- Email: [email protected]
Which statin is most efficient for the treatment of hypercholesterolemia? a cost-effectiveness analysis
CLINICAL
THERAPEUTICSVVOL.
2 1, NO. 11,1999
Which Statin Is Most Efficient for the Treatment of Hypercholesterolemia? A Cost-Effectiveness Analysis Albert Cobos, MD, PhD,’ Albert J. Jove& MD, MSc, PhD? Anna Garcia-Alt&,2 Reina Garcia-Closas, MD, MSc, PhD,3 and Lluis Serra-Majem, MD, MSc, Phil ‘Novartis Farmactfutica, SA, 2Ag&nciad’Avaluacid de TecnologiaMdica, Servei Catalci de la Salut, Department de Sanitat i Seguretat Social, Barcelona, and 3Research Group on Community Nutrition, Universities of Las Palmas de Gran Canaria, La Laguna, and Barcelona, Spain
ABSTRACT A review of the cost-effectivenessliterature indicated that the hydroxymethylglutaryl coenzymeA-reductase inhibitor fluvastatin is more cost-effective for achieving minorto-moderate reductions in low-density lipoprotein cholesterol(LDL-C) levels than 3 other statins: lovastatin, pravastatin, and simvastatin. The main goal of this study was to verify the applicability of theseconclusions to Spanish health care costs and patternsof resourceconsumptionrelated to the treatment of hypercholesterolemia.A stochastic simulation model was used to predict both the costs and effects of treating high-risk hypercholesterolemicpatients with fluvastatin, lovastatin, pravastatin, or simvastatin.Epidemiologic data were used to find a suitable theoretic probability distribution model for baselineLDL-C values Accepted for publication Printed in the USA. Reproduction
1924
in whole
August
8, 1999.
or part is not permitted.
in high-risk hypercholesterolemicpatients. The modelwasthenusedto generate10,000 random observations of baseline LDL-C values; the corresponding LDL-C values after a 2-year treatment period were predicted as a function of the baselinevalue and the percentagereduction expectedwith a particular statin and dose, according to the resultsobtainedin 2 me&analyses.The probability of treatment discontinuation wasalsotaken into accountusingestimates obtained in usual practice. The effects of treatmentwere expressedasthe rate of successin achieving the goal level of LDL-C, as defined in the current Spanish recommendationsfor the treatment of hypercholesterolemia. The average costs of treatment were computed from both the social and public-financing perspectives,including the cost of lipid-lowering drugs, physician visits, laboratory tests, and days off work, as appropriate. The occurrence of nonscheduledvisits and workdays lost becauseof side effects were taken into account to computeindirect costsrelevant to 0149.2918/99/$19.00
A. COBOS ET AL.
the social perspective. The potential costs of treatingsideeffectswere ignored.A costeffectiveness analysis was performed to compare the cost-effectiveness ratios obtained with eachof the 4 statinsconsidered in this study. Model-based predictions of the effects, total costs,and cost-effectiveness ratios were made. Cost-effectivenessratios were interpretedas the cost per patient meetingthe goal of therapy, according to current Spanish recommendations. The data showed that fluvastatin had the lowestcost-effectiveness ratioswhenLDL-C levels required reduction to ~25% of baseline levels. In this situation, fluvastatin was more cost-effective than lovastatin, pravastatin, or simvastatinfrom public-financing and socialperspectives.Key wotrik: cholesterol, health economics,cost-effectiveness analysis,modeling.
INTRODUCTION High levels of blood cholesterolhave been associatedwith an increasedrisk of cardiovascular mortality and morbidity,le3 and cholesterolreduction hasproved beneficial no matter what type of intervention is adopted.1*4Several recommendationsfor the prevention of cardiovascular events have chosencholesterolreduction asa main target.5-7Health care plans have included cholesterolreduction in healthy populations amongtheir goals.8FVoposedmeansof prevention of cardiovascular diseasethrough cholesterolreduction have included dietary changes,exercise,and medications.9,10 Spanish trends indicate increasing use of lipid-lowering drugs.” According to IMS data,‘* statins accountedfor 80% of expendituresfor lipid-lowering drugs prescribed in 1997 (31,085 million pesetas [PTA]), which wasa 22.35%increasecompared with 1996.This cost scenarioshould
cause decision makers and prescribers to seriouslyconsiderthe cost-effectivenessof the different statins as a guiding criterion for determining treatment. In reviewing the published cost-effectiveness studies that compared simvastatin, pravastatin, lovastatin, and fluvastatin,13-*0we concluded that fluvastatin was the preferred alternative for patients needing minor-to-moderate reductions in low-density lipoprotein cholesterol (LDL-C). However, most of these studies were carried out in the United States, and none were conducted in Spain. Consequently, we wanted to determine if the sameresults were valid when using costs and resource consumption patterns in a hypercholesterolemic population in Spain. The aim of this study was to assessthe cost-effectivenessratio of different statins in the clinical management of high-risk hypercholesterolemic patients who need reductions of ~25% of baseline LDL-C levels. A stochastic simulation model was developed and usedto assess costsand effectiveness during a 2-year treatment period with each of the 4 statins considered.
MATERIALS
AND METHODS
The risk of coronary heart diseasehasbeen defined using an algorithm recommended by 3 medical societiesin Spain.7The algorithm takes into account total cholesterol level and the number of other risk factors (ie, age >45 years for men, >55 years for women; smoking; hypertension; diabetes;sedentarylifestyle) and classifies individuals into 1 of 3 categories of cardiovascular risk: mild, moderate, or high (Figure 1). Subjects were treated with drugs if their LDL-C values measuredafter dietary treatment were higher than the drug treatment threshold. 1925
CLINICAL
Total Cholesterol OxW)
Other Risk Factors
Global CV Risk Mild
Moderate
Goal LDL-C Owl)
175
THERAPEUTICS”
Drug Therapy Threshold (w/W 190
155
180
135
160
Figure 1. Hypercholesterolemiatreatmentalgorithm recommendedby 3 Spanishmedicalsocieties.’ Patientshaving low-density lipoprotein cholesterol(LDL-C) levels after dietary treatment that are above the drug-treatmentthresholdshouldbe treated with lipid-lowering drugs until the goal LDL-C level is reached.CV = cardiovascular. The distribution of LDL-C levels in highrisk patientsneedingdrug therapy wasmodeledusingdata from the CatalanNutritional Survey (CNS).21This survey was designed to investigate the dietary habits of a random sampleof 2346 Catalonian subjects; 893 of them consented to a biochemical test. Data on serumlipids and major nonlipid risk factors were available from 888 subjects.Basedon the algorithm described in Figure 1, those patients with a serum cholesterolconcentration >200 mg/dL and a serum triglyceride concentration x200 mg/dL were classified by cardiovascular risk category (mild, moderate, high) according to the presenceor absenceof major nonlipid risk factors. Subjects with LDL-C values above the goal indicated in Figure 1 shouldbe treated with diet modification before drug treatment.We assumed that dietary intervention would reduce LDL-C values by 1O%.22After we applied this 10% reduction to simulatepostdietary LDL-C values,subjectsin the high-risk category who fulfilled the criteria for drug therapy were selectedfor the analysis,and these “postdiet” LDL-C values were consideredto be baselinevalues. 1926
The Pareto distribution23 is a suitable theoretic probability distribution for modeling baseline LDL-C values. The lower boundary of this distribution was setto 160 mg/dL by the algorithm, making this the value of the truncation parameter K. The maximum likelihood estimatefor the second parametercx was obtained as N (z In xi - N In K)-l, where xi is the ith observed value (i = 1, 2, .... N), K is the truncation parametertaking the value 160, and addition is taken over i. The fitted Pareto distribution wasthen usedto generaterandom observationsof baselineLDL-C values in high-risk casesneedingdrug therapy, until 10,000casesneedingreductionsin LDL-C values of ~25% were achieved. In the selected cases,a 2-year treatment period with statins was simulated and the results assessedusing a cost-effectivenessanalysis. Simulation of Therapeutic Processand Effectiveness Data Low, medium, and high doses were consideredfor all statins- 10, 20, and 40 mg for simvastatin and pravastatin and 20, 40, and 80 mg for lovastatin and flu-
A. COBOS ET AL.
vastatin. When a particular dose was not marketed in Spain, we defined the high dose as 2 medium-dose tablets. The number and sequence of physician visits was simulated following current Spanish recommendations7 This consensus report recommends physician visits after 1 month; after 6 weeks; and after 3, 6,9, 12, 18, and 24 months from the start of treatment. At the 6-week visit the effect of therapy was assessed by comparing the simulated LDL-C value with the value defining the goal of therapy. In case of therapeutic success, treatment was maintained at the low dose for the whole simulation period. In case of therapeutic failure, the low dose was titrated to the medium dose. A tolerance margin of 5 mg/dL was allowed in assessing the effectiveness of therapy to avoid doubling the dose when LDL-C values were higher but were close to the treatment goal. A similar strategy was followed at the third visit. In case of therapeutic failure with the medium dose, the highest dose was given and maintained for s2 years. For any of the statins at any of the doses considered, the effect of therapy was simulated as the LDL-C value achieved through the model: yijk = x,(1 - l.+j/lOO) + (oiilOO)ek where yijk is the LDL-C value obtained after treatment with statin i (i = fluvastatin, lovastatin, pravastatin, simvastatin) at dose j (j = low, medium, high) in subject k; xk is the baseline LDL-C value in subject k; l~,~.is the percentage reduction expected wit4 the ith statin at the jth dose; ati is the SD of the percentage reduction in LDL-C obtained with the ith statin at the jth dose; and ek is an N(O,l) random variable representing a “subject” effect.
