7.6 Cost-effectiveness analysis: concepts and applications*
Oxford Textbook of Public Health
Cost-effectiveness analysis: concepts and applications*
Dean T. Jamison
Assessing the cost-effectiveness of intervention
Outcome measurement: QALYs
An application: the World Bank Health Sector Priorities Review
Five strategies of public health intervention
Lessons from the Health Sector Priorities Review
Many of the world’s poorest countries spend less than $10 per person per year on health services. High-income countries spend thousands of dollars per year. Yet across this entire expenditure range of several orders of magnitude questions arise about value gained for money spent on health. Most poor countries suffer huge burdens from some mix of childhood infection, malaria, maternal deaths, tuberculosis, and HIV/AIDS. Highly effective interventions exist to address most (but not all) of these conditions. If a few additional dollars per year were available, misspending those dollars on interventions offering relatively little health gain for the money would entail lost opportunities to postpone many deaths and prevent much serious disability. High-income countries also face choices: Could an improved mix of interventions reduce overall costs (or at least their rate of growth) while maintaining existing levels of health? Which of the effective but usually costly new interventions emerging from the research and development pipeline should public or private insurance plans cover?
Therefore concern with value for money (or cost-effectiveness) spans income levels. Neither is the issue one principally for the private sector, for non-governmental organizations, or for the public sector. Each has a potential interest. Analysts have responded to this interest over a period of several decades and produced a substantial literature on both methods and results. Thus the purpose of this chapter is to introduce the reader to this literature.
The chapter begins with a brief discussion of background and terminology, and then describes methods of cost-effectiveness analysis (CEA). It illustrates use of these methods with an application that provides a sense of results and one way that they can be used. We are all used to hearing statements of the sort ‘prevention is more cost-effective than cure’ or ‘tertiary facilities are not cost-effective in low-income countries’. The application section includes discussion of the cost-effectiveness generalizations that would support or undermine such propositions. More frequently, analysts use CEA to assess options for dealing with a particular problem—options involving scale of intervention or choice of technique. Discussion in this chapter covers these issues as well.
CEA in health comprises one part of a very much larger literature on project appraisal, i.e. on assessment of the economic desirability of alternative ‘projects’ from a social perspective. Antecedent to such analyses is assessment of whether the project should be financed by the public sector (or by the decision-maker using the analysis). The perspective of this chapter is, for the most part, that of a public-sector decision-maker responsible for a broad range of preventive and curative health services. Barr (1993) reviews arguments, mostly around insurance market failure, for including personal clinical services for the entire population in what is to be publicly financed. Musgrove (1995) discusses related issues. Table 1 lists three approaches to the economic appraisal of projects or interventions and indicates their realm of applicability.
Table 1 Choice of economic appraisal techniques
Cost-minimization analysis examines the costs of alternative approaches to achieving a quite specific objective, for example the cost per infant death averted or new HIV infection averted. The purpose is to identify the least-cost method of achieving the objective and to see how both cost and choice of technique vary as the magnitude of the objective varies. For example, if one had very modest goals with respect to the prevention of HIV infection, the least-cost approach might very well be blood screening in hospitals; more substantial goals would entail addition of more costly programmes—treatment of sexually transmitted diseases, condom use, etc.—to achieve the goal at minimum total cost. Note that in this example, as will often be the case, the average cost per HIV infection averted will rise as the target number of infections averted rises. Cost-minimization analysis has the virtues of specificity and of ease of communication concerning results. The disadvantage becomes apparent if there is a need to compare the attractiveness of efforts to reduce infant mortality rate with those to avert HIV infections: costs can be compared but outcomes remain incommensurable.
Cost–benefit analysis, in contrast, allows comparison of projects (or interventions or investments) across the entire economy. It does so by placing monetary values on outcomes as well as inputs. Kilowatt-hours of electricity can be compared with kilograms of rice by multiplying each by its price to obtain a total value. However, the simple word ‘price’ conceals vast complexities, particularly when used to measure social benefits. There is an extensive literature on the theoretical methods as well as applications in different contexts of monetary valuation of benefits or costs when markets are distorted or incomplete. Stern and Ferreira (1997, pp. 567–9) provide a brief overview of these issues and of related issues concerning the distribution of costs and benefits. Benefits and costs occur over time—with benefits usually following costs—and alternative figures of merit (e.g. present value of net benefits or the internal rate-of-return) generate orderings of outcomes by desirability. Squire (1989) and Layard and Glaister (1994) provide excellent overviews of the methods of cost–benefit analysis and the related literature.
Practical difficulties associated with monetary valuation of benefits often lead analysts to utilize the much simpler methods of cost minimization (with the concomitant limits on applicability of the results). In addition to practical difficulties there is the more fundamental problem, in assessing health intervention options, of placing dollar value on human life (or other health outcomes). Sometimes this can be non-controversial, as when Levin et al. (1993) use labour productivity increases associated with reducing anaemia to derive benefit measures to weigh against the costs of anaemia control. Their findings—of high dollar benefits relative to dollar costs—can either be compared with findings for interventions in other sectors or, more importantly, to assess intrinsic value: if benefits exceed costs the intervention is worth doing (ignoring possible public-sector fiscal constraints). If one can overcome practical and other problems with cost–benefit analysis, its results have the virtue of standing alone in the sense of indicating intervention desirability independently of comparison with alternatives.
In the health sector, CEA lies between cost-minimization analysis and cost–benefit analysis (Table 1). CEA rests on a non-financial metric designed to allow comparisons across the health sector. The concept most typically used is that of the quality-adjusted life-year (QALY), which can be measured in many ways, but which then allows costs per QALY gained to be compared for interventions addressing a broad range of problems (by assigning, for example, a QALY value both to an HIV infection averted and to an infant death averted). However, even focusing analysis to within the health sector cannot be completely done by CEA. Some interventions which may be undertaken principally for health reasons, such as reducing ambient air pollution, have other outcomes—in this case reduced pollution-related corrosion and the amenity value of clean air. These outcomes elude the QALY metric but must be explicitly listed as inputs to the decision-making process.
This chapter focuses on CEA. That said, work on cost minimization will often in practice prove essential to CEA. Likewise, empirical observations of what societies appear to be prepared to pay for a QALY, or more frequently to avert a death (which can be converted to QALYs), have increasingly been undertaken. Jones-Lee (1994) reports on 27 valuations of a ‘statistical life’ of which he categorizes 16 as ‘more reliable’. While 10 of the 16 studies estimated a value of over £1 000 000 for a statistical life, one estimate fell under £250 000. These (or other) estimates of the value of a statistical life allow CEAs to be immediately translated into cost–benefit analyses. From experience it can be suggested that an explicit valuation of human life for cost–benefit analysis usually generates reactions that distract from a discussion of improving efficiency of resource allocation in the health sector. The interested reader is referred to Viscusi (1993), Jones-Lee (1994), Tolley et al. (1994), and Pauly (1995) for valuable reviews of the monetary valuation of health outcomes.
Part of the value of undertaking CEA lies in the ability to formulate generalizations—or to indicate their inapplicability. Doing so requires care and consistency concerning the definitions that underlie the generalizations, and in this chapter attempts to be quite explicit. Table 2 provides a number of definitions and distinctions that will be used later in the chapter. Perhaps the central point to note in Table 2 is the distinction between ‘interventions’ per se and the ‘instruments of policy’ that can encourage (or discourage) intervention or intended behaviour change. Although most CEAs concern intervention, some concern instruments of policy (Wells 1999). More is needed concerning the latter, which, after all, is what policy can do to change behaviour. In this context it will be important to consider the potential fungibility of resources if the policy instrument consists of direct investment. Lower-level decision-makers may redirect previously allocated resources away from activities that a higher-level authority is seeking to encourage by funding it (Devarajan et al. 1997).
Table 2 Definition of terms
CEAs in the literature vary substantially in their underlying methodologies and assumptions, and therefore comparisons are frequently difficult. Yet without being able to compare substantial numbers of interventions—a point to which we shall return—the relative attractiveness of individual interventions remains uncertain and generalizations are difficult or impossible. To define best practice in methods and to provide a template for comparative studies, the United States Public Health Service convened a major review panel in 1993. Gold et al. (1996) reported its conclusions. Discussion in this chapter for the most part follows the United States Public Health Service guidelines. More extensive and technical discussion of the theory and methods of CEA than is appropriate for this chapter is provided by Garber et al. (1996) and Garber (2000), and the interested reader is referred to these reviews.
