Cost-Effectiveness of Genotype-Guided and Dual Antiplatelet Therapies in Acute Coronary Syndrome

Context Several options for antiplatelet therapy after percutaneous coronary intervention for acute coronary syndrome are available. Contribution This cost-effectiveness analysis compared drug-only strategies (generic clopidogrel, prasugrel, or ticagrelor) and genotype-guided strategies targeting ticagrelor or prasugrel. Ticagrelor was the most cost-effective strategy. The genotyping-with-prasugrel strategy was superior to giving all patients prasugrel. The genotyping-with-ticagrelor strategy was clinically superior but more expensive than clopidogrel. Caution No randomized trials have directly compared genotyping strategies or prasugrel with ticagrelor. Implication Genotype-guided personalization of antiplatelet therapy could improve cost-effectiveness in some situations, but ticagrelor for all without genotyping also seems reasonable. The Editors Dual antiplatelet therapy combining aspirin with a second agent is the mainstay of therapy after acute coronary syndrome (ACS), particularly among patients who receive a percutaneous coronary intervention (PCI) (1). Antiplatelet agents reduce thrombotic events, such as myocardial infarction (MI) and stent thrombosis, but increase risk for bleeding (2). Approximately one half of the 1.1 million ACS events in the United States every year are treated with a PCI, making the choice of antiplatelet therapy a common and important clinical decision (3, 4). Clopidogrel has been the standard of care after PCI for nearly a decade (5). Until recently, it was the second-largest drug in terms of sales, and much of the $12 billion spent on it each year was for use after ACS (6). However, many patients receiving clopidogrel and aspirin have recurrent cardiovascular events (7, 8), and on-treatment platelet inhibition varies considerably (9, 10). Patients who carry a loss-of-function polymorphism of CYP2C19 (a key enzyme involved in the hepatic activation of clopidogrel) achieve less platelet inhibition with clopidogrel and have more thrombotic events (1113) and less bleeding. However, carriers of gain-of-function alleles of the CYP2C19 enzyme achieve greater platelet inhibition with clopidogrel and have fewer thrombotic events and more bleeding (14, 15). Two new drugs, prasugrel and ticagrelor, are approved for use in patients having PCI for ACS (1619). The greater antiplatelet activity of these agents reduces the rate of MI and cardiovascular death compared with clopidogrel. However, prasugrel increases fatal bleeding so that its net effect on mortality rates is neutral (16, 17). Ticagrelor is dosed twice daily and causes mild to moderate dyspnea in some patients (18, 19), which may adversely affect adherence. Both agents are expensive, particularly when compared with generic formulations of clopidogrel that are now available. Further, commercial availability of genetic testing may allow clinicians to personalize antiplatelet therapy so that the new, more expensive drugs could be selectively prescribed to patients most likely to benefit (11, 12, 20, 21). These recent developments have altered the therapeutic landscape, highlighting the need for a comprehensive evaluation of alternative strategies for dual antiplatelet therapy. There are no head-to-head clinical trials of ticagrelor with prasugrel and no prospective studies of genotype-based treatment decisions. In this article, we present a simulation that addresses uncertainties about the role of genotyping and identifies the most cost-effective strategies for dual antiplatelet therapy after PCI for ACS. Methods We developed a discrete-state Markov model to compare 5 strategies of dual antiplatelet therapy (22). Drug-Only Strategies Drug-only strategies were generic clopidogrel, prasugrel, or ticagrelor. We assumed that generic clopidogrel had the same efficacy as the proprietary formulation. On the basis of the results of TRITON-TIMI 38 (Trial to Assess Improvement in Therapeutic Outcomes by Optimizing Platelet Inhibition With PrasugrelThrombolysis in Myocardial Infarction), we assumed that prasugrel led to fewer cardiovascular deaths but more fatal bleeding compared with clopidogrel (16, 17). On the basis of the PLATO (Platelet Inhibition and Patient Outcomes) study, we assumed that ticagrelor reduced cardiovascular deaths without a corresponding increase in fatal bleeding (18, 19) and that some patients had dyspnea and bradyarrhythmias while on treatment (23, 24). We did not distinguish between patients who presented with or without ST-segment elevations because this feature did not modify the effect of prasugrel or ticagrelor on the primary end point in either TRITON-TIMI 38 or PLATO (16, 18). Genotype-Guided Strategies We modeled the genotype-guided regimens on the basis of the recently published guidelines of the Clinical Pharmacogenetics Implementation Consortium (25) (Table 1 of the Supplement). In the 2 genotype-guided strategies, we assumed that carriers of 1 or 2 loss-of-function alleles would receive prasugrel (genotyping-with-prasugrel strategy) or ticagrelor (genotyping-with-ticagrelor strategy), whereas patients with 2 gain-of-function alleles, 1 gain-of-function allele and 1 wild-type allele, or 2 wild-type alleles would be treated with clopidogrel. Because 1 gain-of-function allele does not completely compensate for 1 loss-of-function allele (25), such persons would receive prasugrel or ticagrelor after genotyping. We did not evaluate strategies using tests of platelet reactivity or clopidogrel dose-escalation because their clinical relevance was unclear (26, 27). Supplement. Modeling Details and Supplementary Results The base case was a hypothetical cohort of 100000 patients aged 65 years with ACS who had PCI with 1 or more drug-eluting stents. All patients received dual antiplatelet therapy with 1 of the previously mentioned agents and aspirin for 12 months after the last PCI or MI and low-dose aspirin daily thereafter unless contraindicated. We assumed the societal perspective (28), considering all direct and induced medical costs and relevant clinical outcomes. Utilities and costs were assigned to each clinical event in 1-month cycles and discounted at 3% annually (29). We conducted extensive deterministic, probabilistic, and scenario-based sensitivity analyses to account for uncertainty in the input variables. We adhered to the recommendations of the Panel on Cost-Effectiveness in Health and Medicine (30). We reported results in 2011 U.S. dollars, quality-adjusted life-years (QALYs), and incremental cost-effectiveness ratios (ICERs) (30). For each ICER evaluation, the comparator was the strategy that produced the next-most QALYs, excluding strategies that cost more (strictly dominated) or had a greater ICER (dominated by extension). Because of the inherent challenges of indirect comparisons between the 2 drugs, we did tiered comparisons: We first compared the drug-only strategies (to distinguish the drug effect from the effect of genetic testing), then we examined the effect of genotyping on prasugrel and ticagrelor separately; finally, we did a global comparison across all 5 strategies. Where required, we applied a willingness-to-pay threshold of $50000 per QALY. Modeling was done using TreeAge Pro 2009 (TreeAge Software, Williamstown, Massachusetts) and Excel 2007 (Microsoft, Redmond, Washington), and statistical analyses were done using Stata, version 11 (StataCorp, College Station, Texas). Model Structure After the initial PCI, patients were at risk for stent thrombosis, nonfatal MI (unrelated to stent thrombosis), percutaneous or surgical revascularization, intracranial and extracranial bleeding, and death of cardiovascular and noncardiovascular causes (Figure 1 of the Supplement). Three additional states were modeled: Post-MI (patients who had an MI after entering the model had an increased risk for future MIs and cardiovascular death); intracranial bleed; and a steady state, into which all patients entered after a coronary artery bypass graft or 4 years after their initial PCI, whichever was sooner. The steady state accounted for age-specific medical costs and QALYs without tracking individual clinical events. Model Inputs Details can be found in the Appendix Table. For patients in the clopidogrel group, we estimated the incidence and management of major coronary events from trials (8, 16, 18, 19, 3137), observational data (4, 3854), U.S. life tables (55), U.S. Food and Drug Administration publications (56), Medicare claims data (57, 58), clinical guidelines (5, 5961), and other publications (48, 62). Event rates in the other groups were estimated using rate ratios relative to patients on clopidogrel (1619, 3335). Long-term survival of patients with ACS was estimated using Medicare claims data from 2002 to 2006 (Figure 2 of the Supplement) (57, 58). See the Supplement for additional information. Appendix Table. Summary of Key Model Variables We estimated the prevalence of loss-of-function polymorphisms from published studies (25, 34, 6365). Although some studies showed that loss-of-function carriers had a greater rate of thrombotic events than noncarriers when treated with clopidogrel (66), 2 recent reviews estimated different degrees of association between carrier states and thrombotic events. In a collaborative, random-effects model meta-analysis of 9 studies including 9685 patients (91% of whom had a PCI), Mega and colleagues (12) found that carriers of 1 or 2 CYP2C19 loss-of-function alleles had a hazard ratio of 1.57 for thrombotic events (95% CI, 1.13 to 2.16) relative to noncarriers. In a fixed-effects model meta-analysis of 42016 patients from 32 clopidogrel trials that were not limited to patients with PCI, Holmes and colleagues (67) found that carriers of loss-of-function alleles had a relative risk of 1.18 (CI, 1.09 to 1.28) for thrombotic events relative to noncarriers. In light of this uncertainty in the ability of loss-of-function alleles to discriminate between high- and low-risk patien