For any subject k, the LDL-C value generated by this model (y,,) would have an expected value equal to the baseline value of subject k affected by the reduction expected for each statin and dose. Also, the subject effect term ensured that the response for subject k would always be in the same percentile for all statins and doses (ie. good or bad responders would be equally good or bad for all statins and doses). Estimates for p,ij and ui. were obtained from 2 recently publishe d meta-analyses on the efficacy of statins.24,25 In some cases, the estimates themselves were not reported, but we were able to compute them from information provided in the publication or from the original publication. Also, the model considered whether compliance with therapy influenced the effectiveness and cost of treatment. To do so, a Bernouilli random variable was generated to indicate compliance with therapy, independent of effectiveness. The probability of discontinuation was estimated as 60% during the first year of treatment and was distributed as 25% at the first month and 50% within the first 3 months, with the remaining 25% being uniformly distributed up to the end of the first year. 26We assumed that lack of compliance was uniformly distributed within each interval between adjacent visits. Lack of compliance was classified as therapeutic failure, and LDL-C values in these cases were set back to baseline levels. Cost Data
We adopted2 perspectivesfor the costeffectiveness analyses:the social perspective (direct and labor costs) and the public-financing perspective (direct costs incurred by the health care system). The 1927
CLINICAL THERAPEUTICS”
following direct costs were considered: scheduled physician visits, unscheduled physician visits because of side effects, lipid-lowering drugs, and laboratory tests used to monitor efficacy and safety of therapy. Unscheduled visits were simulated by generating a Poisson random variable independent of any other simulation event with an intensity of 0.05, 0.34, 0.34, and 0.26 for fluvastatin, lovastatin, pravastatin, and simvastatin, respectively. l9 We assumed that laboratory tests were done at each of the scheduled and unscheduled visits. We did not consider the cost of treating side effects, since they are extremely rare and tend to resolve spontaneously after therapy is discontinued. Drug costs were estimated using mean retail prices for all the forms marketed in Spain.12 Estimates for the unit cost of doctors’ visits and laboratory tests were obtained from a specific health care cost database and fixed at 2.093 and 4.342 PTA, respectively.27 Labor or indirect costs considered in the analysis were computed using the time spent at physician visits and the days of work missed because of side effects. According to published results,19 the number of workdays missed because of side effects was simulated using an independent Poisson random variable with an intensity of 0.11, 0.33, 0.14, and 0.23 for fluvastatin, lovastatin, pravastatin, and simvastatin, respectively. The cost per hour of work missed was estimated from a survey of salaries and services and fixed at 1439 PTA.28 We assumed a loss of 2 hours of work for each visit and 8 hours for each working day. A discount rate of 5% was applied to all costs incurred in the second year of treatment. All costs were expressed in Spanish currency using 1997 values. Currently US $1 equals 150 PTA. 1928
Cost-Effectiveness Analysis Effectiveness was defined as the number of cases meeting the goal of therapy in the 10,000 cases simulated. The analysis carried out from the social perspective included all costs incurred during the 2-year treatment period. The analysis carried out from the health care system or publicfinancing perspective only (including direct costs and drug costs) was estimated according to a patient’s employment status. An active worker pays 40% of the cost of the drug, while a pensioner (a retired worker or a patient receiving disability payments) pays nothing. The percentage of units prescribed to pensioners within the publicly financed health care system was estimated to be 77.53%, according to published data.29 This proportion was used as the parameter of a Bernouilli random variable defining each simulated patient’s health care affiliation status. For each selected perspective, all relevant costs incurred in the 2-year treatment period were used to estimate the costeffectiveness ratios. Costs for the first year of treatment included the dose-titration phase and the specific schedule of visits. Cost-effectiveness ratios were estimated as the cost per patient meeting the therapeutic goal. Incremental analyses were performed and interpreted when appropriate. Several one-way sensitivity analyses were performed to test the robustness of the primary analysis to changes in major assumptions and model parameters. Thus the discontinuation rate was set at 40% and 0%. The intensities of the Poisson variables simulating the number of additional visits and workdays missed because of side effects were set to their corresponding average values to
A. COBOS ET AL.
A
B 1.0
l.O0.90.8-
0.9 0.8 0.7 0.6
b 0
0.70.6-
3 0.5.0 .g 0.4$
i;j/,
,
,
(
,
,
,
,
,
0.30.2-
,
160170
180190200210220230240250
Baseline
LDL-C
,
(
,
(
,
,
,
,
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
Pareto CDF
Figure 2. Fit to a Pareto distribution (dashedline) of the baselinelow-density lipoprotein cholesterol (LDL-C) levels of 53 patients categorized as being at high cardiovascular risk and needing drug therapy (circles). (A) Cumulative distribution function (CDF); (B) quantile-quantile plot. carry out an equal data analysis for all 4 statins. In addition, a 20% deviation in the unit cost of drugs, physician visits, and laboratory tests was considered for sensitivity analyses.Finally, data were analyzed using discount ratesof 3% and 0%.
RESULTS When the algorithm proposedby the Spanish recommendations7was applied to the 888 casesof the CNS*l with available data on lipid profiles, 235 had high LDL-C values such that dietary intervention was indicated.After applying a 10%reduction to the LDL-C values to simulatethe effect of diet, 58 of the 888 caseshad LDL-C values above the drug-treatment threshold. This correspondsto 6.5% of the sample, with a 95% CI from 4.9% to 8.1%. No patient was in the low-risk category, 5 were in the moderate-risk category, and 53 were in the high-risk category. Further analysesfocused only on these 53 highrisk cases.
Figure 2 depicts the fit of a Pareto distribution to the baselineLDL-C values of the 53 high-risk casesneeding drug therapy. Both the cumulative distribution function andthe quantile-quantileplots showed a good fit of the data to the Pareto model. TableI showsthe 2 setsof estimatesused for the parameterspij and uij in our simulation model.Theserepresentthe meanand SD of the percentagereductionsof LDL-C values achieved with different statins and doses.Although mean percentage reduction estimateswere similar in both sets,SD estimateschanged considerably (with the exception of lovastatin) dependingon the source of the data. In general, these estimateswere much lower in the meta-analysis of Delea et a125than in the estimatesof Kong et al.M The difference between estimateswas more evident for simvastatin. Average retail prices were used as cost data for the drugs marketed in Spain (Table II). Annual costs for the therapy rangedfrom 38,560 PTA for fluvastatin 20 mg to 145,531 PTA for simvastatin40 mg. 1929
CLINICAL THERAPEUTICS”
Table I. Estimates of the mean (* SD) percentage reduction of low-density cholesterol induced by 4 statins at the doses used in the simulation.
DW Fluvastatin
Lovastatin
Pravastatin
Dose (mg) 20 40 80* 20 40
80 10 20 40
Simvastatin
10 20 40
lipoprotein
Kong et alz4
Delea et a125
f 12.1 f 11.0 + 10.9 24.9 + 11.7 30.4 f 11.1 39.8 * 11.1 19.3 + 17.1 26.1 zt 12.5 27.2 k 10.8 28.6 + 19.9 34.4 f 15.7 40.7 + 21.9
19.9 24.3
21.0 23.1 33.1
-
zt 2.7 f 5.7
25.6 zt 8.8 31.9 + 10.7 38.8 + 14.9
18.8 + 7.4 + 4.0 f 5.0 28.3 f. 3.2 32.4 + 5.0 42.2 ztz6.4 26.9 30.8
*Estimates for fluvastatin 80 mg were taken from Table 2 in Kong et al (or from references therein) and were used to complete the set of estimates from Delea et al used in the simulation. The meta-analysis by Delea et al did not include fluvastatin 80 mg.
