- Email: [email protected]
Valuing vaccines using value of statistical life measures
Vaccine 32 (2014) 5065–5070
Contents lists available at ScienceDirect
Vaccine journal homepage: www.elsevier.com/locate/vaccine
Valuing vaccines using value of statistical life measures Ramanan Laxminarayan a,b,∗ , Dean T. Jamison c , Alan J. Krupnick d , Ole F. Norheim e a
Center for Disease Dynamics, Economics & Policy, 1616 P Street NW, Suite 430, Washington, DC 20036, USA Princeton University, M43 Guyot Hall, Room 132, Princeton, NJ 08544, USA c University of Washington, Department of Global Health, Ninth and Jefferson Building, 13th Floor, 908 Jefferson Street, Box 359931, Seattle, WA 98104, USA d Resources for the Future, 1616 P Street NW, Suite 600, Washington, DC 20036, USA e University of Bergen, Department of Medical Ethics, Department of Global Public Health and Primary Care, Kalfarveien 31, 5018 Bergen, Norway b
a r t i c l e
i n f o
Article history: Received 11 November 2013 Received in revised form 19 June 2014 Accepted 8 July 2014 Available online 18 July 2014 Keywords: Vaccine priorities Vaccine policy Willingness-to-pay measures Value of statistical life
a b s t r a c t Vaccines are effective tools to improve human health, but resources to pursue all vaccine-related investments are lacking. Benefit–cost and cost-effectiveness analysis are the two major methodological approaches used to assess the impact, efficiency, and distributional consequences of disease interventions, including those related to vaccinations. Childhood vaccinations can have important non-health consequences for productivity and economic well-being through multiple channels, including school attendance, physical growth, and cognitive ability. Benefit–cost analysis would capture such non-health benefits; cost-effectiveness analysis does not. Standard cost-effectiveness analysis may grossly underestimate the benefits of vaccines. A specific willingness-to-pay measure is based on the notion of the value of a statistical life (VSL), derived from trade-offs people are willing to make between fatality risk and wealth. Such methods have been used widely in the environmental and health literature to capture the broader economic benefits of improving health, but reservations remain about their acceptability. These reservations remain mainly because the methods may reflect ability to pay, and hence be discriminatory against the poor. However, willingness-to-pay methods can be made sensitive to income distribution by using appropriate income-sensitive distributional weights. Here, we describe the pros and cons of these methods and how they compare against standard costeffectiveness analysis using pure health metrics, such as quality-adjusted life years (QALYs) and disabilityadjusted life years (DALYs), in the context of vaccine priorities. We conclude that if appropriately used, willingness-to-pay methods will not discriminate against the poor, and they can capture important nonhealth benefits such as financial risk protection, productivity gains, and economic wellbeing. © 2014 Elsevier Ltd. All rights reserved.
1. Introduction Methodological approaches to setting priorities can be broadly divided into benefit–cost analysis (BCA) and cost-effectiveness analysis (CEA). Under BCA, the benefits and costs of a policy are expressed in monetary (dollar) terms. Subtracting the costs of the policy from the benefits gives the net benefit of the policy. Ideally, policies with the greatest net benefit should be the most
∗ Corresponding author at: Center for Disease Dynamics, Economics & Policy, 1616 P Street NW, Suite 430, Washington, DC 20036, USA. Tel.: +1 202 328 5085; fax: +1 202 328 5170. E-mail addresses: [email protected], [email protected] (R. Laxminarayan), [email protected] (D.T. Jamison), [email protected] (A.J. Krupnick), [email protected] (O.F. Norheim). http://dx.doi.org/10.1016/j.vaccine.2014.07.003 0264-410X/© 2014 Elsevier Ltd. All rights reserved.
preferred, all else being equal. BCA is also used to understand the distributional consequences of a policy for sex, race, and age groups. BCA offers comparability with similar analyses outside the health sector. In the context of vaccines, BCA includes non-health benefits like financial risk protection, future learning and productivity gains, which can be substantial, particularly for childhood vaccines. Most non-economists in the health arena are familiar with cost-effectiveness analysis. Here, the metric is a simple ratio of the cost of an intervention and its health impacts (measured in deaths or disability-adjusted life years (DALYs) averted or quality-adjusted life years (QALYs) gained). In CEA, impacts are not expressed in monetary terms, and therefore it is not possible to conclude whether a policy increases social welfare, as is the case with BCA. CEA can help rank alternative interventions by order of cost-effectiveness. CEA analyses may be easier to conduct
5066
R. Laxminarayan et al. / Vaccine 32 (2014) 5065–5070
and communicate than BCA but may miss important non-health benefits of vaccines. The fundamental divide between the two approaches is the following. Under BCA, health outcomes are judged by the extent of their contribution to overall societal wellbeing, measured as the sum of wellbeing of individuals. Under CEA, the objective is to maximize contributions to societal health, measured as the sum of individual health status. CEA approaches do not recognize the benefits of health care in broader welfare terms, or in terms of preferences for health relative to other goods [1,2]. What does this mean in practice? Under BCA, it may be preferable to provide treatment to a person who copes poorly with a disease, or who are thrown into poverty by it, rather than to someone who copes well. This makes sense because, in the latter case, individual wellbeing is not enhanced to as great an extent. In the CEA framework, the value of treatment for the two individuals would be independent of individual preferences, or their wellbeing, under the assumption that all individuals value similar health states similarly [3]. Most importantly, BCA accounts for many kinds of benefits, including non-health benefits and financial risk protection afforded by the intervention. It is in the valuation of health benefits in economic terms that value of a statistical life enters the picture. 2. Value of statistical life1 Although many think the value of a life is infinite, we all make decisions that implicitly place a value on our loss of limb and even life. The simplest example is the decision to cross a busy street against the light. One might get to his or her destination a bit faster—or one might not get there at all. We make a tradeoff between the gain (shorter time to the destination) and the loss (small probability of loss of life or limb). Even a policymaker who allocates resources for health needs, based on his or her perception of costs and benefits, is implicitly and inescapably placing a value on life. Three principal approaches are used to evaluate the value of a statistical life (VSL) and willingness to pay (WTP) for reducing risks to life. The most common approach is based on wage-risk trade-offs—the risk premiums paid to workers who accept jobs with a high risk of death or injury2 . This approach is also called “revealed-preference” because it is based on an examination of how individuals actually behave in the face of job market risks3 . Revealed preference uses labor market data to estimate the effect of morbidity and mortality risk on wage differences between occupations with differing levels of risk, after controlling for other variables that would explain wages4 . For example, all else being equal, a construction worker employed on a high-rise building must be paid more than someone working on a single-story building to compensate him for the greater probability of dying on the job. VSLs are based on the fairly robust theory of compensating differentials—the idea that workers must be paid more to take on tasks that are unpleasant or hazardous5 . At the same time,
1 The UK Green Book refers to the VSL measure as the value of a prevented fatality or prevented injury, which may be more easily understood than the notion of “statisticallives”. 2 Therefore, if lifetime wages for a high-rise construction worker with a 1/10,000 greater probability of death on the job are $500 more than for workers with a similar job but with a lower risk of death, VSL is calculated as $5,000,000. 3 In contrast, in stated-preference methods respondents are asked how much they would hypothetically pay for a lower risk of death. 4 For conceptual and implementation-related critiques of VSL; see [28]. 5 Even drug dealers understand compensating differentials. For instance, Steve asked a gang foot soldier who is normally was paid very little why he was paid 70% more during a gang war. “Would you stand around here when all this s— {shooting}
