The long-term cost-effectiveness of clopidogrel plus aspirin in patients undergoing percutaneous coronary intervention in Sweden

Clinical T h e r a p e u t i c s / V o l u m e

27, N u m b e r 1, 2 0 0 5

The Long-Term Cost-Effectiveness of Clopidogrel Plus Aspirin in Patients Undergoing Percutaneous Coronary Intervention in Sweden Peter Lindgren, MSc, 1'2 UIfStenestrand, MD, PhD, 3 Klas Malmberg, MD, PhD, 4 and BengtJtnsson, P h D s

~Department of Cardiovascufar Epidemiofogy, Institute of Environmentaf Medicine, Karofinska Institute, Stockholm, 2Stockhofm Health Economics, Stockholm, 3Department of Cardiofogy, UniversityHospital of LinkOping~ LinkOping, 4Department of Cardiofogy, Karofinska Hospital, Stockholm, and SCentre for Health Economics, Stockholm School of Economics, Stockholm, Sweden ABSTRACT Background:

The Percutaneous C o r o n a r y Intervention-Clopidogrel in Unstable Angina to Prevent Recurrent Events (PCI-CURE) study, which examined the effect of adding clopidogrel to aspirin versus aspirin alone in patients with unstable coronary artery disease (CAD) undergoing PCI, found a relative risk reduction in cardiovascular deaths and myocardial infarction among those treated with clopidogrel. In addition, a within-trial cost-effectiveness analysis showed favorable costs per event avoided. However, to estimate the long-term effects, a modeling approach is necessary. Objectives: The purpose of this study was to estimate the long-term cost-effectiveness of treating patients undergoing PCI with clopidogrel plus aspirin in Sweden. Methods: A Markov model was developed. Transition probabilities were estimated based on a register of patients treated in the coronary care units at 74 (out of 78) hospitals throughout Sweden. Patients were assumed to be treated for 1 year with an effect based on data from the PCI-CURE study. Costs were collected from published sources and recalculated to year-2004 euros (@1.00 = US $1.24). Life-years gained were used as the measure of effectiveness. The perspective was that of the Swedish society, with a separate analysis using a health care cost perspective. Results: After inclusion and exclusion criteria were applied, 3474 patients were included in the model analysis. The model predicted a net gain in survival of 0.04 year per patient when adding clopidogrel. This yielded a net increase of •449 if only direct costs were included; with indirect costs, the net increase was £332. 100

The resulting cost-effectiveness ratios were £10,993 and £8127 per life-year gained. Conclusions: The predicted cost-effectiveness ratios were well below the threshold values generally considered cost-effective. Adding clopidogrel to aspirin appeared to be cost-effective in this model analysis of patients with unstable CAD undergoing PCI in Sweden. (Cti~z The~ 2005;27:100-110) Copyright © 2005 Excerpta Medica, Inc. Key words: clopidogrel, decision analysis, economic evaluation, Sweden.

INTRODUCTION

The Clopidogrel in Unstable Angina to Prevent Recurrent Events (CURE) trial investigated the use of dopidogrel in addition to aspirin versus aspirin alone in patients with acute coronary syndromes, defined as unstable angina or non-ST-segment elevation myocardial infarction (MI). The intervention was shown to be both welltolerated and effective,1 as well as cost-effective.2,3 The Percntaneous Coronary Intervention (PCI)-CURE study was a substudy investigating the outcomes of patients undergoing PCIs. 4 Patients who had undergone a revascularization procedure within 3 months or received treatment with glycoprotein IIb/IIIa inhibitors 3 days before randomization were excluded, as were patients contraindicated to antithrombotic

Accepted for pubfication November22, 2004. doi: 10.1016/j.clinthera.2005.01.008 0149-2918/05/$19.00

Printed in the USA. Reproduction in whole or part is not permitted. Copyright © 2005 Excerpta Medica, Inc. Volume 27, N u m b e r 1

