Introduction: Statin therapy reduces the rate of coronary heart disease, but high costs in combination with a large population eligible for treatment ask for priority setting. Although trials agree on the size of the benefit, economic analyses of statins report contradictory results. This article reviewed cost effectiveness analyses of statins and sought to synthesise cost effectiveness ratios for categories of risk of coronary heart disease and age.
Methods: The review searched for studies comparing statins with no treatment for the prevention of either cardiovascular or coronary heart disease in men and presenting cost per years of life saved as outcome. Estimates were extracted, standardised for calendar year and currency, and stratified by categories of risk, age, and funding source
Results: 24 studies were included (from 50 retrieved), yielding 216 cost effectiveness ratios. Estimated ratios increase with decreasing risk. After stratification by risk, heterogeneity of ratios is large varying from savings to $59 000 per life year saved in the highest risk category and from $6500 to $490 000 in the lowest category. The pooled estimates show values of $21571 per life year saved for a 10 year coronary heart disease risk of 20% and $16862 per life year saved for 10 year risk of 30%.
Conclusion: Statin therapy is cost effective for high levels of risk, but inconsistencies exist at lower levels. Although the cost effectiveness of statins depends mainly on absolute risk, important heterogeneity remains after adjusting for absolute risk. Economic analyses need to increase their transparency to reduce their vulnerability to bias and increase their reproducibility.
- CHD, coronary heart disease
- CEA, cost effectiveness analyses
- CER, cost effective ratio
- CVD, cardiovascular disease
- cardiovascular disease
- coronary heart disease
- cost effectiveness analysis
Statistics from Altmetric.com
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.
- CHD, coronary heart disease
- CEA, cost effectiveness analyses
- CER, cost effective ratio
- CVD, cardiovascular disease
With the advent of statins, the controversy surrounding cardiovascular risk management has shifted from how to treat hypercholesterolaemia to the proportion of eligible people that may actually be treated because of financial constraints.1 Statins reduce the rate of coronary heart disease (CHD) by more than 30%,2–4 side effects are unusual, and they are well tolerated.5 However, statins’ high costs limit the scope of treatment: society wants the best returns for its investments in health.6 This asks for priority setting aided by economical analyses, commonly cost effectiveness analyses (CEA).7 As relative risk reduction is thought to be constant,2,8,9 benefits are largely determined by absolute risk of CHD. The question is, because of costs, at which level of risk, treatment is cost effective. Different levels are used worldwide, ranging from 20%10 to 30%11 10 year absolute risk of CHD.
Regardless of comparable health effects and drug costs, CEA of statins have reported contradictory results ranging from very low to very high cost effectiveness ratios (CERs). The reasons are unknown and subject to speculation about methodological issues and competing interests.
We reviewed CEA of statins and sought to identify sources of heterogeneity of CERs adjusted for categories of risk of CHD, age, and funding source.
We performed a systematic review of the published statin CEA.
We searched for CEA in English, Spanish, Dutch, or German on statins for the prevention of either CHD or cardiovascular disease (CVD) in adult male populations (>20 years). We selected more than one language to make our search criteria more comprehensive. Reviews and meta-analyses were excluded. Studies needed to compare cost effectiveness of statin therapy with no pharmacological treatment and present cost per years of life gained/saved (YLG, YLS) as outcome. Studies comparing statins with other statins or with other cholesterol lowering drugs were excluded.
We used the databases Medline, the British National Health Service Economic Evaluation Database (NHS EED), Database of Abstracts of Reviews of Effectiveness (DARE), and the Health Technology Assessment database (HTA). We searched for papers published between 1990 and July 2002 and used these terms for searching: Statins OR Hydroxymethylglutaryl-CoA-Reductase Inhibitors AND Cost-effectiveness.
Two independent investigators analysed the abstracts obtained from the databases. All studies that matched our inclusion criteria were retrieved, and their reference lists were checked manually to identify more studies. We contacted experts in the field to find unpublished or ongoing studies.