Effectiveness and maximum dose reached with each statin after the initial dose-titration period are shown in Table III. When estimates reported in the metaanalysis by Delea et a125were used, the highest success rate was achieved with simvastatin, followed by pravastatin, fluvastatin, and lovastatin. All fell between 36% and 40%. However, this ranking changed remarkably when data from Kong et a124were used. In this case, lovastatin was the most effective, followed by fluvastatin, simvastatin, and pravastatin. The total costs, cost-effectiveness ratios, and incremental ratios obtained in the analyses are displayed in Table IV. Fluvastatin was the least expensive and most cost-effective drug from both the social and public-financing perspectives, independent of the meta-analysis data used to estimate pij and crij. Consequently, incremental ratios were computed taking fluvastatin as the reference value for those 1930
Table II. Average retail prices (including value-added tax)* of drug doses marketed in Spain. Dose
DW Fluvastatin Lovastatin
Pravastatin Simvastatin
Costs (Pesetas)+ Annuals
0s)
UniG
20 40
2958 4349
38,560 56,692
20 40
4160 665
54,229 86,701
10
1 4160
20
5978
54,229 77,928
4132 5874 11,164
53,864 76,572 145,531
10 20 40
*These prices were used as cost estimates in the simulation. +IMS, 1997.‘* *Unit costs are the average prices of 1 package; all packages contain 28 tablets. 5Annual cost is based on the assumption that all drugs are prescribed as a 1 tablet/d regimen.
A. COBOS ET AL.
alternatives showing higher effectiveness but costing more. The sensitivity analyses produced similar results. Fluvastatin was always the least expensive alternative. The difference in efficacy compared with the more efficacious therapeutic alternative never exceeded 4% for credible dropout-rate Table III. Effectiveness
scenarios (60% or 40%), or 6% if we assumed a 0 dropout rate. When we used estimates from the meta-analysis of Delea et al,25 fluvastatin always showed the lowest cost-effectiveness ratio despite the higher effectiveness of simvastatin and pravastatin, with incremental ratios ranging, respectively, from 225,734 to 428,423
and maximum dose reached. Dose (%)*
Source Delea et a125
Kong et alN
E (%I
Low
Medium
High
Fluvastatin Lovastatin Pravastatin Simvastatin
37.01
81 83 62 99
9 9 37
10
36.59
Fluvastatin Lovastatin Pravastatin Simvastatin
36.62 38.56 32.59
DWZ
39.19 39.37
0
64 75 56 72
34.21
8 1
1
7 13 21 14
29 12 23 14
E = effectiveness as the percentage of treatment successes in the simulated cohort of 10,000 treated patients. *Dose is the maximum dose reached in the titration period (see Methods for corresponding values in milligrams.)
Table IV. Costs (total cost per patient), cost-effectiveness ratios (C/E), and incremental ratios of total costs to the number of cases reaching the goal of therapies (DJD,) for the base-case analysis. Social Perspective Source*
=‘w
costs (Pesetas)
C/E
Delea et a125 Fluvastatin 86,531 Lovastatin 102,37 1 Pravastatin 106,158 Simvastatin 96,489
233,804 279,778 270,880 245,084
Kong et a124 Fluvastatin Lovastatin Pravastatin Simvastatin
266,481 271,428 369,364 298,419
97,585 104,663 120,376 102,089
Public-Financing Perspective costs (Pesetas)
C/E
68,326 Dominated* 8 1,809
184,616 223,584
Dominated+
85,610 76,752
218,449
792,833 357,016
78,536 83,940
214,462 217,686 302,705 239,385
DC4
900,322 421,984
364,798
Dominated+ 98,65 1 Dominated+ 8 1,894
*Source of the estimates for the pij and aij parameters in the simulation model. tAn alternative is considered dominated when another lower-cost, higher-effectiveness
194,950
alternative
DcDs
278,542
Dominated+ Dominated+ exists.
1931
CLINICAL
PTA and 573,624to 951,405 PTA from the public-financing perspective and from 262,609 to 506,390 PTA and 642,269 to 1,080,397PTA from the socialperspective. With the estimatesof Kong et al,24fluvastatin was again the alternative with the lowest cost-effectivenessratio in all but 2 scenarios.In this sense,only when we assumeda 0 dropout rate and also an equal mean number of additional visits and workdays missedbecauseof side effects, lovastatin was the most cost-effective option, with incrementalratios from 125,872 to 212,173 PTA, depending on the perspective, comparedwith fluvastatin.
DISCUSSION
AND CONCLUSIONS
The results of these analyses show that fluvastatin is more cost-effective than lovastatin, pravastatin, and simvastatin from both a public-financing and social perspective, using cost data from Spain. Earlier cost-effectiveness analyses also found fluvastatin to be the most costeffective statin for patients needingminorto-moderate LDL-C reduction.13~14*1~20 Lovastatin had more favorable costeffectiveness ratios than fluvastatin in only 2 of the 32 scenariosconsidered in our sensitivity analysis. The 2 scenarios were characterized by perfect compliance (0% dropout rate) and an equal mean number of extra visits and workdays missedbecauseof side effects. A 0 dropout rate seemsunrealistic, and previous studies have not confirmed the hypothesis of equality.i9 Although the different statins are often considered equal in terms of safety and tolerability, there is some evidence that fluvastatin is the least likely to interact with drugs metabolized by the cytochrome P-450 3A4 isoenzyme.30-33 1932
THERAPEUTICS”
Several limitations should be kept in mind when interpreting the results of our analyses.First, they do not include other marketed statins, such as atorvastatin and cerivastatin. Although this is an important limitation, the lack of meta-analysesincluding all 6 statinsprevented us from including them in the analyses. Second,there are limitations associated with the method and with the source of data. Limitations associatedwith modeling have been pointed out elsewhere.34To overcome most of these limitations and the potential bias of model assumptions, we used the results of 2 published metaanalysesas estimatesfor the relevant parametersof our simulation model. In addition, a one-way sensitivity analysis was performed modifying the values of major influential parameters,including the rate of compliance, which varied acrossstudies 26~35-37 and the discount rate. Although the primary studies included in the 2 meta-analyseswere not conducted in Spain, it is likely that their primary results (ie, percentagereduction of LDL-C values) are applicable to Spanish hypercholesterolemic patients. Similar data from Spanishpatients are not available in the literature. A further limitation in the method is that our model ignored the potential costs of treating the side effects of statins. Unfortunately, the data neededto model the incidence of side effects as well as their likely treatments are not available in a comparableformat for the different statins (the incidence of adverseevents is not addressedin the only 2 meta-analyses we found comparing statins). However, statins are known to have a particularly good safety profile, with rare and mild sideeffects that typically ceasewhen therapy is stoppedand that do not require spe-
A. COBOS ET AL.
cific treatment. Thus the possibility of a single expensive adverse event that outweighs savings does not seemlikely. A third type of limitation is associated with the quality of the data sources.The fmt of theselimitationsinvolves the LDL-C baseline data used to define the population at high risk for cardiovascular events. Becausethe CNS provided data only from subjects who were followed and consented to a biochemical test, a selection bias could be present due to over-representation of a higher-risk population. However, in a recent investigation conducted in 1 of the 4 provinces of Catalonia,38in which nonresponseswere much more uncommon than they were in the CNS, the mean levels of total cholesterol were higher for all age groups and both sexesthan they were in the CNS. This is inconsistent with the presenceof an important selection bias toward higher-risk cases affecting CNS data. Also, in our analyses,we assumedthat the profile risks of the Catalan population could be extrapolated to the rest of Spain. Another issueto consider in our simulations concerns the variability of the results achieved; these findings depended on the meta-analysisestimatesapplied to our model parameters.The fact that 1 of the meta-analysesfocused on primary hypercholesterolemia25 and the other included both primary and secondary hypercholesterolemia24might be the cause of their different results, along with differences in study selection criteria. Furthermore, the ranking of the 4 statins according to the percentage of successful treatments obtained with the estimatesof Delea et a125was consistent with previous findings regarding the relative potency of these statins, while this was not the case with the estimatesof Kong