P. Lindgren et al.

therapy or those suffering from class IV heart failure. Thirty percent of the patients included in the PCI-CURE study were women, and the mean age was 61 years. Twenty-six percent of patients had suffered a myocardial infarction (MI) previously, and 13% had undergone a revascularization procedure. In the PCI-CURE study, patients were pretreated with aspirin and study drug for a median of 10 days. After PCI, patients in both the active and placebo groups were allowed to receive open-label thienopyridines (ie, clopidogrel and ticlopidine) for a median of 4 weeks. After this period, patients in the active group received treatment for an additional 8 months (mean, 11 months maximum). As many as 1313 patients in the active group and 1345 in the placebo group had -30 days of follow-up, making them eligible for the long-term analysis. Using propensity scores to adjust for covariates that influence the likelihood of undergoing a PCI, long-term treatment (preceded by pretreatment) with clopidogrel showed a relative risk of cardiovascular death and MI of 0.72 (95% CI, 0.53-0.96; P = 0.03). The trial showed no increase in major bleedings or blood transfusions needed for the patients treated with clopidogrel, although there was a slight increase in the number of minor bleeding incidents (P = 0 . 0 3 ) . 4 A within-trial cost-effectiveness analysis of the PCI-CURE study was performed previously, showing favorable costs per event avoided, s However, it is generally accepted that the cost and effects of MI span a longer time frame than can be captured within a trial. To estimate the total long-term consequences on costs, survival, and quality of life, a modeling approach is necessary. The purpose of this study was to estimate the long-term cost-effectiveness of adding clopidogrel to aspirin, using register data for Swedish patients undergoing PCI, as well as data from the PCI-CURE study and other published research. PATIENTS A N D M E T H O D S A computer simulation model was developed, incorporating data from the Swedish Register of Information and Knowledge about Swedish Heart Intensive care Admissions (RIKS-HIA) register on coronary care patients to estimate the risk of an event (ie, cardiovascular death or nonfatal MI) and mortality in a Swedish population, published data on cost of events, and data from the PCI-CURE study to estimate the cost of treatment. The perspective of the analysis was that of the Swedish society. January2005

The model constructed was a Markov state transition model, e>* The model, presented in Figure 1, consisted of 4 states: after PCI (the starting state for all patients), first year after an MI, the second and subsequent years after an MI, and death. Each of the states was associated with defined costs. The model used yearly cycles: each year, patients had a specified probability of moving from their present state to one of the other states in the model or of remaining in the same state. Patients in the post-PCI state had a risk of suffering either a nonfatal MI (which would cause them to move to the first-year-after-an-MI state) or fatal cardiovascular disease (which would cause them to move to the death state). They could also die from other causes. If none of this occurred, they remained in the post-PCI state, where they had a risk of suffering an event in the subsequent cycle. Patients who survived a cycle in the state for the first year after an MI moved to the state for second and subsequent years, where they remained until death. The model was run using the lifetime perspective (ie, until all patients reached death). The model was developed in TreeAge Pro 2004 (TreeAge Software, Inc., Williamstown, Massachusetts). The model was evaluated using second-order Monte Carlo simulation. 9 This means that in each simulation, each parameter (ie, cost or transition probability) in the model was drawn from its underlying distribution. A large number of simulations (ie, 1000) were performed to estimate the mean cost and effect. Uncertainty was reported in the form of a costeffectiveness acceptability curve. I° The distributions used and the assumptions around them are summarized in Table I and are discussed herein. The principal data source for use in the model was RIKS-HIA, a national register that currently covers 74 of the 78 hospitals located throughout Sweden. All patients admitted to the coronary care units at the participating hospitals are included in the register. Data for -100 variables are collected, covering background factors such as age, sex, risk factors, diagnostic measures, and treatments administered (both pharmacologic and nonpharmacologic). The register is continuously merged with the national Patient Administration Register and the National Cause of Death Register. 11 We extracted data from the register for patients who fulfilled the inclusion and exclusion criteria of the PCI-CURE study. Data for patients undergoing a PCI between January 1, 1995, and August 1, 2001 101

Clinical Therapeutics

-~ After PCI

After PCI 1

M I year 1

C

O

Event

O

Other mortality

<~ Death

OSf

0 k

~

<~] MI subsequent years <~ Death

MI subsequent O f f years \ 0

Death

CVdeath <:~ Death

<~ MI subsequent years <~ Death

<1

Figure 1. The Markov model used in the analysis.@ = Markov node; PCI = percutaneous coronary intervention; M I = myocardial infarction; CV = cardiovascular.