D ual antiplatelet therapy combining aspirin with a second agent is the mainstay of therapy after acute coronary syndrome (ACS), particularly among patients who receive a percutaneous coronary intervention (PCI) (1). Antiplatelet agents reduce thrombotic events, such as myocardial infarction (MI) and stent thrombosis, but increase risk for bleeding (2). Approximately one half of the 1.1 million ACS events in the United States every year are treated with a PCI, making the choice of antiplatelet therapy a common and important clinical decision (3,4).
Clopidogrel has been the standard of care after PCI for nearly a decade (5). Until recently, it was the secondlargest drug in terms of sales, and much of the $12 billion spent on it each year was for use after ACS (6). However, many patients receiving clopidogrel and aspirin have recurrent cardiovascular events (7,8), and on-treatment platelet inhibition varies considerably (9,10). Patients who carry a loss-of-function polymorphism of CYP2C19 (a key enzyme involved in the hepatic activation of clopidogrel) achieve less platelet inhibition with clopidogrel and have more thrombotic events (11)(12)(13) and less bleeding. However, carriers of gain-of-function alleles of the CYP2C19 enzyme achieve greater platelet inhibition with clopidogrel and have fewer thrombotic events and more bleeding (14,15).
Two new drugs, prasugrel and ticagrelor, are approved for use in patients having PCI for ACS (16 -19). The greater antiplatelet activity of these agents reduces the rate of MI and cardiovascular death compared with clopidogrel. However, prasugrel increases fatal bleeding so that its net effect on mortality rates is neutral (16,17). Ticagrelor is dosed twice daily and causes mild to moderate dyspnea in some patients (18,19), which may adversely affect adherence. Both agents are expensive, particularly when compared with generic formulations of clopidogrel that are now available.
Further, commercial availability of genetic testing may allow clinicians to personalize antiplatelet therapy so that the new, more expensive drugs could be selectively prescribed to patients most likely to benefit (11,12,20,21). These recent developments have altered the therapeutic landscape, highlighting the need for a comprehensive evaluation of alternative strategies for dual antiplatelet therapy. There are no head-to-head clinical trials of ticagrelor with prasugrel and no prospective studies of genotype-based treatment decisions. In this article, we present a simulation that addresses uncertainties about the role of genotyping and identifies the most cost-effective strategies for dual antiplatelet therapy after PCI for ACS.