Assessing the cost-effectiveness of intervention
This section contains a discussion of general issues associated with choosing interventions, i.e. with criteria for cost-effective choice. The nature of the instruments open to government to promote cost-effective intervention have been outlined in Table 2. The purpose is not to provide an account of the (many) methodological issues associated with economic assessment of intervention options; rather, it is simply to describe the basic concepts being applied, raise a few particular issues, and refer the reader to the relevant literature. In addition to the comprehensive work for the United States Public Health Service previously mentioned, valuable additional background can be found in Weinstein and Stason (1977), Drummond (1987a,b) Johannesson and Weinstein (1993), Sloan and Conover (1995), Weinstein (1995), and Drummond et al. (1997).
As previously indicated it is useful to consider two distinct uses for CEA. One is to inform broad policy generalizations and the other is to help assess the relative attractiveness of changes in the scale of implementation of an intervention or in the technique for addressing a specific problem. In either case the analyst must specify a base case and define the intervention as a change from that base. For policy generalizations it will typically be useful to include consideration of large changes; for addressing specific problems more modest increments will be typical. The natural base case for dealing with specific problems will usually be the status quo, and what is to be considered as ‘given’ for the purpose of analysis will usually be substantial (although dependent on time frame). Establishing a base case for policy generalizations is less obvious. Guidelines being developed at the World Health Organization (WHO) (Murray et al. 1999) suggest using the ‘…null set of related interventions.’ Substantial practical difficulties are likely to be associated with ascertaining the consequences of no intervention, and the utility to policy-makers of trying to imagine a starting point so different from their own may be limited. In most cases a more natural approach will be to identify base cases close to current reality for policy-makers in a number of paradigmatic circumstances. Incremental cost-effectiveness assessments from those bases will then provide more naturally interpretable information. In this context it will often prove important to consider explicitly the effects of doing less than is being done in the base case, thereby generating negative costs and negative effects. Such ‘negative intervention’ may often prove highly cost-effective.
Outcome measurement: QALYs
A critical choice in applications of economic analysis to resource allocation is that of whether to value outcomes because of their economic benefits or because of some more proximal effectiveness measure (Table 1). To provide a clearer sense of the context for CEA it is worth a brief additional discussion of approaches to monetary valuation of health outcomes. To a lesser extent some of these points also relate to cost analysis. When there are good markets for products, benefits can be assessed in monetary terms by using market prices (i.e. willingness of consumers to pay) to value benefits as well as to value costs. Even when willingness-to-pay valuation cannot be assessed directly because of lack of market prices, as is typically true in the health sector, questions in surveys are increasingly being used to elicit information about hypothetical willingness to pay (or contingent valuation). Briscoe and de Ferranti (1988) indicate the potential for contingent valuation in water projects, and there are early applications in education (Jamison and Lumsden 1975, Table 2). Pervasive problems of consumer ignorance of effectiveness of intervention and a widespread tendency for individuals systematically to underestimate risks (Weinstein 1989) suggest that willingness-to-pay assessments will need to be used with caution when applied to health. An alternative approach—sometimes called the human capital approach—is to view health investments as instrumental to improving economic productivity; thus estimates of the effect of a health intervention on productivity provide a lower bound to total benefits. One example comes from assessing the effect on the productivity of rubber plantation workers of correcting iron deficiencies (Basta et al. 1979; Levin et al. 1993); other examples come from assessment of the effect on economic productivity of malaria control efforts. It is worth noting that both the willingness-to-pay and the human capital approaches inevitably imply different values to be attached to the life of different individuals of the same age in the same country—and even greater variation across countries. Phelps and Mushlin (1991), Johanneson and Meltzer (1998), Bleichrodt and Quiggen (1999), and Garber (2000) further discuss the close relation (and sometimes equivalence) between cost-effectiveness and cost–benefit analyses.
More typically, however, outcomes will be assessed in deaths or disability averted, rather than dollars, and the task is to come up with some measure for making such an assessment that allows comparisons across the health sector (i.e. that allows CEA), even if intersectoral comparisons (cost–benefit analyses) remain infeasible or subject to excessive ethical debate. There is now a valuable literature on how effectiveness measures to aggregate the disability-, morbidity-, and premature-mortality-averting effects of interventions across the health sector might be constructed and applied (Zeckhauser and Shepard 1976; Kaplan and Bush 1982; Barnum 1987; Feachem et al. 1992; Fox-Rushby and Hanson 2001). Such measures, in addition to providing the effectiveness measures for CEAs, can be used with epidemiological information to assess the burden of disease in a population, as has been done for the major regions of the world by Murray et al. (1994) (most recently updated in WHO (2000)).
Table 3 sets forth the characteristics of the main approaches to disability weighting that serve as the core of effectiveness measurement. Stouthard et al. (1997) provide a clear exposition of methods for disability weighting with an informative application for The Netherlands. From a practical perspective, the use of ratings based on expert judgement is probably the best that can now be done if the purpose of the analysis is to compare interventions across the sector; however, as Preston (1991) has noted, these measures must be used with care. It is also worth noting that the construction of QALYs requires value judgements, although they are less subject to controversy than is explicit valuation of human life. (Even measures involving mortality only, such as numbers of deaths averted, while they appear to be value-free, if used to measure intervention effectiveness or disease burden, rest on strong value judgements. Minimally, a mortality based measure rests on the implicit value judgement that disability is not a concern.)
Table 3 Alternative approaches to measuring outcomesa
A workable measure for effectiveness for most CEAs will be QALYs gained. The QALY gain associated with averting a death at a given age is, simply, the life expectancy at that age (in the local environment), with life-years gained in the future discounted back to the present (typically at a discount rate of 3 per cent per annum). Unhealthy life-years are given lower weights than healthy ones, depending on the degree of disability (assessed by one of the rating procedures listed in Table 3); therefore the effectiveness of interventions to address morbidity or disability can be measured in terms that allow comparison with interventions that delay mortality. The QALY measures now used are particular forms of the more general concept introduced by Zeckhauser and Shepard (1976). Garber and Phelps (1997) provide the basic theoretical underpinnings for CEAs in health that adjust life-years for quality; in particular, they point to conditions allowing a dollar value to be assigned to a QALY so that, if desired, a CEA can be directly reinterpreted as a cost–benefit analysis.
Timing of outcomes can be dealt with through discounting. Johannesson (1992) provides a general discussion of discounting healthy life-years, and Cropper et al. (1992) report empirical assessments of time preference for saving lives. Most analysts value years of healthy life at all ages equally; however, this assumption can be readily relaxed to give greater weight to those age groups likely, say, to have more dependants (Musgrove 1991). The variant of the QALY known as the ‘disability-adjusted life-year’ (DALY) does weight different age groups differently. DALYs have been used for disease burden assessment and CEA in a number of recent World Bank and WHO documents (World Bank 1993; Murray et al. 1994; WHO 1996, 2000). An overview is given in Chapter 2.9. Sensitivity analyses were undertaken in the initially published disease burden assessment using DALYs (Murray et al. 1994) and it was concluded that results were insensitive to age weights over a broad range.
In principle, QALYs can also be weighted to reflect how equitably they are distributed in ways that are standard in project evaluation outside the health sector (Squire 1989). Wagstaff (1994) provides a valuable overview of this possibility (but concludes that a more general approach, not involving QALYs, would be preferable).
An important outcome of publicly financed intervention, which is usually neglected in the CEA literature, is the contribution to reducing financial risk for households. A tuberculosis intervention and measles immunization may, in the relevant age range, have costs per QALY that are close, but the greater costs of treatment and income loss associated with tuberculosis would entail much greater risk protection from the former. A significance evaluation of the Medicare program in the United States (Skinner and McClellan 2000) explicitly examined this issue. More will need to be done in the future.
Costs of inputs are generally assessed at market prices. However, this simple observation masks much complexity, both conceptually and in practice. Luce et al. (1996) provide a valuable overview of these issues in the specific context of CEA for health. Several important issues are highlighted below, but the interested reader is referred to Luce et al. (1996) for a more thorough treatment.
Tradeable and non-tradeable inputs
Costs for some inputs into health care (e.g. semiskilled labour) may be lower in developing countries. These costs are typically for inputs that cannot be traded internationally, and their existence undermines attempts to estimate costs that are not simply country specific. Squire (1989) provides a general discussion of approaches to dealing with tradeables in project analysis through the use of ‘shadow prices’. His recommendations are more relevant to country-specific assessments than to cross-national comparisons.