Table I. Model inputs. Costs are given as year-2004 euros, e

Model Parameter

Value Used

Distribution

Risk of event

Logistic model

Bootstrapf

Fraction of events being CV death

Logistic model

Bootstrapf

Mortality after MI

Observed rates

Normal

Mortality due to other causes

Logistic model

Bootstrapt

Risk reduction from treatment

0.72 (95% CI, 0.53-0.96)

Normal

Direct costs during first year after MI

571 5 (SD, 7448)

Bootstrap t

Indirect costs during first year after MI

1 2,496 (SD, 11,363)

Bootstrapf

Direct costs during second and subsequent years after MI

835 (SD, 835)

Normal

Indirect costs during second and subsequent years after MI

6565 (SD, 6565)

Normal

Cost of study drug

698

Fixed

Cost reduction during initial hospitalization

1 23

Bootstrapf

CV = cardiovascular; MI = myocardial infarction. ~Conversion rate: ~gl ÷00 = US $1.24. fThe parameter was drawn from 1 of 1000 bootstrap estimates.

102

Volume 27~ Number 1

P. Lindgren et al.

(the same cut-off date as the publication of the CURE trial, to avoid including patients who already received long-term treatment), were extracted from the register. Because all patients in the CURE trial received open-label aspirin, this criterion also had to be fulfilled. Patients included in our model analysis also needed to have a diagnosis of non-ST-elevation MI, unstable angina pectoris, or hospitalization for angina pectoris with ST-segment depression and/or elevation of infarction markers (troponin-T >0.05 gg/L or creatine kinase MB protein concentration :-5 gg/L). Patients undergoing thrombolytic therapy, receiving oral anticoagulants, or suffering from class IV heart failure were excluded, which matched the exclusion criteria of the trial. The trial excluded patients undergoing PCI or coronary artery bypass grafting (CABG) 3 months before the current procedure. Because we could not establish the exact date of previous procedures; we excluded all patients with a previous PCI or CABG, which may have led to a potential underestimation of the risks. With regard to events, patients could be followed until December 31, 2001. One additional year was available for mortality statistics, meaning that with regard to mortality, the data could be analyzed until December 31, 2002. To estimate which risk factors must be considered when predicting probabilities, we performed a Cox proportional hazards analysis including the potential risk factors: age, male sex, diabetes mellitus, previous MI, current smoking status, and whether the patient was receiving treatment for hypertension at admission. The aim of including covariates was to allow us to perform subgroup analyses. The significant risk factors were included in logistic regressions to estimate the yearly risk of an event (or death after an event) to be used in the M a r k o v model. An event was equally defined as in the PCI-CURE study (ie, as the occurrence of a new MI or death from cardiovascular

causes). An MI within 7 days of the index admission was excluded to avoid including the incident that created the need for PCI and to avoid dilution with procedurerelated infarctions. This might have led to an underestimation of the true risk. However, this possibility was tested in a sensitivity analysis. The risk of an event was highest during the first month after an intervention, and decreased gradually with time. After 1 year; the rate was fairly constant. Because the model used

January2005

yearly cycles, we simulated this by estimating 2 risk functions: one for the first year and one for subsequent years. This implied that we assumed there were no time-dependent factors (apart from those included in the regression model, such as age) influencing the risk, other than higher risk during the first year. To decide whether an event was death from cardiovascular causes or a nonfatal MI, a logistic regression predicting the risk of the event being fatal was used. This regression used the same covariates as the regression used in predicting the probability of an event. Too few deaths were available to estimate the survival after an event stratified by subgroups. For this reason, we used the observed risk of death during the first and subsequent years after the event (8.1% during the first year and 3.8 % during subsequent years). The risk of death from other causes was estimated using a single logistic regression. Long-term treatment with clopidogrel plus aspirin led to a relative risk of suffering cardiovascular death or nonfatal MI of 0.72 (95% CI, 0.53-0.96) versus aspirin alone. 4 This reduction was applied to the clopidogrel plus aspirin study arm during the first year, and we assumed no residual effects of treatment. We assumed that patients were treated for 12 months (ie, the longest follow-up time in the PCI-CURE study). Once treatment was discontinued, patients who received clopidogrel plus aspirin were assumed to have the same risk of an event as the aspirin-only patients for the rest of their lives. In the model, this was done by estimating the probability of having an event in the aspirin-only group using logistic regression and then applying the relative risk reduction observed in the trial (in the base case, the resulting risks were 5.0% in the aspirin-only arm and 3.6% in the combination arm). Patients who suffered a nonfatal MI had the same yearly mortality after the event as those in the aspirin-only arm (which was the rate observed in the register). Because the rate of mortality after an MI was higher than the rate of mortality after PCI, there was a slightly higher mean survival for patients in the combination arm (due to fewer events). However; during subsequent years, no risk reduction was applied: patients had the same risk of an event in both treatment arms. This implies that, for some patients, events were not avoided, but instead were only delayed, which moderated the effect on survival. In essence, this means that all probabilities in the model were the same for both arms, except during the

