel) for a range of cost-effectiveness thresholds. This is shown for the patient profile with median cost-effectiveness in the overall licensed ACS population (based on net health benefit at €30,000 per QALY).

In the deterministic overall licensed population sensitivity analysis, alternative model assumptions, including doubling the cost of re-hospitalizations and increasing the disutility associated with bleeds, had little impact on the cost-effectiveness of prasugrel (Table 3). The impact of each sensitivity analysis on costs and QALYs in each arm is consistent across countries, although the impact on incremental cost-effectiveness ratios varies. For Germany, discounting both costs and QALYs at 5% reduced mortality impact of non-fatal events, while adjusting clopidogrel efficacy in the opening 3-day period (‘loading dose adjustment’) and setting drug costs for clopidogrel to zero each saw prasugrel’s ICER rise to ∼€18,000, compared with the base case of €14,350. In Sweden the greatest impact of any sensitivity analysis came in setting re-hospitalization costs to zero, with the ICER rising from a base of €6520 to €9656. In the Netherlands the ICER remained below €13,000, reaching €12,122 when both costs and QALYs were discounted at 5%, compared with a base of €7369. Halving the mortality impact of MI and stroke and the clopidogrel loading dose adjustment increased the ICER in Turkey to €9017 and €9399, respectively, whilst setting drug acquisition costs in the clopidogrel arm to zero increased the ICER to €14,251. In all other cases the ICER in Turkey remained below €8000.

Table 3.  Cost-effectiveness sensitivity analyses for overall licensed population.

	Germany			Sweden			Netherlands			Turkey
	Cost	QALY	ICER (€)	Cost	QALY	ICER (€)	Cost	QALY	ICER (€)	Cost	QALY	ICER (€)
base case	783	0.055	14,350	342	0.052	6520	501	0.068	7369	378	0.052	7294
0% discounting	859	0.089	9607	430	0.086	4992	568	0.088	6442	394	0.082	4824
5% discounting	752	0.041	18,225	308	0.040	7768	490	0.040	12,122	370	0.040	9261
4 × bleed HRQoL decrement	783	0.054	14,379	342	0.052	6534	501	0.068	7381	378	0.052	7309
½ × MI/stroke HRQoL decrement	783	0.055	14,313	342	0.053	6503	501	0.068	7350	378	0.052	7275
2 × MI/stroke HRQoL decrement	783	0.054	14,425	342	0.052	6555	501	0.068	7406	378	0.052	7332
½ × MI/stroke mortality effect	763	0.044	17,497	316	0.042	7522	488	0.054	8947	374	0.042	9017
2 × MI/stroke mortality effect	801	0.064	12,498	366	0.062	5894	512	0.079	6509	381	0.060	6382
zero hospital costs	818	0.055	14,994	507	0.052	9656	546	0.068	8029	364	0.052	7032
½ × rehospitalization costs	801	0.055	14,672	425	0.052	8088	523	0.068	7699	371	0.052	7163
2 × rehospitalization costs	748	0.055	13,706	178	0.052	3385	456	0.068	6708	392	0.052	7556
clopidogrel loading adjustment	766	0.042	18,181	321	0.041	7908	491	0.052	9375	375	0.040	9399
clopidogrel drug cost set at zero	1009	0.055	18,494	371	0.052	7058	519	0.068	7634	739	0.052	14,251

Discussion

In prasugrel’s licensed population (all patients other than those with prior stroke or TIA, including patients who are now recommended to be treated with a 5 mg maintenance dose), drug costs were higher than for clopidogrel; however, cost savings associated with reduced hospitalizations and revascularizations over 1 year partially offset this higher drug acquisition cost. Higher projected hospitalization rates beyond the trial period, driven by a longer life expectancy due to fewer MIs in the prasugrel group over the treatment period, lead to higher total costs for prasugrel. The clinical benefit, associated with therapy, however, translates into additional life-years and QALYs over the longer term.

When a new intervention is more costly than a comparator but also improves health outcomes for patients, any conclusion about whether that intervention is cost-effective requires consideration of the appropriate cost-effectiveness threshold. In principle this threshold represents the opportunity cost of the additional expenditure on the new technology; that is, the magnitude of what is forgone as a result of funding the new technology. In budget-constrained health systems any additional expenditure on a new intervention will usually require other services (often in quite different clinical areas) to be displaced to release the necessary funding, in which case the opportunity cost can be expected to fall on patients’ health39. Decision-makers in one such system—NICE in the UK NHS—have been explicit about their cost-effectiveness threshold—£20,000–£30,000 (∼25,000–37,000 Euros) per QALY gained40. Amongst other health systems routinely using cost-effectiveness analysis to support decisions, few have been so transparent as NICE. If the NICE threshold is used to interpret the cost-effectiveness results presented here, all ICERs from the base-case and sensitivity analyses relating to the overall licensed population fall below the lower bound, suggesting that prasugrel is cost-effective. This conclusion would not hold if thresholds were lower in some of the countries considered here. Turkey’s lower GDP per capita and healthcare expenditure per capita may suggest a lower threshold would apply, but the ICERs for Prasugrel range from €3036–€9371 per QALY in that jurisdiction in the base-case, and up to €14,251 in sensitivity analysis, appreciably lower than the lower bound of the NICE threshold, and within the WHO-CHOICE threshold of 3-times GDP41. Further research is needed to estimate the cost-effectiveness thresholds relevant in specific jurisdictions.