The working conclusion of this chapter is that considerations of cost variability between high- and low-income countries are of minimal significance (relative to other uncertainties)for tradeables (e.g. non-patented drugs, most equipment, and high-level manpower). Real costs for facilities and lower-level manpower do vary across countries, leading some analysts to conclude that costs are most usefully expressed as fractions of local per capita income—a method that assumes that essentially no health sector inputs are internationally tradeable. The CEA for cancer interventions developed by Barnum and Greenberg (1993) is an example of an attempt to divide costs into those for traded goods and those for non-tradeables. Their assessments suggest that local costs will often be important and that those who attempt to assess the cost-effectiveness of intervention in a country-specific context should pay close attention to this issue unless there is a free market for foreign exchange and the costs of non-tradeables are similar to those of the comparator country. It is a matter of judgement about the extent to which costing of non-tradeables undermines efforts to form generalizations across countries. My own view is that such generalizations are both useful and possible, but that they are best done within groups of countries with broadly similar income levels.
Patient and home provider time
Another important issue in cost analysis concerns assessment of the amount and value of time required of patients or caretakers. Attention to time costs is important both for improving cost analyses and because behavioural response to the availability of an intervention may be sensitive to time requirements. The importance of mothers’ time, in particular for compliance with child survival interventions, has been stressed by Leslie (1989). These time costs are potentially difficult to value (Briscoe and de Ferranti 1988) and are often neglected. The United States Public Health Service provides recommendations for subsequent work that would help to redress this omission (Gold et al. 1996). A related issue concerns treatment of costs that will ensue from intervention success. For example, Levin et al. (1993) point out that substantial food costs can result from micronutrient supplementation or parasite control—appetites improve. The existence of such costs suggests the importance, in these cases, of broadening the definition of the intervention.
A final issue concerning cost analysis is that of joint costs, i.e. the situation where several interventions are essentially made available with a (partially) common set of inputs. Over (1988) provides an extended discussion in the context of immunization. Some authors handle this in part by defining interventions in terms of natural packages; for example, Jamison et al. (1993) consider the preventive intervention for poliomyelitis to be diphtheria–pertussis–tetanus vaccine plus poliomyelitis immunization, and to assess the cost-effectiveness of that package, because poliomyelitis immunization would usually be given with the other vaccines. In many cases, however, such packaging would become too unwieldy, and in these cases analysts should note where joint costs would need to be considered in country-specific applications.
Thus, for comparisons across interventions, CEAs use the common metric of dollar cost per QALY gained, with the understanding that incremental costs and cost-effectiveness will probably vary across locales (even after controlling for intervention quality) because of differences in individuals, in epidemiological conditions, in delivery system characteristics, in the initial degree of penetration of the intervention into the population, and in the range of available alternatives. Table 4 lists many important factors that lead to variation in incremental cost-effectiveness, and, to the extent that interventions are first applied where their cost-effectiveness is highest, these factors collectively will lead to rising costs per QALY with increased application of an intervention. Figure 1 illustrates this for control of dengue; up to a point, improved case management is most cost-effective, but beyond that point chemical and then environmental strategies of vector control must be introduced if a higher level of control for dengue is to be sought.
Table 4 Factors influencing variation in cost effectiveness
Fig. 1 Increasing cost per QALY associated with more complete control of dengue.
Intervention specificity and targeting
The phenomenon of rising costs per QALY comes up implicitly in many analyses; the cause of the phenomenon is, frequently, the lack of intervention specificity and, also frequently, the need for costly targeting, case-finding, or compliance monitoring. Intervention specificity refers to what fraction of intervention recipients would benefit assuming that the intervention is applied exactly to the individuals to whom it should be applied. Specificity will be influenced by such factors as ‘prevalence of the condition’, ‘incidence of condition’, and ‘levels of risk factors’ (Table 4). For example, many countries specify that BCG vaccination for tuberculosis be applied to all newborns, but it is a benefit, ex post, only to that tiny fraction of children who would have died in childhood from miliary tuberculosis without it. In contrast, tuberculosis chemotherapy for sputum positives, although costly, will not be applied unless necessary—it is highly specific. Initially targeting BCG or other interventions to populations at highest risk, although inevitably at some cost, will maximize cost-effectiveness while simultaneously advancing equity objectives (Mosley and Jolly 1987). Although the incremental cost per QALY gained by expanding coverage may be rising, sufficient resource availability may justify expansion.
To continue the tuberculosis example, patients who seek care, and who are then compliant with the treatment regimen, cost less than those for whom active case-finding is required or who require careful monitoring for compliance. All these factors lead to another reason for rising costs per incremental QALY gained. To take another example, oral rehydration therapy in the hospital or clinic setting is highly cost-effective; it will only be used for severe cases of diarrhoea, and it is likely to be applied effectively by qualified medical personnel. However, when oral rehydration therapy is taken to the community, cost-effectiveness declines substantially both because of a decrease in intervention specificity (mild cases will be treated unnecessarily) and because home treatment will be applied less effectively than hospital treatment in severe cases.
These points are relatively obvious, but there is often an optimistic bias towards assessing cost-effectiveness under assumptions of favourable targeting and compliance costs and of favourable intervention specificity. One might expect, as previously noted, rising marginal costs and decreasing marginal effectiveness as interventions are extended through populations; these combine to dilute cost-effectiveness. Thus, favourable case cost-effectiveness estimates can be real, but their margin of applicability may be limited. In principle, it is desirable to acquire some sense of the responsiveness of intervention cost-effectiveness to a range of parameters, particularly the extent of application of the intervention. In practice, sensitivity analysis is sometimes possible but often difficult—and comparisons are then made for ‘representative’ estimates of incremental cost-effectiveness to provide general guidance to decision-makers. When there are great differences in the incremental cost-effectiveness of different interventions—as this chapter concludes there to be—this ‘general guidance’ can suggest important redirections of policy.
When an intervention requires large fixed costs, total programme costs need to be weighed against total effects; simple assessment of marginal cost and effectiveness fails to suffice. The fixed costs involved in (to take several examples) investing in major facilities, mounting a media-based health education programme, or devising regulations and procedures can be substantial. Fixed costs need not be financial; managerial or political attention to a problem may have an important fixed-cost element. When fixed cost may be important, understanding the total burden of disease is necessary for estimating potential total intervention effects. By the same token CEAs will need to include consideration of large increments in intervention. (Examples include Barnum et al. (1980) for analysis of simultaneous scaling up of multiple child survival interventions or Watts and Kumaranayake (1999) for a brief discussion of scaling up AIDS control interventions in Africa.)
Disease burden assessment needs can be combined with CEA in an explicit way to help evaluate where there might be large payoffs to research and development investments or to focused political or managerial attention on reallocation of interventions. This requires an analysis, essentially, of whether a major disease burden persists mainly because of a lack of knowledge about the disease and its determinants, a lack of tools, or failure to use the existing tools efficiently. Of course, more than one factor is likely in each case. Where possible, this analysis can be quantitative. Figure 2 illustrates an analytical approach applied recently (WHO 1996). By using data on the efficacy of the available cost-effective interventions and consulting the judgement of field experts on the proportion of the population receiving effective interventions, it is possible to estimate the following.
Fig. 2 Analysing the burden of a health problem to identify control and research needs: R&D, research and development.
What portion of the potential burden of each disease or condition is now being averted.
What could be averted now with better use of existing cost-effective interventions.
What could be averted now, but only with interventions that are not cost-effective.
What cannot be averted with existing interventions but would require new ones.
The analysis is intended to identify where the greatest needs lie, and thereby guide assessment of priorities for different major fixed commitments such as research and development or political attention. The unit of currency employed for this analysis is, once again, the QALY. While such analyses are not intended to suggest that some spurious precision can be achieved in the analysis of need, they do indicate a sense of the approximate distribution of the effort required.
The area of the rectangle in Fig. 2 represents the total estimated disease burden (in QALYs) from a given condition (e.g. diarrhoeal disease), under the counterfactual assumption that current explicit control interventions were not being applied. The horizontal axis represents the extent to which effective treatment is reaching the population, i.e. how far into the population a mix of interventions is penetrating. The vertical axis represents the combined efficacy of this mix. The subdivisions within that square represent different portions of the burden: that which is being averted now by the existing mix of cost-effective interventions among the people that the intervention is reaching, that which could be averted if the existing interventions were used more efficiently, that which could be averted with existing tools, but not cost-effectively, and that which could not be averted with existing interventions. Calculations of the relative share occupied by each subdivision can help to spell out the priorities. For example, where it is calculated that a large portion of the total burden of a certain disease cannot be averted with the existing cost-effective tools, there is a strong case for research and development to develop new ones (if the disease burden is sufficiently large). Where it is calculated that a large fraction of the burden could be averted if existing tools were used more efficiently, and the absolute disease burden is large, there is a strong case for political and managerial attention to achieve fuller employment of available cost-effective interventions. Meltzer (2001) and Kremer (2001) provide valuable discussions of assessing the cost-effectiveness of research and development investments in health.