103

Clinical Therapeutics

first year, when the relative risk reduction from the trial was applied in the combination arm. Within-trial analyses have shown that patients in the active arm have less resource consumption during their initial hospitalization (recorded as a different distribution of diagnoses-related groups [DRGs]). s This is not captured elsewhere in the model; for this reason, we estimated the cost reduction by applying Swedish DRG costs to the observed DRGs and deducted this from the cost of treatment. In a sensitivity analysis, this cost was excluded. Costs during the first year after an MI were taken from a study by Zethraeus et al. 12 This was a retrospective study investigating the costs during I year after various cardiovascular events, in which each patient was used as his or her own control (comparing costs I year before the event to the costs I year after the event). The study included costs for in- and outpatient care, pharmaceuticals, and lost productivity (ie, work absence). Patients' own expenditures were not included, which may have led to a slight underestimation of the costs. As in the case with the transition probabilities, the uncertainty around the costs was taken into consideration by performing 1000 bootstrap replications of the means. One of these replications was drawn at random and used in each of the Monte Carlo simulations. In a sensitivity analysis, we tested the impact of higher and lower costs associated with an MI. We assumed that the costs during the subsequent years were the same as those in a model developed by Johannesson, 13 and used a direct cost of ¢835 and an indirect cost of ¢6565.14 This would imply that roughly half of patients return to full-time work. In a sensitivity analysis, we tested the impact of excluding these costs altogether. Studies have suggested that when performing an analysis from the societal perspective, costs relating to the increased survival of patients outside the health care sector should be included--that is to say, the value of the patients' production (minus consumption) should be subtracted, which would indicate a benefit to society of preventing deaths among the working-age population. Is In our opinion, relatively few studies take this into consideration; to facilitate comparison with other studies, we did not include this cost in the base case. However; it was included in the sensitivity analysis. The values for costs due to increased survival have been estimated by Ekrnanl6: --¢7601 (patients aged 50-64 years), £16,521 (aged 104

6 5 - 7 4 years), £19,956 (aged 75-84 years), and £29,842 (aged ,-85 years). We assumed that these costs were normally distributed with an SD of £890 in the stochastic analysis. All costs have been adjusted to year-2004 Swedish kronors (SEK) using the consumer price index lr and then converted to 2004 euros using the mean exchange rate for the year 2004 (¢1.00 = 9.13 SEK and US $1.24). 18 Based on a search of the English-language literature on MEDLINE (years, 1980-2003; search terms: utility, quality of life, and myocardial i~farction), no good estimates on the reduction in quality of life after MI in the studied population were available, preventing us from performing a cost-utility analysis. Instead, lifeyears gained (LYG) was used as the main measure of effectiveness. In some previous research, a utility reduction of 0.1 was assumed. 1~ We used this value in the present study, assuming the underlying distribution to be normal with an SD equal to the mean, to perform a cost-utility analysis as a sensitivity analysis (ie, we assumed that patients who suffer a MI have their utility reduced with 0.1 each year when we calculated quality-adjusted life-years). Both cost and effect were discounted using a discount rate of 3%, as is recommended by the Swedish guidelines for economic evaluation. 19 The discount rate was varied in a sensitivity analysis. The base case used in the analysis was a population with similar demographic characteristics as the one extracted from the register. This was summarized in Table II. Subgroup analyses were performed, defined by age groups and history of MI.

Table II. Population characteristics in the base case. Diabetes Mellitus Status Diabetes mellitus ~ Previous MI No previous MI No diabetes mellitus f Previous MI No previous MI

Share o f Mean Age, Subpopulation, % y

21.94 78.06

66.3 63.2

16.46 83.54

66.0 61.5

MI - myocardial infarction. ~Represents 14.00% off population overall. tRepresents 86.00% of population overall.