critics point out that while the observed risk premium is based on perceived risks, the calculation is almost always based on actual risks, because the former is unobserved. If these are different, bias will be introduced. A second approach is based on observing how much consumers are willing to pay to lower the risk of death. A growing literature on VSLs has measured these values based on the risk-payoff tradeoff and includes studies on the wages of Sherpas in the Himalayas (the value of climbers’ safety being observable in higher wages for better guides) [4] and the willingness of U.S. states to forgo federal highway construction funds in exchange for higher speed limits [5]. Since VSLs estimated using these two approaches look at how people actually behave, economists see them as relatively complete measures of the economic value of health. Whereas the two approaches described above are based on actual behavior, a third approach, “stated-preference”, relies on survey responses to carefully structured, hypothetical questions about one’s willingness to pay for a lower risk of death or disability. A challenge with “stated-preference” is that individuals may find it hard to provide accurate responses to direct willingness-to-pay questions, especially for unfamiliar options and small changes in risks [6]—but methods have improved substantially over the years, such as validity tests, which are built into the experimental design. Stated-preference studies can obtain VSLs in specific contexts where revealed-preference approaches may not be applicable. For example, stated preference has been shown useful for determining the willingness-to-pay for a hypothetical malaria vaccine in Ethiopia [7]. Sometimes a vaccine introduction could involve tradeoffs between different vaccines that vary in health benefits and target populations. Whereas a separate willingness-to-pay study may be needed for each health effect, a newer approach, called conjoint analysis, asks individuals to choose among different attributes, such as health states, to estimate “prices” for a variety of health attributes. Trade-offs made by study participants are then used to statistically estimate the relative importance of different health attributes. These kinds of studies, if conducted carefully, have the ability to complement VSL studies. Recent World Health Organization (WHO) economic analysis guidelines on hypothetical estimates of willingness-to-pay recommend that, “(empirically-based) estimates of market losses be separately identified and reported from (hypothetically-based) estimates of foregone welfare6 .” These guidelines also rely on the cost-of-illness approach that tries to measure the cost associated with ill health and seeking treatment, but leaves out the costs of pain and suffering associated with illness and is, therefore, inconsistent with the welfare economic approach. 3. Use of VSL in health and other sectors In the United States, the Office of Management and Budget requires that all federal agencies conduct both CEA and BCA on proposed rules and regulations whose annual effect on the economy will exceed $100 million. Recent updates to the guidelines require a CEA for major regulations where a significant emphasis is on health and safety, and BCA is required for major health and safety rulemakings to the extent that the primary health and safety outcomes can be expressed in monetary terms. The rationale underlying this
is going on? No, right? So if I gonna be asked to put my life on the line, then front me the cash, man. Pay me more ‘cause it ain’t worth my time to be here when {the gangs are} warring."[29]. 6 WHO Guide to Identifying the Economic Consequences of Disease and Injury, World Health Organization, Department of Health Systems Financing Health Systems and Services http://www.who.int/choice/publications/ d economic impact guide.pdf.
R. Laxminarayan et al. / Vaccine 32 (2014) 5065–5070
latter requirement is that monetized economic benefits indicate what the public is willing to pay for improvements in health and safety. Among the U.S. Federal Agencies that use VSLs are the Consumer Product Safety Commission, the Food and Drug Administration, the Federal Aviation Administration, the Environmental Protection Agency, and the Food and Nutrition Service. For these agencies, VSL values range from $3 million to $6.3 million (in year 2000 dollars), based on the affected population and underlying assumptions. However, agencies whose sole mandate is the improvement of health, such as the National Institutes of Health and the Food and Drug Administration, use health measures—specifically, QALYs. VSLs are used in Australia, Canada and Japan as part of costbenefit analyses of measures to reduce mortality risk. The United Kingdom also uses VSLs, usually based on the revealed-preference approach. VSLs are also used by many multinational organizations, including the World Bank, the European Commission, the United Nations’ Intergovernmental Panel on Climate Change, and the World Health Organization. The VSL concept has been helpful in studying the benefits of chronic disease interventions [8], water and sanitation improvements [9], and the economic and social aspects of climate change. They have been used to estimate the benefit–cost ratio of the Global Plan to Stop TB, relative to a baseline of directly observed treatment–short course (DOTS) for tuberculosis [10]. 4. VSL compared with health metrics used in CEA Many people object to VSL measures on the grounds that they attempt to place a value on life, a task that is fraught with ethical and moral challenges. Actually, what VSL measures do is aggregate individuals’ willingness to pay for a small change in risk, which is then taken as a measure of the financial trade-offs that individuals are willing to make with respect to risk. QALYs and DALYs may appear not to value life, but in reality they only postpone the necessity of translating health measures into economic measures, thereby, allowing policymakers to make meaningful comparisons across sectors. In addition to assigning a value to life, VSL compared to health metrics used in CEA has other properties of importance. 4.1. Sensitivity to nonhealth benefits Although averted morbidity and mortality are important consequences of early childhood vaccinations, they can have important non-health consequences. For example, averting diarrhea through a rotavirus vaccine may have long-term benefits for productivity and economic well-being through multiple channels, including school attendance, physical growth, and cognitive ability [11,12]. VSL metrics would capture such non-health benefits; cost-effectiveness analysis focuses on health outcomes. The extent to which nonhealth benefits are important should inform the choice of metrics in health priority setting. For vaccines, the insensitivity of standard cost-effectiveness analysis to non-health benefits could be seen as a major limitation. Standard cost-effectiveness analysis may grossly underestimate the benefits of vaccines. 4.2. Sensitivity to income Since VSL measures reflect willingness to pay for better health, they are generally positively correlated with income, both in theory and practice. The link between VSLs and income, to some, reflects discrimination against the poor, who are most likely to benefit from childhood vaccines. Using monetary measures of health effects to make health policy decisions may be more controversial than using physical measures of health outcomes, which do not appear to place a value on life, and therefore, may be perceived as more neutral [13]. The link between VSLs and income could be seen as placing greater
5067
weight on policies that favor the rich than the poor (e.g., favoring an influenza vaccine over a malaria vaccine). However, this bias can and should be corrected. It is common practice to use a mean willingness to pay across income groups within a country to avoid this charge. Others have proposed ways to make willingness-topay methods sensitive to income distribution. Fleurbaey et al., for example, have proposed a method for deriving income-sensitive distributional weights based on the concept of equivalent income [14]. Using country-level VSLs may overstate the willingness-topay for vaccines that are more important for poorer populations, but may be a reflection of mean societal willingness-to-pay. 4.3. Sensitivity to age Most empirical research on VSLs assesses the value of reducing the mortality rate for one year by about one in 10,000 for people in middle age. [15], for example, report that in VSL studies from the U.S., the average remaining life expectancy is about 45 years. A key question in applying studies of the value of mortality reduction then, is how to translate the available empirical information – mostly about middle age – to evaluation of interventions that primarily affect the old or, as in the case of vaccines, the very young. Jamison, Summers et al. find that under standard assumptions for VSLs, the value of a death averted in early childhood is about 70% higher than one in middle age—a number that is both implausible and at odds with the limited available empirical information [16]. For this reason Jamison, Summers et al. explore options to applying the proportionality formula for very young ages, and for their headline numbers they use a value for a death averted in young children that is about 50% of the value for middle age. Health indices, such as QALYs, value extensions to younger peoples’ lives and healthy peoples’ lives more than extending the lives of those who are older (and have shorter life expectancies) or infirm (and have lower health status). Self-described health status measures apply to groups of different ages, and health status can be used to generate age- and health-specific QALY measures (see, e.g., [17]), but making these adjustments detracts from the simplicity of applying health measures. Many studies use the same VSL for all adults, and thus postponing an adult’s death by one year – say, age 40 to 41 – represents no gain within the framework. This problem can be fixed by remaining close to the empirical tradeoffs underlying the VSL. In practice one needs values for small changes in age-specific annual mortality probabilities to undertake BCAs, and this comes directly from the empirical literature. The intermediate construct of the VSL is not needed [16,18]. VSLs have been estimated for different age groups, in particular the very young and the very old. The latter “senior discount” reflects the findings in some studies that older people value reducing their death risks less than younger people. But Krupnick [19] finds that this literature is not robust. Health status has an ambiguous effect on the VSL. In contrast, most health indices treat all life years equally, whether from a younger or older person. However, extending a healthier person’s life implicitly is “valued” more than extending an ill person’s. Determining the value of averting stillbirths or the death of newborns and children – relevant for vaccines like tetanus – through VSLs is another challenge. Here the VSL may be based on the parent’s willingness to pay for saving the child’s life, but there are relatively few such studies. 4.4. Insensitivity to baseline risk, size of risk and perception of risk VSL is not sensitive to the concentration of risk in identified individuals. Some studies indicate that people value the same risk reduction differently according to baseline risk. In policy, the
5068
R. Laxminarayan et al. / Vaccine 32 (2014) 5065–5070