Non-health outcomes of health interventions
An additional problem in applications concerns interventions that have outcomes outside the health sector. Table 5 lists a number of important examples. Obviously, CEA applied to health outcomes only will understate the overall value of these interventions. While cost–benefit analysis would solve this problem, applicability may be difficult for the reasons previously discussed. Under these circumstances a clear listing of costs, probable health effects, and non-health effects will at least inform the analysis.
Table 5 Selected interventions with multiple outcomes
Perhaps the clearest examples are the control of smoking, the promotion of breast feeding, and environmental improvements. Limitation of smoking markedly reduces risk for lung cancer, ischaemic heart disease, and chronic obstructive pulmonary disease; outside the health sector it reduces (at least to some extent) property damage from fire and frees productive resources for alternative use. Likewise, breast feeding has multiple health effects: it enhances child immunity, reduces exposure to infection, provides balanced nutrition, and, by suppressing ovulation, postpones the next pregnancy (Anderson 1990). However, the cost of breast feeding includes, like many health-promoting interventions, substantial amounts of mothers’ time, which is not easily valued in terms, say, of wages forgone (Leslie 1992). Finally, whereas environmental interventions have beneficial health consequences, their main objectives may lie outside the health sector; World Bank (1992) provides a comprehensive discussion.
Thus when interventions for health have a range of non-health outcomes, assessment of the attractiveness of these interventions should, ideally, quantitatively aggregate intervention effects along multiple dimensions. Likewise, for clinical intervention there will frequently be joint costs (associated, for one example, with the availability of diagnostic facilities in a district hospital); again, in country-specific application, these matters can be assessed more quantitatively than they can be in a general overview.
The purpose of this section has been to introduce concepts without attempting to provide a detailed discussion of methods. In the next section an extended example of application of CEA is provided both to convey broad substantive lessons and to indicate how CEA has now become a working tool of the health policy analyst. A number of valuable handbooks on methods exist and, as indicated earlier, this chapter is in the spirit of the United States Public Health Services recommendations. Box 1 encapsulates that perspective.
Box 1 United States Public Health Service recommendations on cost-effectiveness analysis
In 1993 the United States Public Health Service convened a Panel on Cost-Effectiveness in Health and Medicine. The Public Health Service asked the Panel to assess the current state of the art of CEA in health and to provide recommendations for the conduct of future studies. Gold et al. (1996) bring together the Panel’s conclusions, and their Appendix A provides a summary of recommendations. The following extracts provide the highlights of that summary.
Purpose of CEA
CEA evaluates a given health intervention through the use of a ‘cost-effectiveness ratio.’ In this ratio, all health effects of the intervention (relative to a stated alternative) are captured in the denominator, and changes in resource use (relative to the alternative) are captured in the numerator and valued in monetary terms.
CEA is an aid to decision-making, not a complete procedure for making resource allocation decisions in health and medicine, because it cannot incorporate all the values relevant to such decisions.
The major categories of resource use that should be reflected in the numerator of a cost-effectiveness ratio include costs of health-care services, costs of patient time expended for the intervention, costs associated with care-giving (paid or unpaid), other costs associated with illness such as child care or travel expenses, and costs associated with non-health impacts of the intervention (e.g. on the education system or the environment).
Time spent seeking care or undergoing an intervention is a resource and a component of the intervention. It should be valued in monetary terms and incorporated in the numerator of a cost-effectiveness ratio. For individuals in the labour force, wages are generally an acceptable measure of time costs.
In aggregating resource costs across time, CEAs should be conducted in constant dollars that remove general price inflation.
‘Transfer payments’ (e.g. cash transfers from tax payers to welfare recipients) associated with a health intervention redistribute resources from one individual to another. While administrative costs associated with such transfers are included in the numerator of a cost-effectiveness ratio, the transfers themselves are not, as by definition, their impact on the transferer and the recipient cancel out.
Incorporation of morbidity and mortality consequences into a single measure should be accomplished using QALYs. In general, as lives saved or extended by an intervention will not be in perfect health, a saved life-year will count as less than one full QALY.
In general, community preferences for health states are the appropriate ones for use. If distinct subgroup preferences are identified that will markedly affect a cost-effectiveness ratio, the study should provide this information and conduct sensitivity analyses that reflect this difference.
The health-related quality of life of those whose lives have been saved or extended by a health intervention may be influenced by characteristics such as age, gender, or race. This may affect the analysis in ways that are ethically problematic. In these instances, sensitivity analyses should be conducted to indicate explicitly how the results are affected by these characteristics.
Costs and health outcomes should be discounted to present value with the shadow-price-of-capital approach to evaluating public investments. This rate (often termed the social rate of time preference) can be approximated by the real rate of return on long-term government bonds, and a real riskless discount rate of 3 per cent is now appropriate. Because of the large number of previous CEAs that have adhered to a discount rate of 5 per cent, analysts should perform sensitivity analyses using 5 per cent. The discount rate should be subject to review, and possible revision, over time in light of significant changes in the underlying economic data.
Costs and health outcomes should be discounted at the same rate.
At a minimum, univariate (one-way) sensitivity analyses should be conducted in order to determine where uncertainty or lack of agreement about some key parameter’s value could have substantial impact on the CEA’s conclusions.
Where possible, where parameter uncertainty is a major concern, a reasonable confidence interval should be estimated based on either statistical methods or simulation.
An application: the World Bank Health Sector Priorities Review
In this section we summarize the findings of 25 condition-specific analyses, principally relevant to low- and middle-income countries, which were undertaken for the World Bank’s Health Sector Priorities Review (HSPR) (Jamison et al. 1993). Earlier in this chapter it was noted that dividing interventions into two broad categories—public health and clinical—was conducive to discussing policy trade-offs and this section is so divided. (Table 2 defined what is included in each of these categories.) First we deal with public health interventions, and then with clinical interventions. Unless otherwise specified, the assessments are of incremental cost-effectiveness from an implicitly defined typical starting point, and they are designed to reach generalizable conclusions rather than to inform decision-making in a specific context. The need for manageability constrained the range of conditions covered; Jamison (1993, pp. 3–4, 6) discusses selection criteria and outcomes for what was viewed as an initial effort. The HSPR reached a number of substantive conclusions and these are discussed to give a sense of the input CEA can make to informing policy.
Five strategies of public health intervention
Public health interventions were organized into five separate strategies in the HSPR: those designed to change personal behaviour, to control environmental hazards, to immunize, to provide mass chemoprophylaxis, and to establish mechanisms for screening and referral. In reviewing health policies, or intervention alternatives, it will often be useful to do so within each of these five broad strategies because of commonalities of logistics, policy instruments, and approaches within each. (This is true despite the frequently great diversity of conditions to be addressed within any one intervention strategy.)
Before turning to the summary of findings, the issue of joint costs (and multiple outcomes) of interventions in light of conclusions from the HSPR are discussed. The analysis upon which the HSPR was based was structured by diseases (or adverse health conditions more generally), and the issues addressed in the individual analyses thus concern the nature, cost, and effectiveness of the interventions available for dealing with each condition. In many cases, of course, any given intervention will address multiple conditions and, indeed, may well have important effects outside the health sector altogether.
Looking across findings of the individual chapters in the HSPR, it is clear that multiple effects and joint cost problems complicate the task of assessing cost-effectiveness in many important instances; that said, it is more generally true that these problems are relatively minor or can be dealt with by reasonable approximations and simplifications in the analysis.
A few general conclusions on each public health approach emerged from the HSPR.
Personal behaviour change
Some personal behaviour changes that are favourable for health outcomes tend to occur naturally as incomes rise; these include, at least for many cultures, improved hygienic behaviours, increased energy intake and quality in the diet, and decreased crowding. Improvements in these behaviours are typically important for the pre-epidemiological transition diseases and can often be affected by educational interventions even though the main force driving improvements—income increases—is beyond the domain of health policy.