METHODS
We developed a discrete-state Markov model to compare 5 strategies of dual antiplatelet therapy (22).

Drug-Only Strategies
Drug-only strategies were generic clopidogrel, prasugrel, or ticagrelor. We assumed that generic clopidogrel had the same efficacy as the proprietary formulation. On the basis of the results of TRITON-TIMI 38 (Trial to Assess Improvement in Therapeutic Outcomes by Optimizing Platelet Inhibition With Prasugrel-Thrombolysis in Myocardial Infarction), we assumed that prasugrel led to fewer cardiovascular deaths but more fatal bleeding compared with clopidogrel (16,17). On the basis of the PLATO (Platelet Inhibition and Patient Outcomes) study, we assumed that ticagrelor reduced cardiovascular deaths without a corresponding increase in fatal bleeding (18,19) and that some patients had dyspnea and bradyarrhythmias while on treatment (23,24). We did not distinguish between patients who presented with or without ST-segment elevations because this feature did not modify the effect of prasugrel or ticagrelor on the primary end point in either TRITON-TIMI 38 or PLATO (16,18).

Genotype-Guided Strategies
We modeled the genotype-guided regimens on the basis of the recently published guidelines of the Clinical Pharmacogenetics Implementation Consortium (25) (Table 1 of the Supplement, available at www.annals.org). In the 2 genotype-guided strategies, we assumed that carriers of 1 or 2 loss-of-function alleles would receive prasugrel (genotyping-with-prasugrel strategy) or ticagrelor (genotyping-with-ticagrelor strategy), whereas patients with 2 gain-of-function alleles, 1 gain-of-function allele and 1 wild-type allele, or 2 wild-type alleles would be treated with clopidogrel. Because 1 gain-of-function allele does not completely compensate for 1 loss-of-function allele (25), such persons would receive prasugrel or ticagrelor after genotyping. We did not evaluate strategies using tests of platelet reactivity or clopidogrel dose-escalation because their clinical relevance was unclear (26,27).
The base case was a hypothetical cohort of 100 000 patients aged 65 years with ACS who had PCI with 1 or more drug-eluting stents. All patients received dual antiplatelet therapy with 1 of the previously mentioned agents and aspirin for 12 months after the last PCI or MI and low-dose aspirin daily thereafter unless contraindicated. We assumed the societal perspective (28), considering all direct and induced medical costs and relevant clinical outcomes. Utilities and costs were assigned to each clinical event in 1-month cycles and discounted at 3% annually (29). We conducted extensive deterministic, probabilistic, and scenario-based sensitivity analyses to account for uncertainty in the input variables. We adhered to the recommendations of the Panel on Cost-Effectiveness in Health and Medicine (30).
We reported results in 2011 U.S. dollars, qualityadjusted life-years (QALYs), and incremental costeffectiveness ratios (ICERs) (30). For each ICER evaluation, the comparator was the strategy that produced the next-most QALYs, excluding strategies that cost more (strictly dominated) or had a greater ICER (dominated by extension). Because of the inherent challenges of indirect comparisons between the 2 drugs, we did tiered comparisons: We first compared the drug-only strategies (to distinguish the drug effect from the effect of genetic testing), then we examined the effect of genotyping on prasugrel and ticagrelor separately; finally, we did a global comparison across all 5 strategies. Where required, we applied a willingness-to-pay threshold of $50 000 per QALY.

Context
Several options for antiplatelet therapy after percutaneous coronary intervention for acute coronary syndrome are available.