so-called “rule of rescue” is an extreme manifestation of this kind of preference. In other words, society should be (and is) willing to pay whatever it takes to rescue a baby trapped at the bottom of a well without regard to costs. However, CEA is not sensitive to baseline risk in this sense. Most labor market VSL estimates rely on workers’ wage-risk trade-offs to capture the pay differential workers require to incur risks on the job. Therefore, they are not really representative of willingness to pay to avert risk but rather a willingness to accept greater risk in exchange for higher wages. There are moral issues related to allowing people to trade money for the higher risk of morbidity and mortality but the fact remains that this is routinely done in labor markets. When markets are imperfect, or when workers are poorly informed about the risk they face in a particular occupation, VSL estimates could understate the willingness to accept greater risk for higher wages. In practice, willingness to pay (WTP) and willingness to accept (WTA) are reasonably identical for small changes in risk. However, the divergence between them may be significant for large changes in risk, such as for a TB vaccine in a high endemicity setting. Typically, WTP is smaller than WTA. A review of studies over a wide range of commodities, including environmental outcomes and real goods, found an average of the mean ratios of WTA to WTP of 7.2 [20]. If VSL estimates are viewed as willingness-to-accept values and are characterized by a similar ratio, then meta-analysis estimates of the median labor market VSL, pegged at $7 million in year-2000 dollars [21], should be reduced to $1 million. How should the choice of metric be influenced by the dread factor associated with potentially catastrophic and unfamiliar diseases like H1N1? All else being equal, should there be greater weight on diseases that are risky and appear involuntary, such as viral influenza, than on more familiar, more avoidable but possibly more deadly risks, such as tuberculosis? Sudden, catastrophic risk may be seen as less acceptable compared with risk of a disease that involves a long period of suffering, even though both may result in the same number of DALYs. Public tolerance for voluntary risks may be as much as 1000 times that for involuntary risks that offered the same dollar benefit per person [22]. In other words, how should the quality of risk be measured against the quantity of risk? The problem of how to deal with risk perceptions is important in both CEA and VSL literatures but has been better addressed in the context of VSLs. These considerations are particularly important in the context of vaccine-preventable diseases, where risk is not always known. 4.5. Sensitivity to context Most VSL estimates come from high-income countries and are extrapolated to low- and middle-income countries using a method called benefits transfer. This method requires an assumption about the relationship between willingness-to-pay and income. While this is often assumed to be proportional, studies of the VSL in developing countries, compared to similar estimates based on benefit transfer, show that this relationship is less than proportional, i.e., that the VSL estimated in such low income countries is higher than that transferred from high income countries, assuming proportionality of VSL to income. Since VSLs are based on the characteristics of the risk and the affected population, the same VSL should not be used for all contexts. For instance, a VSL measured in the context of work-related health risks may not be applicable to diet-related health risks. Similarly, VSLs based on accidental deaths of healthy, working-age adults may not provide appropriate VSLs for valuing the benefits of vaccines for childhood diseases. How valid are such comparisons? Ideally, the estimates of VSLs for vaccinations or tuberculosis treatment are derived from the
same context in which the benefit–cost ratios are calculated. In practice, that may not always be possible, and the benefits transfer method is a crude second best. Moreover, benefits estimates are not very useful without some idea about the incremental costs of proposed policies. The incremental costs include not only the dollar costs, but also the marginal excess burden (or efficiency loss) imposed to pay for the program. 4.6. Transparency Willingness-to-pay measures for acute effects are more transparent in that they are simply the average willingness to pay to avoid a case of an illness. QALYs may appear more transparent than willingness-to-pay measures because they are the product of two components: a health state score and its duration. However, the information underlying the health state scores in QALYs are not transparent—scores can be taken from different indices, which in turn are developed using different approaches for deriving preference weight. This non-transparency comes with implications for health state scores, in the same way that VSL values can vary depending on whether revealed-preference or stated-preference approaches are used. 4.7. Preferences Finally, should uninformed preferences be weighed against informed ones? Poorly educated or ill-informed people may not know the risks associated with their consumption or employment choices. An important concern about VSL estimates is whether individuals can make informed decisions about safety risks that pose challenges even for experts. The same concern applies to health status preferences used in QALYs and DALYs. 4.8. Metric conversion Could we reconcile VSLs, QALYs, and DALYs by converting CEAs into CBAs using a constant conversion ratio—say, a benchmark of dollars to QALYs? Two decades ago, the World Bank described interventions that cost less than US$150 per DALY saved as highly cost-effective [23]. The report of the WHO’s Commission on Macroeconomics and Health describes interventions whose cost is equivalent to GDP per capita as “very cost-effective”, and interventions costing three times GDP as “cost-effective” [24]. Although they are increasingly used, these conversions have no theoretical basis. Moreover, they are arbitrary because QALYs and willingness-to-pay measures vary with life expectancy, health status, and income [25]. That is, the ratio for converting from QALYs to willingness-to-pay is not uniform across individuals7 . 5. Points to consider for institutions using economic methods for the evaluation of vaccines In a recent study, Ozawa and colleagues calculate that $151 billion in treatment costs and productivity losses could be averted by expanding the delivery of six life-saving vaccines in 72 lowand middle-income countries between 2011 and 2020 [26]. This figure is based on low- and middle-income countries’ estimated VSLs derived from empirical estimates from high-income countries, using an assumption of a constant and percentage proportional relationship between income and VSL. Although there are some
7 For instance, QALYs are unaffected by income, but VSLs increase with income, holding other factors constant. Therefore, an individual’s willingness-to-pay per QALY would be increasing with QALY.
R. Laxminarayan et al. / Vaccine 32 (2014) 5065–5070
valid critiques of this study, analyses like these could be expanded to better understand the broader economic benefits of vaccination. What approach might institutions like the WHO, through groups such as the Strategic Advisory Group of Experts on Immunization (SAGE), take to using evidence from studies that use VSLs and BCAs to set vaccine priorities, such as the one by Ozawa and colleagues8 ? Implicitly, the question is whether VSL measures have advantages over the WHO’s well-established cost-effectiveness analyses for setting disease priorities, and whether SAGE should consider using VSLs in its decision-making. VSLs allow benefits to be expressed in dollar terms in a direct way that is consistent with standard welfare theory. This enables direct comparisons with other health and non-health interventions, whereas health measures like QALYs and DALYs do not. Shortcuts to convert QALYs and DALYs into dollars are arbitrary and lack theoretical basis. VSLs, when based on revealed-preference methods, are constructed from actual, observed behavior. They account for both morbidity and mortality risk and take into account important aspects of vaccines, such as avoidance behavior and nonhealth consequences. VSLs can also include “qualitative attributes” of health states, such as the dread associated with catastrophic influenza outbreaks. VSLs are attractive in theory but face some significant challenges as well. Estimates of the revealed value of a prevented fatality may be imprecise. Transferring VSLs from one country and demographic group to other contexts can be poorly related to the VSL that would be estimated in an original study. Moreover, the idea of valuing life may be uncomfortable to some, even though this is not what is being done. QALYs, too, have their drawbacks, in that the health state scores may be developed using different approaches to deriving preference weights and thus not be consistent, and they grossly underestimate the non-health benefits of vaccines. What are the alternatives? Direct health measures, such as lives saved, are less controversial, but their usefulness depends on the policy context and they cannot be aggregated with other types of health effects or non-health effects. The differences among measures could lead to markedly different recommendations from SAGE and other advisory groups for prioritizing vaccines and, more generally, for valuing the economic benefits of vaccines versus other health and non-health interventions. The following are issues that bodies like the Immunization and Vaccines related Implementation Research (IVIR) and Strategic Advisory Group of Experts (SAGE) at the international level, or National Technical Advisory Group on Immunization in countries may want to consider when deciding on whether or not to introduce a new vaccine. 5.1. What is the nature of the decision to be informed by the analysis? • If the objective is to compare two candidate vaccines for the same disease, CEA using QALYs or DALYs may be adequate. • If the objective is to compare two vaccines that address the same disease and in the same target population, the added value of BCA using VSLs may be limited. • If the objective is to compare two vaccines that target different populations and have different non-health benefits, BCA could be used alongside the current CEA approach. • If the objective is to compare vaccines with other health interventions, a BCA using VSLs may be useful; the following two points should then be considered.