Sensitivity analyses which included halving the mortality impact of stroke and MI and increasing 4-fold the disutility associated with bleeds showed prasugrel’s cost-effectiveness was robust to different assumptions within the model. Prasugrel’s incremental cost-effectiveness was demonstrated in groups of patients with varying underlying event risks with clopidogrel, and in the sub-groups of patients with either UA/NSTEMI or STEMI, and diabetes. Results for STEMI and diabetes sub-groups, populations at high-risk of ischemic events, were particularly favourable.

A further sub-group, the core clinical cohort, has been defined as patients without prior stroke or TIA at baseline, age <75 years, and weight ≥60 kg15,16. Significant interaction effects (treatment effect modifiers) have been reported both by prior stroke or TIA, and the combined characteristics for the core clinical cohort (including prior stroke or TIA)12,16. The present model is designed to evaluate cost-effectiveness in the overall licensed population and, therefore, incorporates the interaction by prior stroke/TIA. To fully reflect the additional benefit seen in the core clinical cohort within a model for the overall licensed population, however, would require separate interaction effects for age ≥75 and weight <60 kg that could not be fitted to the model’s risk equations due to the loss of power in separating prior stroke/TIA from the age and weight characteristics. Consequently, the ratio of 12-month event rates with prasugrel compared with clopidogrel in the core clinical cohort reported in Table 2 (0.79) is comparable to that in the overall licensed population and the hazard ratio for the population without prior history of stroke or TIA as reported in TRITON TIMI-3812. In the model, therefore, the difference between the overall licensed population and the core clinical cohort is seen in terms of lower absolute event rates in both arms, rather than further absolute reductions with prasugrel. As a result the cost per QALY estimates for prasugrel in this cohort are likely to be conservative.

In the TRITON-TIMI 38 trial, approximately one-quarter of patients received study drug prior to PCI and three-quarters received study drug during PCI12. The approved 300 mg loading dose and 75 mg maintenance dose of clopidogrel was studied as the comparator arm to prasugrel 60 mg loading dose and 10 mg maintenance dose. The PRINCIPLE-TIMI 44 trial demonstrated that loading with 60 mg prasugrel resulted in greater platelet inhibition than a 600 mg clopidogrel loading dose13, and in the ACAPULCO study42 maintenance with 10 mg prasugrel resulted in greater platelet inhibition than a 150 mg clopidogrel maintenance dose (following a 900 mg clopidogrel loading dose). This evidence supports that the timing of the loading dose did not substantially influence the overall efficacy43. Nevertheless, we performed a sensitivity analysis assuming the treatment effect of clopidogrel was the same as for prasugrel during the first 3 days of therapy, when the potential benefit of a higher or earlier loading dose might accrue. Even when it was assumed that there was no treatment difference between prasugrel and clopidogrel during the first 3 days, prasugrel’s cost per QALY changed little from the base case analyses, with ICERs ranging from ∼€7900 in Sweden to €18,000 in Germany. Thus, had either a higher clopidogrel loading dose or pre-treatment been employed as the comparator arm to prasugrel, the overall cost-effectiveness results would probably not have been significantly altered.

Our cost-effectiveness results vary to some degree between countries. All analyses share common risk equations, relative risks for all-cause mortality following non-fatal MI and stroke, and HRQoL impacts associated with events. Although costs associated with individual DRGs assigned to re-hospitalizations vary between countries, the aggregate average cost per re-hospitalization is broadly similar across the countries, and the net effect of re-hospitalizations on cost-effectiveness is comparatively minor. The major determinant of international differences in prasugrel’s cost-effectiveness profile in our analysis is drug acquisition cost, although different discount rates also play some part. Whereas prasugrel’s additional cost per day over clopidogrel is ∼€1.50 in Sweden and the Netherlands, and €1.00 in Turkey, prasugrel’s daily cost is ∼€2.30 higher than that of clopidogrel in Germany. We employed drug prices that were current at the time of analyses. However, still lower generic clopidogrel prices may impact prasugrel’s cost-effectiveness. Sensitivity analyses setting drug costs in the clopidogrel arm to zero show the maximum effect (an ICER of €18,494 in Germany) that continued reductions in clopidogrel prices could have on our analyses for the overall license population in TRITON. Thus, prasugrel’s cost per QALY remains below €20,000, irrespective of clopidogrel’s acquisition cost.

Each deterministic sensitivity analysis reflects diversity in patients’ baseline characteristics by re-running the model for all patient profiles in TRITON, and aggregating these individual results. To conduct probabilistic analyses on this basis, however, would be impractical as this would require several million iterations of the model.