Volume 27~ Number 1

P. Lindgren et al.

RES U LTS After applying inclusion and exclusion criteria to the database, 3474 patients were included in the model analysis. Of these, 248 suffered from an event. Table III shows the demographic characteristics of the register sample compared with that of the PCI-CURE study. As can be seen, the register population was fairly similar to that of the clinical trial. Three risk factors were significant in the Cox proportional hazards model: age, diabetes mellitus, and previous MI. Therefore, these 3 variables were included in the logistic regression used to predict transition probabilities. Table IV shows the coefficients for the

risk functions used in the model. Higher age was associated with a higher risk of events and higher mortality rate. The same was true for patients with diabetes mellitus and patients with an MI any time before admission to the coronary care unit. The model predicted that the probability of having an event during the first year was 5 %. Furthermore, it predicted a discounted mean survival of 14.12 years for patients treated with aspirin only; patients who received combination therapy were estimated to have a 3.6% probability of experiencing a cardiovascular death or nonfatal MI during the first year after PCI and a mean survival of 14.16 years, a predicted increase of

Table III. Demographic characteristics. PCI-CURE 4

Variable

Register Population (n - 3474)

Aspirin Only Arm (n - 1345)

Clopidogrel + Aspirin Arm (n - 1313)

62.4 (10.9)

61.4 (10.9)

61.6 (11.2)

2363 (68.0)

940 (69.9)

914 (69.6)

Age, mean (SD), y Male sex, no. (%) Diabetes mellitus, no. (%)

486 (14.0)* 587 (16.9)*

Previous MI, no. (%) Current smoker, no. (%)

858 (24.7)* 1149(33.1)

Treated for hypertension, no. (%)

2SS (19.0)

249 (19.0)

349 (25.9) 396 (29.4)

3S9 (27.3) 406 (30.9)

PCI-CURE - Percutaneous Coronary Intervention-Clopidogrel in Unstable Angina to Prevent Recurrent Events study; MI - myocardial infarction. ~Three patients lacked information on diabetes mellitus status. fSixty four patients lacked data regarding history of MI÷ ::One hundred sixty patients lacked information on their smoking history.

Table IV. Coefficients for the risk functions used to calculate transition probabilities in the model. Time of Event*

Coefficient Age (95% CI) Diabetic (95% CI)

Previous MI (9S% CI) Constant

First Year After M I

Subsequent Year After M I

CV Death

Death Due to Other Causes

0.057 (0.042 to 0.0738) 0.552

0.054 (0.018 to 0.090) 0.667

0.081 (0.051 to 0.111 ) 0.1 656

0.0646 (0.046 to 0.083) 0.462

(0.188 to 0.915) 0.220 (-0.136 to 0.577) 6.761

( 0.026 to 1.308) 0.621 (0.058 to 1.1 84) 8.185

( 0.498 to 0.830) 0.289 (-0.338 to 0.915) 6.41 7

(0.029 to 0.898) 0.365 (-0.012 to 0.741) 8.442

MI - myocardial infarction; CV- cardiovascular. ~CV death or nonfatal MI.

January200S

105

Clinical Therapeutics

0.04 year per patient. Table V shows the predicted costs and incremental cost-effectiveness for the base case. Figure 2 shows the cost-effectiveness acceptability curve for the base-case analysis. Such a curve can be used for hypothesis testing, giving the P value for a 1-sided test of cost-effectiveness for different threshold values of the willingness to pay for an LYG. More elegantly, from a Bayesian perspective, the curve can be interpreted as the probability that the treatment strategy is cost-effective given the data at different threshold values. 2° Table VI shows the results from the subgroup analysis. As expected, patient categories with higher

risk (ie, diabetes mellitus) had more favorable costeffectiveness ratios. The predicted gain in survival was higher among older patients (driven by their higher risk and thus higher absolute risk reduction from treatment). Younger patients showed more savings, because the impact of savings in indirect costs was present in this group. If only direct costs were considered (data not shown), cost-effectiveness ratios were higher in these groups. Table VII shows the results from the sensitivity analysis. The resuks were insensitive to changes in the costs for the first year after an MI. If costs for the sec-

Table V. Predicted costs in year-2004 euros* and incremental cost-effectiveness ratios (ICERs) when analyzing a population similar to that of the register population. Costs perTreatment Arm, Mean (SD) Clopidogrel Variable Direct costs Indirect costs Total

LYG,

ICER

Aspirin

+ Aspirin

Net

Mean (SD)

(~/LYG)

2277 (1478) 523 (1 74) 2799 (1494)

2726 (1220) 282 (1 79) 3132 (1253)

449 (180) 11 7 (124) 332 (276)

0.04 (0.05)

10,993

0.04 (0.05)

8127

LYG = life year gained. ~Conversion rate: ~1.00 = US $1.24.