et a1.24Moreover, the dose distribution produced by the estimatesin the study by Delea et al closely reflects market data,‘* where the 20-mg form of fluvastatin accounted for >80% of the total number of units sold; again, this is not the casewith the estimatesof Kong et al. In short, both the effectiveness and drug-consumption patterns obtained with our simulation model were consistent with external data when the model was fed estimatesfrom Delea et al but not when estimatesfrom Kong et al were adopted. The influence that SD estimates of the treatment response had in our simulation results showsthe importance of the estimatesof treatment responsevariability; however, few efficacy studies have reported this kind of data. Dropouts may have a major impact on the resultsof cost-effectivenessanalyses.In our model,the dropout rate was setat 60%, according to data obtained from current clinical practice.26*35*36 The model assumed independencebetween treatment response and treatment discontinuation,which is an unlikely assumption, but data needed to model compliance as a function of effectiveness were unavailable. On the other hand, there is no evidence that treatment with different statins resulted in different ratesof treatment discontinuation.26,36 The cost data used in the analysis were intended to reflect the cost of resource consumption in Spain, which is not necessarily similar to the costs in other countries, and was taken from one of the most reliable sourcesof this type of data that is available in Spain.27 Finally, a further limitation of the analyseswas that in modeling the process of care of hypercholesterolemic patients, we assumeda clinical practice recommendation basedon consensus,but evidence of 1933
CLINICAL
its adoption by general practitioners is lacking.40 However, previous recommendations on the managementof hypercholesterolemia had been broadly adopted.41 Our selection of the effectiveness measure was determined by the lack of studies on cardiovascular end points for all statins and doses. Despite the above-mentioned limitations, this cost-effectiveness analysis can be helpful in guiding clinical decisionsfor selected hypercholesterolemic patients needing drug treatment with statins. Our resultsseemto agreewith reports by other authors.13,14,19*20 Further researchis needed to assessthe comparative effectivenessof statins in primary and secondary prevention of cardiovascular events. The economic impact of suchtherapy requiresfurther assessment of the efftciency of clinical decisionson prevention of cardiovascular morbidity and mortality. ACKNOWLEDGMENTS We thank Drs Ricard Tresserrasand LourdesRibas for providing the necessarydata from the Catalan Nutritional Survey.21 Address correspondence to: Albert Cobos, MD, Tarragona 84, 2” la, 08015 Barcelona, Spain. REFERENCES 1. Law MR, Wald NJ, Thompson SG. By how much and how quickly does reduction in serum cholesterol concentration lower risk of ischaemic heart disease? Br Med J. 1994;308:367-372. 2. Holme I. Cholesterol reduction and its im-
pact on coronaryartery diseaseand total mortality.AmJ Cardiol. 1995;76: lOC-17C. 1934
THERAPEUTICS”
3. Levine GN, Keaney JN, Vita JA. Cholesterol reduction in cardiovascular disease. N Engl J Med. 1995;332:512-521. 4. Gould AL, Rossouw JE, Santanello NC, et al. Cholesterol reduction yields clinical benefit: A new look at old data. Circulation. 1995;91:2274-2282. 5. National Cholesterol Education Program. Second report of the Expert Panel on Detection, Evaluation, and Treatment of High Blood Cholesterol in Adults (Adult Treatment Panel II). Circulation. 199489: 1329-1345. 6. Pyorala K, De Baker G, Graham I, et al, on the behalf of the Task Force. Prevention of coronary heart disease in clinical practice. Recommendations of the Task Force of the European Society of Cardiology, European Atherosclerosis Society, and European Society of Hypertension. Eur Heart J. 1994;15:1300-1331. 7. Sociedad Espaiiola de Medicina Intema y Liga para la Lucha contra la Hipertension Arterial. Recomendaciones para la prevencion primaria de la enfermedad cardiovascular. Clin Invest Arterioscler. 1994;6:62-102. 8. Estudio Descriptive de 10s Planes de Salud en Espafia (Descriptive study of the Health Plans in Spain). Madrid: Ministerio de Sanidad y Consumo, Direction General de Alta Inspection y Relaciones Institucionales; 1994. 9. LA Declaracid de Woria Sobre Salat Cardiovascular. Barcelona: Departament de Sanitat i Seguretat Social; 1995. 10. L.a Declaracid de Catahnya: Invertir en Salut Cardiovascular. Barcelona: Departament de Sanitat i Seguretat Social; 1996. 11. De Abajo FJ, Madurga M, Montero D, et al. Trends in the supply and use of lipid-
A. COBOS ET AL.
lowering drugs in Spain, 1983 through 1991. Therupie. 1993;48: 145-149. 12. IMS. The Pharmaceutical Market, Spain. IMS: Iberica S.A., Spain; 1997. 13. Blum CB. Comparison of properties of four inhibitors of 3-hydroxy-3-methylglutaryl-coenzyme A reductase. Am .I Cardial. 1994;73:3D-LlD. 14. Martens LL, Guibert R. Cost-effectiveness analysis of lipid-modifying therapy in Canada: Comparison of HMG-CoA reductase inhibitors in the primary prevention of coronary heart disease. Clin Ther. 1994;16:1052-1062. 15. Morris S, McGuire A, Caro J, et al. Strategies for the management of hypercholesterolemia: A systematic review of the costeffectiveness literature. J Health Serv Res Policy. 1997;2:23 l-250. 16. Jacobson TA. Cost-effectiveness of 3-hydroxy-methylglutaryl-coenzyme A (HMGCoA) reductase inhibitor therapy in the managed care era. Am J Cardiol. 1996; 7832-41. 17. Morris S, McGuire A, Hogan T. The economics of low density lipoprotein cholesterol reduction with fluvastatin sodium. Rev Contemp Pharmacother. 1996;7: 191-201. 18. Hsu I, Spinler SA, Johnson NE. Comparative evaluation of the safety and efficacy of HMG-CoA reductase inhibitor monotherapy in the treatment of primary hypercholesterolemia. Ann Pharmacother. 1995;29: 743-759. 19. Spearman ME, Summers K, Moore V, et al. Cost-effectiveness of initial therapy with 3-hydroxy-3-methylglutaryl coenzyme A reductase inhibitors to treat hypercholesterolemia in a primary care setting of a managed-care organization. Clin Ther. 1997;19:582-599.
20. Kong SX, Gandhi SK, Crawford Pharmacoeconomic evaluation CoA reductase inhibitors in the of hypercholesterolemia. Am J Cure. 1996;2: 1055-1073.
SY, et al. of HMGtreatment Managed
21 Serra LI, Ribas L, Garcia R, et al. Avuluacid de PEstat Nutritional de la Poblacid Catalana (1992-93). (la edicio). Barcelona: Departament de Sanitat i Seguretat Social. Generalitat de Catalunya; 1996:27-31, 78-80. 22. Clarke R, Frost C, Collins R, et al. Dietary lipids and blood cholesterol: A quantitative meta-analysis of metabolic ward studies. Br Med J. 1997;314:112-117. 23. Rothschild V, Logothetis N. Probability Distributions. New York: John Wiley & Sons; 1986:62-63. 24. Kong SX, Crawford SY, Gandhi SK, et al. Efficacy of 3-hydroxy-3-methylglutaryl coenzyme A reductase inhibitors in the treatment of patients with hypercholesterolemia: A meta-analysis of clinical trials. Clin Ther 1997;19:778-797. 25. Delea TE, Hogan TJ, Gasper MC, et al. Cost-effectiveness of cholesterol-lowering drugs: A re-examination of treatment alternatives. Value Health. In press. 26. Simons A, Levis G, Simons J. Apparent discontinuation rates in patients prescribed lipid-lowering drugs. Med J Allina. 1996;164:208-211. 27. Gisbert R, Brosa M. Base de datos de castes sanitarios (v 1.0) [database on CDROM]. Barcelona, Spain: Soikos, 1998. 28. Boletin Mensual de Estadfstica. Madrid: Instituto National de Estadistica; August-September 1997. 29. Indicadores de la Prescripcidn Farmaceutica en el Sistema National de Salud. Vol. 15, Enero-Junio de 1997. Madrid:
1935
CLINICAL
Subdireccidn General de Coordinacio’n Administrativa, Institute National de Salud; 1997. 30. Regazzi MB, Icona I, Campana C, et al. Altered disposition of pravastatin following concomitant drug therapy with cyclosporin A in transplant recipients. Transplant Proc. 1993;25:2732-2734. 31. Kantola ‘I, Kivisto KT, Neuvonen PJ. Gmpefruit juice greatly increases serum concentrations of lovastatin and lovastatin acid. Clin Pharmacol Ther. 1998;63:397-2. 32. Neuvonen PJ, Kantola T, Kivisto KT. Simvastatin but not pravastatin is very susceptible to interaction with the CYP3A4 inhibitor itraconazoIe. Clin Pharmacol Ther. 1998;63:332-341. 33. Nordin C, Dahl ML, Eriksson M, et al. Is the cholesterol-lowering effect of simvastatin influenced by CYP2D6 polymorphism? L.uncet. 1997;350:29-30. 34. Sheldon TA. Problems of using modelling in the economic evaluation of health care. Health &on. 1996;56: l-l 1. 35. O’Connor PJ, Rush WA, Trence DL. Relative effectiveness of niacin and lovastatin for treatment of dyslipidemias in a
1936
THERAPEUTICS”
health maintenance organization. Pratt. 1997;44:462467.
J Fam
36. Fumarga EM. Pharmacist management of a hyperlipidemia clinic. Am J Hosp Pharm. 1993;50:91-95. 37. Kruse W, Schlierf G. Monitoring compliance in clinical trials. Lkzncet. 1990;335: 803-804. 38. Macia R, Pena A, Marrugat J, et al. High prevalence of cardiovascular risk factors in Gerona, Spain, a province with low myocardial infarction incidence. J Epidemiol Community Health. In press. 39. Illingworth DR, Erkelens DW, Keller U, et al. Defined daily doses in relation to hypolipidaemic efficacy of lovastatin, pravastatin and simvastatin. Lancet. 1994; 343:1554-1555. 40. Masana L, Pedro-Botet J, Roe E, et al. Impact0 de1 consenso para el control de la colesterolemia en Espafia. Med C&z. 1998; 110:598-599. 41. Brotons C, Server M, Pinto X, et al. Impacto de1 consenso para el control de la colesterolemia en Espatia. Med Clin. 1996; 108:9-15.