8
SAGE is the principal advisory group to WHO for vaccines and immunization.
5069
5.2. What are the belief systems of decisionmakers and of the people subject to these decisions? In one study in Australia, individuals rejected the goal of maximizing QALYs in allocating health resources [27]. Would individuals support the assumptions of life-years saved as the main objective, or would they prefer a metric that reflects risk trade-offs based on actual behavior? Both types of health measures have equity biases (which can be adjusted by an appropriate choice of equity weights), and the choice between them depends on the belief system of the decisionmakers. 5.3. What level of cost and ease of application are decisionmakers looking for? Willingness-to-pay studies based on VSLs can provide useful information on the social value of health interventions but may not be applicable if the context differs from that in which the VSLs were measured. Moreover, relatively few studies have been conducted in low- and middle-income countries, although that is now changing. Revealed-preference studies based on observing how people trade risk for money in labor and consumer markets are now being supplemented by well-designed stated-preference studies. The stated-preference studies have the advantage of being able to focus on population subsets (such as the elderly) that would not normally participate in labor markets. 6. Conclusions CEAs based on QALYs and DALYs will continue to be important and enjoy the confidence of many in the medical and public health communities as a tool for prioritizing interventions. They are easy to administer and require relatively few surveys to develop weights that apply to many types of health states and describe specific health effects. Renaming VSL as value of a prevented fatality (VPF) or value of prevented morbidity (VPM), as is now done by the UK government, may help increase their acceptability outside the domain of economics. VSLs have been commonly used to guide regulatory decisionmaking in the fields of environment, health and safety and transportation. Despite empirical challenges and the paucity of studies in appropriate settings, they are being increasingly used to assist in setting health policy in developing countries. They help provide greater clarity on whether a specific action should be taken from an economic benefit–cost standpoint and, with certain caveats, can usefully complement existing cost-effectiveness methods. If appropriately used, BCA and willingness-to-pay methods will not discriminate against the poor, and they capture important non-health benefits from vaccines such as financial risk protection, education and productivity gains, and economic wellbeing. Acknowledgements We are grateful to two anonymous reviewers at the journal Vaccine, Raymond Hutubessy, John Edmunds, David Bloom and other members of the Immunization and Vaccines related Implementation Research Advisory Committee at the World Health Organization for useful comments. Any errors that remain are solely the responsibility of the authors. References [1] Brouwer WB, Culyer AJ, Van Exel NJ, Rutten FF. Welfarism vs. extra-welfarism. J Health Econ 2008;27(2):325–38. [2] Coast J, Smith RD, Lorgelly P. Welfarism, extra-welfarism and capability: the spread of ideas in health economics. Soc Sci Med 2008;67(7):1190–8.
5070
R. Laxminarayan et al. / Vaccine 32 (2014) 5065–5070
[3] Gyrd-Hansen D. Willingness to pay for a QALY: theoretical and methodological issues. Pharmacoeconomics 2005;23(5):423–32. [4] Goucher J, Horrace WC. The value of life: real risks and safety-related productivity in the Himalaya. Labour Econ 2012;19(1):27–32. [5] Ashenfelter O, Greenstone M. Using mandated speed limits to measure the VSL. J Pol Econ 2004;112:S226–67. [6] Hammitt JK, Graham JD. Willingness to pay for health protection: inadequate sensitivity to probability? J Risk Uncertainty 1999;18(1):33–62. [7] Cropper M, Haile M, Lampetti J, Poulos C, Whittington D. The demand for a malaria vaccine—evidence from Ethiopia. J Dev Econ 2004;75:303–18. [8] WHO. Preventing chronic disease: a vital investment. Geneva: World Health Organization; 2005. [9] WHO. Global costs and benefits of drinking-water supply and sanitation interventions to reach the MDG target and universal coverage. Geneva: World Health Organization; 2012. [10] Laxminarayan R, Klein EY, Darley S, Adeyi O. Global investments in TB control: economic benefits. Health Aff (Millwood) 2009;28(4):w730–42. [11] Lorntz B, Soares AM, Moore SR, Pinkerton R, Gansneder B, Bovbjerg VE, Guyatt H, Lima AM, Guerrant RL. Early childhood diarrhea predicts impaired school performance. Pediatr Infect Dis J 2006;25(6):513–20. [12] Niehaus MD, Moore SR, Patrick PD, Derr LL, Lorntz B, Lima AA, Guerrant RL. Early childhood diarrhea is associated with diminished cognitive function 4 to 7 years later in children in a northeast Brazilian shantytown. Am J Trop Med Hyg 2002;66(5):590–3. [13] Krupnick AJ. Valuing health outcomes: policy choices and technical issues. Washington, DC: Resources for the Future; 2004. [14] Fleurbaey M, Luchini S, Muller C, Schokkaert E. Equivalent income and fair evaluation of health care. Health Econ 2012;22(6):S226–67. [15] Hammitt JK, Robinson LA. The income elasticity of the value per statistical life: transferring estimates between high and low income populations. J Benefit Cost Anal 2011;2:1–27.
[16] Jamison DT, Summers LW, Alleyne G, Arrow KJ, Berkley S, Binagwaho A, Yamey G. Appendix 3 Global Health 2035: a world converging within a generation. The Lancet 2013;382:1898–955. [17] Cutler D, Richardson E. Measuring the health of the United States population. Brookings papers on economic activity. Microeconomics 1997:217–72. [18] Cameron TA. Euthanizing the value of a statistical life. Rev Environ Econ Policy 2010;4:161–78. [19] Krupnick A. Mortality-risk valuation and age stated preference evidence. Rev Environ Econ Policy 2007;1(2):261–82. [20] Horowitz JK, McConnell KE. A review of WTA/WTP studies. J Environ Econ Manage 2002;44:426–47. [21] Viscusi WK, Aldy J. The value of a statistical life: a critical review of market estimates throughout the world. J Risk Uncertainty 2003;27(1):5–76. [22] Starr C. Social benefit versus technological risk. Science 1969;165(3899): 1232–8. [23] World Bank. World development report: investing in health. New York, NY: Oxford University Press; 1993. [24] WHO. Macroeconomics and health. In: Report of the commission on macroeconomics and health. Geneva: World Health Organization; 2001. [25] Hammitt JK. QALYs versus WTP. Risk Anal 2002;22(5):985–1001 (an official publication of the Society for Risk Analysis). [26] Ozawa S, Stack ML, Bishai DM, Mirelman A, Friberg IK, Niessen L, Walker DG, Levine OS. During the decade of vaccines, the lives of 6.4 million children valued at $231 billion could be saved. Health Aff 2011;30(6):1010–20. [27] Nord E, Richardson J, Street A, Kuhse H, Singer P. Maximizing health benefits vs egalitarianism: an Australian survey of health issues. Soc Sci Med 1995;41(10):1429–37. [28] Grüne-Yanoff T. Mismeasuring the value of statistical life. J Econ Methodol 2009;16(2). [29] Levitt SD, Venkatesh SA. An economic analysis of a drug-selling gang’s finances. Q J Econ 2000;115(3):755–89.