Other behaviours are likely either to be less dependent on income levels (e.g. breast-feeding behaviour, sexual practices) or to be adversely influenced by income increases, at least for a period of time (e.g. dietary excess, sedentary lifestyle, smoking, alcohol consumption). Most of these are risk behaviours for post-transition conditions. Although the natural course of development may well improve these behaviours, the HSPR found scope for affordable government policy to influence them. Regulatory policies and, in particular, taxation policies for tobacco, alcohol, and fatty meats show great promise for inducing behavioural change and, currently, are very much underused. Education of elites and the public are complementary instruments, not least because they generate the political will and popular support for regulation and taxation. The extremely high cost-effectiveness of smoking control makes it, perhaps, the top priority for governmental action.
Environmental hazards control
Rising incomes help with improving water supply and sanitation, which are likely to be important in the prevention of a broad range of infectious and parasitic diseases. Specific investments in water supply and sanitation are unlikely because of high costs to be justified in terms of health benefits alone. However, vector control is at least marginally cost-effective for a number of conditions (malaria, onchocerciasis, dengue) in some environments. Industrialization introduces new hazards into the environment (lead, mercury, etc.) that can produce severe lifetime disability if not effectively controlled. Cleaner fuels and improvements in ventilation of indoor fireplaces and cookstoves can substantially reduce risks for chronic obstructive pulmonary disease; and occupational and transport safety measures are important in many specific instances. In principle, protective measures can be delivered through environmental intervention; water fluoridation for the prevention of caries is one example. Another problem is lead toxicity resulting from excess use of lead-based paints and combustion of gasoline with high lead content. Some research (reviewed by Pollitt (1990)) indicates that lead toxicity may be far more important than previously thought as a determinant of slow development and impaired mental functioning.
Immunization, mass chemoprophylaxis, and screening
Interventions that can be characterized under the headings immunization, mass chemoprophylaxis, and screening all share certain common characteristics: they involve the direct administration or application of a specific technical intervention to individuals on a one-by-one basis, they are directed to certain target populations, and the coverage of the target population is important to produce the desired effect. Technically, each of these intervention strategies is highly efficacious when correctly applied to a compliant subject, but their actual effectiveness in developing country settings is strongly conditioned by the local administrative, managerial, and logistical capabilities, as well as by traditional cultural constraints.
Most immunization interventions are highly cost-effective, and many of them address highly prevalent conditions. Measles and tetanus vaccination appear particularly cost-effective and worthy of relatively greater attention within immunization programmes. Far more could be efficiently spent on immunization than is now being spent and, even though costs of delivery tend to rise as more marginal populations are reached, extending immunization programmes to virtually universal coverage is likely to prove both cost-effective and a practical way of significantly improving the health of the poor.
One particularly promising application of mass chemoprophylaxis is the administration of antihelminthic medication and micronutrient supplements to school-age children. Here, cost-effectiveness appears quite high for conditions that, although of extremely high prevalence, have only recently been seen to be of substantial importance for intellectual and physical development. A programme of chemoprophylaxis for school-age children could, like the Expanded Programme on Immunization for younger children, be expected to serve as the starting point for an ultimately much expanded capacity to deal with the health needs of this age group.
Perhaps the most significant cancers for which treatment may be cost-effective (breast, cervical) are those for which early screening and referral are important; therefore, as non-communicable diseases become increasingly significant, this strategy will become increasingly relevant. The emerging strategies for the treatment of acute respiratory infections in children all rely heavily on community-based programmes for early detection and quick referral; with increased experience, improvements in the capacity for cost-effective screening and referral programmes can be expected to develop.
Facilities to provide clinical intervention vary continuously in size, in the degree of complexity (and range) of the conditions that they address, in the sophistication of their facilities and equipment, and in the training and skill of their staff. Nonetheless, for conducting comparable CEAs it is useful to use generally accepted terminology in categorizing facilities into three groups (clinic level, district hospitals, and referral hospitals), while recognizing that this categorization involves much simplification and that the appropriate classification structure will vary substantially from country to country. Table 6 indicates (in a very general way), for each of these three levels of facility, examples of the kinds of interventions they might address and the capacity such a facility might have for primary modes of diagnostic and therapeutic intervention.
Table 6 Clinical intervention: level of facility and mode of intervention
One lesson that emerged from this HPSR is that currently CEA is severely constrained by the paucity of data relating to the effect and cost of clinical interventions in low- and middle-income environments. In the absence of such analyses, it is perhaps natural for developing countries to import, to the extent that resources permit, the methods of case management used or being developed in high-income countries. Of course, the key phrase here is, ‘to the extent that resources permit.’ Available resources permit the import of high-cost interventions for only a tiny proportion of a developing country’s population. In order to extend access to services for the rapidly emerging epidemic of AIDS as well as for the impending epidemic of non-communicable disease, radically lower cost methods of case management will need to be developed from the rich range of technologies and procedures that now exist, or that are coming into being. Several additional observations can be made.
Curative care for tuberculosis and the sexually transmitted diseases appears to be extremely cost-effective; further, such care is not now being provided to anything like the extent it should be, given the high burden of morbidity and mortality resulting from these conditions. The surgical treatment of cataracts is also highly cost-effective.
The extremely diverse range of clinical interventions of moderate cost-effectiveness (medical management of angina or diabetes are examples as is surgical management of cervical cancer) suggests that country-specific analyses of these conditions are required and that facilities capable of competently handling diverse conditions will need to be developed.
The cost is sufficiently high for some clinical interventions to imply that, even if they are effective (as is the case with coronary artery bypass grafting to deal with angina), their marginal cost-effectiveness (in this case relative to medical management) is so poor that their use should be actively discouraged until other more cost-effective interventions can be delivered to their appropriate potential.
Control of pain from terminal cancer could benefit perhaps 1.5 million individuals annually at acceptable costs; current legislation and standard practices greatly limit what is done in relation to what potentially could be done.
Rehabilitation (in particular from leprosy, poliomyelitis, and injury) shows promise of being extremely cost-effective, but very little attention has been accorded to rehabilitation and little is known about how best to provide services on a population basis or what might be expected in terms of effectiveness and cost.
Again, as with the discussion of public health interventions, one theme that emerges from this review of clinical intervention cost-effectiveness is that of complexity and diversity. Many interventions are clearly not cost-effective, and public policy should make every effort to discourage their use. However, the available evidence does suggest that a broad range of interventions, addressing a similarly broad range of conditions, will prove cost-effective. Many of these interventions are not now being used to anything like the extent that they should be. Likewise, much of what is currently undertaken by the clinical system is misdirected (towards interventions of low cost-effectiveness) or simply inefficiently used. The redirection of substantial resources from interventions of low cost-effectiveness towards those with very high cost-effectiveness is clearly possible; a central task of health policy must be to design implementation strategies and government policy instruments that can promote these potential efficiency gains.
Lessons from the Health Sector Priorities Review
Five very broad conclusions can be drawn from the HSPR—one methodological and the other four substantive. The methodological conclusion is that it is feasible, on a broad scale, systematically to assess intervention cost-effectiveness in the health sector in a way that can provide broad policy guidance. The effort required is substantial, but results that allow broad intrasectoral assessment of intervention priorities can be obtained.
One substantive conclusion is that the available evidence points to great variation, across interventions, in marginal cost-effectiveness. Table 7 and Table 8 summarize this evidence by grouping interventions into ranges of marginal cost per QALY for different intervention objectives (Table 7) and for different public health and clinical approaches (Table 8). The challenge ahead is that of designing and implementing instruments of government policy that will greatly expand use of the interventions in the first several rows of these tables while decreasing use of interventions, like many of those in the last row of the tables, that provide very little value for money.
Table 7 Intervention cost-effectiveness by objective
Table 8 Intervention cost-effectiveness by public health and clinical approaches
Garber and Phelps (1997) calculate that under a reasonable range of assumptions it will make economic sense to pay for QALYs up to a cost of about twice the level of per capita income; this leads to a second substantive conclusion from Table 7 and Table 8, which is that, in many countries, quite a broad array of specific additional intervention is likely to prove attractive by any reasonable economic standard. (Such intervention could be financed either by reallocation from non-cost-effective interventions within the health sector or from resources outside the sector.)
The third substantive conclusion concerns the extent to which public health as opposed to clinical strategies tend to be more cost-effective and the extent to which seeking primary preventive objectives will tend to be more cost-effective than seeking other objectives (see Table 2 for relevant definitions). Again, Table 7 and Table 8 summarize material in ways that allow these questions to be addressed. Although there are some patterns (in particular, primary prevention via immunization accounts for many highly cost-effective interventions), in general it can be concluded that there is no particularly strong general tendency for primary prevention or public health interventions to have superior cost-effectiveness.