When the decision of inclusion could not be achieved by reading the abstracts, the papers were retrieved for analysis.
Two independent investigators read each one of the articles retrieved and selected the studies to include based on the inclusion criteria. A third investigator was contacted to reach a decision in case of disagreement.
A data extraction form was elaborated based on prior knowledge and literature. Using the form, two independent investigators (which could be the same as the ones in the study selection process) collected the data from the articles.
Variables included in the form were: publication date, date and country used to calculate costs, annual drug costs, type of model, category of prevention, mean or range of age at the start of treatment, annual level of absolute risk of CHD (fatal and non-fatal events) at the start of treatment, time horizon of treatment, method for effect calculation, economic perspective, discount factor, funding source, and final outcome (CERs).
Model type could be primary or secondary modelling. Primary modelling studies were conducted parallel to clinical trials. Secondary modelling studies used data from clinical trials or other sources to model the effect of statins. Studies were also divided in two categories of effect calculation: studies that used levels of CHD risk reduction to model the effect, or studies that used intermediate effects (for example, % cholesterol level reduction) for the modelling procedure.
The category of prevention was either primary, for populations free of either CHD or CVD, or secondary, for CHD or CVD patients.
Time horizon of treatment refers to the observed or assumed duration of treatment period. Treatment period can be restricted to the observed trial period (which in the case of statins is generally around five years), or can be assumed to last (that is, after the trial) for a limited or unlimited (lifelong) period. We classified time horizon of treatment in three categories: five years, 10 years, or >10 years/lifetime.
Discount factors are used to weight the value of future benefits and costs; four categories were used: 3%, 5%, 6%, or none.
Funding source of the studies was classified in two categories: funding provided by the industry and funding provided by others (academic or governmental institutions or none).
For each CER we wanted absolute risk of CHD of the study population before treatment. If not stated in the paper, we estimated risk by the D’Agostino CHD function 12 for primary and secondary prevention respectively, using these variables: levels of total cholesterol (or LDL), HDL cholesterol, smoking, blood pressure, diabetes history, and personal history of CHD.
The CERs were stratified by categories of risk, age groups, and funding source, and standardised.
To standardise we used only ratios reported as cost per YLG or YLS that were calculated using net costs and with future costs and benefits weighted with the same discount factor. As the CERs were reported using different currencies and dates, we standardised by calendar year and currency. Firstly, to correct for inflation, we converted all currencies into the same date: 30 Deceember 2001, using the correspondent consumer price index (CPI). Afterwards all currencies were converted into US dollars of the date 30 December 2001. Data on CPI and conversions were taken from different sources: Federal Reserve Bank of Indianapolis USA, Bank of Canada, The European Union (publication: statistics in focus, economy and finance: prices and purchasing power parities, theme 2-48/2002), and the University of Exeter England.
After standardisation, CERs were stratified by categories of risk of CHD, by age groups at the start of treatment and by funding source.
The categories of annual risk of CHD at the start of treatment were: <1%, 1%–<2%, 2%–<3%, 3%–4%, and >4%.
Age groups used were: <45 years, 45–65 years, and >65 years.
Outcomes were compared by whether the study was funded by pharmaceutical companies or by others.
Once stratified, the CERs were plotted using spreadsheets. For each category of risk the distribution of the outcomes was calculated (median, quartiles, and centiles) and compared.