We therefore report limited probabilistic analyses in Figure 2 based on analyses of the patient profile in each group with respect to cost-effectiveness based on a threshold cost-effectiveness value of €30,000 per QALY. The probabilities of prasugrel being cost-effective at given thresholds for cost per QALY are shown in Figure 2. These analyses show prasugrel has very strong probabilities of being cost-effective for these illustrative patient profiles. Although the probabilistic analyses for Germany show prasugrel to be cost-effective at a threshold of €30,000, the cost-effectiveness acceptability curve is less emphatic than for other countries. We explored this further, and found that substituting German drug costs with those for the Netherlands results in similar probabilities for cost-effectiveness as in the Dutch analysis. A limitation of the present analysis is the inability to perform probabilistic sensitivity analyses on the whole patient population or across the whole of any of the sub-groups. However, the base-case analyses have the advantage of being able to take account of the differing characteristics and risks, and to some extent treatment effects, seen in the different groups.

Prasugrel was associated with an increased risk of bleeding in TRITON-TIMI 38. All fatal bleeds and cardiovascular deaths observed in the trial are directly reflected in the model. Patient’s life expectancy beyond the trial period is, however, assumed to be unaffected by serious non-fatal bleeds within the trial. Furthermore, as patients cease both treatments by ∼12 months after PCI, the risk of further new bleeding events is minimal. Were the mortality risks associated with non-fatal serious bleeding within the trial to persist over the long-term (i.e., the 40 year time horizon of the model), the consequences of the additional non-fatal bleeding with prasugrel would be under-estimated. Analysis by Hochholzer et al.44, however, suggests the assumption of no long-term elevation in mortality risks after bleeding is reasonable, based on the estimated impact of serious bleeding on mortality in TRITON-TIMI 38. This showed a declining hazard ratio over time following a non-fatal serious bleed, stabilizing after 40 days at ∼1.38 (0.72–2.66). Even this non-significant elevation in mortality risk, however, is likely to overstate the long-term impact of serious bleeding on mortality due in part to the underlying conditions contributing to the bleeding event (e.g., colorectal cancer), and to residual confounding such as subsequent cardiovascular deaths, which are already accounted for in the model. We explored our assumptions around bleeding in the sensitivity analysis; a 4-fold increase in the disutility associated with bleeds had little impact on the cost-effectiveness.

There are a number of differences between this analysis and the TRITON-TIMI 38 analysis of Mahoney et al.17 that evaluated prasugrel from the perspective of the US healthcare system. In their base-case analysis, prasugrel was shown to be dominant over clopidogrel (prasugrel achieved greater clinical benefit at lower costs). That analysis differed from the present analysis in that it did not include any cardiovascular healthcare costs accrued beyond the period covered by the trial, although these on-going costs were recognized in sensitivity analyses. The most important difference between the present model and the analysis by Mahoney et al.17 is the much higher cost of PCI in the US compared with Europe. Mauskopf et al.18 recently analysed prasugrel’s cost-effectiveness from the perspective of a US Managed Care Organization. This analysis also estimated re-hospitalization costs to be substantially lower with prasugrel, although in this case savings were also attributed to lower thienopyridine costs. Re-hospitalization savings amounted to ∼$100 per patient, whereas our analysis results in additional re-hospitalization costs, but these are modest and lead to prasugrel being cost-effective without dominating clopidogrel. Both the US and European analyses show prasugrel to be cost-effective compared with clopidogrel, the current study doing so despite the now much reduced acquisition costs of clopidogrel.

Conclusion

Among patients undergoing PCI for ACS, treatment with prasugrel compared with clopidogrel resulted in favourable cost-effectiveness profiles for the licensed population from the perspective of the healthcare systems considered. A range of sensitivity analyses indicates that this conclusion is robust to the range of uncertainties that exist in generalizing the results of a multi-national clinical trial to routine clinical practice in a number of countries.

Transparency

Declaration of funding

This study was funded by Daiichi Sankyo Company, Limited and Eli Lilly and Company.

Declaration of financial/other relationships

A. Barrett and P. Graham-Clarke have disclosed that they are employees of Eli Lilly and Company; C. Schmitt is a former employee of Eli Lilly. A. Bakhai has disclosed that he recruits to and manages clinical trials and registries with commercial sponsorship from Lilly, Roche, GSK, and Health-Smart and has received honoraria for educational activities and grants from Takeda, AstraZeneca, Sanofi Aventis, and Lilly. A. Davies and M. Sculpher have disclosed that they are employees of Oxford Outcomes, a company that received funding from Eli Lilly for economic analyses. JME Peer Reviewers on this manuscript have no relevant financial relationships to disclose.

Acknowledgements

The authors thank Antje Tockhorn, Pat McCollam, Johannes Clouth, and Jay Bae for their assistance in the preparation of this manuscript; Dr Heiko Friedel and Ms Dana Trauvetter from Germany and the Mentor R&D Training and Consultancy, Turkey who reviewed the DRG unit costs and assigned local unit costs of German DRG to match US DRGs. The authors also thank Dr Stephen Palmer for assistance in the design of the analysis and Sabina Murphy for assistance in preparing the manuscript.

Notice of Correction

The version of this article published online ahead of print on 12 Feb 2013 contained an error on page 7. The error has been corrected for this version.