Contribution
This cost-effectiveness analysis compared drug-only strategies (generic clopidogrel, prasugrel, or ticagrelor) and genotype-guided strategies targeting ticagrelor or prasugrel. Ticagrelor was the most cost-effective strategy. The genotyping-with-prasugrel strategy was superior to giving all patients prasugrel. The genotyping-with-ticagrelor strategy was clinically superior but more expensive than clopidogrel.

Caution
No randomized trials have directly compared genotyping strategies or prasugrel with ticagrelor.

Implication
Genotype-guided personalization of antiplatelet therapy could improve cost-effectiveness in some situations, but ticagrelor for all without genotyping also seems reasonable.

-The Editors
Original Research Antiplatelet Therapy for Acute Coronary Syndrome ses were done using Stata, version 11 (StataCorp, College Station, Texas).

Model Structure
After the initial PCI, patients were at risk for stent thrombosis, nonfatal MI (unrelated to stent thrombosis), percutaneous or surgical revascularization, intracranial and extracranial bleeding, and death of cardiovascular and noncardiovascular causes (Figure 1 of the Supplement). Three additional states were modeled: Post-MI (patients who had an MI after entering the model had an increased risk for future MIs and cardiovascular death); intracranial bleed; and a "steady state," into which all patients entered after a coronary artery bypass graft or 4 years after their initial PCI, whichever was sooner. The steady state accounted for age-specific medical costs and QALYs without tracking individual clinical events.
We estimated the prevalence of loss-of-function polymorphisms from published studies (25,34,(63)(64)(65). Although some studies showed that loss-of-function carriers had a greater rate of thrombotic events than noncarriers when treated with clopidogrel (66), 2 recent reviews estimated different degrees of association between carrier states and thrombotic events. In a collaborative, random-effects model meta-analysis of 9 studies including 9685 patients (91% of whom had a PCI), Mega and colleagues (12) found that carriers of 1 or 2 CYP2C19 loss-of-function alleles had a hazard ratio of 1.57 for thrombotic events (95% CI, 1.13 to 2.16) relative to noncarriers. In a fixedeffects model meta-analysis of 42 016 patients from 32 clopidogrel trials that were not limited to patients with PCI, Holmes and colleagues (67) found that carriers of loss-of-function alleles had a relative risk of 1.18 (CI, 1.09 to 1.28) for thrombotic events relative to noncarriers. In light of this uncertainty in the ability of loss-of-function alleles to discriminate between high-and low-risk patients, we modeled 2 scenarios (66). In the base-case or lowdiscrimination scenario, we modeled conservative correlations as seen by Holmes and colleagues among all patients treated with clopidogrel, including patients who had not had PCI (67). In a sensitivity analysis, we modeled a highdiscrimination scenario on the basis of the associations seen by Mega and colleagues (12) in the cohort of patients treated with clopidogrel after PCI. In both cases, we assumed that carriers of loss-of-function alleles had a lower risk for bleeding than noncarriers (67).
Carriers of gain-of-function alleles achieved a greater degree of platelet inhibition than patients with wild-type alleles treated with clopidogrel, which translated into fewer thrombotic events and increased bleeding (14,15). Because some evaluations suggested that this correlation may be partly due to linkage disequilibrium with loss-offunction alleles, we conducted a sensitivity analysis that assumed no correlation between gain-of-function alleles and outcomes (25). We assumed that genotyping was 100% sensitive and 99.3% specific in detecting CYP2C19 alleles (21) but varied these assumptions in sensitivity analyses. The pharmacologic effects of ticagrelor and prasugrel are unaffected by genotype (34,35,68,69), so the model assumed that carriers and noncarriers have similar outcomes when treated with 1 of these drugs.

Quality-of-Life Estimates
We estimated age-specific quality of life (70), which we also adjusted for adverse clinical events or invasive procedures (71)(72)(73)(74)(75)(76). We assumed that patients who had an MI or stent thrombosis had a 12% permanent quality-of-life decrement relative to their age-matched counterparts (77), patients who had a nonfatal intracranial hemorrhage had a 61% permanent quality-of-life decrement (78), and patients with ticagrelor-associated dyspnea had a quality-oflife decrement equal to that of patients with a history of angina (79).

Costs
We included direct medical costs (such as inpatient admissions, procedures, outpatient visits, and drugs) and induced costs (such as cost of procedural complications) but not indirect costs (such as lost wages and caregiver costs). We estimated acute event costs from Medicare reimbursement rates, the Nationwide Inpatient Sample, and the published literature (74, 80 -82). We estimated agespecific costs of outpatient and total medical care from the Agency for Healthcare Research and Quality's Medical Expenditure Panel Survey (83). All costs were converted to 2011 dollars using the U.S. gross domestic product deflator (84).
We assumed a base-case cost of $30 per month for generic clopidogrel and included the current average wholesale price of the proprietary formulation ($218 per month) in the sensitivity analyses (82). We assumed the costs of prasugrel and ticagrelor to equal their average wholesale price ($220 and $261 per month for prasugrel and ticagrelor, respectively) (82). We estimated the cost of genotyping from a survey of retail prices of commercially available tests but included the estimated unit cost of point-of-care tests in the range tested in sensitivity analyses.