THERAPEUTICSVVOL.
2 1, NO. 11,1999
Which Statin Is Most Efficient for the Treatment of Hypercholesterolemia? A Cost-Effectiveness Analysis Albert Cobos, MD, PhD,’ Albert J. Jove& MD, MSc, PhD? Anna Garcia-Alt&,2 Reina Garcia-Closas, MD, MSc, PhD,3 and Lluis Serra-Majem, MD, MSc, Phil ‘Novartis Farmactfutica, SA, 2Ag&nciad’Avaluacid de TecnologiaMdica, Servei Catalci de la Salut, Department de Sanitat i Seguretat Social, Barcelona, and 3Research Group on Community Nutrition, Universities of Las Palmas de Gran Canaria, La Laguna, and Barcelona, Spain
ABSTRACT A review of the cost-effectivenessliterature indicated that the hydroxymethylglutaryl coenzymeA-reductase inhibitor fluvastatin is more cost-effective for achieving minorto-moderate reductions in low-density lipoprotein cholesterol(LDL-C) levels than 3 other statins: lovastatin, pravastatin, and simvastatin. The main goal of this study was to verify the applicability of theseconclusions to Spanish health care costs and patternsof resourceconsumptionrelated to the treatment of hypercholesterolemia.A stochastic simulation model was used to predict both the costs and effects of treating high-risk hypercholesterolemicpatients with fluvastatin, lovastatin, pravastatin, or simvastatin.Epidemiologic data were used to find a suitable theoretic probability distribution model for baselineLDL-C values Accepted for publication Printed in the USA. Reproduction
1924
in whole
August
8, 1999.
or part is not permitted.
in high-risk hypercholesterolemicpatients. The modelwasthenusedto generate10,000 random observations of baseline LDL-C values; the corresponding LDL-C values after a 2-year treatment period were predicted as a function of the baselinevalue and the percentagereduction expectedwith a particular statin and dose, according to the resultsobtainedin 2 me&analyses.The probability of treatment discontinuation wasalsotaken into accountusingestimates obtained in usual practice. The effects of treatmentwere expressedasthe rate of successin achieving the goal level of LDL-C, as defined in the current Spanish recommendationsfor the treatment of hypercholesterolemia. The average costs of treatment were computed from both the social and public-financing perspectives,including the cost of lipid-lowering drugs, physician visits, laboratory tests, and days off work, as appropriate. The occurrence of nonscheduledvisits and workdays lost becauseof side effects were taken into account to computeindirect costsrelevant to 0149.2918/99/$19.00
A. COBOS ET AL.
the social perspective. The potential costs of treatingsideeffectswere ignored.A costeffectiveness analysis was performed to compare the cost-effectiveness ratios obtained with eachof the 4 statinsconsidered in this study. Model-based predictions of the effects, total costs,and cost-effectiveness ratios were made. Cost-effectivenessratios were interpretedas the cost per patient meetingthe goal of therapy, according to current Spanish recommendations. The data showed that fluvastatin had the lowestcost-effectiveness ratioswhenLDL-C levels required reduction to ~25% of baseline levels. In this situation, fluvastatin was more cost-effective than lovastatin, pravastatin, or simvastatinfrom public-financing and socialperspectives.Key wotrik: cholesterol, health economics,cost-effectiveness analysis,modeling.
INTRODUCTION High levels of blood cholesterolhave been associatedwith an increasedrisk of cardiovascular mortality and morbidity,le3 and cholesterolreduction hasproved beneficial no matter what type of intervention is adopted.1*4Several recommendationsfor the prevention of cardiovascular events have chosencholesterolreduction asa main target.5-7Health care plans have included cholesterolreduction in healthy populations amongtheir goals.8FVoposedmeansof prevention of cardiovascular diseasethrough cholesterolreduction have included dietary changes,exercise,and medications.9,10 Spanish trends indicate increasing use of lipid-lowering drugs.” According to IMS data,‘* statins accountedfor 80% of expendituresfor lipid-lowering drugs prescribed in 1997 (31,085 million pesetas [PTA]), which wasa 22.35%increasecompared with 1996.This cost scenarioshould
cause decision makers and prescribers to seriouslyconsiderthe cost-effectivenessof the different statins as a guiding criterion for determining treatment. In reviewing the published cost-effectiveness studies that compared simvastatin, pravastatin, lovastatin, and fluvastatin,13-*0we concluded that fluvastatin was the preferred alternative for patients needing minor-to-moderate reductions in low-density lipoprotein cholesterol (LDL-C). However, most of these studies were carried out in the United States, and none were conducted in Spain. Consequently, we wanted to determine if the sameresults were valid when using costs and resource consumption patterns in a hypercholesterolemic population in Spain. The aim of this study was to assessthe cost-effectivenessratio of different statins in the clinical management of high-risk hypercholesterolemic patients who need reductions of ~25% of baseline LDL-C levels. A stochastic simulation model was developed and usedto assess costsand effectiveness during a 2-year treatment period with each of the 4 statins considered.
MATERIALS
AND METHODS
The risk of coronary heart diseasehasbeen defined using an algorithm recommended by 3 medical societiesin Spain.7The algorithm takes into account total cholesterol level and the number of other risk factors (ie, age >45 years for men, >55 years for women; smoking; hypertension; diabetes;sedentarylifestyle) and classifies individuals into 1 of 3 categories of cardiovascular risk: mild, moderate, or high (Figure 1). Subjects were treated with drugs if their LDL-C values measuredafter dietary treatment were higher than the drug treatment threshold. 1925
CLINICAL
Total Cholesterol OxW)
Other Risk Factors
Global CV Risk Mild
Moderate
Goal LDL-C Owl)
175
THERAPEUTICS”
Drug Therapy Threshold (w/W 190
155
180
135
160
Figure 1. Hypercholesterolemiatreatmentalgorithm recommendedby 3 Spanishmedicalsocieties.’ Patientshaving low-density lipoprotein cholesterol(LDL-C) levels after dietary treatment that are above the drug-treatmentthresholdshouldbe treated with lipid-lowering drugs until the goal LDL-C level is reached.CV = cardiovascular. The distribution of LDL-C levels in highrisk patientsneedingdrug therapy wasmodeledusingdata from the CatalanNutritional Survey (CNS).21This survey was designed to investigate the dietary habits of a random sampleof 2346 Catalonian subjects; 893 of them consented to a biochemical test. Data on serumlipids and major nonlipid risk factors were available from 888 subjects.Basedon the algorithm described in Figure 1, those patients with a serum cholesterolconcentration >200 mg/dL and a serum triglyceride concentration x200 mg/dL were classified by cardiovascular risk category (mild, moderate, high) according to the presenceor absenceof major nonlipid risk factors. Subjects with LDL-C values above the goal indicated in Figure 1 shouldbe treated with diet modification before drug treatment.We assumed that dietary intervention would reduce LDL-C values by 1O%.22After we applied this 10% reduction to simulatepostdietary LDL-C values,subjectsin the high-risk category who fulfilled the criteria for drug therapy were selectedfor the analysis,and these “postdiet” LDL-C values were consideredto be baselinevalues. 1926
The Pareto distribution23 is a suitable theoretic probability distribution for modeling baseline LDL-C values. The lower boundary of this distribution was setto 160 mg/dL by the algorithm, making this the value of the truncation parameter K. The maximum likelihood estimatefor the second parametercx was obtained as N (z In xi - N In K)-l, where xi is the ith observed value (i = 1, 2, .... N), K is the truncation parametertaking the value 160, and addition is taken over i. The fitted Pareto distribution wasthen usedto generaterandom observationsof baselineLDL-C values in high-risk casesneedingdrug therapy, until 10,000casesneedingreductionsin LDL-C values of ~25% were achieved. In the selected cases,a 2-year treatment period with statins was simulated and the results assessedusing a cost-effectivenessanalysis. Simulation of Therapeutic Processand Effectiveness Data Low, medium, and high doses were consideredfor all statins- 10, 20, and 40 mg for simvastatin and pravastatin and 20, 40, and 80 mg for lovastatin and flu-
A. COBOS ET AL.
vastatin. When a particular dose was not marketed in Spain, we defined the high dose as 2 medium-dose tablets. The number and sequence of physician visits was simulated following current Spanish recommendations7 This consensus report recommends physician visits after 1 month; after 6 weeks; and after 3, 6,9, 12, 18, and 24 months from the start of treatment. At the 6-week visit the effect of therapy was assessed by comparing the simulated LDL-C value with the value defining the goal of therapy. In case of therapeutic success, treatment was maintained at the low dose for the whole simulation period. In case of therapeutic failure, the low dose was titrated to the medium dose. A tolerance margin of 5 mg/dL was allowed in assessing the effectiveness of therapy to avoid doubling the dose when LDL-C values were higher but were close to the treatment goal. A similar strategy was followed at the third visit. In case of therapeutic failure with the medium dose, the highest dose was given and maintained for s2 years. For any of the statins at any of the doses considered, the effect of therapy was simulated as the LDL-C value achieved through the model: yijk = x,(1 - l.+j/lOO) + (oiilOO)ek where yijk is the LDL-C value obtained after treatment with statin i (i = fluvastatin, lovastatin, pravastatin, simvastatin) at dose j (j = low, medium, high) in subject k; xk is the baseline LDL-C value in subject k; l~,~.is the percentage reduction expected wit4 the ith statin at the jth dose; ati is the SD of the percentage reduction in LDL-C obtained with the ith statin at the jth dose; and ek is an N(O,l) random variable representing a “subject” effect.