90 80 70 60

v

~J

50

DJ

40

~j

30 20 10 0

i

i

i

i

i

i

i

i

i

i

i

i

i

i

i

i

i

i

@

@

@

@

@

@

@

@

@

@

@

@

@

@

@

@

@

@

WTP (~) Figure 2. Cost-effectiveness acceptability curve for the model analysis, showing the proportion of simulations considered to be cost-effective at different threshold values of willingness to pay (VVTP) to gain I life-year. ~ I . 0 0 = US $1.24.

106

Volume 27~ Number 1

P. Lindgren ond year were excluded, the cost-effectiveness ratio increased, but it was still acceptable. As expected, including costs with added years of life increased the cost-effectiveness ratio. Excluding the savings from the initial hospitalization yielded a somewhat higher cost-effectiveness ratio. The cost-utility analysis indicated a cost-effectiveness ratio of C6506

et

al.

per quality-adjusted life-year. Assumptions about discounting had no major impact on the results. However, including MIs that occurred within 7 days of admission had a substantial effect on the results. The 1-year probability of suffering a cardiovascular death or nonfatal MI was predicted to be 14.0%, and the predicted cost savings from avoided events were

Table Vl. Subgroup analysis: Predicted net costs in year-2004 euros,* effectiveness, and incremental costeffective ness. Scenario

Net Total Cost (SD)

LYG (SD)

ICER (@:/LYG)

Aged S0 years

1 6 (818)

0.03 (0.0S)

Dominance

Aged 60 years

72 (594)

0.04 (0.06)

1969

371 (266) 374 (275)

0.0S (0.08) 0.09 (0.11 )

7213 3961

Diabetes mellitus

Aged 70 years Aged 80 years No diabetes mellitus Aged S0 years

211 (491)

0.03 (0.03)

7243

Aged 60 years

261 (366)

0.04 (0.04)

6929

Aged 70 years Aged 80 years

436 (1 95) 430 (1 97)

0.05 (0.06) 0.09 (0.09)

7937 4609

LYG - life-year gained; ICER- incremental cost-effectiveness ratio. ~Conversion rate: @1.00 - US $1.24.

Table VII. Sensitivity analysis: Predicted net costs in year-2004 euros,* effectiveness, and incremental costeffectiveness. Scenario

Net Total Cost (SD)

LYG (SD)

307 (298) 319 (287) 345 (265

0.04 (0,0s)

7s08

0.04 (0.05) 0.04 (0.05) 0.04 (0.05)

7820 8444 8756

ICER (@:/LYG)

Variation in first year costs

+20% +10% 1 o~ 203 No second-year cost Including cost in added years o f life No saving during initial hospitalization Cost utility analysis Discounting

0% 5% Including MIs during 7 days after admission

338 446 942 456 332

(2ss (141 (848 (269 (276

0.04 0.04 0.04 o.os

(0.05) (0.0s) (0.0S) (0.06)+

289 (343 351 (249

0.05 (0.08) 0.04(0.05)

287 (948

0.08 (0.1 2)

10,929 23,0s3 11,161 6306* 5702

9982 Dominance

LYG- life-year gained; ICER- incremental cost-effectiveness ratio; MI - myocardial infarction. ~Conversion rate: ~1.00 - US $1.24. fQuali~y adjusted life years. ¢Cost per quality adjusted life years gained.

January200s

107

Clinical Therapeutics

greater than the base case, indicating a net total cost reduction of £289 overall. DISCUSSION Our model predicted a slight increase in survival when adding clopidogrel to aspirin therapy for patients undergoing PCI, but at the expense of higher costs. Roughly half the costs of the drug were offset by savings when considering direct and indirect costs. If only considering direct costs, this offset was roughly 15%. The results of this analysis showed higher costeffectiveness ratios than a previous analysis of the entire CURE trial population. 2 This may be explained by the lower gain in survival predicted by the present model, which in turn may be explained by the lower observed risk and longer predicted survival following an event. Nevertheless, the results of the present analysis were well below the levels of what is generally considered to be cost-effective; for example, the World Health Organization has defined a cost-effective strategy as one with a threshold value of 1 to 3 times the gross domestic product (GDP) per capita for a disability-adjusted LYG or quality-adjusted LYG. 21 With a GDP per capita of roughly E25,000 in Sweden, 17 our results fall well below this threshold. Apart from comparing the cost-effectiveness with threshold values, it is also useful to compare the costeffectiveness of this intervention with that of other established interventions. The results predicted by our model were quite similar to those estimated by M a r k et a122 when studying the cost-effectiveness of glycoprotein IIb/IIIa inhibition with eptifibatide in acute coronary syndromes, which was found to be £16,491 per LYG (induding only direct costs). Analyses of the Scandinavian Simvastatin Survival Study on lipid lowering in secondary prevention indicated cost-effectiveness of £3630 and £11,495 per LYG, respectively, for men and women aged 59 years with different levels of cholesterol. 23 Our results are of similar magnitude. The present model predicted fewer events with aspirin alone than were reported in the PCI-CURE study4: 5% compared to almost 15%. One explanation for this is that we did not include events occurring within the first 7 days from the index admission to avoid including events that initiated the PCI. Some of the excluded events should probably be included in the risk calculations; therefore, we may be overestimating the cost-effectiveness ratios. Including all events would 108