Role of the Funding Source
This study was funded by the American Heart Association, U.S. Department of Veterans Affairs, Stanford University, and the University of California San Francisco. The funding source had no role in the design, conduct, or interpretation of the study or the preparation, review, or approval of the manuscript.  Table 1).

Drug-Only Strategies
Both prasugrel and ticagrelor reduced thrombotic events relative to clopidogrel, but patients receiving prasugrel had substantially greater fatal bleeding ( Table 2 of the Supplement). As a result, prasugrel was relatively expensive, with an ICER of $124 400 per QALY relative to clopidogrel, whereas ticagrelor had a lower ICER of $22 800 per QALY relative to prasugrel. Thus, prasugrel was eliminated by extended dominance, and ticagrelor had an ICER of $40 300 per QALY relative to clopidogrel.

All Strategies
In the base case, we assumed that loss-of-function alleles were only modestly correlated with thrombotic out- † Incremental cost-effectiveness for each strategy was measured relative to the next-best strategy that had not been eliminated by dominance and was rounded to the closest $100 to reflect the precision in the model. ‡ Proportion of patients who die of a cardiovascular cause or fatal bleed in the first 4 y after index percutaneous coronary intervention. § The ICER of prasugrel relative to clopidogrel ($124 400/QALY) was greater than the ICER of ticagrelor, relative to prasugrel ($22 800/QALY). Prasugrel was therefore eliminated from the comparison by the principle of extended dominance, and ticagrelor was compared directly with clopidogrel.
In the genotyping with prasugrel strategy, carriers of 1 or 2 loss-of-function polymorphisms in CYP2C19 were treated with prasugrel; the others received generic clopidogrel. ¶ In the low-discrimination scenario, prasugrel cost $1076 more than the genotyping-with-prasugrel strategy and produced 0.015 fewer QALYs. It was therefore eliminated from the evaluation (strictly dominated), and genotyping with prasugrel was compared with clopidogrel. ** In the low-discrimination scenario, the ICER of genotyping with prasugrel relative to clopidogrel ($35 800/QALY) was less favorable than the ICER of genotyping with ticagrelor relative to genotyping with prasugrel ($22 800/QALY). Therefore, genotyping with prasugrel was eliminated from the comparison by the principle of extended dominance, and genotyping with ticagrelor was compared directly with clopidogrel. † † In the high-discrimination scenario, prasugrel cost $727 more than genotyping with prasugrel and produced 0.042 fewer QALYs. Prasugrel was therefore eliminated from the evaluation (strictly dominated), and genotyping with prasugrel was compared with clopidogrel. ‡ ‡ In the high-discrimination scenario, the ICER of genotyping with prasugrel relative to clopidogrel ($25 600/QALY) was less favorable than the ICER of genotyping with ticagrelor relative to genotyping with prasugrel ($22 800/QALY). Therefore, genotyping with prasugrel was eliminated from the comparison by the principle of extended dominance, and genotyping with ticagrelor was compared directly with clopidogrel. § § In the genotyping-with-ticagrelor strategy, carriers of 1 or 2 loss-of-function polymorphisms in CYP2C19 were treated with ticagrelor; the others received generic clopidogrel.
Original Research Antiplatelet Therapy for Acute Coronary Syndrome comes (12). When all 5 strategies were considered in order of increasing QALYs gained and compared incrementally ( Table 1 and Figure 1), the prasugrel strategy was more expensive and had worse outcomes than genotyping with prasugrel and was therefore eliminated ("dominated"). Next, the ICER for genotyping with prasugrel relative to clopidogrel ($35 800 per QALY) was less favorable than the ICER for genotyping with ticagrelor relative to genotyping with prasugrel ($22 800 per QALY); genotyping with prasugrel was therefore inside the "cost-effectiveness frontier" and was eliminated (Figure 1). Genotyping with ticagrelor was therefore compared directly with clopidogrel (the next-best, nondominated strategy) and yielded an ICER of $30 200 per QALY. The ticagrelor-for-all strategy produced the highest QALYs but was also the most expensive with a less favorable ICER ($52 600 per QALY relative to genotyping with ticagrelor).

High-Discrimination Scenario
Assuming stronger associations between loss-offunction genotype and thrombotic outcomes greatly increased the cost-effectiveness of genotyping-based strategies ( Table 1 and Figure 1) (12). In this setting, genotyping with ticagrelor was the most cost-effective strategy, with an ICER of $24 700 per QALY. Treating all patients with ticagrelor produced 0.02 additional QALYs but was eco-nomically unattractive, with an ICER of $104 800 per QALY relative to genotyping with ticagrelor.