Contents lists available at ScienceDirect
Vaccine journal homepage: www.elsevier.com/locate/vaccine
Valuing vaccines using value of statistical life measures Ramanan Laxminarayan a,b,∗ , Dean T. Jamison c , Alan J. Krupnick d , Ole F. Norheim e a
Center for Disease Dynamics, Economics & Policy, 1616 P Street NW, Suite 430, Washington, DC 20036, USA Princeton University, M43 Guyot Hall, Room 132, Princeton, NJ 08544, USA c University of Washington, Department of Global Health, Ninth and Jefferson Building, 13th Floor, 908 Jefferson Street, Box 359931, Seattle, WA 98104, USA d Resources for the Future, 1616 P Street NW, Suite 600, Washington, DC 20036, USA e University of Bergen, Department of Medical Ethics, Department of Global Public Health and Primary Care, Kalfarveien 31, 5018 Bergen, Norway b
a r t i c l e
i n f o
Article history: Received 11 November 2013 Received in revised form 19 June 2014 Accepted 8 July 2014 Available online 18 July 2014 Keywords: Vaccine priorities Vaccine policy Willingness-to-pay measures Value of statistical life
a b s t r a c t Vaccines are effective tools to improve human health, but resources to pursue all vaccine-related investments are lacking. Benefit–cost and cost-effectiveness analysis are the two major methodological approaches used to assess the impact, efficiency, and distributional consequences of disease interventions, including those related to vaccinations. Childhood vaccinations can have important non-health consequences for productivity and economic well-being through multiple channels, including school attendance, physical growth, and cognitive ability. Benefit–cost analysis would capture such non-health benefits; cost-effectiveness analysis does not. Standard cost-effectiveness analysis may grossly underestimate the benefits of vaccines. A specific willingness-to-pay measure is based on the notion of the value of a statistical life (VSL), derived from trade-offs people are willing to make between fatality risk and wealth. Such methods have been used widely in the environmental and health literature to capture the broader economic benefits of improving health, but reservations remain about their acceptability. These reservations remain mainly because the methods may reflect ability to pay, and hence be discriminatory against the poor. However, willingness-to-pay methods can be made sensitive to income distribution by using appropriate income-sensitive distributional weights. Here, we describe the pros and cons of these methods and how they compare against standard costeffectiveness analysis using pure health metrics, such as quality-adjusted life years (QALYs) and disabilityadjusted life years (DALYs), in the context of vaccine priorities. We conclude that if appropriately used, willingness-to-pay methods will not discriminate against the poor, and they can capture important nonhealth benefits such as financial risk protection, productivity gains, and economic wellbeing. © 2014 Elsevier Ltd. All rights reserved.
1. Introduction Methodological approaches to setting priorities can be broadly divided into benefit–cost analysis (BCA) and cost-effectiveness analysis (CEA). Under BCA, the benefits and costs of a policy are expressed in monetary (dollar) terms. Subtracting the costs of the policy from the benefits gives the net benefit of the policy. Ideally, policies with the greatest net benefit should be the most
∗ Corresponding author at: Center for Disease Dynamics, Economics & Policy, 1616 P Street NW, Suite 430, Washington, DC 20036, USA. Tel.: +1 202 328 5085; fax: +1 202 328 5170. E-mail addresses: [email protected], [email protected] (R. Laxminarayan), [email protected] (D.T. Jamison), [email protected] (A.J. Krupnick), [email protected] (O.F. Norheim). http://dx.doi.org/10.1016/j.vaccine.2014.07.003 0264-410X/© 2014 Elsevier Ltd. All rights reserved.
preferred, all else being equal. BCA is also used to understand the distributional consequences of a policy for sex, race, and age groups. BCA offers comparability with similar analyses outside the health sector. In the context of vaccines, BCA includes non-health benefits like financial risk protection, future learning and productivity gains, which can be substantial, particularly for childhood vaccines. Most non-economists in the health arena are familiar with cost-effectiveness analysis. Here, the metric is a simple ratio of the cost of an intervention and its health impacts (measured in deaths or disability-adjusted life years (DALYs) averted or quality-adjusted life years (QALYs) gained). In CEA, impacts are not expressed in monetary terms, and therefore it is not possible to conclude whether a policy increases social welfare, as is the case with BCA. CEA can help rank alternative interventions by order of cost-effectiveness. CEA analyses may be easier to conduct
5066
R. Laxminarayan et al. / Vaccine 32 (2014) 5065–5070
and communicate than BCA but may miss important non-health benefits of vaccines. The fundamental divide between the two approaches is the following. Under BCA, health outcomes are judged by the extent of their contribution to overall societal wellbeing, measured as the sum of wellbeing of individuals. Under CEA, the objective is to maximize contributions to societal health, measured as the sum of individual health status. CEA approaches do not recognize the benefits of health care in broader welfare terms, or in terms of preferences for health relative to other goods [1,2]. What does this mean in practice? Under BCA, it may be preferable to provide treatment to a person who copes poorly with a disease, or who are thrown into poverty by it, rather than to someone who copes well. This makes sense because, in the latter case, individual wellbeing is not enhanced to as great an extent. In the CEA framework, the value of treatment for the two individuals would be independent of individual preferences, or their wellbeing, under the assumption that all individuals value similar health states similarly [3]. Most importantly, BCA accounts for many kinds of benefits, including non-health benefits and financial risk protection afforded by the intervention. It is in the valuation of health benefits in economic terms that value of a statistical life enters the picture. 2. Value of statistical life1 Although many think the value of a life is infinite, we all make decisions that implicitly place a value on our loss of limb and even life. The simplest example is the decision to cross a busy street against the light. One might get to his or her destination a bit faster—or one might not get there at all. We make a tradeoff between the gain (shorter time to the destination) and the loss (small probability of loss of life or limb). Even a policymaker who allocates resources for health needs, based on his or her perception of costs and benefits, is implicitly and inescapably placing a value on life. Three principal approaches are used to evaluate the value of a statistical life (VSL) and willingness to pay (WTP) for reducing risks to life. The most common approach is based on wage-risk trade-offs—the risk premiums paid to workers who accept jobs with a high risk of death or injury2 . This approach is also called “revealed-preference” because it is based on an examination of how individuals actually behave in the face of job market risks3 . Revealed preference uses labor market data to estimate the effect of morbidity and mortality risk on wage differences between occupations with differing levels of risk, after controlling for other variables that would explain wages4 . For example, all else being equal, a construction worker employed on a high-rise building must be paid more than someone working on a single-story building to compensate him for the greater probability of dying on the job. VSLs are based on the fairly robust theory of compensating differentials—the idea that workers must be paid more to take on tasks that are unpleasant or hazardous5 . At the same time,
1 The UK Green Book refers to the VSL measure as the value of a prevented fatality or prevented injury, which may be more easily understood than the notion of “statisticallives”. 2 Therefore, if lifetime wages for a high-rise construction worker with a 1/10,000 greater probability of death on the job are $500 more than for workers with a similar job but with a lower risk of death, VSL is calculated as $5,000,000. 3 In contrast, in stated-preference methods respondents are asked how much they would hypothetically pay for a lower risk of death. 4 For conceptual and implementation-related critiques of VSL; see [28]. 5 Even drug dealers understand compensating differentials. For instance, Steve asked a gang foot soldier who is normally was paid very little why he was paid 70% more during a gang war. “Would you stand around here when all this s— {shooting}