Data were analysed using a linear mixed effect model. The log of the CER was taken as continuous response value, because errors were assumed to become larger with larger CERs. The studies were put at the first level and CERs at the second level. The result of this design is that the power of characteristics of the study is limited to the number of studies and that the power of predictors that differ within the studies is increased by the extraction of the variance between studies from the random error. We included 11 variables in the analysis: absolute risk (continuous), age (<45, 45–65, and >65), treatment duration (<5years, 5–10 years, and >10 years (lifetime)), country used as source of costs (USA, Canada, and Europe), type of model (primary and secondary), effect calculation (direct reduction of risk and indirect risk reduction through lipid lowering), category of prevention (primary and secondary), perspective (societal and third party payer), funding source (pharmaceutical companies and others), year of publication (<1996 and ⩾1996), and discount factor (<5%, ⩾5%). Drug cost was not included as an explanatory variable because of its high correlation with the CER. First we performed a univariate analysis to find overall predictors, and then an analysis of interactions with risk. This last analysis shows the effect modification by variables over changing risk.
Variables were considered significant at a two sided p value <0.05.
We used S-PLUS (version 6.0 for Windows) for the analyses.
Study identification and selection
We found 308 references using the databases: 235 with Medline and 73 with the NHS EED, DARE, and HTA databases. From the 308 references, 186 were rejected because they were not CEA. Of the 122 abstracts selected, 50 studies were retrieved. By checking reference lists we found four new references, but they did not match our inclusion criteria. Finally 24 studies were selected for the analysis.17–40
Reasons for exclusion were: seven studies were reviews or meta-analyses,41–47 seven compared statins with statins or with other lipid lowering drugs,48–54 three were cost-utility analyses,55–57 three compared strategies of cardiovascular risk management,58–60 three did not use as outcome cost/YLS,61–63 two were descriptive methodological papers,64,65 and one study was excluded because it was not possible to determine absolute risk before treatment.66
Descriptive of the studies
All the studies compared statin therapy with no treatment of CHD or CVD risk and were conducted in developed countries.
Mean annual drug cost was US$ 7687 (ranged from 399 to 1670). Pharmaceutical companies funded 13 studies. Only one study presented the estimated CERs without any discount factor, the rest used predominantly 5% (table 1).
The outcome selected was cost per YLS standardised by calendar year and currency.
From the 24 studies selected we identified 216 CERs.
The ratios reported ranged over an enormous range: from savings to $489 000/YLS (table 2). However there were no negative CERs (suggesting that the statins were worse than the comparator) in any of the studies selected.
Outcomes were first stratified by categories of risk at the start of treatment as absolute risk is an important determinant of the cost and benefits of statin therapy. We found a strong inverse relation between absolute risk at baseline and the CER (fig 1).
Even though there was an external consistency between the studies, the variability among the ratios extracted was high, even after standardisation and stratification. In all categories of risk there were cost effective, moderate, and expensive ratios (table 2).
For levels of annual absolute risk over 4% most of the CERs agreed in showing statins as a cost effective option, at least 90% of the ratios were not expensive. For the lowest level of risk (<1%) the agreement of the studies was also clear, but in showing statin therapy as an expensive option. For the rest of the levels (1% to 4%) there was no clear agreement.
Stratified by risk, the difference between the 10th and 90th centile of the CERs was still fourfold to sevenfold.
A comparison between the age groups was not possible as estimates for younger (<45) and older ages (>65) were few.
Estimated CERs from studies funded by others were generally more expensive than estimates from studies funded by pharmaceutical companies. This difference is striking at lower levels of risk (<2%), corresponding to primary prevention (fig 2).
Multilevel linear regression
Our aim was evaluate the degree to which explanatory variables account for the observed variability in CERs between the studies. In the univariate analysis (second column, table 3) only absolute risk and category of prevention are significant. For levels of 2% and 3% annual risk of CHD the corresponding average CERs were $21571/YLS and $16 862/YLS. As annual absolute risk increases by 1% the CER is expected to decrease by 21.8%.
Compared with primary prevention, secondary prevention represents a decrease of 62% in the CER: category of prevention is highly correlated with absolute risk.
The third column of table 3 shows the interaction effects between absolute risk and the other explanatory variables. Here we compared the original (univariate) effect of 1% increase in absolute risk (21.8 % decrease in CERs) in each variable category. For age, cost country, category of prevention, funding source, year of publication, and discount factor the effect of absolute risk in the CER differed significantly between categories. As a result of the multilevel analysis each CER is compared with other CER within the same study, which gives more discriminating power to find interactions.