For any subject k, the LDL-C value generated by this model (y,,) would have an expected value equal to the baseline value of subject k affected by the reduction expected for each statin and dose. Also, the subject effect term ensured that the response for subject k would always be in the same percentile for all statins and doses (ie. good or bad responders would be equally good or bad for all statins and doses). Estimates for p,ij and ui. were obtained from 2 recently publishe d meta-analyses on the efficacy of statins.24,25 In some cases, the estimates themselves were not reported, but we were able to compute them from information provided in the publication or from the original publication. Also, the model considered whether compliance with therapy influenced the effectiveness and cost of treatment. To do so, a Bernouilli random variable was generated to indicate compliance with therapy, independent of effectiveness. The probability of discontinuation was estimated as 60% during the first year of treatment and was distributed as 25% at the first month and 50% within the first 3 months, with the remaining 25% being uniformly distributed up to the end of the first year. 26We assumed that lack of compliance was uniformly distributed within each interval between adjacent visits. Lack of compliance was classified as therapeutic failure, and LDL-C values in these cases were set back to baseline levels. Cost Data
We adopted2 perspectivesfor the costeffectiveness analyses:the social perspective (direct and labor costs) and the public-financing perspective (direct costs incurred by the health care system). The 1927
CLINICAL THERAPEUTICS”
following direct costs were considered: scheduled physician visits, unscheduled physician visits because of side effects, lipid-lowering drugs, and laboratory tests used to monitor efficacy and safety of therapy. Unscheduled visits were simulated by generating a Poisson random variable independent of any other simulation event with an intensity of 0.05, 0.34, 0.34, and 0.26 for fluvastatin, lovastatin, pravastatin, and simvastatin, respectively. l9 We assumed that laboratory tests were done at each of the scheduled and unscheduled visits. We did not consider the cost of treating side effects, since they are extremely rare and tend to resolve spontaneously after therapy is discontinued. Drug costs were estimated using mean retail prices for all the forms marketed in Spain.12 Estimates for the unit cost of doctors’ visits and laboratory tests were obtained from a specific health care cost database and fixed at 2.093 and 4.342 PTA, respectively.27 Labor or indirect costs considered in the analysis were computed using the time spent at physician visits and the days of work missed because of side effects. According to published results,19 the number of workdays missed because of side effects was simulated using an independent Poisson random variable with an intensity of 0.11, 0.33, 0.14, and 0.23 for fluvastatin, lovastatin, pravastatin, and simvastatin, respectively. The cost per hour of work missed was estimated from a survey of salaries and services and fixed at 1439 PTA.28 We assumed a loss of 2 hours of work for each visit and 8 hours for each working day. A discount rate of 5% was applied to all costs incurred in the second year of treatment. All costs were expressed in Spanish currency using 1997 values. Currently US $1 equals 150 PTA. 1928
Cost-Effectiveness Analysis Effectiveness was defined as the number of cases meeting the goal of therapy in the 10,000 cases simulated. The analysis carried out from the social perspective included all costs incurred during the 2-year treatment period. The analysis carried out from the health care system or publicfinancing perspective only (including direct costs and drug costs) was estimated according to a patient’s employment status. An active worker pays 40% of the cost of the drug, while a pensioner (a retired worker or a patient receiving disability payments) pays nothing. The percentage of units prescribed to pensioners within the publicly financed health care system was estimated to be 77.53%, according to published data.29 This proportion was used as the parameter of a Bernouilli random variable defining each simulated patient’s health care affiliation status. For each selected perspective, all relevant costs incurred in the 2-year treatment period were used to estimate the costeffectiveness ratios. Costs for the first year of treatment included the dose-titration phase and the specific schedule of visits. Cost-effectiveness ratios were estimated as the cost per patient meeting the therapeutic goal. Incremental analyses were performed and interpreted when appropriate. Several one-way sensitivity analyses were performed to test the robustness of the primary analysis to changes in major assumptions and model parameters. Thus the discontinuation rate was set at 40% and 0%. The intensities of the Poisson variables simulating the number of additional visits and workdays missed because of side effects were set to their corresponding average values to
A. COBOS ET AL.
A
B 1.0
l.O0.90.8-
0.9 0.8 0.7 0.6
b 0
0.70.6-
3 0.5.0 .g 0.4$
i;j/,
,
,
(
,
,
,
,
,
0.30.2-
,
160170
180190200210220230240250
Baseline
LDL-C
,
(
,
(
,
,
,
,
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
Pareto CDF
Figure 2. Fit to a Pareto distribution (dashedline) of the baselinelow-density lipoprotein cholesterol (LDL-C) levels of 53 patients categorized as being at high cardiovascular risk and needing drug therapy (circles). (A) Cumulative distribution function (CDF); (B) quantile-quantile plot. carry out an equal data analysis for all 4 statins. In addition, a 20% deviation in the unit cost of drugs, physician visits, and laboratory tests was considered for sensitivity analyses.Finally, data were analyzed using discount ratesof 3% and 0%.
RESULTS When the algorithm proposedby the Spanish recommendations7was applied to the 888 casesof the CNS*l with available data on lipid profiles, 235 had high LDL-C values such that dietary intervention was indicated.After applying a 10%reduction to the LDL-C values to simulatethe effect of diet, 58 of the 888 caseshad LDL-C values above the drug-treatment threshold. This correspondsto 6.5% of the sample, with a 95% CI from 4.9% to 8.1%. No patient was in the low-risk category, 5 were in the moderate-risk category, and 53 were in the high-risk category. Further analysesfocused only on these 53 highrisk cases.
Figure 2 depicts the fit of a Pareto distribution to the baselineLDL-C values of the 53 high-risk casesneeding drug therapy. Both the cumulative distribution function andthe quantile-quantileplots showed a good fit of the data to the Pareto model. TableI showsthe 2 setsof estimatesused for the parameterspij and uij in our simulation model.Theserepresentthe meanand SD of the percentagereductionsof LDL-C values achieved with different statins and doses.Although mean percentage reduction estimateswere similar in both sets,SD estimateschanged considerably (with the exception of lovastatin) dependingon the source of the data. In general, these estimateswere much lower in the meta-analysis of Delea et a125than in the estimatesof Kong et al.M The difference between estimateswas more evident for simvastatin. Average retail prices were used as cost data for the drugs marketed in Spain (Table II). Annual costs for the therapy rangedfrom 38,560 PTA for fluvastatin 20 mg to 145,531 PTA for simvastatin40 mg. 1929
CLINICAL THERAPEUTICS”
Table I. Estimates of the mean (* SD) percentage reduction of low-density cholesterol induced by 4 statins at the doses used in the simulation.
DW Fluvastatin
Lovastatin
Pravastatin
Dose (mg) 20 40 80* 20 40
80 10 20 40
Simvastatin
10 20 40
lipoprotein
Kong et alz4
Delea et a125
f 12.1 f 11.0 + 10.9 24.9 + 11.7 30.4 f 11.1 39.8 * 11.1 19.3 + 17.1 26.1 zt 12.5 27.2 k 10.8 28.6 + 19.9 34.4 f 15.7 40.7 + 21.9
19.9 24.3
21.0 23.1 33.1
-
zt 2.7 f 5.7
25.6 zt 8.8 31.9 + 10.7 38.8 + 14.9
18.8 + 7.4 + 4.0 f 5.0 28.3 f. 3.2 32.4 + 5.0 42.2 ztz6.4 26.9 30.8
*Estimates for fluvastatin 80 mg were taken from Table 2 in Kong et al (or from references therein) and were used to complete the set of estimates from Delea et al used in the simulation. The meta-analysis by Delea et al did not include fluvastatin 80 mg.