The fourth substantive conclusion from the HSPR is that few cost-effective interventions in low- and middle-income countries require more specialized facilities than those available at district hospitals. Thus, even though one cannot argue, in general, in favour of prevention over cure or public health over clinical intervention, one can, at least tentatively, conclude that district hospitals and lower-level facilities potentially offer almost all attractive interventions. A strong caveat here is that relatively few surgical interventions were assessed. Many of the more cost-effective surgical interventions can be done in a district hospital, but some may require referral facilities.
Multiple methods can provide decision-makers with insights into resource allocation in health; for example, cost-minimization analysis, CEA, and cost–benefit analysis. Methods for undertaking these analyses are now mature, although controversy continues on specific points. Extensive efforts over many years have yielded a large harvest of results. Among the methods in use, CEA appears most relevant for many purposes, but little additional effort may be required to recast results in terms of cost-minimization or cost–benefit analyses. In short, CEA and its relatives have been tested as working tools for the analyst.
That said, much remains to be done that goes beyond specific individual applications, important as those remain. Parallel analyses of a broad array of interventions provide information more than in proportion to the number of interventions. Much of what has caught political attention in CEA has resulted from these larger efforts, although only few exist. Further investment in large comparative studies (taking a number of paradigmatic environments as the base case) will both generate valuable insights directly and serve as solid starting points for more tailored country-specific efforts.
An example of negative intervention may be useful. Many countries now place individuals with severe mental illness in specialized mental hospitals that provide very long-term (and hence expensive) care. An increasingly advocated alternative would be short-term inpatient care in general hospitals combined with long-term medical management on an outpatient basis. Scaling back or closing mental hospitals would gain dollars, possibly at the cost of QALYs. From the perspective of a national decision-maker, assessing the cost-effectiveness of closing down existing facilities is likely to prove more salient than would an exercise that hypothesizes no intervention as the base case and concludes that the health system should have avoided building mental hospitals in the first place. The widespread existence of mental hospitals for long-term care makes generic analysis of the desirability of closing them down valuable (perhaps for several paradigmatic environments).
Most procedures for measuring QALYs result in an interval scale of measurement, i.e. a scale unique up to an affine transformation. That is, if q1 is a utility function resulting from the measurement process, then q2 will equally well represent that measurement process if q2 = a + bq1, b > 0. Incremental CEA utilizing interval scales will preserve cost-effectiveness ratios under permissible transformations of the utility function. Any attempt at assessing cost-effectiveness in a more absolute way (e.g. not with respect to a stated starting point) will require a scale of measurement that is stronger in the sense that it will need to be unique up to a similarity transformation (q2 = bq1, b > 0) if cost-effectiveness ratios are to be preserved. Such a scale, which has a natural zero that interval scales lack, is called a ratio scale. Use of QALYs or DALYs to measure burden of disease also requires a ratio scale of measurement. The existing literature on utility measurement in health lacks an axiomatic formulation of the conditions under which such a scale will exist and, until such a formulation is undertaken, the theoretical foundation for disease burden measurement will remain shaky. Krantz et al. (1971) give a thorough discussion of measurement theory, including a discussion of conditions under which two differently established interval scales on a set of outcomes can be used to generate an underlying ratio scale. These are conditions that indicate when, in the health context, utility measures generated by the time–trade-off method and the standard gamble method on the same set of outcomes would suffice to identify a ratio scale.
Existing disease burden studies (and CEAs) discount life-years lost from the life expectancy at the age of death to the present. For reasonable discount rates, this implies that the QALY or DALY loss associated with a death just after birth lies within 20 to 30 per cent of the loss associated with a death at age 20. This ratio differs substantially from the factor of 2 to 4 that has been obtained in the limited number of empirical assessments reported, such as Institute of Medicine (1986). At the same time, deaths before birth are treated as having no loss—at patent variation with human reaction and social willingness to pay to avert late fetal death. This issue is quantitatively important in that there are about 4 million stillbirths annually (2 million of which are in the 12 hours before the expected time of birth). A conceptual approach to dealing with these two problems, and a related complete recalculation of the global burden of disease, appears in Jamison et al. (2002). See also Musgrove (1991).
Garber (2000) discusses the question of what cost to assign pharmaceuticals (or devices) that are covered by patent. Patents confer temporary monopolies on the patent holders that allow prices to be set at levels often far above the marginal cost of production and packaging. This provides incentives for new product development. If a CEA uses the market price (i.e. monopoly price) of a patented drug as its measure of cost, clearly it cannot properly be considered an incremental CEA. Garber argues that if the CEA is undertaken from a consumer perspective, the practical approach will nonetheless be to use market prices (or whatever price can be negotiated by an influential purchaser) for costs. Pharmaceutical companies often adopt ‘tiered’ pricing regimes that result in lower prices in low-income countries. This will be profit-maximizing from the company’s perspective and will result in patented drug prices in developing countries being much closer to the marginal cost of production, thereby attenuating the problem that Garber raises. For this reason CEAs from a low-income country perspective should not treat patented drugs as tradeables.
Practical work in CEA often devotes substantial effort to defining and structuring the set of alternatives (Garber 2000, pp. 193–6). One result will often be to demonstrate that one or more alternatives are in some sense dominated by other alternatives under consideration. Which techniques should be chosen early (i.e. under very tight budget constraints) and which ones should be added later can be assessed. Finally, only in the context of considering closely related options can the attractiveness of a more costly but better technique be assessed. An example comes from an analysis of the attractiveness of coronary artery bypass grafts (CABG) in Brazil (Briscoe 1990), which concluded that CABG for disease in the left main coronary artery was a ‘good buy’ because the cost per QALY was only about 25 per cent of Brazil’s GDP per capita. However, this was the cost per QALY of CABG relative to doing nothing. Medical management and (now) angioplasty are less costly but nonetheless effective alternatives to CABG. The right way to think about CABG is in terms of how much more it would cost than one of these alternatives and how many more QALYs it would buy. It is likely that considered as incremental to alternatives the cost per QALY for CABG would be far higher than the original estimate. Thus the cost-effectiveness of any one intervention can be highly sensitive to the range of alternatives being considered.
The extent to which environmental interventions are justified on health grounds varies. While some discussions of air quality, for example, place importance on the amenity value of clean air, others emphasize health consequences. A particularly important example of the need to consider non-health outcomes, in the context of very poor environments, concerns improving water supplies (from the collection of surface water, say, to wells serving a community). Unclean and inadequate water supplies undoubtedly contribute substantially to risks of diarrhoeal and other diseases which kill millions of people every year. Increased quantities of cleaner water will have important health benefits. However, improving water supplies is very costly, and in most circumstances would appear to be non-cost-effective relative to public health or clinical interventions to reduce child mortality; that is, they would appear non-cost-effective if there were no other benefits. Other benefits include time savings (usually for women) in fetching water and the amenity value (beyond the sanitary value) of the cleaner bodies, clothing, and dwellings that improved water supplies facilitate. A cost–benefit analysis, if it were feasible, would place a monetary value on all benefits that would allow combining them. If a cost–benefit analysis cannot be done in an acceptable way, can CEA help to inform decisions? This is probably best done through sensitivity analysis. If all the non-health benefits can be given monetary values, one can calculate the dollar value per QALY that would be required for a satisfactory rate of return to the investment in water supplies. A high value would suggest that the water supply intervention was unattractive. Alternatively, one can calculate the cost of the intervention that would make the cost per QALY of improved water competitive with alternatives for reducing child mortality. If the calculated cost is much less than the actual cost, this would suggest that primary justification for the water-supply investment should be for its other benefits, not its health benefits, even if the other benefits cannot be valued in monetary terms.
In many ways the World Bank review is very much in the spirit of several previous reviews (Walsh and Warren 1979, 1986; Ghana Health Assessment Project Team 1981; Walsh 1988). These reviews provide assessments of priorities for the control of communicable childhood diseases in developing countries. The World Bank’s effort in 1993 involved more extensive use of economic analysis and covered a much broader range of conditions. Other recent work in this comparative spirit, but emphasizing effectiveness, includes Amler and Dull (1987) and the United States Department of Health and Human Services (1991), which reviewed a broad range of preventive intervention policies for the United States, and, more for clinical preventive services, the United States Preventive Services Task Force (1989) review of the effectiveness of 169 interventions. The state of Oregon in the United States rank ordered over 700 interventions, using cost-effectiveness and other criteria, for the purpose of rationing limited public resources to provide health care for the poor; Strosberg et al. (1992) discuss many facets of the Oregon Plan. Patel (1989) reviewed estimates of cost and effectiveness for a range of health interventions for UNICEF, and Jha et al. (1998) assessed the relative cost-effectiveness of 40 potentially important interventions in the West African context. The Harvard ‘life-saving’ project assessed cost per life saved of several hundred preventive options (Tengs et al. 1995; Tengs 1996). Udvarhelyi et al. (1992) provide a comprehensive review of medical cost-effectiveness and cost–benefit studies from the perspective of their methodological adequacy. All these approaches to the analytic evaluation of health practices fall within the general area of CEA.