Lower (more negative) effects imply steeper decreases of cost effectiveness by increasing absolute risk; therefore the impact of absolute risk on the CER is larger. In secondary prevention, the gradient of absolute risk is less important, as it is always high. At lower ages (<45 years), which also constitutes generally populations at lower levels of risk, increases in the level absolute risk has a higher impact than in older populations with higher levels of risk. Studies funded by pharmaceutical companies tended to report lower CERs for the same levels of absolute risk (fig 2) and were also less responsive to the effect of absolute risk (table 3). Studies published before 1996, generally responded more dramatically to the effect of absolute risk as they generally used populations at lower levels of risk that were classified according to their individual risk factors, and not based on comprehensive CHD risk. High discount rates increase the effect of absolute risk at start of treatment: increases of absolute risk further in time/age are devalued more by higher discount rates.
What this paper adds
Statin therapy is cost effective for high levels of risk, but inconsistencies exist at lower levels. Although the cost effectiveness of statins depends mainly on absolute risk, important heterogeneity remains after adjusting for absolute risk.
The most probable explanation for this ample heterogeneity is different methodology in the CEA. However, the impact of funding source found suggests the potential for some estimates to be biased.
Economic analyses need to increase their transparency to reduce their vulnerability to bias and facilitate their evaluation.
In randomised trials, meta-analysis pool empirical results, reducing the level of random error and bias. In CEA, systematic reviews may identify agreement and disagreement. As prognosis is such an important predictor of absolute effects, the model used standardises for known determinants of health outcomes. Cost effectiveness of statin treatment is strongly related to absolute risk of CHD. Our pooled estimates show values of $21 571/YLS for an annual CHD risk of 2% and $16 862/YLS for annual risk of 3%. Most studies agree that statin treatment is cost effective for high risk patients (annual absolute risk >4%) but not cost effective for low risk patients (annual risk <1%). For medium risk patients (annual absolute risk 1 to 4%) the decision of whether treatment is cost effective depends on the choice of the study. After adjusting for levels of absolute risk, large heterogeneity in the estimated CERs remained. Most of the remaining variability can be explained by interactions between absolute risk and age, cost country, category of prevention, funding source, year of publication, and discount factor. We were not able to identify further methodological differences, assumptions, or model design that caused the remaining heterogeneity. Variability in the risk reduction with statins underlying each of the studies could not explain the variability in the CERs estimates, as there is general concordance among studies with respect to effect size, and we standardised for absolute risk. The role of costs in the CER was evaluated standardising or adjusting for the year costs, year of publication, country of costs, drug costs, economic perspective, and currency used.
Economic analyses need to increase their transparency to reduce their vulnerability to bias and facilitate their evaluation. New policies to standardise these types of analyses are mandatory.
Statin therapy is a cost effective intervention to treat populations at high risk of cardiovascular disease (>4%/year) and it is not cost effective for populations at low risk (<1%/year). For populations with moderate risk (1 to 4%/year) the available evidence is contradictory and new transparent evaluations are needed.
Diverse levels of survival and prevalences of risk can explain the different effects of absolute risk by age groups. For age group <45, a population at generally low levels of risk, high risks at start of treatment represent low CERs. In contrast, age group >65 with generally high levels of risk, benefits are limited by a shorter life expectancy and the relative effect of absolute risk on the CERs is lower. However, the effect of age is difficult to disentangle, as it is included in absolute risk. Age and absolute risk are highly correlated, but risk (corrected for age) is negatively correlated with CER and age (corrected for risk) is positively related with CER.
The different effects of absolute risk by cost country could be caused by methodological differences, gaps in interventions’ prices, different practice patterns specific to each country, and the characteristics of the populations studied. It could be inferred from the observed differences between countries that the generalisability of CERs to additional geographical areas is limited and the estimates might be specific to the country or area used in the original analyses.