Effectiveness and maximum dose reached with each statin after the initial dose-titration period are shown in Table III. When estimates reported in the metaanalysis by Delea et a125were used, the highest success rate was achieved with simvastatin, followed by pravastatin, fluvastatin, and lovastatin. All fell between 36% and 40%. However, this ranking changed remarkably when data from Kong et a124were used. In this case, lovastatin was the most effective, followed by fluvastatin, simvastatin, and pravastatin. The total costs, cost-effectiveness ratios, and incremental ratios obtained in the analyses are displayed in Table IV. Fluvastatin was the least expensive and most cost-effective drug from both the social and public-financing perspectives, independent of the meta-analysis data used to estimate pij and crij. Consequently, incremental ratios were computed taking fluvastatin as the reference value for those 1930
Table II. Average retail prices (including value-added tax)* of drug doses marketed in Spain. Dose
DW Fluvastatin Lovastatin
Pravastatin Simvastatin
Costs (Pesetas)+ Annuals
0s)
UniG
20 40
2958 4349
38,560 56,692
20 40
4160 665
54,229 86,701
10
1 4160
20
5978
54,229 77,928
4132 5874 11,164
53,864 76,572 145,531
10 20 40
*These prices were used as cost estimates in the simulation. +IMS, 1997.‘* *Unit costs are the average prices of 1 package; all packages contain 28 tablets. 5Annual cost is based on the assumption that all drugs are prescribed as a 1 tablet/d regimen.
A. COBOS ET AL.
alternatives showing higher effectiveness but costing more. The sensitivity analyses produced similar results. Fluvastatin was always the least expensive alternative. The difference in efficacy compared with the more efficacious therapeutic alternative never exceeded 4% for credible dropout-rate Table III. Effectiveness
scenarios (60% or 40%), or 6% if we assumed a 0 dropout rate. When we used estimates from the meta-analysis of Delea et al,25 fluvastatin always showed the lowest cost-effectiveness ratio despite the higher effectiveness of simvastatin and pravastatin, with incremental ratios ranging, respectively, from 225,734 to 428,423
and maximum dose reached. Dose (%)*
Source Delea et a125
Kong et alN
E (%I
Low
Medium
High
Fluvastatin Lovastatin Pravastatin Simvastatin
37.01
81 83 62 99
9 9 37
10
36.59
Fluvastatin Lovastatin Pravastatin Simvastatin
36.62 38.56 32.59
DWZ
39.19 39.37
0
64 75 56 72
34.21
8 1
1
7 13 21 14
29 12 23 14
E = effectiveness as the percentage of treatment successes in the simulated cohort of 10,000 treated patients. *Dose is the maximum dose reached in the titration period (see Methods for corresponding values in milligrams.)
Table IV. Costs (total cost per patient), cost-effectiveness ratios (C/E), and incremental ratios of total costs to the number of cases reaching the goal of therapies (DJD,) for the base-case analysis. Social Perspective Source*
=‘w
costs (Pesetas)
C/E
Delea et a125 Fluvastatin 86,531 Lovastatin 102,37 1 Pravastatin 106,158 Simvastatin 96,489
233,804 279,778 270,880 245,084
Kong et a124 Fluvastatin Lovastatin Pravastatin Simvastatin
266,481 271,428 369,364 298,419
97,585 104,663 120,376 102,089
Public-Financing Perspective costs (Pesetas)
C/E
68,326 Dominated* 8 1,809
184,616 223,584
Dominated+
85,610 76,752
218,449
792,833 357,016
78,536 83,940
214,462 217,686 302,705 239,385
DC4
900,322 421,984
364,798
Dominated+ 98,65 1 Dominated+ 8 1,894
*Source of the estimates for the pij and aij parameters in the simulation model. tAn alternative is considered dominated when another lower-cost, higher-effectiveness
194,950
alternative
DcDs
278,542
Dominated+ Dominated+ exists.
1931
CLINICAL
PTA and 573,624to 951,405 PTA from the public-financing perspective and from 262,609 to 506,390 PTA and 642,269 to 1,080,397PTA from the socialperspective. With the estimatesof Kong et al,24fluvastatin was again the alternative with the lowest cost-effectivenessratio in all but 2 scenarios.In this sense,only when we assumeda 0 dropout rate and also an equal mean number of additional visits and workdays missedbecauseof side effects, lovastatin was the most cost-effective option, with incrementalratios from 125,872 to 212,173 PTA, depending on the perspective, comparedwith fluvastatin.
DISCUSSION
AND CONCLUSIONS
The results of these analyses show that fluvastatin is more cost-effective than lovastatin, pravastatin, and simvastatin from both a public-financing and social perspective, using cost data from Spain. Earlier cost-effectiveness analyses also found fluvastatin to be the most costeffective statin for patients needingminorto-moderate LDL-C reduction.13~14*1~20 Lovastatin had more favorable costeffectiveness ratios than fluvastatin in only 2 of the 32 scenariosconsidered in our sensitivity analysis. The 2 scenarios were characterized by perfect compliance (0% dropout rate) and an equal mean number of extra visits and workdays missedbecauseof side effects. A 0 dropout rate seemsunrealistic, and previous studies have not confirmed the hypothesis of equality.i9 Although the different statins are often considered equal in terms of safety and tolerability, there is some evidence that fluvastatin is the least likely to interact with drugs metabolized by the cytochrome P-450 3A4 isoenzyme.30-33 1932
THERAPEUTICS”
Several limitations should be kept in mind when interpreting the results of our analyses.First, they do not include other marketed statins, such as atorvastatin and cerivastatin. Although this is an important limitation, the lack of meta-analysesincluding all 6 statinsprevented us from including them in the analyses. Second,there are limitations associated with the method and with the source of data. Limitations associatedwith modeling have been pointed out elsewhere.34To overcome most of these limitations and the potential bias of model assumptions, we used the results of 2 published metaanalysesas estimatesfor the relevant parametersof our simulation model. In addition, a one-way sensitivity analysis was performed modifying the values of major influential parameters,including the rate of compliance, which varied acrossstudies 26~35-37 and the discount rate. Although the primary studies included in the 2 meta-analyseswere not conducted in Spain, it is likely that their primary results (ie, percentagereduction of LDL-C values) are applicable to Spanish hypercholesterolemic patients. Similar data from Spanishpatients are not available in the literature. A further limitation in the method is that our model ignored the potential costs of treating the side effects of statins. Unfortunately, the data neededto model the incidence of side effects as well as their likely treatments are not available in a comparableformat for the different statins (the incidence of adverseevents is not addressedin the only 2 meta-analyses we found comparing statins). However, statins are known to have a particularly good safety profile, with rare and mild sideeffects that typically ceasewhen therapy is stoppedand that do not require spe-
A. COBOS ET AL.
cific treatment. Thus the possibility of a single expensive adverse event that outweighs savings does not seemlikely. A third type of limitation is associated with the quality of the data sources.The fmt of theselimitationsinvolves the LDL-C baseline data used to define the population at high risk for cardiovascular events. Becausethe CNS provided data only from subjects who were followed and consented to a biochemical test, a selection bias could be present due to over-representation of a higher-risk population. However, in a recent investigation conducted in 1 of the 4 provinces of Catalonia,38in which nonresponseswere much more uncommon than they were in the CNS, the mean levels of total cholesterol were higher for all age groups and both sexesthan they were in the CNS. This is inconsistent with the presenceof an important selection bias toward higher-risk cases affecting CNS data. Also, in our analyses,we assumedthat the profile risks of the Catalan population could be extrapolated to the rest of Spain. Another issueto consider in our simulations concerns the variability of the results achieved; these findings depended on the meta-analysisestimatesapplied to our model parameters.The fact that 1 of the meta-analysesfocused on primary hypercholesterolemia25 and the other included both primary and secondary hypercholesterolemia24might be the cause of their different results, along with differences in study selection criteria. Furthermore, the ranking of the 4 statins according to the percentage of successful treatments obtained with the estimatesof Delea et a125was consistent with previous findings regarding the relative potency of these statins, while this was not the case with the estimatesof Kong