Efficacy and Safety Variables
The cost-effectiveness of genotyping with ticagrelor was sensitive to modest changes in assumptions about the efficacy and safety of ticagrelor relative to clopidogrel and the association between thrombotic events in loss-offunction carriers relative to noncarriers ( Table 3 of the Supplement and Appendix Figures 1 and 2, available at www.annals.org). For instance, if the rate of cardiovascular death among patients treated on ticagrelor decreased by 1.3% or the rate of fatal bleeding by 38.0%, treating all patients with ticagrelor became the most cost-effective strategy. In contrast, in the high-discrimination scenario, the optimal strategy-genotyping with ticagrelor-was robust to wide variations in underlying assumptions ( Table 3 of the Supplement).

Ticagrelor-Associated Dyspnea
The choice of optimal therapy was sensitive to the decrement in the patient's quality of life from ticagrelorassociated dyspnea (Figure 3 of the Supplement). A utility decrement of greater than 0.049 (Ն6% of baseline quality of life at the age of 65 years) made genotyping with prasugrel the most cost-effective therapy.

Allelic Frequency
The population frequency of loss-of-function alleles varied substantially by race and ethnicity, and it was considerably greater in South Asia (35%), East Asia (40%), and Oceania (76%) than in Europe (16%) or Africa (16%) (25). As the proportion of carriers 1 or 2 loss-of-function alleles increased, both genotyping with ticagrelor and ticagrelor became increasingly cost-effective (Appendix Figure 3, available at www.annals.org). Treating all patients with ticagrelor was the most cost-effective therapy when loss-of-function carriers constituted more than 52.7% of the population. In contrast, increasing population frequency of the gain-of-function allele did not materially affect the cost-effectiveness of genotyping but made the ticagrelor-for-all strategy less cost-effective (Figure 4 of the Supplement).

Accuracy of Genetic Testing
The ICER for ticagrelor relative to genotyping with ticagrelor was affected by the accuracy of genotyping (Figure 5 of the Supplement), and declining accuracy favored treating all patients with ticagrelor independent of genotype. For instance, if the sensitivity and specificity of the test were 95% (instead of the base case of 100% sensitivity and 99.3% specificity), the ICER for ticagrelor would decrease to $51 500 per QALY and the ICER for genotyping with ticagrelor would increase to $31 500 per QALY.

Cost of Genetic Testing
In the low-discrimination scenario, it was costeffective to treat all patients with ticagrelor regardless of genotype if genetic testing cost more than $358 per patient ( Figure 6 of the Supplement). In the high-discrimination scenario, genotyping with ticagrelor was the most costeffective strategy until the cost of genetic testing exceeded $1355.

Drug Costs
The choice of optimal antiplatelet therapy was sensitive to the difference in the monthly cost of ticagrelor and clopidogrel: Smaller differences in cost made both ticagrelor and genotyping with ticagrelor more cost-effective (Figure 7 of the Supplement). In the low-discrimination scenario, treating all patients with ticagrelor was the most cost-effective strategy when the difference in monthly cost of ticagrelor and clopidogrel decreased from a base case of $231 to $215, either because ticagrelor was less expensive or clopidogrel was more expensive than the base case. In the high-discrimination scenario, the difference had to decrease to $93 or less to make ticagrelor cost-effective at a threshold of $50 000 per QALY.

Duration of Dual Antiplatelet Therapy
The absolute cardiovascular risk was greatest in the first year after PCI, whereas bleeding risk and drug costs persisted for the entire duration of antiplatelet therapy. Therefore, dual antiplatelet therapy became less economically attractive as the duration of treatment increased from 12 to 36 months. The genotyping-with-ticagrelor strategy remained the most cost-effective alternative for dual antiplatelet therapy after PCI for ACS, with an ICER less than $50 000 per QALY (Figure 8 of the Supplement).

Probabilistic Sensitivity Analysis
We performed 10 000 microsimulations where all input variables were varied simultaneously along prespecified distributions. In the low-discrimination scenario, genotyping with ticagrelor was the preferred strategy in 39% of the simulations and ticagrelor in 42% of the simulations (Figure 2). In the high-discrimination scenario, the preferred strategy was genotyping with ticagrelor in 63% of the simulations, ticagrelor in 19%, and genotyping with prasugrel in 13% (Figure 2). Ticagrelor was the preferred strategy in more than 50% of simulations at thresholds greater than $54 500 per QALY in the low-discrimination scenario and $98 000 per QALY in the high-discrimination scenario.

Scenario Analyses
The optimal strategies for dual antiplatelet therapy under different clinical scenarios in which ticagrelor or prasugrel may not be indicated (for example, among patients with a history of a transient ischemic attack) are presented in Table 2. Additional sensitivity analyses are presented in Tables 4 to 7 and Figures 9 to 13 of the Supplement.