produce a similar probability of suffering an event as in the trial, but then we would be likely to include too many events. The cost-effectiveness ratio estimated from this calculation is thus likely an underestimation, with the true estimate somewhere between the ratios predicted using the 2 different criteria in the risk calculation. This may also explain why our model predicts a lower survival gain than a study by Weintraub et al, 24 which (based on Framingham data) estimated the gain in survival to be 0.10 year. One limitation of our study is that the cost data was somewhat old and based on a small sample. However; sensitivity analysis showed that variations in these figures had no major impact on the results because the key cost driver was the cost of the drug. If newer treatments made available since the time period for data collection for the cost-of-illness study are more expensive than older treatments, our cost estimates would represent an underestimation of the potential cost savings. Therefore, our cost-effectiveness ratios may be, if anything, somewhat inflated. One exclusion criterion in the PCI-CURE study was that patients who had undergone a PCI or CABG 3 months before the index PCI were not permitted to participate. We had no opportunity to verify the timing of previous procedures; therefore, we excluded all patients with a previous PCI or CABG. This most likely led to a slight underestimation of the risk of an event in the model, because risk factors such as diabetes mellitus and previous MI were less prevalent in our sample than in the population of the PCICURE study. Fewer patients with diabetes mellitus or MI may also have led to an underestimation of the risk, and thus to overestimated cost-effectiveness ratios, another argument that our estimates were conservative ones. The model was quite robust in changes to key parameters such as costs. Changes in discounting had no major impact on the results. The gold standard for performing health economic evaluations, recommended by most guidelines, is a cost-utility analysis taking into account potential gains in both survival and quality of life. Unfortunately, no good estimates on the reduction of quality of life (ie, suitable for economic evaluation) were available at the time of the present analysis. When using a reduction of 0.1 in a sensitivity analysis, cost-effectiveness ratios are improved; thus, potential reductions in quality of life may be important. Therefore, further research in the Volume 27~ Number 1

P. Lindgren et al. relationship between cardiovascular events and reductions in quality of life is merited. CONCLUSIONS

Treatment with clopidogrel plus aspirin, as in the PCI-CURE study, appeared to be cost-effective in this model analysis of patients with unstable CAD undergoing PCI in Sweden. The predicted cost-effectiveness ratios were well below the threshold values generally considered cost-effective. AC K N O W L E DG M ENTS

This work was funded in part by a grant from SanofiSynthelabo, Paris, France. Mr. Lindgren and Dr. J6nsson have worked as consultants to Sanofi and Bristol-Myers Squibb. REFERENCES 1. YusufS, Zhao F, Mehta SR, et al, for the Clopidogrel in Unstable Angina to Prevent Recurrent Events Trial Investigators+ Effects ofclopidogrel in addition to aspirin in patients with acute coronary syndromes without STsegment elevation [published corrections appear in N ErGI J Med. 2001 ;345:1 71 6 and N En~J Med. 2001 ;345:1 506]. N En~j Med. 2001 ;345:494-502. 2. Lindgren P, JOnsson B, Yusuf S. Cost-effectiveness of clopidogrel in acute coronary syndromes in Sweden: A long, term model based on the CURE trial+J Intern Med+ 2004; 255:562-570. 3. Lamy A, JOnsson B, Weintraub WS, et al. The costeffectiveness of the use of clopidogrel in acute coronary syndromes in five countries based upon the CURE study. EurJ Cordiovosc Pr~v Rehobil. 2004;11:460-465. 4. Mehta SR, YusufS, Peters RJ, et al, for the Clopidogrel in Unstable Angina to Prevent Recurrent Events Trial (CURE) Investigators+ Effects of pretreatment with clopidogrel and aspirin followed by long term therapy in patients undergoing percutaneous coronary intervention: The PCI CURE study. Lancet. 2001;358:527 533+ 5. Mehta SR, Weintraub WS, JOnsson B, et al. Incremental cost effectiveness of early and long term clopidogrel in patients undergoing percutaneous coronary intervention in the CURE trial: The PCI-CURE economic analysis. Poster presented at: S2nd Scientific Session of the Ameri can College of Cardiology; March 30-April 4, 2003; Ch icag% III. 6. Naimark D, Krahn MD, Naglie G, et al. Primeron medical decision analysis: Part 5 Working with Markov process es+ Med Decis Making. 1 997;17:152 159+ 7+ Markov modeling, analysis and microstimulation with the health care module. Tr~eAge Pro HeMth Core Users MonuM. Williamstown, Mass: TreeAge Software, Inc; 2004:59-128.