critics point out that while the observed risk premium is based on perceived risks, the calculation is almost always based on actual risks, because the former is unobserved. If these are different, bias will be introduced. A second approach is based on observing how much consumers are willing to pay to lower the risk of death. A growing literature on VSLs has measured these values based on the risk-payoff tradeoff and includes studies on the wages of Sherpas in the Himalayas (the value of climbers’ safety being observable in higher wages for better guides) [4] and the willingness of U.S. states to forgo federal highway construction funds in exchange for higher speed limits [5]. Since VSLs estimated using these two approaches look at how people actually behave, economists see them as relatively complete measures of the economic value of health. Whereas the two approaches described above are based on actual behavior, a third approach, “stated-preference”, relies on survey responses to carefully structured, hypothetical questions about one’s willingness to pay for a lower risk of death or disability. A challenge with “stated-preference” is that individuals may find it hard to provide accurate responses to direct willingness-to-pay questions, especially for unfamiliar options and small changes in risks [6]—but methods have improved substantially over the years, such as validity tests, which are built into the experimental design. Stated-preference studies can obtain VSLs in specific contexts where revealed-preference approaches may not be applicable. For example, stated preference has been shown useful for determining the willingness-to-pay for a hypothetical malaria vaccine in Ethiopia [7]. Sometimes a vaccine introduction could involve tradeoffs between different vaccines that vary in health benefits and target populations. Whereas a separate willingness-to-pay study may be needed for each health effect, a newer approach, called conjoint analysis, asks individuals to choose among different attributes, such as health states, to estimate “prices” for a variety of health attributes. Trade-offs made by study participants are then used to statistically estimate the relative importance of different health attributes. These kinds of studies, if conducted carefully, have the ability to complement VSL studies. Recent World Health Organization (WHO) economic analysis guidelines on hypothetical estimates of willingness-to-pay recommend that, “(empirically-based) estimates of market losses be separately identified and reported from (hypothetically-based) estimates of foregone welfare6 .” These guidelines also rely on the cost-of-illness approach that tries to measure the cost associated with ill health and seeking treatment, but leaves out the costs of pain and suffering associated with illness and is, therefore, inconsistent with the welfare economic approach. 3. Use of VSL in health and other sectors In the United States, the Office of Management and Budget requires that all federal agencies conduct both CEA and BCA on proposed rules and regulations whose annual effect on the economy will exceed $100 million. Recent updates to the guidelines require a CEA for major regulations where a significant emphasis is on health and safety, and BCA is required for major health and safety rulemakings to the extent that the primary health and safety outcomes can be expressed in monetary terms. The rationale underlying this
is going on? No, right? So if I gonna be asked to put my life on the line, then front me the cash, man. Pay me more ‘cause it ain’t worth my time to be here when {the gangs are} warring."[29]. 6 WHO Guide to Identifying the Economic Consequences of Disease and Injury, World Health Organization, Department of Health Systems Financing Health Systems and Services http://www.who.int/choice/publications/ d economic impact guide.pdf.
R. Laxminarayan et al. / Vaccine 32 (2014) 5065–5070
latter requirement is that monetized economic benefits indicate what the public is willing to pay for improvements in health and safety. Among the U.S. Federal Agencies that use VSLs are the Consumer Product Safety Commission, the Food and Drug Administration, the Federal Aviation Administration, the Environmental Protection Agency, and the Food and Nutrition Service. For these agencies, VSL values range from $3 million to $6.3 million (in year 2000 dollars), based on the affected population and underlying assumptions. However, agencies whose sole mandate is the improvement of health, such as the National Institutes of Health and the Food and Drug Administration, use health measures—specifically, QALYs. VSLs are used in Australia, Canada and Japan as part of costbenefit analyses of measures to reduce mortality risk. The United Kingdom also uses VSLs, usually based on the revealed-preference approach. VSLs are also used by many multinational organizations, including the World Bank, the European Commission, the United Nations’ Intergovernmental Panel on Climate Change, and the World Health Organization. The VSL concept has been helpful in studying the benefits of chronic disease interventions [8], water and sanitation improvements [9], and the economic and social aspects of climate change. They have been used to estimate the benefit–cost ratio of the Global Plan to Stop TB, relative to a baseline of directly observed treatment–short course (DOTS) for tuberculosis [10]. 4. VSL compared with health metrics used in CEA Many people object to VSL measures on the grounds that they attempt to place a value on life, a task that is fraught with ethical and moral challenges. Actually, what VSL measures do is aggregate individuals’ willingness to pay for a small change in risk, which is then taken as a measure of the financial trade-offs that individuals are willing to make with respect to risk. QALYs and DALYs may appear not to value life, but in reality they only postpone the necessity of translating health measures into economic measures, thereby, allowing policymakers to make meaningful comparisons across sectors. In addition to assigning a value to life, VSL compared to health metrics used in CEA has other properties of importance. 4.1. Sensitivity to nonhealth benefits Although averted morbidity and mortality are important consequences of early childhood vaccinations, they can have important non-health consequences. For example, averting diarrhea through a rotavirus vaccine may have long-term benefits for productivity and economic well-being through multiple channels, including school attendance, physical growth, and cognitive ability [11,12]. VSL metrics would capture such non-health benefits; cost-effectiveness analysis focuses on health outcomes. The extent to which nonhealth benefits are important should inform the choice of metrics in health priority setting. For vaccines, the insensitivity of standard cost-effectiveness analysis to non-health benefits could be seen as a major limitation. Standard cost-effectiveness analysis may grossly underestimate the benefits of vaccines. 4.2. Sensitivity to income Since VSL measures reflect willingness to pay for better health, they are generally positively correlated with income, both in theory and practice. The link between VSLs and income, to some, reflects discrimination against the poor, who are most likely to benefit from childhood vaccines. Using monetary measures of health effects to make health policy decisions may be more controversial than using physical measures of health outcomes, which do not appear to place a value on life, and therefore, may be perceived as more neutral [13]. The link between VSLs and income could be seen as placing greater
5067
weight on policies that favor the rich than the poor (e.g., favoring an influenza vaccine over a malaria vaccine). However, this bias can and should be corrected. It is common practice to use a mean willingness to pay across income groups within a country to avoid this charge. Others have proposed ways to make willingness-topay methods sensitive to income distribution. Fleurbaey et al., for example, have proposed a method for deriving income-sensitive distributional weights based on the concept of equivalent income [14]. Using country-level VSLs may overstate the willingness-topay for vaccines that are more important for poorer populations, but may be a reflection of mean societal willingness-to-pay. 4.3. Sensitivity to age Most empirical research on VSLs assesses the value of reducing the mortality rate for one year by about one in 10,000 for people in middle age. [15], for example, report that in VSL studies from the U.S., the average remaining life expectancy is about 45 years. A key question in applying studies of the value of mortality reduction then, is how to translate the available empirical information – mostly about middle age – to evaluation of interventions that primarily affect the old or, as in the case of vaccines, the very young. Jamison, Summers et al. find that under standard assumptions for VSLs, the value of a death averted in early childhood is about 70% higher than one in middle age—a number that is both implausible and at odds with the limited available empirical information [16]. For this reason Jamison, Summers et al. explore options to applying the proportionality formula for very young ages, and for their headline numbers they use a value for a death averted in young children that is about 50% of the value for middle age. Health indices, such as QALYs, value extensions to younger peoples’ lives and healthy peoples’ lives more than extending the lives of those who are older (and have shorter life expectancies) or infirm (and have lower health status). Self-described health status measures apply to groups of different ages, and health status can be used to generate age- and health-specific QALY measures (see, e.g., [17]), but making these adjustments detracts from the simplicity of applying health measures. Many studies use the same VSL for all adults, and thus postponing an adult’s death by one year – say, age 40 to 41 – represents no gain within the framework. This problem can be fixed by remaining close to the empirical tradeoffs underlying the VSL. In practice one needs values for small changes in age-specific annual mortality probabilities to undertake BCAs, and this comes directly from the empirical literature. The intermediate construct of the VSL is not needed [16,18]. VSLs have been estimated for different age groups, in particular the very young and the very old. The latter “senior discount” reflects the findings in some studies that older people value reducing their death risks less than younger people. But Krupnick [19] finds that this literature is not robust. Health status has an ambiguous effect on the VSL. In contrast, most health indices treat all life years equally, whether from a younger or older person. However, extending a healthier person’s life implicitly is “valued” more than extending an ill person’s. Determining the value of averting stillbirths or the death of newborns and children – relevant for vaccines like tetanus – through VSLs is another challenge. Here the VSL may be based on the parent’s willingness to pay for saving the child’s life, but there are relatively few such studies. 4.4. Insensitivity to baseline risk, size of risk and perception of risk VSL is not sensitive to the concentration of risk in identified individuals. Some studies indicate that people value the same risk reduction differently according to baseline risk. In policy, the
5068
R. Laxminarayan et al. / Vaccine 32 (2014) 5065–5070