DISCUSSION
Nearly 500 000 patients in the United States face the choice of dual antiplatelet therapy after PCI for ACS every year. This choice has substantial clinical and economic implications and entails a marked difference in drug costs as well as a tradeoff between thrombotic events and major bleeding. Our analysis suggests that genotype-guided personalization of therapy may improve the cost-effectiveness of the newer, more expensive antiplatelet agents. The targeted use of prasugrel in carriers of CYP2C19 loss-offunction alleles consistently decreased costs and improved outcomes relative to treating all patients with prasugrel, making genotyping before treatment with prasugrel the clinically and economically superior strategy. The selective use of ticagrelor in CYP2C19 loss-of-function carriers and clopidogrel in noncarriers was the most cost-effective strategy when genotyping discriminates well between patients at high and low risk for thrombotic events (that is, where there is a strong association between genotype and clinical outcomes). If genotype were only modestly predictive of thrombotic outcomes, ticagrelor for all patients independent of genotype would be the most cost-effective strategy for dual antiplatelet therapy after PCI for ACS.
Genotype-guided therapy aims to reduce costs and improve outcomes by targeting the use of the more expensive Original Research Antiplatelet Therapy for Acute Coronary Syndrome drugs to patients most likely to benefit from them. Contrary to concerns that cost-effectiveness considerations encourage a "1-size-fits-all" approach (85) or "stymie progress in personalized medicine" (86), models such as ours that estimate clinically meaningful outcomes among genetic subgroups of patients can help clarify the potential value of individualized therapeutics (87). Further, sensitivity analyses help quantify the effect of uncertainty on clinical and policy-level decision making and identify the knowledge gaps that should be addressed in future research. Our study highlights that a well-designed cost-effectiveness analysis can both support and guide innovation in personalized medicine (87).
Our results suggest 4 key considerations in the choice of antiplatelet therapy after PCI for ACS. First, clinicians should consider genotyping all patients before using prasugrel because targeted use of prasugrel therapy among lossof-function carriers seems to reduce costs and improve clinical outcomes. Second, clinicians should be cognizant of the effect of ticagrelor-associated dyspnea on the pa- Results of the probabilistic sensitivity analysis are illustrated as acceptability curves, which plot the proportion of simulations in which a certain strategy is "optimal" (or most cost-effective) against the amount one is willing to pay per QALY gained. In the low-discrimination scenario, genotyping with ticagrelor is the preferred strategy in 42.3% of the simulations at a willingness-to-pay threshold of $50 000/QALY (green vertical line) and ticagrelor is the preferred strategy in 32% of the simulations, reflecting the underlying uncertainty. Greater thresholds make ticagrelor more economically attractive.
In the high-discrimination scenario, which assumes stronger associations between loss-of-function genotype and the rate of thrombotic events, genotyping with ticagrelor is the optimal strategy in 63.4% of the simulations at a threshold of $50 000/QALY. QALY ϭ quality-adjusted life-year.