January2005

8. Sonnenberg FA, BeckJR. Markov models in medical decision making: A practical guide. Med Decis Making+ 1 993; 13:322 338. 9. Halpern EF, Weinstein MC, Hunink MG, Gazelle GS. Representing both first and second order uncertainties by Monte Carlo simulation for groups of patients. Med Decis Making. 2000;20:314-322. 10. Lothgren M, Zethraeus N. Definition, interpretation and calculation of cost-effectiveness acceptability curves. Heolth Econ. 2000;9:623 630. 11. Swedish Register of Information and Knowledge about Swedish Heart Intensive care Admissions Web site. Available at: http://www.riks hia+se.Accessed August 1, 2003. 12. Zethraeus N, Molin T, Henriksson P, Jonsson B. Costs of coronary heart disease and stroke: The case of Sweden+ J Intern Med. 1 999;246:1 51-1 59. 13. ]ohannesson M. At what coronary risk level is it cost effective to initiate cholesterol lowering drug treatment in primary prevention? Eur HeartJ. 2001 ;22:919-925. 14. ]onsson B. Economics of drug treatment: For which patients is it cost-effective to lower cholesterol? Loncet. 2001 ;358:1251-1256. 15. Meltzer D+ Accounting for future costs in medical cost effectiveness analysis. J Heolth Econ. 1997;1 6:33-64. 1 6. Ekman M. Two Essays in Health Economics: Consumption and Production by Axe,, with Emphasis on Health Care Expenditures ond Economic Evo&ation oflBeto-Blocker Therapy in Heort Failure. Stockholm, Sweden: Stockholm School of Econom ics; 2001. 17. Statistics Sweden Web site+Available at: http://www.scb+se+ Accessed September 24, 2003. 18. The Bank of Sweden. Interest and exchange rates. Available at: htrp://www.riksban ken.se+AccessedSeptember24, 2003+ 19. Generol Guidelines for Economic Evo&ations from the Pharmaceutical Benefits Board [in Swedish]+ Stockholm, Sweden: Pharmaceutical Benefits Board; 2003. 20. Eenwick E, O'Brien BJ, Briggs A. Cost effectiveness accept ability curves facts, fallacies and frequently asked ques tions. HeMth Econ. 2004;13:405-41 5. 21+ Murray CJ, LauerJA, Hutubessy RC, et al. Effectiveness and costs of interventions to lower systolic blood pressure and cholesterol: A global and regional analysis on reduction of cardiovasculaFdisease risk+ Lancet. 2003;361: 717-725. 22. Mark DB, Harrington RA, Lincoff AM, et al+ Cost effectiveness of platelet glycoprotein lib/Ilia inhibition with eptifibatide in patients with non-ST-elevation acute coronary syndromes+ Circulation+ 2000;101:366 371. 23. Johannesson M, JOnsson B, Kjekshus J, et al, for the Scandinavian Simvastatin Survival Study Group+ Cost effectiveness ofsimvastatin treatment to lower cholesterol levels in patients with coronary heart disease. N EngljMed. 1997;336:332 336.

109

Clinical Therapeutics

24. Weintraub WS) Mahoney EM, Mehta S, et al. Long term cost effectiveness of clopidogrel in pa tients having percutaneous coronary intervention early after acute coronary syndrome: Results from PCI CURE. j Am Coil Cardiol. 2004;43(Suppl 2): A296. Abstract 11 37 77+

Address correspondence to: Peter Lindgren, MSc, D e p a r t m e n t of Cardiovascular Epiderniology, Institute of Environmental Medicine, Box 210, 171 77 Stockholm, Sweden. E-rnaih [email protected] 110

Volume 27, Number 1