et a1.24Moreover, the dose distribution produced by the estimatesin the study by Delea et al closely reflects market data,‘* where the 20-mg form of fluvastatin accounted for >80% of the total number of units sold; again, this is not the casewith the estimatesof Kong et al. In short, both the effectiveness and drug-consumption patterns obtained with our simulation model were consistent with external data when the model was fed estimatesfrom Delea et al but not when estimatesfrom Kong et al were adopted. The influence that SD estimates of the treatment response had in our simulation results showsthe importance of the estimatesof treatment responsevariability; however, few efficacy studies have reported this kind of data. Dropouts may have a major impact on the resultsof cost-effectivenessanalyses.In our model,the dropout rate was setat 60%, according to data obtained from current clinical practice.26*35*36 The model assumed independencebetween treatment response and treatment discontinuation,which is an unlikely assumption, but data needed to model compliance as a function of effectiveness were unavailable. On the other hand, there is no evidence that treatment with different statins resulted in different ratesof treatment discontinuation.26,36 The cost data used in the analysis were intended to reflect the cost of resource consumption in Spain, which is not necessarily similar to the costs in other countries, and was taken from one of the most reliable sourcesof this type of data that is available in Spain.27 Finally, a further limitation of the analyseswas that in modeling the process of care of hypercholesterolemic patients, we assumeda clinical practice recommendation basedon consensus,but evidence of 1933
CLINICAL
its adoption by general practitioners is lacking.40 However, previous recommendations on the managementof hypercholesterolemia had been broadly adopted.41 Our selection of the effectiveness measure was determined by the lack of studies on cardiovascular end points for all statins and doses. Despite the above-mentioned limitations, this cost-effectiveness analysis can be helpful in guiding clinical decisionsfor selected hypercholesterolemic patients needing drug treatment with statins. Our resultsseemto agreewith reports by other authors.13,14,19*20 Further researchis needed to assessthe comparative effectivenessof statins in primary and secondary prevention of cardiovascular events. The economic impact of suchtherapy requiresfurther assessment of the efftciency of clinical decisionson prevention of cardiovascular morbidity and mortality. ACKNOWLEDGMENTS We thank Drs Ricard Tresserrasand LourdesRibas for providing the necessarydata from the Catalan Nutritional Survey.21 Address correspondence to: Albert Cobos, MD, Tarragona 84, 2” la, 08015 Barcelona, Spain. REFERENCES 1. Law MR, Wald NJ, Thompson SG. By how much and how quickly does reduction in serum cholesterol concentration lower risk of ischaemic heart disease? Br Med J. 1994;308:367-372. 2. Holme I. Cholesterol reduction and its im-
pact on coronaryartery diseaseand total mortality.AmJ Cardiol. 1995;76: lOC-17C. 1934
THERAPEUTICS”
3. Levine GN, Keaney JN, Vita JA. Cholesterol reduction in cardiovascular disease. N Engl J Med. 1995;332:512-521. 4. Gould AL, Rossouw JE, Santanello NC, et al. Cholesterol reduction yields clinical benefit: A new look at old data. Circulation. 1995;91:2274-2282. 5. National Cholesterol Education Program. Second report of the Expert Panel on Detection, Evaluation, and Treatment of High Blood Cholesterol in Adults (Adult Treatment Panel II). Circulation. 199489: 1329-1345. 6. Pyorala K, De Baker G, Graham I, et al, on the behalf of the Task Force. Prevention of coronary heart disease in clinical practice. Recommendations of the Task Force of the European Society of Cardiology, European Atherosclerosis Society, and European Society of Hypertension. Eur Heart J. 1994;15:1300-1331. 7. Sociedad Espaiiola de Medicina Intema y Liga para la Lucha contra la Hipertension Arterial. Recomendaciones para la prevencion primaria de la enfermedad cardiovascular. Clin Invest Arterioscler. 1994;6:62-102. 8. Estudio Descriptive de 10s Planes de Salud en Espafia (Descriptive study of the Health Plans in Spain). Madrid: Ministerio de Sanidad y Consumo, Direction General de Alta Inspection y Relaciones Institucionales; 1994. 9. LA Declaracid de Woria Sobre Salat Cardiovascular. Barcelona: Departament de Sanitat i Seguretat Social; 1995. 10. L.a Declaracid de Catahnya: Invertir en Salut Cardiovascular. Barcelona: Departament de Sanitat i Seguretat Social; 1996. 11. De Abajo FJ, Madurga M, Montero D, et al. Trends in the supply and use of lipid-
A. COBOS ET AL.
lowering drugs in Spain, 1983 through 1991. Therupie. 1993;48: 145-149. 12. IMS. The Pharmaceutical Market, Spain. IMS: Iberica S.A., Spain; 1997. 13. Blum CB. Comparison of properties of four inhibitors of 3-hydroxy-3-methylglutaryl-coenzyme A reductase. Am .I Cardial. 1994;73:3D-LlD. 14. Martens LL, Guibert R. Cost-effectiveness analysis of lipid-modifying therapy in Canada: Comparison of HMG-CoA reductase inhibitors in the primary prevention of coronary heart disease. Clin Ther. 1994;16:1052-1062. 15. Morris S, McGuire A, Caro J, et al. Strategies for the management of hypercholesterolemia: A systematic review of the costeffectiveness literature. J Health Serv Res Policy. 1997;2:23 l-250. 16. Jacobson TA. Cost-effectiveness of 3-hydroxy-methylglutaryl-coenzyme A (HMGCoA) reductase inhibitor therapy in the managed care era. Am J Cardiol. 1996; 7832-41. 17. Morris S, McGuire A, Hogan T. The economics of low density lipoprotein cholesterol reduction with fluvastatin sodium. Rev Contemp Pharmacother. 1996;7: 191-201. 18. Hsu I, Spinler SA, Johnson NE. Comparative evaluation of the safety and efficacy of HMG-CoA reductase inhibitor monotherapy in the treatment of primary hypercholesterolemia. Ann Pharmacother. 1995;29: 743-759. 19. Spearman ME, Summers K, Moore V, et al. Cost-effectiveness of initial therapy with 3-hydroxy-3-methylglutaryl coenzyme A reductase inhibitors to treat hypercholesterolemia in a primary care setting of a managed-care organization. Clin Ther. 1997;19:582-599.
20. Kong SX, Gandhi SK, Crawford Pharmacoeconomic evaluation CoA reductase inhibitors in the of hypercholesterolemia. Am J Cure. 1996;2: 1055-1073.
SY, et al. of HMGtreatment Managed
21 Serra LI, Ribas L, Garcia R, et al. Avuluacid de PEstat Nutritional de la Poblacid Catalana (1992-93). (la edicio). Barcelona: Departament de Sanitat i Seguretat Social. Generalitat de Catalunya; 1996:27-31, 78-80. 22. Clarke R, Frost C, Collins R, et al. Dietary lipids and blood cholesterol: A quantitative meta-analysis of metabolic ward studies. Br Med J. 1997;314:112-117. 23. Rothschild V, Logothetis N. Probability Distributions. New York: John Wiley & Sons; 1986:62-63. 24. Kong SX, Crawford SY, Gandhi SK, et al. Efficacy of 3-hydroxy-3-methylglutaryl coenzyme A reductase inhibitors in the treatment of patients with hypercholesterolemia: A meta-analysis of clinical trials. Clin Ther 1997;19:778-797. 25. Delea TE, Hogan TJ, Gasper MC, et al. Cost-effectiveness of cholesterol-lowering drugs: A re-examination of treatment alternatives. Value Health. In press. 26. Simons A, Levis G, Simons J. Apparent discontinuation rates in patients prescribed lipid-lowering drugs. Med J Allina. 1996;164:208-211. 27. Gisbert R, Brosa M. Base de datos de castes sanitarios (v 1.0) [database on CDROM]. Barcelona, Spain: Soikos, 1998. 28. Boletin Mensual de Estadfstica. Madrid: Instituto National de Estadistica; August-September 1997. 29. Indicadores de la Prescripcidn Farmaceutica en el Sistema National de Salud. Vol. 15, Enero-Junio de 1997. Madrid:
1935
CLINICAL
Subdireccidn General de Coordinacio’n Administrativa, Institute National de Salud; 1997. 30. Regazzi MB, Icona I, Campana C, et al. Altered disposition of pravastatin following concomitant drug therapy with cyclosporin A in transplant recipients. Transplant Proc. 1993;25:2732-2734. 31. Kantola ‘I, Kivisto KT, Neuvonen PJ. Gmpefruit juice greatly increases serum concentrations of lovastatin and lovastatin acid. Clin Pharmacol Ther. 1998;63:397-2. 32. Neuvonen PJ, Kantola T, Kivisto KT. Simvastatin but not pravastatin is very susceptible to interaction with the CYP3A4 inhibitor itraconazoIe. Clin Pharmacol Ther. 1998;63:332-341. 33. Nordin C, Dahl ML, Eriksson M, et al. Is the cholesterol-lowering effect of simvastatin influenced by CYP2D6 polymorphism? L.uncet. 1997;350:29-30. 34. Sheldon TA. Problems of using modelling in the economic evaluation of health care. Health &on. 1996;56: l-l 1. 35. O’Connor PJ, Rush WA, Trence DL. Relative effectiveness of niacin and lovastatin for treatment of dyslipidemias in a
1936
THERAPEUTICS”
health maintenance organization. Pratt. 1997;44:462467.
J Fam
36. Fumarga EM. Pharmacist management of a hyperlipidemia clinic. Am J Hosp Pharm. 1993;50:91-95. 37. Kruse W, Schlierf G. Monitoring compliance in clinical trials. Lkzncet. 1990;335: 803-804. 38. Macia R, Pena A, Marrugat J, et al. High prevalence of cardiovascular risk factors in Gerona, Spain, a province with low myocardial infarction incidence. J Epidemiol Community Health. In press. 39. Illingworth DR, Erkelens DW, Keller U, et al. Defined daily doses in relation to hypolipidaemic efficacy of lovastatin, pravastatin and simvastatin. Lancet. 1994; 343:1554-1555. 40. Masana L, Pedro-Botet J, Roe E, et al. Impact0 de1 consenso para el control de la colesterolemia en Espafia. Med C&z. 1998; 110:598-599. 41. Brotons C, Server M, Pinto X, et al. Impacto de1 consenso para el control de la colesterolemia en Espatia. Med Clin. 1996; 108:9-15.