so-called “rule of rescue” is an extreme manifestation of this kind of preference. In other words, society should be (and is) willing to pay whatever it takes to rescue a baby trapped at the bottom of a well without regard to costs. However, CEA is not sensitive to baseline risk in this sense. Most labor market VSL estimates rely on workers’ wage-risk trade-offs to capture the pay differential workers require to incur risks on the job. Therefore, they are not really representative of willingness to pay to avert risk but rather a willingness to accept greater risk in exchange for higher wages. There are moral issues related to allowing people to trade money for the higher risk of morbidity and mortality but the fact remains that this is routinely done in labor markets. When markets are imperfect, or when workers are poorly informed about the risk they face in a particular occupation, VSL estimates could understate the willingness to accept greater risk for higher wages. In practice, willingness to pay (WTP) and willingness to accept (WTA) are reasonably identical for small changes in risk. However, the divergence between them may be significant for large changes in risk, such as for a TB vaccine in a high endemicity setting. Typically, WTP is smaller than WTA. A review of studies over a wide range of commodities, including environmental outcomes and real goods, found an average of the mean ratios of WTA to WTP of 7.2 [20]. If VSL estimates are viewed as willingness-to-accept values and are characterized by a similar ratio, then meta-analysis estimates of the median labor market VSL, pegged at $7 million in year-2000 dollars [21], should be reduced to $1 million. How should the choice of metric be influenced by the dread factor associated with potentially catastrophic and unfamiliar diseases like H1N1? All else being equal, should there be greater weight on diseases that are risky and appear involuntary, such as viral influenza, than on more familiar, more avoidable but possibly more deadly risks, such as tuberculosis? Sudden, catastrophic risk may be seen as less acceptable compared with risk of a disease that involves a long period of suffering, even though both may result in the same number of DALYs. Public tolerance for voluntary risks may be as much as 1000 times that for involuntary risks that offered the same dollar benefit per person [22]. In other words, how should the quality of risk be measured against the quantity of risk? The problem of how to deal with risk perceptions is important in both CEA and VSL literatures but has been better addressed in the context of VSLs. These considerations are particularly important in the context of vaccine-preventable diseases, where risk is not always known. 4.5. Sensitivity to context Most VSL estimates come from high-income countries and are extrapolated to low- and middle-income countries using a method called benefits transfer. This method requires an assumption about the relationship between willingness-to-pay and income. While this is often assumed to be proportional, studies of the VSL in developing countries, compared to similar estimates based on benefit transfer, show that this relationship is less than proportional, i.e., that the VSL estimated in such low income countries is higher than that transferred from high income countries, assuming proportionality of VSL to income. Since VSLs are based on the characteristics of the risk and the affected population, the same VSL should not be used for all contexts. For instance, a VSL measured in the context of work-related health risks may not be applicable to diet-related health risks. Similarly, VSLs based on accidental deaths of healthy, working-age adults may not provide appropriate VSLs for valuing the benefits of vaccines for childhood diseases. How valid are such comparisons? Ideally, the estimates of VSLs for vaccinations or tuberculosis treatment are derived from the
same context in which the benefit–cost ratios are calculated. In practice, that may not always be possible, and the benefits transfer method is a crude second best. Moreover, benefits estimates are not very useful without some idea about the incremental costs of proposed policies. The incremental costs include not only the dollar costs, but also the marginal excess burden (or efficiency loss) imposed to pay for the program. 4.6. Transparency Willingness-to-pay measures for acute effects are more transparent in that they are simply the average willingness to pay to avoid a case of an illness. QALYs may appear more transparent than willingness-to-pay measures because they are the product of two components: a health state score and its duration. However, the information underlying the health state scores in QALYs are not transparent—scores can be taken from different indices, which in turn are developed using different approaches for deriving preference weight. This non-transparency comes with implications for health state scores, in the same way that VSL values can vary depending on whether revealed-preference or stated-preference approaches are used. 4.7. Preferences Finally, should uninformed preferences be weighed against informed ones? Poorly educated or ill-informed people may not know the risks associated with their consumption or employment choices. An important concern about VSL estimates is whether individuals can make informed decisions about safety risks that pose challenges even for experts. The same concern applies to health status preferences used in QALYs and DALYs. 4.8. Metric conversion Could we reconcile VSLs, QALYs, and DALYs by converting CEAs into CBAs using a constant conversion ratio—say, a benchmark of dollars to QALYs? Two decades ago, the World Bank described interventions that cost less than US$150 per DALY saved as highly cost-effective [23]. The report of the WHO’s Commission on Macroeconomics and Health describes interventions whose cost is equivalent to GDP per capita as “very cost-effective”, and interventions costing three times GDP as “cost-effective” [24]. Although they are increasingly used, these conversions have no theoretical basis. Moreover, they are arbitrary because QALYs and willingness-to-pay measures vary with life expectancy, health status, and income [25]. That is, the ratio for converting from QALYs to willingness-to-pay is not uniform across individuals7 . 5. Points to consider for institutions using economic methods for the evaluation of vaccines In a recent study, Ozawa and colleagues calculate that $151 billion in treatment costs and productivity losses could be averted by expanding the delivery of six life-saving vaccines in 72 lowand middle-income countries between 2011 and 2020 [26]. This figure is based on low- and middle-income countries’ estimated VSLs derived from empirical estimates from high-income countries, using an assumption of a constant and percentage proportional relationship between income and VSL. Although there are some
7 For instance, QALYs are unaffected by income, but VSLs increase with income, holding other factors constant. Therefore, an individual’s willingness-to-pay per QALY would be increasing with QALY.
R. Laxminarayan et al. / Vaccine 32 (2014) 5065–5070
valid critiques of this study, analyses like these could be expanded to better understand the broader economic benefits of vaccination. What approach might institutions like the WHO, through groups such as the Strategic Advisory Group of Experts on Immunization (SAGE), take to using evidence from studies that use VSLs and BCAs to set vaccine priorities, such as the one by Ozawa and colleagues8 ? Implicitly, the question is whether VSL measures have advantages over the WHO’s well-established cost-effectiveness analyses for setting disease priorities, and whether SAGE should consider using VSLs in its decision-making. VSLs allow benefits to be expressed in dollar terms in a direct way that is consistent with standard welfare theory. This enables direct comparisons with other health and non-health interventions, whereas health measures like QALYs and DALYs do not. Shortcuts to convert QALYs and DALYs into dollars are arbitrary and lack theoretical basis. VSLs, when based on revealed-preference methods, are constructed from actual, observed behavior. They account for both morbidity and mortality risk and take into account important aspects of vaccines, such as avoidance behavior and nonhealth consequences. VSLs can also include “qualitative attributes” of health states, such as the dread associated with catastrophic influenza outbreaks. VSLs are attractive in theory but face some significant challenges as well. Estimates of the revealed value of a prevented fatality may be imprecise. Transferring VSLs from one country and demographic group to other contexts can be poorly related to the VSL that would be estimated in an original study. Moreover, the idea of valuing life may be uncomfortable to some, even though this is not what is being done. QALYs, too, have their drawbacks, in that the health state scores may be developed using different approaches to deriving preference weights and thus not be consistent, and they grossly underestimate the non-health benefits of vaccines. What are the alternatives? Direct health measures, such as lives saved, are less controversial, but their usefulness depends on the policy context and they cannot be aggregated with other types of health effects or non-health effects. The differences among measures could lead to markedly different recommendations from SAGE and other advisory groups for prioritizing vaccines and, more generally, for valuing the economic benefits of vaccines versus other health and non-health interventions. The following are issues that bodies like the Immunization and Vaccines related Implementation Research (IVIR) and Strategic Advisory Group of Experts (SAGE) at the international level, or National Technical Advisory Group on Immunization in countries may want to consider when deciding on whether or not to introduce a new vaccine. 5.1. What is the nature of the decision to be informed by the analysis? • If the objective is to compare two candidate vaccines for the same disease, CEA using QALYs or DALYs may be adequate. • If the objective is to compare two vaccines that address the same disease and in the same target population, the added value of BCA using VSLs may be limited. • If the objective is to compare two vaccines that target different populations and have different non-health benefits, BCA could be used alongside the current CEA approach. • If the objective is to compare vaccines with other health interventions, a BCA using VSLs may be useful; the following two points should then be considered.