Original Research
Antiplatelet tient's quality of life. Among patients with a moderate to severe decrement in quality of life due to ticagrelorassociated dyspnea (Ն6% reduction in on-treatment quality of life relative to baseline), genotyping with prasugrel is the most cost-effective strategy (that is, loss-of-function carriers should receive prasugrel, and noncarriers should receive clopidogrel). Third, genotype-guided antiplatelet therapy may be less attractive in populations or regions with a high prevalence of loss-of-function alleles, where treating all patients with ticagrelor may be the most costeffective strategy. Future research should specifically examine the role of genotyping among patients with ancestry in South and East Asia and Oceania, in whom the population frequency of loss-of-function alleles is considerably greater than among patients with European, American, or African ancestry (25). Fourth, genotyping may be less economically attractive in health care markets where the monthly cost of ticagrelor is closer to the monthly cost of generic clopidogrel because the cost-effectiveness of genotyping is sensitive to cost differences between the drugs. Treating all patients with ticagrelor independent of genotype becomes the most cost-effective strategy when the difference in monthly cost of ticagrelor and clopidogrel is less than $215. There are several limitations to this study. Estimated differences in outcomes between various CYP2C19 genotypes are largely based on post hoc analyses of randomized trials. Systematic reviews of the literature have yielded variable results depending upon studies included, definition of end points, and statistical models used. We address this uncertainty by presenting both a lowdiscrimination scenario that assumes a modest ability to discriminate between high-and low-risk patients on the basis of genotype, as well as a high-discrimination scenario, which assumes a stronger association between genotype and thrombotic outcomes. Future randomized trials of genotype-tailored strategies, either alone or in combination with phenotype-based strategies (for example, based on the measurement of on-treatment platelet reactivity), should help further clarify the role of personalization in optimizing antiplatelet therapy after PCI (88).
Estimates of the efficacy and safety of prasugrel and ticagrelor are based on only 1 large, randomized, clinical trial of each drug versus clopidogrel (16,18). The indirect comparison of ticagrelor with prasugrel inherent in the structure of the model is limited by structural differences in the design and execution of the PLATO and TRITON-TIMI 38 clinical trials, as well as any clinical differences in the patients enrolled in these trials. Although a definitive, large, randomized, clinical trial comparing various strategies for dual antiplatelet therapy among real-world patients would be ideal, the prohibitive logistics of such a trial argue for comparative effectiveness studies of ticagrelor and prasugrel on the basis of a large, observational study or pragmatic clinical trial. Until either is done, models such as ours that incorporate a wide range of sensitivity analyses facilitate a systematic synthesis of published data. To alleviate confounding arising from interstudy variations, we used data from previously published trials and observational analyses to model baseline event rates in the clopidogrel group and used rate ratios from TRITON-TIMI 38 and PLATO to model event rates among patients on prasugrel and ticagrelor, respectively.
A post hoc analysis of patients receiving prasugrel in TRITON-TIMI 38 (16) found an increase in bleeding among patients with a history of stroke or transient ischemic attack; prasugrel is contraindicated in this subgroup. It is possible that prasugrel compares more favorably with clopidogrel in patients without a history of stroke or transient ischemic attack than indicated by the full-trial estimates used in this model. A subgroup analysis of patients receiving ticagrelor in PLATO found that patients recruited in North America had worse outcomes than patients from other geographic regions. This may represent a chance finding, a dose-dependent interaction with aspirin, or a real discrepancy arising from international differences in treatment algorithms (89). In line with the U.S. Food and Drug Administration approval of the drug, we as- † Ticagrelor is contraindicated in patients with a history of hemorrhagic stroke. In a small group of patients, ticagrelor produces a syndrome of subjective dyspnea that may last several months. In some patients, this may be severe enough to result in discontinuation of the medication. ‡ In the genotyping-with-prasugrel strategy, patients with 1 or 2 loss-of-function polymorphisms in CYP2C19 were treated with prasugrel; the others received clopidogrel. § Prasugrel is contraindicated in patients with a history of stroke or transient ischemic attack. Caution is also advised for patients weighing less than 60 kg and those who are aged Ն75 y.
In the genotyping-with-ticagrelor strategy, patients with 1 or 2 loss-of-function polymorphisms in CYP2C19 were treated with ticagrelor; the others received clopidogrel. If a threshold of $50 000 per quality-adjusted life-year was assumed, then ticagrelor was the most cost-effective therapy if the monthly price difference between ticagrelor and clopidogrel was less than $215/mo (low-discrimination scenario) or $93/mo (high-discrimination scenario).
Original Research Antiplatelet Therapy for Acute Coronary Syndrome sumed that the clinical outcomes seen in PLATO can be achieved in U.S. patients on low-dose aspirin therapy, but this will need to be confirmed in future studies. Short-term clinical trials may not adequately define all potential safety concerns with a drug (23,90,91). Pursuant to the "lifecycle approach" to drug safety recommended by the Institute of Medicine (92), our analysis should be updated when safety and efficacy data from phase 4 trials or registries become available.
Our results are broadly concordant with previously published analyses that have found genotyping-based personalization of antiplatelet therapy to be cost-effective in other health systems (93,94) but are more conservative than those reported by the trialists themselves (95)(96)(97)(98). For instance, the investigators of TRITON-TIMI 38 concluded that treating patients with prasugrel after PCI for ACS was cost-effective at $50 000 per life-year gained, largely because of a substantially greater gain in life expectancy with prasugrel treatment in their model than seen in our analysis (96). This is probably the result of key methodological differences between the 2 studies-for instance, the trialists estimated life expectancy from a data set of patients who underwent angioplasty in Saskatchewan, Canada, between 1985 and 1995 (before the widespread adoption of intracoronary stenting) (96), whereas we based our estimate on U.S. Medicare beneficiaries who had a PCI for ACS between 2002 and 2005. Nevertheless, the results of our sensitivity analyses underscore the need to accurately define the long-term effect of newer antiplatelet agents on mortality rates, which would define their relative costeffectiveness in the real world.
Based on currently available evidence, genotyping patients having PCI for ACS, followed by the targeted use of ticagrelor in carriers of loss-of-function CYP2C19 alleles and clopidogrel in noncarriers is economically attractive compared with treating all patients with the newer agents or clopidogrel. However, ticagrelor for all patients independent of genotype may be an economically reasonable alternative in some populations and settings. Future studies should directly compare prasugrel with ticagrelor, assess the effect of ticagrelor-associated dyspnea on quality of life, and prospectively establish the role of personalization of antiplatelet therapy after PCI for ACS.

Ticagrelor
In 2-way sensitivity analyses, we simultaneously varied the rate of cardiovascular death and fatal bleeding among patients receiving ticagrelor (relative to patients receiving clopidogrel), holding constant the event rates among patients receiving prasugrel. In the low-discrimination scenario and at a willingness-to-pay threshold of $50 000/quality-adjusted life-year, genotyping with ticagrelor was the most cost-effective strategy at baseline (dotted lines), but relatively small improvements in the efficacy or safety of ticagrelor (e.g., 1.3% decrease in cardiovascular mortality rates) made treating all patients with ticagrelor the most cost-effective option. In the high-discrimination scenario, genotyping with ticagrelor was robust to large changes in the efficacy and safety of ticagrelor. CV ϭ cardiovascular; HR ϭ hazard ratio.