Once somewhat comparable cost-effectiveness assessments are available for a range of interventions, analyses focusing on only a limited set of interventions can be put into the context provided by existing studies. For example, careful analysis for the sub-Saharan Africa context of malaria control (Goodman et al. 1999) and HIV-1 transmission interruption (Kumarayake and Watts 2000) both benefit from and contribute to an increasing understanding of intervention cost-effectiveness in Africa.
Health-care expenditures of over $1000 billion in 1997 for the 270 million people of the United States well exceeded the GNP of China ($924 billion) with a population at the time of 1.24 billion. It was over triple the combined GNPs of all the World Bank member countries of sub-Saharan Africa, which have a total population of about 630 million and a combined GNP of only about $323 billion.
A separate line of evidence, albeit only suggestive, for inefficiency resulting from variation in marginal cost-effectiveness is the very high degree of observed variation in procedure frequency in somewhat similar environments (Sanders et al. 1989).
Details are provided in Jamison (1993, Annex tables 1A–3 to 1A–6).
*Parts of this chapter are revised and updated from Jamison (1993).
Amler, R.W. and Dull, H.B. (ed.) (1987). Closing the gap: the burden of unnecessary illness. Oxford University Press, New York.
Anderson, M.A. (1990). Nature and magnitude of the problem of suboptimal breastfeeding practices. Presented at the International Policymakers Conference on Breastfeeding, Florence..
Barnum, H. (1987). Evaluating healthy days of life gained from health projects. Social Science and Medicine, 24, 833–41.
Barnum, H. and Greenberg, E.R. (1993). Cancers. In Disease control priorities in developing countries (ed. D.T. Jamison, W.H. Mosley, A.R. Measham, and J.L. Bobadilla), pp. 529–60. Oxford University Press for the World Bank.
Barnum, H, Barlow, R., Fajardo, L., and Pradilla, A. (1980). A resource allocation model for child survival. Oeldeschlager, Gunn and Hain, Cambridge, MA.
Barr, N. (1993). The economics of the welfare state (2nd edn). Stanford University Press.
Basta, S.S., Soekirman, D.K., and Scrimshaw, N.S. (1979). Iron deficiency anemia and the productivity of adult males in Indonesia. American Journal of Clinical Nutrition, 32, 916–25.
Bleichrodt, H. and Quiggen, J. (1999). Life-cycle preferences over consumption and health. When is cost-effectiveness analysis equivalent to cost–benefit analysis? Journal of Health Economics, 18, 681–708.
Briscoe, J. (1990). Brazil: the new challenge of adult health. World Bank, Washington, DC.
Briscoe, J. and de Ferranti, D. (1988). Water for rural communities. World Bank, Washington, DC.
Cairns, J. and Johnston, K. (1991). Condition-specific outcome measures as an alternative to across-programme QALYs. Health Economics Unit, University of Aberdeen.
Cropper, M.L., Aydede, S.K., and Portney, PR. (1992). Rates of time preference for saving lives. American Economic Review, 82, 469–72.
Devarajan, S., Squire, L., and Sutchiwart-Naruput, S. (1997). Beyond rate of return: reorienting project appraisal. World Bank Research Observer, 12, 35–46.
Drummond, M.F. (ed.) (1987a). Economic appraisal of health technology in the European Community. Oxford University Press.
Drummond, M.F. (1987b). Methods for economic appraisal of health technology. In Economic appraisal of health technology in the European Community (ed. M.F. Drummond). Oxford University Press.
Drummond, M.F., O’Brien, B.J., Stoddart, G.L., and Torrance, G.W. (1997). Methods for the economic evaluation of health care programs (2nd edn). Oxford University Press.
EuroQol Group (1990). EuroQol—a new facility for the measurement of health-related quality of life. Health Policy, 16, 199–208.
Fallowfield, L. (1990). The quality of life. Souvenir Press, London.
Feachem, R.G.A., Kjellstrom, T., Murray, C.J.L., Over, M., and Phillips, M.A. (1992). The health of adults in the developing world. Oxford University Press, New York.
Fox-Rushby, J.A. and Hanson, K. (2001). Calculating and presenting disability-adjusted life years in cost-effectiveness analysis. Health Policy and Planning, 16, 326–31.
Garber, A.M. (2000). Advances in CE analysis. In Handbook of health economics (ed. J.P. Newhouse and A.J. Culyer), pp. 181–221. Elsevier, Amsterdam.
Garber, A.M. and Phelps, C.E. (1997). Economic foundations of cost-effectiveness analysis. Journal of Health Economics, 16, 1–31.
Garber, A.M., Weinstein, M.C., Torrance, G.W., and Kamlet, M.S. (1996). Theoretical foundations of cost-effectiveness analysis. In Cost-effectiveness in health and medicine (ed. M.R. Gold, J.E. Siegel, L.B. Russell, and M.C. Weinstein), pp. 25–53. Oxford University Press, New York.
Ghana Health Assessment Project Team (1981). Quantitative method of assessing the health impact of different diseases in less developed countries. International Journal of Epidemiology, 10, 73–80.
Gold, M.R., Siegel, J.E., Russell, L.B., and Weinstein, M.C. (ed.) (1996). Cost-effectiveness in health and medicine. Oxford University Press, New York.
Goodman, C.A., Coleman, P.G., and Mills, A.J. (1999). Cost-effectiveness of malaria control in sub-Saharan Africa. Lancet, 354, 378–85.
Institute of Medicine (1986). New vaccine development, establishing priorities, Vols 1 and 2. National Academy Press, Washington, DC.
Jamison, D.T. (1993). Disease control priorities in developing countries: an overview. In Disease control priorities in developing countries (ed. D.T. Jamison, W.H. Mosley, A.R. Measham, and J.L. Bobadilla), pp. 3–34. Oxford University Press for the World Bank.
Jamison, D.T. and Lumsden, K.G. (1975). Television and efficiency in higher education. Management Science, 21, 920–30.
Jamison, D.T., Mosley, W.H., Measham, A.R., and Bobadilla, J.L. (ed.) (1993). Disease control priorities in developing countries. Oxford University Press for the World Bank.
Jamison, D.T., Jamison, J., Bobadilla, J.L., and Zupan, J. (2002). A recalculation of the global burden of disease: estimates are highly sensitive to assumptions about events around the time of birth. In preparation.
Jha, P., Bangura, O., and Ransom, K. (1998). The cost-effectiveness of forty health interventions in Guinea. Health Policy and Planning, 13, 249–62.
Johannesson, M. (1992). On the discounting of gained life-years in cost-effectiveness analysis. International Journal of Technology Assessment in Health Care, 8, 359–64.
Johannesson, M. and Meltzer, D. (1998). Some reflections on cost-effectiveness analysis. Health Economics, 7, 1–7.
Johannesson, M. and Weinstein, M.C. (1993). On the decision rules of cost-effectiveness analysis. Journal of Health Economics, 12, 459–67.
Jones-Lee, M.W. (1994). Safety and the saving of life, the economics of safety and physical risk. In Cost–benefit analysis (ed. R. Layard and S. Glaister), pp. 290–318. Cambridge University Press.
Kaplan, R.M. and Bush, J. (1982). Health-related quality of life measurement for evaluation research and policy analysis. Health Psychology, 2, 61–80.
Krantz, D.H., Luce, D.R., Suppes, P., and Tversky, A. (1971). Foundations of measurement. Vol. I, Additive and polynomial representations. Academic Press, New York.
Kremer, M. (2001). Public policies to stimulate development of vaccines and drugs for neglected diseases. Department of Economics, Harvard University.
Kumaranayake, L. and Watts, C. (2000). Economic costs of HIV/AIDS prevention activities in sub-Saharan Africa. AIDS, 14 (Supplement 3), S239–52.
Last, J.M. (ed.) (1988). A dictionary of epidemiology (2nd edn). Oxford University Press, New York, for the International Epidemiological Association.