8
SAGE is the principal advisory group to WHO for vaccines and immunization.
5069
5.2. What are the belief systems of decisionmakers and of the people subject to these decisions? In one study in Australia, individuals rejected the goal of maximizing QALYs in allocating health resources [27]. Would individuals support the assumptions of life-years saved as the main objective, or would they prefer a metric that reflects risk trade-offs based on actual behavior? Both types of health measures have equity biases (which can be adjusted by an appropriate choice of equity weights), and the choice between them depends on the belief system of the decisionmakers. 5.3. What level of cost and ease of application are decisionmakers looking for? Willingness-to-pay studies based on VSLs can provide useful information on the social value of health interventions but may not be applicable if the context differs from that in which the VSLs were measured. Moreover, relatively few studies have been conducted in low- and middle-income countries, although that is now changing. Revealed-preference studies based on observing how people trade risk for money in labor and consumer markets are now being supplemented by well-designed stated-preference studies. The stated-preference studies have the advantage of being able to focus on population subsets (such as the elderly) that would not normally participate in labor markets. 6. Conclusions CEAs based on QALYs and DALYs will continue to be important and enjoy the confidence of many in the medical and public health communities as a tool for prioritizing interventions. They are easy to administer and require relatively few surveys to develop weights that apply to many types of health states and describe specific health effects. Renaming VSL as value of a prevented fatality (VPF) or value of prevented morbidity (VPM), as is now done by the UK government, may help increase their acceptability outside the domain of economics. VSLs have been commonly used to guide regulatory decisionmaking in the fields of environment, health and safety and transportation. Despite empirical challenges and the paucity of studies in appropriate settings, they are being increasingly used to assist in setting health policy in developing countries. They help provide greater clarity on whether a specific action should be taken from an economic benefit–cost standpoint and, with certain caveats, can usefully complement existing cost-effectiveness methods. If appropriately used, BCA and willingness-to-pay methods will not discriminate against the poor, and they capture important non-health benefits from vaccines such as financial risk protection, education and productivity gains, and economic wellbeing. Acknowledgements We are grateful to two anonymous reviewers at the journal Vaccine, Raymond Hutubessy, John Edmunds, David Bloom and other members of the Immunization and Vaccines related Implementation Research Advisory Committee at the World Health Organization for useful comments. Any errors that remain are solely the responsibility of the authors. References [1] Brouwer WB, Culyer AJ, Van Exel NJ, Rutten FF. Welfarism vs. extra-welfarism. J Health Econ 2008;27(2):325–38. [2] Coast J, Smith RD, Lorgelly P. Welfarism, extra-welfarism and capability: the spread of ideas in health economics. Soc Sci Med 2008;67(7):1190–8.
5070
R. Laxminarayan et al. / Vaccine 32 (2014) 5065–5070
[3] Gyrd-Hansen D. Willingness to pay for a QALY: theoretical and methodological issues. Pharmacoeconomics 2005;23(5):423–32. [4] Goucher J, Horrace WC. The value of life: real risks and safety-related productivity in the Himalaya. Labour Econ 2012;19(1):27–32. [5] Ashenfelter O, Greenstone M. Using mandated speed limits to measure the VSL. J Pol Econ 2004;112:S226–67. [6] Hammitt JK, Graham JD. Willingness to pay for health protection: inadequate sensitivity to probability? J Risk Uncertainty 1999;18(1):33–62. [7] Cropper M, Haile M, Lampetti J, Poulos C, Whittington D. The demand for a malaria vaccine—evidence from Ethiopia. J Dev Econ 2004;75:303–18. [8] WHO. Preventing chronic disease: a vital investment. Geneva: World Health Organization; 2005. [9] WHO. Global costs and benefits of drinking-water supply and sanitation interventions to reach the MDG target and universal coverage. Geneva: World Health Organization; 2012. [10] Laxminarayan R, Klein EY, Darley S, Adeyi O. Global investments in TB control: economic benefits. Health Aff (Millwood) 2009;28(4):w730–42. [11] Lorntz B, Soares AM, Moore SR, Pinkerton R, Gansneder B, Bovbjerg VE, Guyatt H, Lima AM, Guerrant RL. Early childhood diarrhea predicts impaired school performance. Pediatr Infect Dis J 2006;25(6):513–20. [12] Niehaus MD, Moore SR, Patrick PD, Derr LL, Lorntz B, Lima AA, Guerrant RL. Early childhood diarrhea is associated with diminished cognitive function 4 to 7 years later in children in a northeast Brazilian shantytown. Am J Trop Med Hyg 2002;66(5):590–3. [13] Krupnick AJ. Valuing health outcomes: policy choices and technical issues. Washington, DC: Resources for the Future; 2004. [14] Fleurbaey M, Luchini S, Muller C, Schokkaert E. Equivalent income and fair evaluation of health care. Health Econ 2012;22(6):S226–67. [15] Hammitt JK, Robinson LA. The income elasticity of the value per statistical life: transferring estimates between high and low income populations. J Benefit Cost Anal 2011;2:1–27.
[16] Jamison DT, Summers LW, Alleyne G, Arrow KJ, Berkley S, Binagwaho A, Yamey G. Appendix 3 Global Health 2035: a world converging within a generation. The Lancet 2013;382:1898–955. [17] Cutler D, Richardson E. Measuring the health of the United States population. Brookings papers on economic activity. Microeconomics 1997:217–72. [18] Cameron TA. Euthanizing the value of a statistical life. Rev Environ Econ Policy 2010;4:161–78. [19] Krupnick A. Mortality-risk valuation and age stated preference evidence. Rev Environ Econ Policy 2007;1(2):261–82. [20] Horowitz JK, McConnell KE. A review of WTA/WTP studies. J Environ Econ Manage 2002;44:426–47. [21] Viscusi WK, Aldy J. The value of a statistical life: a critical review of market estimates throughout the world. J Risk Uncertainty 2003;27(1):5–76. [22] Starr C. Social benefit versus technological risk. Science 1969;165(3899): 1232–8. [23] World Bank. World development report: investing in health. New York, NY: Oxford University Press; 1993. [24] WHO. Macroeconomics and health. In: Report of the commission on macroeconomics and health. Geneva: World Health Organization; 2001. [25] Hammitt JK. QALYs versus WTP. Risk Anal 2002;22(5):985–1001 (an official publication of the Society for Risk Analysis). [26] Ozawa S, Stack ML, Bishai DM, Mirelman A, Friberg IK, Niessen L, Walker DG, Levine OS. During the decade of vaccines, the lives of 6.4 million children valued at $231 billion could be saved. Health Aff 2011;30(6):1010–20. [27] Nord E, Richardson J, Street A, Kuhse H, Singer P. Maximizing health benefits vs egalitarianism: an Australian survey of health issues. Soc Sci Med 1995;41(10):1429–37. [28] Grüne-Yanoff T. Mismeasuring the value of statistical life. J Econ Methodol 2009;16(2). [29] Levitt SD, Venkatesh SA. An economic analysis of a drug-selling gang’s finances. Q J Econ 2000;115(3):755–89.