Studies funded by pharmaceutical companies generally showed more cost effective CERs. This might be because of the interests that pharmaceutical companies may have in expanding treatment thresholds to lower levels of absolute risks. We found no clear differences in the methods used, in the populations’ characteristics, or in the levels of risk. The differences were striking at low levels of risk, representing large eligible populations. It is tempting to suggest competing interests as an explanation, although we could not prove this.
The possibility of studies that could have been missed exists, but the likelihood that these would change our conclusions is low. The outcome for cost utility studies is years of life gained adjusted by quality weights (QALYs) mortality in treated and untreated cohorts. QALYs add value judgements on the utility lost by decreased health in different disease stages. The different sources of weight values lead to extra potential sources of heterogeneity, which are often not well described in the articles, and interpretation of that heterogeneity is extremely difficult. Therefore, we excluded three cost utility analyses that did not present YLS, the sole outcome of this pooling YLS are solely derived from age specific study.
The primary objective of this study was to evaluate levels of heterogeneity and predictors on CEA of statins; we decided to include only studies reporting results for male populations. Studies of female populations are rarer and more difficult to interpret, as women at the same age run lower heart disease risks. While the absolute CERs will differ by sex, we do not expect the levels of heterogeneity to do so. Of the studies included in this study, 14 separately reported CERs for women. All of them consistently show higher CERs (compared with men).
Not all the studies used absolute risk for their analyses; therefore risk factor combinations were translated into absolute risk of CHD. This approach was not validated and some might have been misclassified. Residual confounding particularly in the first risk group exists (there are great relative differences possible in the group <1%). However, the variability is seen in all risk groups and misclassification will be limited (people at lower levels of risk factors will be at lower levels of risk).
Other methods exist to standardise CER to a same cost year, but this will not change our results and conclusions.
In conclusion, this review confirms how the cost effectiveness of statins treatment in the prevention of CHD is related to the absolute risk of CHD, but shows that within risk strata there still exists large variability in cost effectiveness estimates. Nearly all studies agree that treatment at high levels of risk is cost effective and at low levels is expensive. But in practice, it is not difficult to find CERs that fit any decision for the population at large with intermediate annual risk of CHD (1% to 4%). The most probable explanation for these differences is different methodology in the CEA, and the impact of funding source suggests the potential for some estimates to be biased. It was very difficult, even impossible to pinpoint all the divergent model assumptions that caused the variability.
Further standardisation of methodology for economic analyses and greater transparency in the presentation of results would aid in the evaluation of potential sources of bias, as well as facilitate the evaluation of reproducibility.
The authors thank Arno der Kinderen MSc for his economic advice.
CONTRIBUTORS Oscar H Franco, Anna Peeters, Caspar Looman, and Luc Bonneux; participated actively in all and each of the following aspects for this article: conception and design, or analysis and interpretation of data; drafting the article or revising it critically for important intellectual content; and final approval of the version to be published. Oscar H Franco as guarantor of this paper accepts full responsibility for the integrity of the data and the accuracy of the data analysis, had full access to all the data in the study, and controlled the decision to publish.
Funding: OHF, AP, CWNL, and LB were partly funded by the Netherlands Organisation for Scientific Research (ZON-MW). This funding organisation did not participate in the design and conduct of the study, collection, management, analysis, and interpretation of the data; and preparation, review, or approval of the manuscript.
Competing interests: none.
This manuscript has been presented before (as an abstract) in a poster in the 2003 Congress of the European Society of Cardiology (Vienna, Austria, September 2003), in an oral presentation in the 2004 World Congress of Public Health (Brighton, UK, 20 April 2004), and in an oral presentation during the conference of the Health Technology Assessment International group 2004 (Krakow, Poland, 1 June 2004).
Ethical approval was not required as this was a secondary data analysis.