Layard, R. and Glaister, S. (1994). Introduction. In Cost–benefit analysis (ed. R. Layard and S. Glaister), pp. 1–56. Cambridge University Press.
Leslie, J. (1989). Women’s time: a factor in the use of child survival technologies? Health Policy and Planning, 4, 1–16.
Leslie, J. (1992). Women’s time and the use of health services. IDS Bulletin, 23, 4–7.
Levin, H.M., Pollitt, E., Galloway, R., and McGuire, J. (1993). Micronutrient deficiency disorders. In Disease control priorities in developing countries (ed. D.T. Jamison, W.H. Mosley, A.R. Measham, and J.L. Bobadilla), pp. 421–54. Oxford University Press for the World Bank.
Luce, B.R., Manning, W.G., Siegel, J.E., and Lipscomb, J. (1996). Estimating costs in cost-effectiveness analysis. In Cost-effectiveness in health and medicine (ed. M.R. Gold, L.B. Russell, J.E. Siegel, and M.C. Weinstein), pp. 176–213. Oxford University Press, New York.
Meltzer, D. (2001). Addressing uncertainty in medical cost-effectiveness analysis. Implications of expected utility maximization for methods to perform sensitivity analysis and the use of cost-effectiveness to set priorities for research. Journal of Health Economics, 20, 109–29.
MMWR (1992). Years of potential life lost before ages 65 and 85—United States, 1989–1990. Morbidity and Mortality Weekly Report, 41, 313–15.
Mosley, W.H. and Jolly, R. (1987). Health policy and program options, compensating for the negative effects of economic adjustment. In Adjustment with a human face (ed. G.A. Cornea, R. Jolly, and F. Stewart). Clarendon Press, Oxford.
Murray, C.J., Evans, D., Acharya, A., and Baltussen, R. (1999). Development of WHO guidelines on generalised cost-effectiveness analysis. GPE Discussion Paper 4, WHO, Geneva.
Murray, C. J., Jamison, D.T., and Lopez, A.D. (1994). The global burden of disease in 1990: summary results, sensitivity analysis and future directions. Bulletin of the World Health Organization, 72, 495–509.
Musgrove, P. (1991). The burden of death at different ages: assumptions, parameters and values. Occasional Paper 12, Latin America and the Caribbean Regional Office, Human Resources Division, Technical Department, World Bank, Washington, DC.
Musgrove, P. (1995). Cost-effectiveness and the socialization of health care. Journal of Health Policy, 32, 111–23.
Nord, E. (1991). The relevance of QALYs in prioritizing between different patients. Presented at the 12th Nordic HESG meeting, Copenhagen.
Over, M. (1988). Cost-effective integration of immunization and basic health services in developing countries: the problem of joint costs. Working Paper 23, World Bank, Washington, DC.
Patel, M.S. (1989). Eliminating social distance between North and South: cost-effective goals for the 1990s. Staff Working Paper 5, UNICEF, New York.
Pauly, M.V. (1995). Valuing health care benefits in money terms. In Valuing health care: costs, benefits, and effectiveness of pharmaceuticals and other medical technologies (ed. F.A. Sloan), pp. 99–124. Cambridge University Press.
Phelps, C.E. and Mushlin, A.I. (1991). On the (near) equivalence of cost-effectiveness and cost–benefit analysis. International Journal of Technology Assessment in Health Care, 7, 12–21.
Pollitt, E. (1990). Malnutrition and infection in the classroom. UNESCO, Paris.
Preston, S.H. (1991). Health indexes and health sector planning. Presented at the Workshop on the Policy and Planning Implications of the Epidemiological Transition in Developing Countries, National Research Council, Washington, DC.
Rosser, R.M. and Kind, P. (1978). A scale of valuations of states of illness, is there a social consensus? International Journal of Epidemiology, 7, 347–58.
Sanders, D., Coulter, A., and McPherson, K. (1989). Variation in hospital admission rates: a review of the literature. Paper 79, Kings Fund, London.
Skinner, J. and McClellan, M. (2000). The incidence of Medicaid. Working paper, National Bureau of Economic Research, New York.
Sloan, F.A. and Conover, C.J. (1995). The use of cost-effectiveness/cost–benefit analysis in actual decision making, current status and prospects. In Valuing health care, costs, benefits, and effectiveness of pharmaceuticals and other medical technologies (ed. F.A. Sloan), pp. 207–32. Cambridge University Press.
Squire, L. (1989). Project evaluation in theory and practice. In Handbook of development economics, Vol. 2 (ed. H.B. Chenery and T.N. Srinivasan). North-Holland, Amsterdam.
Stern, N. and Ferreira, F. (1997). The World Bank as intellectual actor. In The World Bank: its first half century. Vol. 2, Perspectives (ed. D. Kapur, J.P. Lewis, and R. Webb), pp. 523–609. Brookings Institution, Washington, DC.
Strosberg, M.A., Weiner, J.M., Baker, R., Fein, I.A. (ed.) (1992). Rationing America’s medical care: the Oregon Plan and beyond. Brookings Institution, Washington, DC.
Stouthard, M.E., et al. (1997). Disability weights for diseases in The Netherlands. Department of Public Health, Erasmus University, Rotterdam.
Tan-Torres, T. (1990). Comparison of different methods of eliciting utilities for outcome states in leprosy. Clinical Epidemiology Unit, Department of Medicine, University of the Philippines, Manila.
Tengs, T.O. (1996). Enormous variation in the cost-effectiveness of prevention: implications for public policy. Public Health, 2, 13–17.
Tengs, T.O., Adams, M.E., and Pliskin, J.S. (1995). Five-hundred life-saving interventions and their cost-effectiveness. Risk Analysis, 15, 369–90.
Tolley, G., Kenkel, D., and Fabian, R. (ed.) (1994). Valuing health for policy: an economic approach. University of Chicago Press.
Udvarhelyi, I.S., Colvitz, G.A., Rai, A., and Epstein, A.M. (1992). Cost-effectiveness and cost–benefit analyses in the medical literature. Annals of Internal Medicine, 116, 238–44.
USDHHS (US Department of Health and Human Services) (1991). Healthy people 2000: national health promotion and disease prevention objectives. US Government Printing Office, Washington, DC.
US Preventive Services Task Force (Chairman, R.S. Lawrence) (1989). Guide to clinical preventive services. Williams & Wilkins, Baltimore, MD.
Vicusi, W.K. (1993). The value of risks to life and health. Journal of Economic Literature, 31, 1912–46.
Wagstaff, A. (1994). Health care, QALYs and the equity–efficiency tradeoff. In Cost–benefit analysis (ed. R. Layard and S. Glaister), pp. 428–47. Cambridge University Press.
Walsh, J.A. (1988). Establishing health priorities in the developing world. Adams, Boston, MA, for the United Nations Development Program.
Walsh, J.A. and Warren, K.S. (1979). Selective primary health care—an interim strategy for disease control in developing countries. New England Journal of Medicine, 301, 967–74.
Walsh, J.A. and Warren, K.S. (ed.) (1986). Strategies for primary health care: technologies appropriate for the control of disease in the developing world. University of Chicago Press.
Watts, C. and Kumaranayake, L. (1999). Thinking big: scaling up HIV-1 interventions in sub-Saharan Africa. Lancet, 354, 1492.
Weinstein, M.C. (1995). From cost-effectiveness ratios to resource allocation: where to draw the line? In Valuing health care: costs, benefits, and effectiveness of pharmaceuticals and other medical technologies (ed. F.A. Sloan), pp. 77–97. Cambridge University Press.
Weinstein, M.C. and Stason, W.B. (1977). Foundations of cost-effectiveness analysis for health and medical practices. New England Journal of Medicine, 296, 716–21.
Weinstein, N.D. (1989). Optimistic biases about personal risks. Science, 246, 1232–33.
Wells, K.B. (1999). The design of partners in care: evaluating the cost-effectiveness of improving care for depression in primary care. Social Psychiatry and Psychiatric Epidemiology, 34, 20–9.
WHO (1996). Investing in health research and development. Report of the Ad Hoc Committee on Health Research Relating to Future Intervention Options. Document TDR/Gen/96.1, WHO, Geneva.
WHO (2000). World health report 2000. Health systems: improving performance. WHO, Geneva.
World Bank (1992). World development report 1992. Development and the environment. World Bank, Washington, DC.
World Bank (1993). World development report 1993 Investing in health. World Bank, Washington, DC.
Zeckhauser, R. and Shepard, D. (1976). Where now for saving lives? Law and Contemporary Problems, 40, 5–45.